[llvm] [AMDGPU] Support double type in atomic optimizer. (PR #84307)
Pravin Jagtap via llvm-commits
llvm-commits at lists.llvm.org
Sun Mar 17 22:04:53 PDT 2024
https://github.com/pravinjagtap updated https://github.com/llvm/llvm-project/pull/84307
>From f79911d89007f089614378f5b3985ac0475ca398 Mon Sep 17 00:00:00 2001
From: Pravin Jagtap <Pravin.Jagtap at amd.com>
Date: Thu, 7 Mar 2024 06:18:25 -0500
Subject: [PATCH 1/2] [AMDGPU] Support double type in atomic optimizer.
---
.../Target/AMDGPU/AMDGPUAtomicOptimizer.cpp | 5 +-
.../AMDGPU/GlobalISel/fp64-atomics-gfx90a.ll | 274 +-
.../CodeGen/AMDGPU/fp64-atomics-gfx90a.ll | 270 +-
.../AMDGPU/global_atomics_scan_fadd.ll | 5578 +++++++++++++++++
.../AMDGPU/global_atomics_scan_fmax.ll | 3960 ++++++++++++
.../AMDGPU/global_atomics_scan_fmin.ll | 3960 ++++++++++++
.../AMDGPU/global_atomics_scan_fsub.ll | 5576 ++++++++++++++++
7 files changed, 19513 insertions(+), 110 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
index 9ba74a23e8afcb..a6013176676bde 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
@@ -209,8 +209,9 @@ void AMDGPUAtomicOptimizerImpl::visitAtomicRMWInst(AtomicRMWInst &I) {
break;
}
- // Only 32-bit floating point atomic ops are supported.
- if (AtomicRMWInst::isFPOperation(Op) && !I.getType()->isFloatTy()) {
+ // Only 32 and 64 bit floating point atomic ops are supported.
+ if (AtomicRMWInst::isFPOperation(Op) &&
+ !(I.getType()->isFloatTy() || I.getType()->isDoubleTy())) {
return;
}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/fp64-atomics-gfx90a.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/fp64-atomics-gfx90a.ll
index 255c6dedbd6e1e..1a76f8cf87ffbe 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/fp64-atomics-gfx90a.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/fp64-atomics-gfx90a.ll
@@ -1090,18 +1090,29 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat(ptr addrspace(1) %ptr) #1 {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: s_mov_b32 s4, s3
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB39_3
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX90A-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX90A-NEXT: v_mul_f64 v[4:5], v[0:1], 4.0
; GFX90A-NEXT: s_mov_b64 s[2:3], 0
-; GFX90A-NEXT: v_mov_b32_e32 v4, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX90A-NEXT: v_mov_b32_e32 v6, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], s[4:5], s[4:5] op_sel:[0,1]
-; GFX90A-NEXT: .LBB39_1: ; %atomicrmw.start
+; GFX90A-NEXT: .LBB39_2: ; %atomicrmw.start
; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], 4.0
+; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
; GFX90A-NEXT: buffer_wbl2
-; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_invl2
; GFX90A-NEXT: buffer_wbinvl1_vol
@@ -1109,20 +1120,31 @@ define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat(ptr addrspace(1) %pt
; GFX90A-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[0:1], v[0:1] op_sel:[0,1]
; GFX90A-NEXT: s_andn2_b64 exec, exec, s[2:3]
-; GFX90A-NEXT: s_cbranch_execnz .LBB39_1
-; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX90A-NEXT: s_cbranch_execnz .LBB39_2
+; GFX90A-NEXT: .LBB39_3:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: s_mov_b32 s4, s3
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB39_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: v_mov_b32_e32 v2, 0
; GFX940-NEXT: buffer_wbl2 sc0 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1] sc1
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc0 sc1
+; GFX940-NEXT: .LBB39_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 seq_cst
@@ -1132,26 +1154,47 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_agent(ptr addrspace(1) %ptr) #1 {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat_agent:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: s_mov_b32 s4, s3
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB40_2
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX90A-NEXT: v_mov_b32_e32 v0, 0
-; GFX90A-NEXT: v_mov_b32_e32 v1, 0x40100000
+; GFX90A-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_wbinvl1_vol
+; GFX90A-NEXT: .LBB40_2:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat_agent:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: s_mov_b32 s4, s3
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB40_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: v_mov_b32_e32 v2, 0
; GFX940-NEXT: buffer_wbl2 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc1
+; GFX940-NEXT: .LBB40_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("agent") seq_cst
@@ -1161,18 +1204,29 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_system(ptr addrspace(1) %ptr) #1 {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat_system:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: s_mov_b32 s4, s3
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB41_3
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX90A-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX90A-NEXT: v_mul_f64 v[4:5], v[0:1], 4.0
; GFX90A-NEXT: s_mov_b64 s[2:3], 0
-; GFX90A-NEXT: v_mov_b32_e32 v4, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX90A-NEXT: v_mov_b32_e32 v6, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], s[4:5], s[4:5] op_sel:[0,1]
-; GFX90A-NEXT: .LBB41_1: ; %atomicrmw.start
+; GFX90A-NEXT: .LBB41_2: ; %atomicrmw.start
; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], 4.0
+; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
; GFX90A-NEXT: buffer_wbl2
-; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_invl2
; GFX90A-NEXT: buffer_wbinvl1_vol
@@ -1180,20 +1234,31 @@ define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_system(ptr addrspace
; GFX90A-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[0:1], v[0:1] op_sel:[0,1]
; GFX90A-NEXT: s_andn2_b64 exec, exec, s[2:3]
-; GFX90A-NEXT: s_cbranch_execnz .LBB41_1
-; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX90A-NEXT: s_cbranch_execnz .LBB41_2
+; GFX90A-NEXT: .LBB41_3:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat_system:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: s_mov_b32 s4, s3
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB41_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: v_mov_b32_e32 v2, 0
; GFX940-NEXT: buffer_wbl2 sc0 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1] sc1
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc0 sc1
+; GFX940-NEXT: .LBB41_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("one-as") seq_cst
@@ -1203,26 +1268,47 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_flush(ptr addrspace(1) %ptr) #0 {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat_flush:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: s_mov_b32 s4, s3
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB42_2
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX90A-NEXT: v_mov_b32_e32 v0, 0
-; GFX90A-NEXT: v_mov_b32_e32 v1, 0x40100000
+; GFX90A-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_wbinvl1_vol
+; GFX90A-NEXT: .LBB42_2:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat_flush:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: s_mov_b32 s4, s3
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB42_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: v_mov_b32_e32 v2, 0
; GFX940-NEXT: buffer_wbl2 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc1
+; GFX940-NEXT: .LBB42_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("agent") seq_cst
@@ -1394,37 +1480,59 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_agent_safe(ptr addrspace(1) %ptr) {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat_agent_safe:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: s_mov_b32 s4, s3
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB49_3
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX90A-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX90A-NEXT: v_mul_f64 v[4:5], v[0:1], 4.0
; GFX90A-NEXT: s_mov_b64 s[2:3], 0
-; GFX90A-NEXT: v_mov_b32_e32 v4, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX90A-NEXT: v_mov_b32_e32 v6, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], s[4:5], s[4:5] op_sel:[0,1]
-; GFX90A-NEXT: .LBB49_1: ; %atomicrmw.start
+; GFX90A-NEXT: .LBB49_2: ; %atomicrmw.start
; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], 4.0
-; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_wbinvl1_vol
; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
; GFX90A-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[0:1], v[0:1] op_sel:[0,1]
; GFX90A-NEXT: s_andn2_b64 exec, exec, s[2:3]
-; GFX90A-NEXT: s_cbranch_execnz .LBB49_1
-; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX90A-NEXT: s_cbranch_execnz .LBB49_2
+; GFX90A-NEXT: .LBB49_3:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat_agent_safe:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: s_mov_b32 s4, s3
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB49_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: v_mov_b32_e32 v2, 0
; GFX940-NEXT: buffer_wbl2 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc1
+; GFX940-NEXT: .LBB49_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("agent") seq_cst
@@ -1866,23 +1974,44 @@ main_body:
define amdgpu_kernel void @local_atomic_fadd_f64_noret_pat(ptr addrspace(3) %ptr) #1 {
; GFX90A-LABEL: local_atomic_fadd_f64_noret_pat:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: s_mov_b32 s4, s3
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB65_2
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dword s0, s[0:1], 0x24
-; GFX90A-NEXT: v_mov_b32_e32 v0, 0
-; GFX90A-NEXT: v_mov_b32_e32 v1, 0x40100000
+; GFX90A-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_mov_b32_e32 v2, s0
; GFX90A-NEXT: ds_add_f64 v2, v[0:1]
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
+; GFX90A-NEXT: .LBB65_2:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: local_atomic_fadd_f64_noret_pat:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: s_mov_b32 s4, s3
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB65_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dword s0, s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: v_mov_b32_e32 v2, s0
; GFX940-NEXT: ds_add_f64 v2, v[0:1]
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
+; GFX940-NEXT: .LBB65_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(3) %ptr, double 4.0 seq_cst
@@ -1892,23 +2021,44 @@ main_body:
define amdgpu_kernel void @local_atomic_fadd_f64_noret_pat_flush(ptr addrspace(3) %ptr) #0 {
; GFX90A-LABEL: local_atomic_fadd_f64_noret_pat_flush:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: s_mov_b32 s4, s3
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB66_2
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dword s0, s[0:1], 0x24
-; GFX90A-NEXT: v_mov_b32_e32 v0, 0
-; GFX90A-NEXT: v_mov_b32_e32 v1, 0x40100000
+; GFX90A-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_mov_b32_e32 v2, s0
; GFX90A-NEXT: ds_add_f64 v2, v[0:1]
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
+; GFX90A-NEXT: .LBB66_2:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: local_atomic_fadd_f64_noret_pat_flush:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: s_mov_b32 s4, s3
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB66_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dword s0, s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: v_mov_b32_e32 v2, s0
; GFX940-NEXT: ds_add_f64 v2, v[0:1]
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
+; GFX940-NEXT: .LBB66_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(3) %ptr, double 4.0 seq_cst
@@ -1918,44 +2068,66 @@ main_body:
define amdgpu_kernel void @local_atomic_fadd_f64_noret_pat_flush_safe(ptr addrspace(3) %ptr) #4 {
; GFX90A-LABEL: local_atomic_fadd_f64_noret_pat_flush_safe:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: s_mov_b32 s4, s3
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB67_3
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dword s0, s[0:1], 0x24
+; GFX90A-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT: v_mov_b32_e32 v2, s0
-; GFX90A-NEXT: ds_read_b64 v[0:1], v2
+; GFX90A-NEXT: v_mov_b32_e32 v4, s0
+; GFX90A-NEXT: ds_read_b64 v[2:3], v4
; GFX90A-NEXT: s_mov_b64 s[0:1], 0
-; GFX90A-NEXT: .LBB67_1: ; %atomicrmw.start
+; GFX90A-NEXT: .LBB67_2: ; %atomicrmw.start
; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT: v_add_f64 v[4:5], v[0:1], 4.0
-; GFX90A-NEXT: ds_cmpst_rtn_b64 v[4:5], v2, v[0:1], v[4:5]
+; GFX90A-NEXT: v_add_f64 v[6:7], v[2:3], v[0:1]
+; GFX90A-NEXT: ds_cmpst_rtn_b64 v[6:7], v4, v[2:3], v[6:7]
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[0:1]
+; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
; GFX90A-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
-; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[4:5] op_sel:[0,1]
+; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[6:7], v[6:7] op_sel:[0,1]
; GFX90A-NEXT: s_andn2_b64 exec, exec, s[0:1]
-; GFX90A-NEXT: s_cbranch_execnz .LBB67_1
-; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX90A-NEXT: s_cbranch_execnz .LBB67_2
+; GFX90A-NEXT: .LBB67_3:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: local_atomic_fadd_f64_noret_pat_flush_safe:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: s_mov_b32 s4, s3
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s4, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB67_3
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dword s0, s[0:1], 0x24
+; GFX940-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
-; GFX940-NEXT: v_mov_b32_e32 v2, s0
-; GFX940-NEXT: ds_read_b64 v[0:1], v2
+; GFX940-NEXT: v_mov_b32_e32 v4, s0
+; GFX940-NEXT: ds_read_b64 v[2:3], v4
; GFX940-NEXT: s_mov_b64 s[0:1], 0
-; GFX940-NEXT: .LBB67_1: ; %atomicrmw.start
+; GFX940-NEXT: .LBB67_2: ; %atomicrmw.start
; GFX940-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
-; GFX940-NEXT: v_add_f64 v[4:5], v[0:1], 4.0
-; GFX940-NEXT: ds_cmpst_rtn_b64 v[4:5], v2, v[0:1], v[4:5]
+; GFX940-NEXT: v_add_f64 v[6:7], v[2:3], v[0:1]
+; GFX940-NEXT: ds_cmpst_rtn_b64 v[6:7], v4, v[2:3], v[6:7]
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
-; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[0:1]
+; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
; GFX940-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], v[4:5]
+; GFX940-NEXT: v_mov_b64_e32 v[2:3], v[6:7]
; GFX940-NEXT: s_andn2_b64 exec, exec, s[0:1]
-; GFX940-NEXT: s_cbranch_execnz .LBB67_1
-; GFX940-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX940-NEXT: s_cbranch_execnz .LBB67_2
+; GFX940-NEXT: .LBB67_3:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(3) %ptr, double 4.0 seq_cst
diff --git a/llvm/test/CodeGen/AMDGPU/fp64-atomics-gfx90a.ll b/llvm/test/CodeGen/AMDGPU/fp64-atomics-gfx90a.ll
index 767d347bcfaa15..a948fab8f1c1b9 100644
--- a/llvm/test/CodeGen/AMDGPU/fp64-atomics-gfx90a.ll
+++ b/llvm/test/CodeGen/AMDGPU/fp64-atomics-gfx90a.ll
@@ -1181,18 +1181,28 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat(ptr addrspace(1) %ptr) #1 {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB42_3
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX90A-NEXT: s_bcnt1_i32_b64 s6, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s6
; GFX90A-NEXT: s_mov_b64 s[2:3], 0
-; GFX90A-NEXT: v_mov_b32_e32 v4, 0
+; GFX90A-NEXT: v_mul_f64 v[4:5], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX90A-NEXT: v_mov_b32_e32 v6, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], s[4:5], s[4:5] op_sel:[0,1]
-; GFX90A-NEXT: .LBB42_1: ; %atomicrmw.start
+; GFX90A-NEXT: .LBB42_2: ; %atomicrmw.start
; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], 4.0
+; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
; GFX90A-NEXT: buffer_wbl2
-; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_invl2
; GFX90A-NEXT: buffer_wbinvl1_vol
@@ -1200,20 +1210,30 @@ define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat(ptr addrspace(1) %pt
; GFX90A-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[0:1], v[0:1] op_sel:[0,1]
; GFX90A-NEXT: s_andn2_b64 exec, exec, s[2:3]
-; GFX90A-NEXT: s_cbranch_execnz .LBB42_1
-; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX90A-NEXT: s_cbranch_execnz .LBB42_2
+; GFX90A-NEXT: .LBB42_3:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB42_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
; GFX940-NEXT: v_mov_b32_e32 v2, 0
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: buffer_wbl2 sc0 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1] sc1
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc0 sc1
+; GFX940-NEXT: .LBB42_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 seq_cst
@@ -1223,26 +1243,45 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_agent(ptr addrspace(1) %ptr) #1 {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat_agent:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB43_2
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX90A-NEXT: v_mov_b32_e32 v0, 0
+; GFX90A-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
-; GFX90A-NEXT: v_mov_b32_e32 v1, 0x40100000
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_wbinvl1_vol
+; GFX90A-NEXT: .LBB43_2:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat_agent:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB43_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
; GFX940-NEXT: v_mov_b32_e32 v2, 0
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: buffer_wbl2 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc1
+; GFX940-NEXT: .LBB43_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("agent") seq_cst
@@ -1252,18 +1291,28 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_system(ptr addrspace(1) %ptr) #1 {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat_system:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB44_3
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX90A-NEXT: s_bcnt1_i32_b64 s6, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s6
; GFX90A-NEXT: s_mov_b64 s[2:3], 0
-; GFX90A-NEXT: v_mov_b32_e32 v4, 0
+; GFX90A-NEXT: v_mul_f64 v[4:5], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX90A-NEXT: v_mov_b32_e32 v6, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], s[4:5], s[4:5] op_sel:[0,1]
-; GFX90A-NEXT: .LBB44_1: ; %atomicrmw.start
+; GFX90A-NEXT: .LBB44_2: ; %atomicrmw.start
; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], 4.0
+; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
; GFX90A-NEXT: buffer_wbl2
-; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_invl2
; GFX90A-NEXT: buffer_wbinvl1_vol
@@ -1271,20 +1320,30 @@ define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_system(ptr addrspace
; GFX90A-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[0:1], v[0:1] op_sel:[0,1]
; GFX90A-NEXT: s_andn2_b64 exec, exec, s[2:3]
-; GFX90A-NEXT: s_cbranch_execnz .LBB44_1
-; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX90A-NEXT: s_cbranch_execnz .LBB44_2
+; GFX90A-NEXT: .LBB44_3:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat_system:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB44_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
; GFX940-NEXT: v_mov_b32_e32 v2, 0
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: buffer_wbl2 sc0 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1] sc1
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc0 sc1
+; GFX940-NEXT: .LBB44_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("one-as") seq_cst
@@ -1294,26 +1353,45 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_flush(ptr addrspace(1) %ptr) #0 {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat_flush:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB45_2
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX90A-NEXT: v_mov_b32_e32 v0, 0
+; GFX90A-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
; GFX90A-NEXT: v_mov_b32_e32 v2, 0
-; GFX90A-NEXT: v_mov_b32_e32 v1, 0x40100000
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_wbinvl1_vol
+; GFX90A-NEXT: .LBB45_2:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat_flush:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB45_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
; GFX940-NEXT: v_mov_b32_e32 v2, 0
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: buffer_wbl2 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc1
+; GFX940-NEXT: .LBB45_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("agent") seq_cst
@@ -1485,37 +1563,57 @@ main_body:
define amdgpu_kernel void @global_atomic_fadd_f64_noret_pat_agent_safe(ptr addrspace(1) %ptr) {
; GFX90A-LABEL: global_atomic_fadd_f64_noret_pat_agent_safe:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB52_3
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX90A-NEXT: s_bcnt1_i32_b64 s6, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s6
; GFX90A-NEXT: s_mov_b64 s[2:3], 0
-; GFX90A-NEXT: v_mov_b32_e32 v4, 0
+; GFX90A-NEXT: v_mul_f64 v[4:5], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX90A-NEXT: v_mov_b32_e32 v6, 0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], s[4:5], s[4:5] op_sel:[0,1]
-; GFX90A-NEXT: .LBB52_1: ; %atomicrmw.start
+; GFX90A-NEXT: .LBB52_2: ; %atomicrmw.start
; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], 4.0
-; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX90A-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
; GFX90A-NEXT: s_waitcnt vmcnt(0)
; GFX90A-NEXT: buffer_wbinvl1_vol
; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
; GFX90A-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[0:1], v[0:1] op_sel:[0,1]
; GFX90A-NEXT: s_andn2_b64 exec, exec, s[2:3]
-; GFX90A-NEXT: s_cbranch_execnz .LBB52_1
-; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX90A-NEXT: s_cbranch_execnz .LBB52_2
+; GFX90A-NEXT: .LBB52_3:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: global_atomic_fadd_f64_noret_pat_agent_safe:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB52_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX940-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s2
; GFX940-NEXT: v_mov_b32_e32 v2, 0
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: buffer_wbl2 sc1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: global_atomic_add_f64 v2, v[0:1], s[0:1]
; GFX940-NEXT: s_waitcnt vmcnt(0)
; GFX940-NEXT: buffer_inv sc1
+; GFX940-NEXT: .LBB52_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("agent") seq_cst
@@ -2020,23 +2118,42 @@ main_body:
define amdgpu_kernel void @local_atomic_fadd_f64_noret_pat(ptr addrspace(3) %ptr) #1 {
; GFX90A-LABEL: local_atomic_fadd_f64_noret_pat:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB70_2
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dword s0, s[0:1], 0x24
-; GFX90A-NEXT: v_mov_b32_e32 v0, 0
-; GFX90A-NEXT: v_mov_b32_e32 v1, 0x40100000
+; GFX90A-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_mov_b32_e32 v2, s0
; GFX90A-NEXT: ds_add_f64 v2, v[0:1]
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
+; GFX90A-NEXT: .LBB70_2:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: local_atomic_fadd_f64_noret_pat:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB70_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dword s0, s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: v_mov_b32_e32 v2, s0
; GFX940-NEXT: ds_add_f64 v2, v[0:1]
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
+; GFX940-NEXT: .LBB70_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(3) %ptr, double 4.0 seq_cst
@@ -2046,23 +2163,42 @@ main_body:
define amdgpu_kernel void @local_atomic_fadd_f64_noret_pat_flush(ptr addrspace(3) %ptr) #0 {
; GFX90A-LABEL: local_atomic_fadd_f64_noret_pat_flush:
; GFX90A: ; %bb.0: ; %main_body
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB71_2
+; GFX90A-NEXT: ; %bb.1:
; GFX90A-NEXT: s_load_dword s0, s[0:1], 0x24
-; GFX90A-NEXT: v_mov_b32_e32 v0, 0
-; GFX90A-NEXT: v_mov_b32_e32 v1, 0x40100000
+; GFX90A-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
; GFX90A-NEXT: v_mov_b32_e32 v2, s0
; GFX90A-NEXT: ds_add_f64 v2, v[0:1]
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
+; GFX90A-NEXT: .LBB71_2:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: local_atomic_fadd_f64_noret_pat_flush:
; GFX940: ; %bb.0: ; %main_body
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB71_2
+; GFX940-NEXT: ; %bb.1:
; GFX940-NEXT: s_load_dword s0, s[0:1], 0x24
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], 4.0
+; GFX940-NEXT: s_bcnt1_i32_b64 s1, s[2:3]
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s1
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
; GFX940-NEXT: v_mov_b32_e32 v2, s0
; GFX940-NEXT: ds_add_f64 v2, v[0:1]
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
+; GFX940-NEXT: .LBB71_2:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(3) %ptr, double 4.0 seq_cst
@@ -2072,46 +2208,66 @@ main_body:
define amdgpu_kernel void @local_atomic_fadd_f64_noret_pat_flush_safe(ptr addrspace(3) %ptr) #4 {
; GFX90A-LABEL: local_atomic_fadd_f64_noret_pat_flush_safe:
; GFX90A: ; %bb.0: ; %main_body
-; GFX90A-NEXT: s_load_dword s2, s[0:1], 0x24
-; GFX90A-NEXT: s_mov_b64 s[0:1], 0
+; GFX90A-NEXT: s_mov_b64 s[2:3], exec
+; GFX90A-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX90A-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX90A-NEXT: s_cbranch_execz .LBB72_3
+; GFX90A-NEXT: ; %bb.1:
+; GFX90A-NEXT: s_load_dword s4, s[0:1], 0x24
+; GFX90A-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT: v_mov_b32_e32 v0, s2
-; GFX90A-NEXT: ds_read_b64 v[0:1], v0
-; GFX90A-NEXT: v_mov_b32_e32 v2, s2
-; GFX90A-NEXT: .LBB72_1: ; %atomicrmw.start
+; GFX90A-NEXT: v_mov_b32_e32 v0, s4
+; GFX90A-NEXT: ds_read_b64 v[2:3], v0
+; GFX90A-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX90A-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
+; GFX90A-NEXT: s_mov_b64 s[0:1], 0
+; GFX90A-NEXT: v_mov_b32_e32 v4, s4
+; GFX90A-NEXT: .LBB72_2: ; %atomicrmw.start
; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT: v_add_f64 v[4:5], v[0:1], 4.0
-; GFX90A-NEXT: ds_cmpst_rtn_b64 v[4:5], v2, v[0:1], v[4:5]
+; GFX90A-NEXT: v_add_f64 v[6:7], v[2:3], v[0:1]
+; GFX90A-NEXT: ds_cmpst_rtn_b64 v[6:7], v4, v[2:3], v[6:7]
; GFX90A-NEXT: s_waitcnt lgkmcnt(0)
-; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[0:1]
+; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
; GFX90A-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
-; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[4:5] op_sel:[0,1]
+; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[6:7], v[6:7] op_sel:[0,1]
; GFX90A-NEXT: s_andn2_b64 exec, exec, s[0:1]
-; GFX90A-NEXT: s_cbranch_execnz .LBB72_1
-; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX90A-NEXT: s_cbranch_execnz .LBB72_2
+; GFX90A-NEXT: .LBB72_3:
; GFX90A-NEXT: s_endpgm
;
; GFX940-LABEL: local_atomic_fadd_f64_noret_pat_flush_safe:
; GFX940: ; %bb.0: ; %main_body
-; GFX940-NEXT: s_load_dword s2, s[0:1], 0x24
-; GFX940-NEXT: s_mov_b64 s[0:1], 0
+; GFX940-NEXT: s_mov_b64 s[2:3], exec
+; GFX940-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX940-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX940-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX940-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX940-NEXT: s_cbranch_execz .LBB72_3
+; GFX940-NEXT: ; %bb.1:
+; GFX940-NEXT: s_load_dword s4, s[0:1], 0x24
+; GFX940-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
-; GFX940-NEXT: v_mov_b32_e32 v0, s2
-; GFX940-NEXT: ds_read_b64 v[0:1], v0
-; GFX940-NEXT: v_mov_b32_e32 v2, s2
-; GFX940-NEXT: .LBB72_1: ; %atomicrmw.start
+; GFX940-NEXT: v_mov_b32_e32 v0, s4
+; GFX940-NEXT: ds_read_b64 v[2:3], v0
+; GFX940-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX940-NEXT: v_mul_f64 v[0:1], v[0:1], 4.0
+; GFX940-NEXT: s_mov_b64 s[0:1], 0
+; GFX940-NEXT: v_mov_b32_e32 v4, s4
+; GFX940-NEXT: .LBB72_2: ; %atomicrmw.start
; GFX940-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
-; GFX940-NEXT: v_add_f64 v[4:5], v[0:1], 4.0
-; GFX940-NEXT: ds_cmpst_rtn_b64 v[4:5], v2, v[0:1], v[4:5]
+; GFX940-NEXT: v_add_f64 v[6:7], v[2:3], v[0:1]
+; GFX940-NEXT: ds_cmpst_rtn_b64 v[6:7], v4, v[2:3], v[6:7]
; GFX940-NEXT: s_waitcnt lgkmcnt(0)
-; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[0:1]
+; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
; GFX940-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
-; GFX940-NEXT: v_mov_b64_e32 v[0:1], v[4:5]
+; GFX940-NEXT: v_mov_b64_e32 v[2:3], v[6:7]
; GFX940-NEXT: s_andn2_b64 exec, exec, s[0:1]
-; GFX940-NEXT: s_cbranch_execnz .LBB72_1
-; GFX940-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX940-NEXT: s_cbranch_execnz .LBB72_2
+; GFX940-NEXT: .LBB72_3:
; GFX940-NEXT: s_endpgm
main_body:
%ret = atomicrmw fadd ptr addrspace(3) %ptr, double 4.0 seq_cst
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fadd.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fadd.ll
index 96c615b974ce17..4f00d48551bed0 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fadd.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fadd.ll
@@ -13,6 +13,7 @@
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizer-strategy=DPP -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX1132-DPP %s
declare float @div.float.value()
+declare double @div.double.value()
define amdgpu_kernel void @global_atomic_fadd_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
; GFX7LESS-LABEL: global_atomic_fadd_uni_address_uni_value_agent_scope_unsafe:
@@ -5408,6 +5409,5583 @@ define amdgpu_kernel void @global_atomic_fadd_uni_address_div_value_defalut_scop
ret void
}
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s3
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s2
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], exec
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s0, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s1, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB9_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x9
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[2:3], s[36:37], 0x0
+; GFX7LESS-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX7LESS-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX7LESS-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], 0
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, s2
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, s3
+; GFX7LESS-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[0:1], v[41:42]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s34, 44
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:8
+; GFX7LESS-NEXT: s_addc_u32 s9, s35, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(2)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b32 s12, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s36
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s37
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB9_2
+; GFX7LESS-NEXT: .LBB9_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s42, -1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], exec
+; GFX9-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 0
+; GFX9-NEXT: s_add_u32 s40, s40, s3
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v0
+; GFX9-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-NEXT: s_mov_b32 s33, s2
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB9_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX9-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX9-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX9-NEXT: s_add_u32 s8, s34, 44
+; GFX9-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-NEXT: s_mov_b32 s12, s33
+; GFX9-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-NEXT: s_cbranch_execnz .LBB9_2
+; GFX9-NEXT: .LBB9_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s42, -1
+; GFX1064-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-NEXT: s_mov_b32 s33, s2
+; GFX1064-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB9_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1064-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1064-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-NEXT: s_mov_b32 s12, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1064-NEXT: .LBB9_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s33, s2
+; GFX1032-NEXT: s_mov_b32 s2, exec_lo
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1032-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s42, -1
+; GFX1032-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-NEXT: s_mov_b32 s38, 0
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB9_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1032-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-NEXT: s_mov_b32 s12, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1032-NEXT: .LBB9_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s33, s2
+; GFX1164-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-NEXT: s_cbranch_execz .LBB9_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1164-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1164-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b32 s12, s33
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1164-NEXT: .LBB9_3:
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b32 s2, exec_lo
+; GFX1132-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s38, 0
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-NEXT: s_cbranch_execz .LBB9_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1132-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b32 s12, s33
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1132-NEXT: .LBB9_3:
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s42, -1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX9-DPP-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 0
+; GFX9-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v0
+; GFX9-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-DPP-NEXT: s_mov_b32 s33, s2
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX9-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX9-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: s_mov_b32 s12, s33
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX9-DPP-NEXT: .LBB9_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1064-DPP-NEXT: .LBB9_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1032-DPP-NEXT: s_mov_b32 s2, exec_lo
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1032-DPP-NEXT: .LBB9_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1164-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1164-DPP-NEXT: .LBB9_3:
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b32 s2, exec_lo
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1132-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1132-DPP-NEXT: .LBB9_3:
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s50, -1
+; GFX7LESS-NEXT: s_mov_b32 s51, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s48, s48, s9
+; GFX7LESS-NEXT: s_addc_u32 s49, s49, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s8
+; GFX7LESS-NEXT: s_mov_b32 s40, s7
+; GFX7LESS-NEXT: s_mov_b32 s41, s6
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX7LESS-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX7LESS-NEXT: s_load_dwordx2 s[44:45], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s47, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s46, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v42, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v42
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: v_mov_b32_e32 v41, v1
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[0:1], off, s[44:47], 0
+; GFX7LESS-NEXT: s_mov_b64 s[42:43], 0
+; GFX7LESS-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[0:1], v[40:41]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[48:51], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:8
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(2)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v42
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s44
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s45
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[48:51], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[42:43], vcc, s[42:43]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[42:43]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB10_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s50, -1
+; GFX9-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-NEXT: s_add_u32 s48, s48, s9
+; GFX9-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-NEXT: s_mov_b32 s33, s8
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_mov_b32 s40, s7
+; GFX9-NEXT: s_mov_b32 s41, s6
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: v_mov_b32_e32 v43, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: v_mov_b32_e32 v41, v1
+; GFX9-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-NEXT: s_cbranch_execnz .LBB10_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s50, -1
+; GFX1064-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-NEXT: s_mov_b32 s33, s8
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_mov_b32 s40, s7
+; GFX1064-NEXT: s_mov_b32 s41, s6
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mov_b32_e32 v43, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: v_mov_b32_e32 v41, v1
+; GFX1064-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s50, -1
+; GFX1032-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-NEXT: s_mov_b32 s33, s8
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_mov_b32 s40, s7
+; GFX1032-NEXT: s_mov_b32 s41, s6
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: v_mov_b32_e32 v43, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: v_mov_b32_e32 v41, v1
+; GFX1032-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: s_mov_b32 s44, 0
+; GFX1032-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-NEXT: s_mov_b32 s33, s8
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: v_mov_b32_e32 v42, v0
+; GFX1164-NEXT: s_mov_b32 s40, s7
+; GFX1164-NEXT: s_mov_b32 s41, s6
+; GFX1164-NEXT: v_mov_b32_e32 v43, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: v_mov_b32_e32 v41, v1
+; GFX1164-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v42
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s41
+; GFX1164-NEXT: s_mov_b32 s13, s40
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-NEXT: s_mov_b32 s40, s14
+; GFX1132-NEXT: s_mov_b32 s41, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: v_dual_mov_b32 v42, v0 :: v_dual_mov_b32 v43, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, v1
+; GFX1132-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1132-NEXT: s_mov_b32 s44, 0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v42 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s41
+; GFX1132-NEXT: s_mov_b32 s13, s40
+; GFX1132-NEXT: s_mov_b32 s14, s33
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s42
+; GFX1132-NEXT: v_dual_mov_b32 v3, s43 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s50, -1
+; GFX9-DPP-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX9-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-DPP-NEXT: s_mov_b32 s33, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_mov_b32 s40, s7
+; GFX9-DPP-NEXT: s_mov_b32 s41, s6
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX9-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1064-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1032-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1032-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v42, v0
+; GFX1164-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1164-DPP-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-DPP-NEXT: s_mov_b32 s40, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s41, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v42, v0 :: v_dual_mov_b32 v43, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, v1
+; GFX1132-DPP-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1132-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v42 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s42
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s43 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.float.value()
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %divValue syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp(ptr addrspace(1) %ptr) #1 {
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s12, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s14, -1
+; GFX7LESS-NEXT: s_mov_b32 s15, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s12, s12, s3
+; GFX7LESS-NEXT: s_addc_u32 s13, s13, 0
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], exec
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s2, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s3, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB11_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7LESS-NEXT: s_bcnt1_i32_b64 s6, s[2:3]
+; GFX7LESS-NEXT: s_mov_b32 s7, 0x43300000
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], 0
+; GFX7LESS-NEXT: s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT: v_add_f64 v[0:1], s[6:7], v[0:1]
+; GFX7LESS-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s8
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s9
+; GFX7LESS-NEXT: s_mov_b32 s2, -1
+; GFX7LESS-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v2
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v1
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v0
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[0:3], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
+; GFX7LESS-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[4:5]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB11_2
+; GFX7LESS-NEXT: .LBB11_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s10, -1
+; GFX9-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9-NEXT: s_add_u32 s8, s8, s3
+; GFX9-NEXT: s_mov_b64 s[2:3], exec
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX9-NEXT: s_addc_u32 s9, s9, 0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB11_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-NEXT: s_mov_b32 s3, 0x43300000
+; GFX9-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT: s_cbranch_execnz .LBB11_2
+; GFX9-NEXT: .LBB11_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s10, -1
+; GFX1064-NEXT: s_mov_b32 s11, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s8, s8, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-NEXT: s_addc_u32 s9, s9, 0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB11_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-NEXT: s_mov_b32 s3, 0x43300000
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-NEXT: v_add_f64 v[0:1], 0xc3300000, s[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1064-NEXT: .LBB11_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s10, -1
+; GFX1032-NEXT: s_mov_b32 s11, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s8, s8, s3
+; GFX1032-NEXT: s_mov_b32 s3, exec_lo
+; GFX1032-NEXT: s_addc_u32 s9, s9, 0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
+; GFX1032-NEXT: s_mov_b32 s2, 0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB11_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_bcnt1_i32_b32 s4, s3
+; GFX1032-NEXT: s_mov_b32 s5, 0x43300000
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-NEXT: v_add_f64 v[0:1], 0xc3300000, s[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1032-NEXT: .LBB11_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_bcnt1_i32_b64 s2, exec
+; GFX1164-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-NEXT: v_mov_b32_e32 v1, s2
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1164-NEXT: scratch_store_b32 off, v1, off
+; GFX1164-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-NEXT: s_cbranch_execz .LBB11_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1164-NEXT: .LBB11_3:
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_bcnt1_i32_b32 s2, exec_lo
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: v_dual_mov_b32 v0, 0x43300000 :: v_dual_mov_b32 v1, s2
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-NEXT: s_mov_b32 s2, 0
+; GFX1132-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1132-NEXT: scratch_store_b32 off, v1, off
+; GFX1132-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-NEXT: s_cbranch_execz .LBB11_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-NEXT: v_mov_b32_e32 v6, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1132-NEXT: .LBB11_3:
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s10, -1
+; GFX9-DPP-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX9-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-DPP-NEXT: s_mov_b32 s3, 0x43300000
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX9-DPP-NEXT: .LBB11_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s10, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s11, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-DPP-NEXT: s_mov_b32 s3, 0x43300000
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1064-DPP-NEXT: .LBB11_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s10, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s11, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX1032-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_bcnt1_i32_b32 s4, s3
+; GFX1032-DPP-NEXT: s_mov_b32 s5, 0x43300000
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1032-DPP-NEXT: .LBB11_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_bcnt1_i32_b64 s2, exec
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s2
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v1, off
+; GFX1164-DPP-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1164-DPP-NEXT: .LBB11_3:
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_bcnt1_i32_b32 s2, exec_lo
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v0, 0x43300000 :: v_dual_mov_b32 v1, s2
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v1, off
+; GFX1132-DPP-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1132-DPP-NEXT: .LBB11_3:
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp(ptr addrspace(1) %ptr) #1 {
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s32, 0
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s9
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s14, s8
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s39, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s38, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s2, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s3, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[2:3]
+; GFX7LESS-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v31, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s12, s6
+; GFX7LESS-NEXT: s_mov_b32 s13, s7
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[4:5], off, s[36:39], 0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], 0
+; GFX7LESS-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v5
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v4
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v2
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[36:39], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[4:5]
+; GFX7LESS-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB12_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s38, -1
+; GFX9-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-NEXT: s_add_u32 s36, s36, s9
+; GFX9-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-NEXT: s_mov_b32 s14, s8
+; GFX9-NEXT: s_add_u32 s8, s2, 44
+; GFX9-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-NEXT: s_getpc_b64 s[2:3]
+; GFX9-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-NEXT: s_mov_b32 s12, s6
+; GFX9-NEXT: s_mov_b32 s13, s7
+; GFX9-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT: s_mov_b32 s32, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-NEXT: s_cbranch_execnz .LBB12_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s38, -1
+; GFX1064-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-NEXT: s_mov_b32 s14, s8
+; GFX1064-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s6
+; GFX1064-NEXT: s_mov_b32 s13, s7
+; GFX1064-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-NEXT: s_mov_b32 s32, 0
+; GFX1064-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s38, -1
+; GFX1032-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-NEXT: s_mov_b32 s14, s8
+; GFX1032-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s6
+; GFX1032-NEXT: s_mov_b32 s13, s7
+; GFX1032-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-NEXT: s_mov_b32 s32, 0
+; GFX1032-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-NEXT: s_mov_b32 s0, 0
+; GFX1032-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s14, s8
+; GFX1164-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s32, 0
+; GFX1164-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-NEXT: s_mov_b32 s0, 0
+; GFX1132-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s38, -1
+; GFX9-DPP-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX9-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-DPP-NEXT: s_mov_b32 s14, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-DPP-NEXT: s_getpc_b64 s[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b32 s12, s6
+; GFX9-DPP-NEXT: s_mov_b32 s13, s7
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b32 s32, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1064-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1032-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1032-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1132-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value() strictfp
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %divValue syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp(ptr addrspace(1) %ptr) #2{
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s12, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s14, -1
+; GFX7LESS-NEXT: s_mov_b32 s15, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s12, s12, s3
+; GFX7LESS-NEXT: s_addc_u32 s13, s13, 0
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], exec
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s2, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s3, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB13_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7LESS-NEXT: s_bcnt1_i32_b64 s6, s[2:3]
+; GFX7LESS-NEXT: s_mov_b32 s7, 0x43300000
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], 0
+; GFX7LESS-NEXT: s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT: v_add_f64 v[0:1], s[6:7], v[0:1]
+; GFX7LESS-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s8
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s9
+; GFX7LESS-NEXT: s_mov_b32 s2, -1
+; GFX7LESS-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v2
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v1
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v0
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[0:3], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
+; GFX7LESS-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[4:5]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB13_2
+; GFX7LESS-NEXT: .LBB13_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s10, -1
+; GFX9-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9-NEXT: s_add_u32 s8, s8, s3
+; GFX9-NEXT: s_mov_b64 s[2:3], exec
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX9-NEXT: s_addc_u32 s9, s9, 0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB13_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-NEXT: s_mov_b32 s3, 0x43300000
+; GFX9-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT: s_cbranch_execnz .LBB13_2
+; GFX9-NEXT: .LBB13_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s10, -1
+; GFX1064-NEXT: s_mov_b32 s11, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s8, s8, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-NEXT: s_addc_u32 s9, s9, 0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB13_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-NEXT: s_mov_b32 s3, 0x43300000
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-NEXT: v_add_f64 v[0:1], 0xc3300000, s[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1064-NEXT: .LBB13_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s10, -1
+; GFX1032-NEXT: s_mov_b32 s11, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s8, s8, s3
+; GFX1032-NEXT: s_mov_b32 s3, exec_lo
+; GFX1032-NEXT: s_addc_u32 s9, s9, 0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
+; GFX1032-NEXT: s_mov_b32 s2, 0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB13_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_bcnt1_i32_b32 s4, s3
+; GFX1032-NEXT: s_mov_b32 s5, 0x43300000
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-NEXT: v_add_f64 v[0:1], 0xc3300000, s[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1032-NEXT: .LBB13_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_bcnt1_i32_b64 s2, exec
+; GFX1164-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-NEXT: v_mov_b32_e32 v1, s2
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1164-NEXT: scratch_store_b32 off, v1, off
+; GFX1164-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-NEXT: s_cbranch_execz .LBB13_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1164-NEXT: .LBB13_3:
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_bcnt1_i32_b32 s2, exec_lo
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: v_dual_mov_b32 v0, 0x43300000 :: v_dual_mov_b32 v1, s2
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-NEXT: s_mov_b32 s2, 0
+; GFX1132-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1132-NEXT: scratch_store_b32 off, v1, off
+; GFX1132-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-NEXT: s_cbranch_execz .LBB13_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-NEXT: v_mov_b32_e32 v6, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1132-NEXT: .LBB13_3:
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s10, -1
+; GFX9-DPP-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX9-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-DPP-NEXT: s_mov_b32 s3, 0x43300000
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX9-DPP-NEXT: .LBB13_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s10, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s11, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-DPP-NEXT: s_mov_b32 s3, 0x43300000
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1064-DPP-NEXT: .LBB13_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s10, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s11, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX1032-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_bcnt1_i32_b32 s4, s3
+; GFX1032-DPP-NEXT: s_mov_b32 s5, 0x43300000
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1032-DPP-NEXT: .LBB13_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_bcnt1_i32_b64 s2, exec
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s2
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v1, off
+; GFX1164-DPP-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1164-DPP-NEXT: .LBB13_3:
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_bcnt1_i32_b32 s2, exec_lo
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v0, 0x43300000 :: v_dual_mov_b32 v1, s2
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v1, off
+; GFX1132-DPP-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1132-DPP-NEXT: .LBB13_3:
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s32, 0
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s9
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s14, s8
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s39, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s38, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s2, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s3, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[2:3]
+; GFX7LESS-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v31, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s12, s6
+; GFX7LESS-NEXT: s_mov_b32 s13, s7
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[4:5], off, s[36:39], 0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], 0
+; GFX7LESS-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v5
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v4
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v2
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[36:39], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[4:5]
+; GFX7LESS-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB14_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s38, -1
+; GFX9-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-NEXT: s_add_u32 s36, s36, s9
+; GFX9-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-NEXT: s_mov_b32 s14, s8
+; GFX9-NEXT: s_add_u32 s8, s2, 44
+; GFX9-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-NEXT: s_getpc_b64 s[2:3]
+; GFX9-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-NEXT: s_mov_b32 s12, s6
+; GFX9-NEXT: s_mov_b32 s13, s7
+; GFX9-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT: s_mov_b32 s32, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-NEXT: s_cbranch_execnz .LBB14_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s38, -1
+; GFX1064-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-NEXT: s_mov_b32 s14, s8
+; GFX1064-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s6
+; GFX1064-NEXT: s_mov_b32 s13, s7
+; GFX1064-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-NEXT: s_mov_b32 s32, 0
+; GFX1064-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s38, -1
+; GFX1032-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-NEXT: s_mov_b32 s14, s8
+; GFX1032-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s6
+; GFX1032-NEXT: s_mov_b32 s13, s7
+; GFX1032-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-NEXT: s_mov_b32 s32, 0
+; GFX1032-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-NEXT: s_mov_b32 s0, 0
+; GFX1032-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s14, s8
+; GFX1164-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s32, 0
+; GFX1164-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-NEXT: s_mov_b32 s0, 0
+; GFX1132-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s38, -1
+; GFX9-DPP-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX9-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-DPP-NEXT: s_mov_b32 s14, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-DPP-NEXT: s_getpc_b64 s[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b32 s12, s6
+; GFX9-DPP-NEXT: s_mov_b32 s13, s7
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b32 s32, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1064-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1032-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1032-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1132-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value()
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %divValue syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp(ptr addrspace(1) %ptr) #1 {
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s32, 0
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s9
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s14, s8
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s39, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s38, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s2, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s3, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[2:3]
+; GFX7LESS-NEXT: s_add_u32 s2, s2, div.float.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s3, s3, div.float.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v31, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s12, s6
+; GFX7LESS-NEXT: s_mov_b32 s13, s7
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[4:5], off, s[36:39], 0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], 0
+; GFX7LESS-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v5
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v4
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v2
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[36:39], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[4:5]
+; GFX7LESS-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB15_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s38, -1
+; GFX9-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-NEXT: s_add_u32 s36, s36, s9
+; GFX9-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-NEXT: s_mov_b32 s14, s8
+; GFX9-NEXT: s_add_u32 s8, s2, 44
+; GFX9-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-NEXT: s_getpc_b64 s[2:3]
+; GFX9-NEXT: s_add_u32 s2, s2, div.float.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s3, s3, div.float.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-NEXT: s_mov_b32 s12, s6
+; GFX9-NEXT: s_mov_b32 s13, s7
+; GFX9-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT: s_mov_b32 s32, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-NEXT: s_cbranch_execnz .LBB15_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s38, -1
+; GFX1064-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-NEXT: s_mov_b32 s14, s8
+; GFX1064-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s6
+; GFX1064-NEXT: s_mov_b32 s13, s7
+; GFX1064-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-NEXT: s_mov_b32 s32, 0
+; GFX1064-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s38, -1
+; GFX1032-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-NEXT: s_mov_b32 s14, s8
+; GFX1032-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s6
+; GFX1032-NEXT: s_mov_b32 s13, s7
+; GFX1032-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-NEXT: s_mov_b32 s32, 0
+; GFX1032-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-NEXT: s_mov_b32 s0, 0
+; GFX1032-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s14, s8
+; GFX1164-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s32, 0
+; GFX1164-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1132-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-NEXT: s_mov_b32 s0, 0
+; GFX1132-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s38, -1
+; GFX9-DPP-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX9-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-DPP-NEXT: s_mov_b32 s14, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-DPP-NEXT: s_getpc_b64 s[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s2, s2, div.float.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s3, s3, div.float.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b32 s12, s6
+; GFX9-DPP-NEXT: s_mov_b32 s13, s7
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b32 s32, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-DPP-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1064-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-DPP-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1032-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1032-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-DPP-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-DPP-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1132-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[2:3], v[4:5], v[0:1]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.float.value() strictfp
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %divValue syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp(ptr addrspace(1) %ptr) #2 {
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s3
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s2
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], exec
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s0, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s1, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB16_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x9
+; GFX7LESS-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s1, 0x43300000
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[2:3], s[36:37], 0x0
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX7LESS-NEXT: v_add_f64 v[0:1], s[0:1], v[0:1]
+; GFX7LESS-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], 0
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, s2
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, s3
+; GFX7LESS-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[0:1], v[41:42]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s34, 44
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:8
+; GFX7LESS-NEXT: s_addc_u32 s9, s35, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(2)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b32 s12, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s36
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s37
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB16_2
+; GFX7LESS-NEXT: .LBB16_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s42, -1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], exec
+; GFX9-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 0
+; GFX9-NEXT: s_add_u32 s40, s40, s3
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v0
+; GFX9-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-NEXT: s_mov_b32 s33, s2
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB16_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-NEXT: s_mov_b32 s1, 0x43300000
+; GFX9-NEXT: v_add_f64 v[0:1], s[0:1], v[0:1]
+; GFX9-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX9-NEXT: s_add_u32 s8, s34, 44
+; GFX9-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-NEXT: s_mov_b32 s12, s33
+; GFX9-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-NEXT: s_cbranch_execnz .LBB16_2
+; GFX9-NEXT: .LBB16_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s42, -1
+; GFX1064-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-NEXT: s_mov_b32 s33, s2
+; GFX1064-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB16_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1064-NEXT: s_mov_b32 s1, 0x43300000
+; GFX1064-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-NEXT: v_add_f64 v[0:1], 0xc3300000, s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-NEXT: s_mov_b32 s12, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1064-NEXT: .LBB16_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s33, s2
+; GFX1032-NEXT: s_mov_b32 s2, exec_lo
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1032-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s42, -1
+; GFX1032-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-NEXT: s_mov_b32 s38, 0
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB16_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1032-NEXT: s_mov_b32 s1, 0x43300000
+; GFX1032-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-NEXT: v_add_f64 v[0:1], 0xc3300000, s[0:1]
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-NEXT: s_mov_b32 s12, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1032-NEXT: .LBB16_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-NEXT: s_bcnt1_i32_b64 s0, exec
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b32 off, v0, off offset:20
+; GFX1164-NEXT: scratch_store_b32 off, v1, off offset:16
+; GFX1164-NEXT: scratch_load_b64 v[0:1], off, off offset:16
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-NEXT: s_cbranch_execz .LBB16_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-NEXT: s_mov_b32 s33, s2
+; GFX1164-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b32 s12, s33
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1164-NEXT: .LBB16_3:
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-NEXT: s_bcnt1_i32_b32 s0, exec_lo
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b32 off, v0, off offset:20
+; GFX1132-NEXT: scratch_store_b32 off, v1, off offset:16
+; GFX1132-NEXT: scratch_load_b64 v[0:1], off, off offset:16
+; GFX1132-NEXT: s_mov_b32 s38, 0
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-NEXT: s_cbranch_execz .LBB16_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b32 s12, s33
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1132-NEXT: .LBB16_3:
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s42, -1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX9-DPP-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 0
+; GFX9-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v0
+; GFX9-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-DPP-NEXT: s_mov_b32 s33, s2
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-DPP-NEXT: s_mov_b32 s1, 0x43300000
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], s[0:1], v[0:1]
+; GFX9-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX9-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: s_mov_b32 s12, s33
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX9-DPP-NEXT: .LBB16_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1064-DPP-NEXT: s_mov_b32 s1, 0x43300000
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1064-DPP-NEXT: .LBB16_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1032-DPP-NEXT: s_mov_b32 s2, exec_lo
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1032-DPP-NEXT: s_mov_b32 s1, 0x43300000
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[0:1]
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1032-DPP-NEXT: .LBB16_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-DPP-NEXT: s_bcnt1_i32_b64 s0, exec
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v0, off offset:20
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v1, off offset:16
+; GFX1164-DPP-NEXT: scratch_load_b64 v[0:1], off, off offset:16
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1164-DPP-NEXT: .LBB16_3:
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-DPP-NEXT: s_bcnt1_i32_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v0, off offset:20
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v1, off offset:16
+; GFX1132-DPP-NEXT: scratch_load_b64 v[0:1], off, off offset:16
+; GFX1132-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[41:42]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1132-DPP-NEXT: .LBB16_3:
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double 4.0 monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp(ptr addrspace(1) %ptr) #2 {
+; GFX7LESS-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s50, -1
+; GFX7LESS-NEXT: s_mov_b32 s51, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s48, s48, s9
+; GFX7LESS-NEXT: s_addc_u32 s49, s49, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s8
+; GFX7LESS-NEXT: s_mov_b32 s40, s7
+; GFX7LESS-NEXT: s_mov_b32 s41, s6
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX7LESS-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX7LESS-NEXT: s_load_dwordx2 s[44:45], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s47, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s46, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v42, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v42
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: v_mov_b32_e32 v41, v1
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[0:1], off, s[44:47], 0
+; GFX7LESS-NEXT: s_mov_b64 s[42:43], 0
+; GFX7LESS-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[0:1], v[40:41]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[48:51], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:8
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(2)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v42
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s44
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s45
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[48:51], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[42:43], vcc, s[42:43]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[42:43]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB17_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s50, -1
+; GFX9-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-NEXT: s_add_u32 s48, s48, s9
+; GFX9-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-NEXT: s_mov_b32 s33, s8
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_mov_b32 s40, s7
+; GFX9-NEXT: s_mov_b32 s41, s6
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: v_mov_b32_e32 v43, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: v_mov_b32_e32 v41, v1
+; GFX9-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-NEXT: s_cbranch_execnz .LBB17_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s50, -1
+; GFX1064-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-NEXT: s_mov_b32 s33, s8
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_mov_b32 s40, s7
+; GFX1064-NEXT: s_mov_b32 s41, s6
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mov_b32_e32 v43, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: v_mov_b32_e32 v41, v1
+; GFX1064-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s50, -1
+; GFX1032-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-NEXT: s_mov_b32 s33, s8
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_mov_b32 s40, s7
+; GFX1032-NEXT: s_mov_b32 s41, s6
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: v_mov_b32_e32 v43, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: v_mov_b32_e32 v41, v1
+; GFX1032-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: s_mov_b32 s44, 0
+; GFX1032-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-NEXT: s_mov_b32 s33, s8
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: v_mov_b32_e32 v42, v0
+; GFX1164-NEXT: s_mov_b32 s40, s7
+; GFX1164-NEXT: s_mov_b32 s41, s6
+; GFX1164-NEXT: v_mov_b32_e32 v43, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: v_mov_b32_e32 v41, v1
+; GFX1164-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v42
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s41
+; GFX1164-NEXT: s_mov_b32 s13, s40
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-NEXT: s_mov_b32 s40, s14
+; GFX1132-NEXT: s_mov_b32 s41, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: v_dual_mov_b32 v42, v0 :: v_dual_mov_b32 v43, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, v1
+; GFX1132-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1132-NEXT: s_mov_b32 s44, 0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v42 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s41
+; GFX1132-NEXT: s_mov_b32 s13, s40
+; GFX1132-NEXT: s_mov_b32 s14, s33
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s42
+; GFX1132-NEXT: v_dual_mov_b32 v3, s43 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s50, -1
+; GFX9-DPP-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX9-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-DPP-NEXT: s_mov_b32 s33, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_mov_b32 s40, s7
+; GFX9-DPP-NEXT: s_mov_b32 s41, s6
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX9-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1064-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1032-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1032-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v42, v0
+; GFX1164-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1164-DPP-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fadd_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-DPP-NEXT: s_mov_b32 s40, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s41, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v42, v0 :: v_dual_mov_b32 v43, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, v1
+; GFX1132-DPP-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1132-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[3:4], v[1:2], v[40:41]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v42 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s42
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s43 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.float.value() strictfp
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %divValue monotonic, align 4
+ ret void
+}
+
attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #1 = { strictfp "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #2 = { strictfp}
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmax.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmax.ll
index 3cc5a4cd1d0aa1..622be43e744291 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmax.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmax.ll
@@ -13,6 +13,7 @@
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizer-strategy=DPP -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX1132-DPP %s
declare float @div.float.value()
+declare float @div.double.value()
define amdgpu_kernel void @global_atomic_fmax_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
; GFX7LESS-LABEL: global_atomic_fmax_uni_address_uni_value_agent_scope_unsafe:
@@ -3550,6 +3551,3965 @@ define amdgpu_kernel void @global_atomic_fmax_uni_address_div_value_defalut_scop
ret void
}
+define amdgpu_kernel void @global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s3
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, exec_lo, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, exec_hi, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB6_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_mov_b32 s33, s2
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x9
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], 0
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, s0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, s1
+; GFX7LESS-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[2:3], v[0:1], v[0:1]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s34, 44
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], 4.0
+; GFX7LESS-NEXT: s_addc_u32 s9, s35, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0 offset:8
+; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b32 s12, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s36
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s37
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB6_2
+; GFX7LESS-NEXT: .LBB6_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s42, -1
+; GFX9-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-NEXT: s_add_u32 s40, s40, s3
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB6_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-NEXT: s_mov_b32 s33, s2
+; GFX9-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX9-NEXT: s_add_u32 s8, s34, 44
+; GFX9-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-NEXT: s_mov_b32 s12, s33
+; GFX9-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-NEXT: s_cbranch_execnz .LBB6_2
+; GFX9-NEXT: .LBB6_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s42, -1
+; GFX1064-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB6_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-NEXT: s_mov_b32 s33, s2
+; GFX1064-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-NEXT: s_mov_b32 s12, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1064-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1064-NEXT: .LBB6_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s42, -1
+; GFX1032-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-NEXT: s_mov_b32 s38, 0
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB6_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-NEXT: s_mov_b32 s33, s2
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-NEXT: s_mov_b32 s12, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1032-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1032-NEXT: .LBB6_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-NEXT: s_cbranch_execz .LBB6_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-NEXT: s_mov_b32 s33, s2
+; GFX1164-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b32 s12, s33
+; GFX1164-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1164-NEXT: .LBB6_3:
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s38, 0
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-NEXT: s_cbranch_execz .LBB6_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b32 s12, s33
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX1132-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1132-NEXT: .LBB6_3:
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s42, -1
+; GFX9-DPP-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-DPP-NEXT: s_mov_b32 s33, s2
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX9-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: s_mov_b32 s12, s33
+; GFX9-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX9-DPP-NEXT: .LBB6_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1064-DPP-NEXT: .LBB6_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1032-DPP-NEXT: .LBB6_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1164-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1164-DPP-NEXT: .LBB6_3:
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX1132-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1132-DPP-NEXT: .LBB6_3:
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double 4.0 syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s50, -1
+; GFX7LESS-NEXT: s_mov_b32 s51, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s48, s48, s9
+; GFX7LESS-NEXT: s_addc_u32 s49, s49, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s8
+; GFX7LESS-NEXT: s_mov_b32 s40, s7
+; GFX7LESS-NEXT: s_mov_b32 s41, s6
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX7LESS-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX7LESS-NEXT: s_load_dwordx2 s[44:45], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s47, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s46, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v40, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[2:3], off, s[44:47], 0
+; GFX7LESS-NEXT: s_mov_b64 s[42:43], 0
+; GFX7LESS-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX7LESS-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s44
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s45
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX7LESS-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7LESS-NEXT: s_or_b64 s[42:43], vcc, s[42:43]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[42:43]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB7_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s50, -1
+; GFX9-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-NEXT: s_add_u32 s48, s48, s9
+; GFX9-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-NEXT: s_mov_b32 s33, s8
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_mov_b32 s40, s7
+; GFX9-NEXT: s_mov_b32 s41, s6
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: v_mov_b32_e32 v41, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX9-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX9-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX9-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-NEXT: s_cbranch_execnz .LBB7_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s50, -1
+; GFX1064-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-NEXT: s_mov_b32 s33, s8
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_mov_b32 s40, s7
+; GFX1064-NEXT: s_mov_b32 s41, s6
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mov_b32_e32 v41, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1064-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1064-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1064-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1064-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s50, -1
+; GFX1032-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-NEXT: s_mov_b32 s33, s8
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_mov_b32 s40, s7
+; GFX1032-NEXT: s_mov_b32 s41, s6
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: v_mov_b32_e32 v41, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1032-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1032-NEXT: s_mov_b32 s44, 0
+; GFX1032-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1032-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1032-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-NEXT: s_mov_b32 s33, s8
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: s_mov_b32 s40, s7
+; GFX1164-NEXT: s_mov_b32 s41, s6
+; GFX1164-NEXT: v_mov_b32_e32 v41, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1164-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1164-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s41
+; GFX1164-NEXT: s_mov_b32 s13, s40
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1164-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-NEXT: s_mov_b32 s40, s14
+; GFX1132-NEXT: s_mov_b32 s41, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1132-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1132-NEXT: s_mov_b32 s44, 0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v40
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s41
+; GFX1132-NEXT: s_mov_b32 s13, s40
+; GFX1132-NEXT: s_mov_b32 s14, s33
+; GFX1132-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1132-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v0, 8 :: v_dual_mov_b32 v1, 0
+; GFX1132-NEXT: v_dual_mov_b32 v2, s42 :: v_dual_mov_b32 v3, s43
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s50, -1
+; GFX9-DPP-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX9-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-DPP-NEXT: s_mov_b32 s33, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_mov_b32 s40, s7
+; GFX9-DPP-NEXT: s_mov_b32 s41, s6
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX9-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX9-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1064-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1064-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1032-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1032-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1032-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1164-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1164-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-DPP-NEXT: s_mov_b32 s40, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s41, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1132-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v0, 8 :: v_dual_mov_b32 v1, 0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s42 :: v_dual_mov_b32 v3, s43
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value()
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %divValue syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, exec_lo, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, exec_hi, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB8_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], 0
+; GFX7LESS-NEXT: s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s6
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s7
+; GFX7LESS-NEXT: s_mov_b32 s2, -1
+; GFX7LESS-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v2
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, v1
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, v0
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[4:7], off, s[0:3], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[2:3]
+; GFX7LESS-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, v4
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, v5
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[4:5]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB8_2
+; GFX7LESS-NEXT: .LBB8_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB8_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT: s_cbranch_execnz .LBB8_2
+; GFX9-NEXT: .LBB8_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB8_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1064-NEXT: .LBB8_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-NEXT: s_mov_b32 s2, 0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_and_saveexec_b32 s3, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB8_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1032-NEXT: .LBB8_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-NEXT: s_cbranch_execz .LBB8_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1164-NEXT: .LBB8_3:
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-NEXT: s_mov_b32 s2, 0
+; GFX1132-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-NEXT: s_cbranch_execz .LBB8_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1132-NEXT: .LBB8_3:
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX9-DPP-NEXT: .LBB8_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1064-DPP-NEXT: .LBB8_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s3, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1032-DPP-NEXT: .LBB8_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1164-DPP-NEXT: .LBB8_3:
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[0:1], 4.0
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1132-DPP-NEXT: .LBB8_3:
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double 4.0 syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s32, 0
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s9
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s14, s8
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s39, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s38, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s2, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s3, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[2:3]
+; GFX7LESS-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v31, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s12, s6
+; GFX7LESS-NEXT: s_mov_b32 s13, s7
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[2:3], off, s[36:39], 0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], 0
+; GFX7LESS-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX7LESS-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v2
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v1
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v0
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[36:39], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
+; GFX7LESS-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB9_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s38, -1
+; GFX9-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-NEXT: s_add_u32 s36, s36, s9
+; GFX9-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-NEXT: s_mov_b32 s14, s8
+; GFX9-NEXT: s_add_u32 s8, s2, 44
+; GFX9-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-NEXT: s_getpc_b64 s[2:3]
+; GFX9-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-NEXT: s_mov_b32 s12, s6
+; GFX9-NEXT: s_mov_b32 s13, s7
+; GFX9-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT: s_mov_b32 s32, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX9-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-NEXT: s_cbranch_execnz .LBB9_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s38, -1
+; GFX1064-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-NEXT: s_mov_b32 s14, s8
+; GFX1064-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s6
+; GFX1064-NEXT: s_mov_b32 s13, s7
+; GFX1064-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-NEXT: s_mov_b32 s32, 0
+; GFX1064-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX1064-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s38, -1
+; GFX1032-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-NEXT: s_mov_b32 s14, s8
+; GFX1032-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s6
+; GFX1032-NEXT: s_mov_b32 s13, s7
+; GFX1032-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-NEXT: s_mov_b32 s32, 0
+; GFX1032-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX1032-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1032-NEXT: s_mov_b32 s0, 0
+; GFX1032-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s14, s8
+; GFX1164-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s32, 0
+; GFX1164-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-NEXT: global_load_b64 v[2:3], v40, s[34:35]
+; GFX1164-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1164-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-NEXT: global_load_b64 v[2:3], v40, s[34:35]
+; GFX1132-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1132-NEXT: s_mov_b32 s0, 0
+; GFX1132-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s38, -1
+; GFX9-DPP-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX9-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-DPP-NEXT: s_mov_b32 s14, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-DPP-NEXT: s_getpc_b64 s[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b32 s12, s6
+; GFX9-DPP-NEXT: s_mov_b32 s13, s7
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b32 s32, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX9-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1064-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX1064-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1032-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX1032-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1032-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-DPP-NEXT: global_load_b64 v[2:3], v40, s[34:35]
+; GFX1164-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-DPP-NEXT: global_load_b64 v[2:3], v40, s[34:35]
+; GFX1132-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1132-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value()
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %divValue syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s3
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, exec_lo, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, exec_hi, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB10_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_mov_b32 s33, s2
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x9
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], 0
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, s0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, s1
+; GFX7LESS-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[2:3], v[0:1], v[0:1]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s34, 44
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], 4.0
+; GFX7LESS-NEXT: s_addc_u32 s9, s35, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0 offset:8
+; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b32 s12, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s36
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s37
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB10_2
+; GFX7LESS-NEXT: .LBB10_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s42, -1
+; GFX9-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-NEXT: s_add_u32 s40, s40, s3
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB10_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-NEXT: s_mov_b32 s33, s2
+; GFX9-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX9-NEXT: s_add_u32 s8, s34, 44
+; GFX9-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-NEXT: s_mov_b32 s12, s33
+; GFX9-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-NEXT: s_cbranch_execnz .LBB10_2
+; GFX9-NEXT: .LBB10_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s42, -1
+; GFX1064-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB10_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-NEXT: s_mov_b32 s33, s2
+; GFX1064-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-NEXT: s_mov_b32 s12, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1064-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1064-NEXT: .LBB10_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s42, -1
+; GFX1032-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-NEXT: s_mov_b32 s38, 0
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB10_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-NEXT: s_mov_b32 s33, s2
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-NEXT: s_mov_b32 s12, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1032-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1032-NEXT: .LBB10_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-NEXT: s_cbranch_execz .LBB10_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-NEXT: s_mov_b32 s33, s2
+; GFX1164-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b32 s12, s33
+; GFX1164-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1164-NEXT: .LBB10_3:
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s38, 0
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-NEXT: s_cbranch_execz .LBB10_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b32 s12, s33
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX1132-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1132-NEXT: .LBB10_3:
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s42, -1
+; GFX9-DPP-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-DPP-NEXT: s_mov_b32 s33, s2
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX9-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: s_mov_b32 s12, s33
+; GFX9-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX9-DPP-NEXT: .LBB10_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1064-DPP-NEXT: .LBB10_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1032-DPP-NEXT: .LBB10_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1164-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1164-DPP-NEXT: .LBB10_3:
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmax_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX1132-DPP-NEXT: v_max_f64 v[3:4], v[3:4], 4.0
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1132-DPP-NEXT: .LBB10_3:
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double 4.0 monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s50, -1
+; GFX7LESS-NEXT: s_mov_b32 s51, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s48, s48, s9
+; GFX7LESS-NEXT: s_addc_u32 s49, s49, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s8
+; GFX7LESS-NEXT: s_mov_b32 s40, s7
+; GFX7LESS-NEXT: s_mov_b32 s41, s6
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX7LESS-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX7LESS-NEXT: s_load_dwordx2 s[44:45], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s47, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s46, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v40, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[2:3], off, s[44:47], 0
+; GFX7LESS-NEXT: s_mov_b64 s[42:43], 0
+; GFX7LESS-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX7LESS-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s44
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s45
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX7LESS-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7LESS-NEXT: s_or_b64 s[42:43], vcc, s[42:43]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[42:43]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB11_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s50, -1
+; GFX9-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-NEXT: s_add_u32 s48, s48, s9
+; GFX9-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-NEXT: s_mov_b32 s33, s8
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_mov_b32 s40, s7
+; GFX9-NEXT: s_mov_b32 s41, s6
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: v_mov_b32_e32 v41, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX9-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX9-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX9-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-NEXT: s_cbranch_execnz .LBB11_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s50, -1
+; GFX1064-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-NEXT: s_mov_b32 s33, s8
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_mov_b32 s40, s7
+; GFX1064-NEXT: s_mov_b32 s41, s6
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mov_b32_e32 v41, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1064-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1064-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1064-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1064-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s50, -1
+; GFX1032-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-NEXT: s_mov_b32 s33, s8
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_mov_b32 s40, s7
+; GFX1032-NEXT: s_mov_b32 s41, s6
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: v_mov_b32_e32 v41, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1032-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1032-NEXT: s_mov_b32 s44, 0
+; GFX1032-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1032-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1032-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-NEXT: s_mov_b32 s33, s8
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: s_mov_b32 s40, s7
+; GFX1164-NEXT: s_mov_b32 s41, s6
+; GFX1164-NEXT: v_mov_b32_e32 v41, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1164-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1164-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s41
+; GFX1164-NEXT: s_mov_b32 s13, s40
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1164-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-NEXT: s_mov_b32 s40, s14
+; GFX1132-NEXT: s_mov_b32 s41, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1132-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1132-NEXT: s_mov_b32 s44, 0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v40
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s41
+; GFX1132-NEXT: s_mov_b32 s13, s40
+; GFX1132-NEXT: s_mov_b32 s14, s33
+; GFX1132-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1132-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v0, 8 :: v_dual_mov_b32 v1, 0
+; GFX1132-NEXT: v_dual_mov_b32 v2, s42 :: v_dual_mov_b32 v3, s43
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s50, -1
+; GFX9-DPP-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX9-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-DPP-NEXT: s_mov_b32 s33, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_mov_b32 s40, s7
+; GFX9-DPP-NEXT: s_mov_b32 s41, s6
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX9-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX9-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1064-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1064-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1032-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1032-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1032-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1164-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1164-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmax_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-DPP-NEXT: s_mov_b32 s40, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s41, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1132-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[0:1], v[41:42]
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v0, 8 :: v_dual_mov_b32 v1, 0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s42 :: v_dual_mov_b32 v3, s43
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value()
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %divValue monotonic, align 4
+ ret void
+}
+
attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
!llvm.module.flags = !{!0}
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmin.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmin.ll
index 314c52a71d938f..49d415c9eed779 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmin.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmin.ll
@@ -13,6 +13,7 @@
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizer-strategy=DPP -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX1132-DPP %s
declare float @div.float.value()
+declare float @div.double.value()
define amdgpu_kernel void @global_atomic_fmin_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
; GFX7LESS-LABEL: global_atomic_fmin_uni_address_uni_value_agent_scope_unsafe:
@@ -3550,6 +3551,3965 @@ define amdgpu_kernel void @global_atomic_fmin_uni_address_div_value_defalut_scop
ret void
}
+define amdgpu_kernel void @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s3
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, exec_lo, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, exec_hi, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB6_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_mov_b32 s33, s2
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x9
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], 0
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, s0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, s1
+; GFX7LESS-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[2:3], v[0:1], v[0:1]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s34, 44
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_min_f64 v[0:1], v[2:3], 4.0
+; GFX7LESS-NEXT: s_addc_u32 s9, s35, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0 offset:8
+; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b32 s12, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s36
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s37
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB6_2
+; GFX7LESS-NEXT: .LBB6_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s42, -1
+; GFX9-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-NEXT: s_add_u32 s40, s40, s3
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB6_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-NEXT: s_mov_b32 s33, s2
+; GFX9-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX9-NEXT: s_add_u32 s8, s34, 44
+; GFX9-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-NEXT: s_mov_b32 s12, s33
+; GFX9-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-NEXT: s_cbranch_execnz .LBB6_2
+; GFX9-NEXT: .LBB6_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s42, -1
+; GFX1064-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB6_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-NEXT: s_mov_b32 s33, s2
+; GFX1064-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-NEXT: s_mov_b32 s12, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1064-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1064-NEXT: .LBB6_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s42, -1
+; GFX1032-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-NEXT: s_mov_b32 s38, 0
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB6_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-NEXT: s_mov_b32 s33, s2
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-NEXT: s_mov_b32 s12, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1032-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1032-NEXT: .LBB6_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-NEXT: s_cbranch_execz .LBB6_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-NEXT: s_mov_b32 s33, s2
+; GFX1164-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b32 s12, s33
+; GFX1164-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1164-NEXT: .LBB6_3:
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s38, 0
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-NEXT: s_cbranch_execz .LBB6_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b32 s12, s33
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX1132-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1132-NEXT: .LBB6_3:
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s42, -1
+; GFX9-DPP-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-DPP-NEXT: s_mov_b32 s33, s2
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX9-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: s_mov_b32 s12, s33
+; GFX9-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX9-DPP-NEXT: .LBB6_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1064-DPP-NEXT: .LBB6_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1032-DPP-NEXT: .LBB6_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1164-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1164-DPP-NEXT: .LBB6_3:
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB6_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB6_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX1132-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB6_2
+; GFX1132-DPP-NEXT: .LBB6_3:
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double 4.0 syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s50, -1
+; GFX7LESS-NEXT: s_mov_b32 s51, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s48, s48, s9
+; GFX7LESS-NEXT: s_addc_u32 s49, s49, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s8
+; GFX7LESS-NEXT: s_mov_b32 s40, s7
+; GFX7LESS-NEXT: s_mov_b32 s41, s6
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX7LESS-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX7LESS-NEXT: s_load_dwordx2 s[44:45], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s47, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s46, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v40, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[2:3], off, s[44:47], 0
+; GFX7LESS-NEXT: s_mov_b64 s[42:43], 0
+; GFX7LESS-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX7LESS-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s44
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s45
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX7LESS-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7LESS-NEXT: s_or_b64 s[42:43], vcc, s[42:43]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[42:43]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB7_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s50, -1
+; GFX9-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-NEXT: s_add_u32 s48, s48, s9
+; GFX9-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-NEXT: s_mov_b32 s33, s8
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_mov_b32 s40, s7
+; GFX9-NEXT: s_mov_b32 s41, s6
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: v_mov_b32_e32 v41, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX9-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX9-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX9-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-NEXT: s_cbranch_execnz .LBB7_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s50, -1
+; GFX1064-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-NEXT: s_mov_b32 s33, s8
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_mov_b32 s40, s7
+; GFX1064-NEXT: s_mov_b32 s41, s6
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mov_b32_e32 v41, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1064-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1064-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1064-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1064-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s50, -1
+; GFX1032-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-NEXT: s_mov_b32 s33, s8
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_mov_b32 s40, s7
+; GFX1032-NEXT: s_mov_b32 s41, s6
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: v_mov_b32_e32 v41, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1032-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1032-NEXT: s_mov_b32 s44, 0
+; GFX1032-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1032-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1032-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-NEXT: s_mov_b32 s33, s8
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: s_mov_b32 s40, s7
+; GFX1164-NEXT: s_mov_b32 s41, s6
+; GFX1164-NEXT: v_mov_b32_e32 v41, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1164-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1164-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s41
+; GFX1164-NEXT: s_mov_b32 s13, s40
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1164-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-NEXT: s_mov_b32 s40, s14
+; GFX1132-NEXT: s_mov_b32 s41, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1132-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1132-NEXT: s_mov_b32 s44, 0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v40
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s41
+; GFX1132-NEXT: s_mov_b32 s13, s40
+; GFX1132-NEXT: s_mov_b32 s14, s33
+; GFX1132-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1132-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v0, 8 :: v_dual_mov_b32 v1, 0
+; GFX1132-NEXT: v_dual_mov_b32 v2, s42 :: v_dual_mov_b32 v3, s43
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s50, -1
+; GFX9-DPP-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX9-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-DPP-NEXT: s_mov_b32 s33, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_mov_b32 s40, s7
+; GFX9-DPP-NEXT: s_mov_b32 s41, s6
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX9-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX9-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1064-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1064-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1032-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1032-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1032-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1164-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1164-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-DPP-NEXT: s_mov_b32 s40, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s41, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1132-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB7_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v0, 8 :: v_dual_mov_b32 v1, 0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s42 :: v_dual_mov_b32 v3, s43
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB7_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value()
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %divValue syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, exec_lo, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, exec_hi, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB8_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], 0
+; GFX7LESS-NEXT: s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s6
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s7
+; GFX7LESS-NEXT: s_mov_b32 s2, -1
+; GFX7LESS-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX7LESS-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v2
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, v1
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, v0
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[4:7], off, s[0:3], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[2:3]
+; GFX7LESS-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, v4
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, v5
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[4:5]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB8_2
+; GFX7LESS-NEXT: .LBB8_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB8_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT: s_cbranch_execnz .LBB8_2
+; GFX9-NEXT: .LBB8_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB8_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1064-NEXT: .LBB8_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-NEXT: s_mov_b32 s2, 0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_and_saveexec_b32 s3, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB8_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1032-NEXT: .LBB8_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-NEXT: s_cbranch_execz .LBB8_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1164-NEXT: .LBB8_3:
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-NEXT: s_mov_b32 s2, 0
+; GFX1132-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-NEXT: s_cbranch_execz .LBB8_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1132-NEXT: .LBB8_3:
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-DPP-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX9-DPP-NEXT: .LBB8_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-DPP-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1064-DPP-NEXT: .LBB8_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s3, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-DPP-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1032-DPP-NEXT: .LBB8_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1164-DPP-NEXT: .LBB8_3:
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_one_as_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB8_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-DPP-NEXT: .LBB8_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-DPP-NEXT: v_min_f64 v[0:1], v[0:1], 4.0
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v4, v[0:3], s[0:1] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB8_2
+; GFX1132-DPP-NEXT: .LBB8_3:
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double 4.0 syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s32, 0
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s9
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s14, s8
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s39, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s38, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s2, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s3, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[2:3]
+; GFX7LESS-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v31, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s12, s6
+; GFX7LESS-NEXT: s_mov_b32 s13, s7
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[2:3], off, s[36:39], 0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], 0
+; GFX7LESS-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX7LESS-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX7LESS-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v2
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v1
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v0
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[36:39], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
+; GFX7LESS-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB9_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s38, -1
+; GFX9-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-NEXT: s_add_u32 s36, s36, s9
+; GFX9-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-NEXT: s_mov_b32 s14, s8
+; GFX9-NEXT: s_add_u32 s8, s2, 44
+; GFX9-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-NEXT: s_getpc_b64 s[2:3]
+; GFX9-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-NEXT: s_mov_b32 s12, s6
+; GFX9-NEXT: s_mov_b32 s13, s7
+; GFX9-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT: s_mov_b32 s32, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX9-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-NEXT: s_cbranch_execnz .LBB9_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s38, -1
+; GFX1064-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-NEXT: s_mov_b32 s14, s8
+; GFX1064-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s6
+; GFX1064-NEXT: s_mov_b32 s13, s7
+; GFX1064-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-NEXT: s_mov_b32 s32, 0
+; GFX1064-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX1064-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s38, -1
+; GFX1032-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-NEXT: s_mov_b32 s14, s8
+; GFX1032-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s6
+; GFX1032-NEXT: s_mov_b32 s13, s7
+; GFX1032-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-NEXT: s_mov_b32 s32, 0
+; GFX1032-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX1032-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1032-NEXT: s_mov_b32 s0, 0
+; GFX1032-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s14, s8
+; GFX1164-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s32, 0
+; GFX1164-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-NEXT: global_load_b64 v[2:3], v40, s[34:35]
+; GFX1164-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1164-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-NEXT: global_load_b64 v[2:3], v40, s[34:35]
+; GFX1132-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1132-NEXT: s_mov_b32 s0, 0
+; GFX1132-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s38, -1
+; GFX9-DPP-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX9-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-DPP-NEXT: s_mov_b32 s14, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-DPP-NEXT: s_getpc_b64 s[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b32 s12, s6
+; GFX9-DPP-NEXT: s_mov_b32 s13, s7
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b32 s32, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX9-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1064-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX1064-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1032-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[2:3], v40, s[34:35]
+; GFX1032-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1032-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-DPP-NEXT: global_load_b64 v[2:3], v40, s[34:35]
+; GFX1164-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_one_as_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-DPP-NEXT: global_load_b64 v[2:3], v40, s[34:35]
+; GFX1132-DPP-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1132-DPP-NEXT: .LBB9_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[4:5]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v40, v[0:3], s[34:35] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB9_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value()
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %divValue syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s3
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, exec_lo, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, exec_hi, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB10_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_mov_b32 s33, s2
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x9
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], 0
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, s0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, s1
+; GFX7LESS-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[2:3], v[0:1], v[0:1]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s34, 44
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_min_f64 v[0:1], v[2:3], 4.0
+; GFX7LESS-NEXT: s_addc_u32 s9, s35, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0 offset:8
+; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b32 s12, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s36
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s37
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB10_2
+; GFX7LESS-NEXT: .LBB10_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s42, -1
+; GFX9-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-NEXT: s_add_u32 s40, s40, s3
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB10_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-NEXT: s_mov_b32 s33, s2
+; GFX9-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX9-NEXT: s_add_u32 s8, s34, 44
+; GFX9-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-NEXT: s_mov_b32 s12, s33
+; GFX9-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-NEXT: s_cbranch_execnz .LBB10_2
+; GFX9-NEXT: .LBB10_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s42, -1
+; GFX1064-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB10_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-NEXT: s_mov_b32 s33, s2
+; GFX1064-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-NEXT: s_mov_b32 s12, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1064-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1064-NEXT: .LBB10_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s42, -1
+; GFX1032-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-NEXT: s_mov_b32 s38, 0
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB10_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-NEXT: s_mov_b32 s33, s2
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-NEXT: s_mov_b32 s12, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1032-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1032-NEXT: .LBB10_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-NEXT: s_cbranch_execz .LBB10_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-NEXT: s_mov_b32 s33, s2
+; GFX1164-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b32 s12, s33
+; GFX1164-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1164-NEXT: .LBB10_3:
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s38, 0
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-NEXT: s_cbranch_execz .LBB10_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b32 s12, s33
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX1132-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1132-NEXT: .LBB10_3:
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s42, -1
+; GFX9-DPP-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX9-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-DPP-NEXT: s_mov_b32 s33, s2
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX9-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: s_mov_b32 s12, s33
+; GFX9-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX9-DPP-NEXT: .LBB10_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1064-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1064-DPP-NEXT: .LBB10_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1032-DPP-NEXT: .LBB10_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1164-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1164-DPP-NEXT: .LBB10_3:
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmin_double_uni_address_uni_value_defalut_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB10_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB10_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_max_f64 v[3:4], v[1:2], v[1:2]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_4)
+; GFX1132-DPP-NEXT: v_min_f64 v[3:4], v[3:4], 4.0
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB10_2
+; GFX1132-DPP-NEXT: .LBB10_3:
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double 4.0 monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s50, -1
+; GFX7LESS-NEXT: s_mov_b32 s51, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s48, s48, s9
+; GFX7LESS-NEXT: s_addc_u32 s49, s49, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s8
+; GFX7LESS-NEXT: s_mov_b32 s40, s7
+; GFX7LESS-NEXT: s_mov_b32 s41, s6
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX7LESS-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX7LESS-NEXT: s_load_dwordx2 s[44:45], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s47, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s46, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v40, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[2:3], off, s[44:47], 0
+; GFX7LESS-NEXT: s_mov_b64 s[42:43], 0
+; GFX7LESS-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX7LESS-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s44
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s45
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX7LESS-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7LESS-NEXT: s_or_b64 s[42:43], vcc, s[42:43]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[42:43]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB11_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s50, -1
+; GFX9-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-NEXT: s_add_u32 s48, s48, s9
+; GFX9-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-NEXT: s_mov_b32 s33, s8
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_mov_b32 s40, s7
+; GFX9-NEXT: s_mov_b32 s41, s6
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: v_mov_b32_e32 v41, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX9-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX9-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX9-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-NEXT: s_cbranch_execnz .LBB11_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s50, -1
+; GFX1064-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-NEXT: s_mov_b32 s33, s8
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_mov_b32 s40, s7
+; GFX1064-NEXT: s_mov_b32 s41, s6
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mov_b32_e32 v41, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1064-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1064-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1064-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1064-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s50, -1
+; GFX1032-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-NEXT: s_mov_b32 s33, s8
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_mov_b32 s40, s7
+; GFX1032-NEXT: s_mov_b32 s41, s6
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: v_mov_b32_e32 v41, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1032-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1032-NEXT: s_mov_b32 s44, 0
+; GFX1032-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1032-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1032-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-NEXT: s_mov_b32 s33, s8
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: s_mov_b32 s40, s7
+; GFX1164-NEXT: s_mov_b32 s41, s6
+; GFX1164-NEXT: v_mov_b32_e32 v41, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1164-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1164-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s41
+; GFX1164-NEXT: s_mov_b32 s13, s40
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1164-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-NEXT: s_mov_b32 s40, s14
+; GFX1132-NEXT: s_mov_b32 s41, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1132-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1132-NEXT: s_mov_b32 s44, 0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v40
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s41
+; GFX1132-NEXT: s_mov_b32 s13, s40
+; GFX1132-NEXT: s_mov_b32 s14, s33
+; GFX1132-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1132-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v0, 8 :: v_dual_mov_b32 v1, 0
+; GFX1132-NEXT: v_dual_mov_b32 v2, s42 :: v_dual_mov_b32 v3, s43
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s50, -1
+; GFX9-DPP-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX9-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-DPP-NEXT: s_mov_b32 s33, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_mov_b32 s40, s7
+; GFX9-DPP-NEXT: s_mov_b32 s41, s6
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX9-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX9-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1064-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1064-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1032-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: v_or3_b32 v40, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[2:3], v41, s[42:43]
+; GFX1032-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1032-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v0, off, s[48:51], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v3, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v41, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1164-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1164-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fmin_double_uni_address_div_value_defalut_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-DPP-NEXT: s_mov_b32 s40, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s41, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: global_load_b64 v[2:3], v41, s[42:43]
+; GFX1132-DPP-NEXT: v_max_f64 v[41:42], v[0:1], v[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB11_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT: v_min_f64 v[0:1], v[0:1], v[41:42]
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[2:3], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[0:1], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v0, 8 :: v_dual_mov_b32 v1, 0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s42 :: v_dual_mov_b32 v3, s43
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[2:3], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB11_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value()
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %divValue monotonic, align 4
+ ret void
+}
+
attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
!llvm.module.flags = !{!0}
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fsub.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fsub.ll
index bc9125e326c4d9..7a7ddbe618b0bc 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fsub.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fsub.ll
@@ -13,6 +13,7 @@
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizer-strategy=DPP -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX1132-DPP %s
declare float @div.float.value()
+declare double @div.double.value()
define amdgpu_kernel void @global_atomic_fsub_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
; GFX7LESS-LABEL: global_atomic_fsub_uni_address_uni_value_agent_scope_unsafe:
@@ -5616,6 +5617,5581 @@ define amdgpu_kernel void @global_atomic_fsub_uni_address_div_value_defalut_scop
ret void
}
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s3
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s2
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], exec
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s0, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s1, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB9_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x9
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[2:3], s[36:37], 0x0
+; GFX7LESS-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX7LESS-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX7LESS-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], 0
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, s2
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, s3
+; GFX7LESS-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[0:1], -v[41:42]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s34, 44
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:8
+; GFX7LESS-NEXT: s_addc_u32 s9, s35, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(2)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b32 s12, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s36
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s37
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB9_2
+; GFX7LESS-NEXT: .LBB9_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s42, -1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], exec
+; GFX9-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 0
+; GFX9-NEXT: s_add_u32 s40, s40, s3
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v0
+; GFX9-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-NEXT: s_mov_b32 s33, s2
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB9_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX9-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX9-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX9-NEXT: s_add_u32 s8, s34, 44
+; GFX9-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-NEXT: s_mov_b32 s12, s33
+; GFX9-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-NEXT: s_cbranch_execnz .LBB9_2
+; GFX9-NEXT: .LBB9_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s42, -1
+; GFX1064-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-NEXT: s_mov_b32 s33, s2
+; GFX1064-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB9_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1064-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1064-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-NEXT: s_mov_b32 s12, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1064-NEXT: .LBB9_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s33, s2
+; GFX1032-NEXT: s_mov_b32 s2, exec_lo
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1032-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s42, -1
+; GFX1032-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-NEXT: s_mov_b32 s38, 0
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB9_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1032-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-NEXT: s_mov_b32 s12, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1032-NEXT: .LBB9_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s33, s2
+; GFX1164-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-NEXT: s_cbranch_execz .LBB9_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1164-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1164-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b32 s12, s33
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1164-NEXT: .LBB9_3:
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b32 s2, exec_lo
+; GFX1132-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s38, 0
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-NEXT: s_cbranch_execz .LBB9_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1132-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b32 s12, s33
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1132-NEXT: .LBB9_3:
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s42, -1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX9-DPP-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 0
+; GFX9-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v0
+; GFX9-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-DPP-NEXT: s_mov_b32 s33, s2
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX9-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX9-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: s_mov_b32 s12, s33
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX9-DPP-NEXT: .LBB9_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1064-DPP-NEXT: .LBB9_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1032-DPP-NEXT: s_mov_b32 s2, exec_lo
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1032-DPP-NEXT: .LBB9_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1164-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1164-DPP-NEXT: .LBB9_3:
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b32 s2, exec_lo
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB9_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1132-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-DPP-NEXT: v_cvt_f64_u32_e32 v[0:1], s0
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_mul_f64 v[41:42], v[0:1], 4.0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB9_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB9_2
+; GFX1132-DPP-NEXT: .LBB9_3:
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double 4.0 syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s50, -1
+; GFX7LESS-NEXT: s_mov_b32 s51, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s48, s48, s9
+; GFX7LESS-NEXT: s_addc_u32 s49, s49, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s8
+; GFX7LESS-NEXT: s_mov_b32 s40, s7
+; GFX7LESS-NEXT: s_mov_b32 s41, s6
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX7LESS-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX7LESS-NEXT: s_load_dwordx2 s[44:45], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s47, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s46, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v42, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v42
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: v_mov_b32_e32 v41, v1
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[0:1], off, s[44:47], 0
+; GFX7LESS-NEXT: s_mov_b64 s[42:43], 0
+; GFX7LESS-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[0:1], -v[40:41]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[48:51], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:8
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(2)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v42
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s44
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s45
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[48:51], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[42:43], vcc, s[42:43]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[42:43]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB10_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s50, -1
+; GFX9-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-NEXT: s_add_u32 s48, s48, s9
+; GFX9-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-NEXT: s_mov_b32 s33, s8
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_mov_b32 s40, s7
+; GFX9-NEXT: s_mov_b32 s41, s6
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: v_mov_b32_e32 v43, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: v_mov_b32_e32 v41, v1
+; GFX9-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-NEXT: s_cbranch_execnz .LBB10_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s50, -1
+; GFX1064-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-NEXT: s_mov_b32 s33, s8
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_mov_b32 s40, s7
+; GFX1064-NEXT: s_mov_b32 s41, s6
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mov_b32_e32 v43, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: v_mov_b32_e32 v41, v1
+; GFX1064-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s50, -1
+; GFX1032-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-NEXT: s_mov_b32 s33, s8
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_mov_b32 s40, s7
+; GFX1032-NEXT: s_mov_b32 s41, s6
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: v_mov_b32_e32 v43, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: v_mov_b32_e32 v41, v1
+; GFX1032-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: s_mov_b32 s44, 0
+; GFX1032-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-NEXT: s_mov_b32 s33, s8
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: v_mov_b32_e32 v42, v0
+; GFX1164-NEXT: s_mov_b32 s40, s7
+; GFX1164-NEXT: s_mov_b32 s41, s6
+; GFX1164-NEXT: v_mov_b32_e32 v43, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: v_mov_b32_e32 v41, v1
+; GFX1164-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v42
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s41
+; GFX1164-NEXT: s_mov_b32 s13, s40
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-NEXT: s_mov_b32 s40, s14
+; GFX1132-NEXT: s_mov_b32 s41, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: v_dual_mov_b32 v42, v0 :: v_dual_mov_b32 v43, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, v1
+; GFX1132-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1132-NEXT: s_mov_b32 s44, 0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v42 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s41
+; GFX1132-NEXT: s_mov_b32 s13, s40
+; GFX1132-NEXT: s_mov_b32 s14, s33
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s42
+; GFX1132-NEXT: v_dual_mov_b32 v3, s43 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s50, -1
+; GFX9-DPP-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX9-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-DPP-NEXT: s_mov_b32 s33, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_mov_b32 s40, s7
+; GFX9-DPP-NEXT: s_mov_b32 s41, s6
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX9-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1064-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1032-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1032-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v42, v0
+; GFX1164-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1164-DPP-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_align4_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-DPP-NEXT: s_mov_b32 s40, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s41, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v42, v0 :: v_dual_mov_b32 v43, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, v1
+; GFX1132-DPP-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1132-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB10_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v42 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s42
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s43 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB10_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.float.value()
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %divValue syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp(ptr addrspace(1) %ptr) #1 {
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s12, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s14, -1
+; GFX7LESS-NEXT: s_mov_b32 s15, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s12, s12, s3
+; GFX7LESS-NEXT: s_addc_u32 s13, s13, 0
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], exec
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s2, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s3, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB11_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7LESS-NEXT: s_bcnt1_i32_b64 s6, s[2:3]
+; GFX7LESS-NEXT: s_mov_b32 s7, 0x43300000
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], 0
+; GFX7LESS-NEXT: s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT: v_add_f64 v[0:1], s[6:7], v[0:1]
+; GFX7LESS-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s8
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s9
+; GFX7LESS-NEXT: s_mov_b32 s2, -1
+; GFX7LESS-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v2
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v1
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v0
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[0:3], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
+; GFX7LESS-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[4:5]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB11_2
+; GFX7LESS-NEXT: .LBB11_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s10, -1
+; GFX9-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9-NEXT: s_add_u32 s8, s8, s3
+; GFX9-NEXT: s_mov_b64 s[2:3], exec
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX9-NEXT: s_addc_u32 s9, s9, 0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB11_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-NEXT: s_mov_b32 s3, 0x43300000
+; GFX9-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT: s_cbranch_execnz .LBB11_2
+; GFX9-NEXT: .LBB11_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s10, -1
+; GFX1064-NEXT: s_mov_b32 s11, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s8, s8, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-NEXT: s_addc_u32 s9, s9, 0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB11_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-NEXT: s_mov_b32 s3, 0x43300000
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-NEXT: v_add_f64 v[0:1], 0xc3300000, s[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1064-NEXT: .LBB11_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s10, -1
+; GFX1032-NEXT: s_mov_b32 s11, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s8, s8, s3
+; GFX1032-NEXT: s_mov_b32 s3, exec_lo
+; GFX1032-NEXT: s_addc_u32 s9, s9, 0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
+; GFX1032-NEXT: s_mov_b32 s2, 0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB11_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_bcnt1_i32_b32 s4, s3
+; GFX1032-NEXT: s_mov_b32 s5, 0x43300000
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-NEXT: v_add_f64 v[0:1], 0xc3300000, s[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1032-NEXT: .LBB11_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_bcnt1_i32_b64 s2, exec
+; GFX1164-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-NEXT: v_mov_b32_e32 v1, s2
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1164-NEXT: scratch_store_b32 off, v1, off
+; GFX1164-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-NEXT: s_cbranch_execz .LBB11_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1164-NEXT: .LBB11_3:
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_bcnt1_i32_b32 s2, exec_lo
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: v_dual_mov_b32 v0, 0x43300000 :: v_dual_mov_b32 v1, s2
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-NEXT: s_mov_b32 s2, 0
+; GFX1132-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1132-NEXT: scratch_store_b32 off, v1, off
+; GFX1132-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-NEXT: s_cbranch_execz .LBB11_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-NEXT: v_mov_b32_e32 v6, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1132-NEXT: .LBB11_3:
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s10, -1
+; GFX9-DPP-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX9-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-DPP-NEXT: s_mov_b32 s3, 0x43300000
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX9-DPP-NEXT: .LBB11_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s10, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s11, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-DPP-NEXT: s_mov_b32 s3, 0x43300000
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1064-DPP-NEXT: .LBB11_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s10, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s11, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX1032-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_bcnt1_i32_b32 s4, s3
+; GFX1032-DPP-NEXT: s_mov_b32 s5, 0x43300000
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1032-DPP-NEXT: .LBB11_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_bcnt1_i32_b64 s2, exec
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s2
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v1, off
+; GFX1164-DPP-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1164-DPP-NEXT: .LBB11_3:
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_one_as_scope_unsafe_structfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_bcnt1_i32_b32 s2, exec_lo
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v0, 0x43300000 :: v_dual_mov_b32 v1, s2
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v1, off
+; GFX1132-DPP-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB11_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-DPP-NEXT: .LBB11_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB11_2
+; GFX1132-DPP-NEXT: .LBB11_3:
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double 4.0 syncscope("one-as") monotonic
+ ret void
+}
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp(ptr addrspace(1) %ptr) #1 {
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s32, 0
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s9
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s14, s8
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s39, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s38, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s2, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s3, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[2:3]
+; GFX7LESS-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v31, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s12, s6
+; GFX7LESS-NEXT: s_mov_b32 s13, s7
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[4:5], off, s[36:39], 0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], 0
+; GFX7LESS-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v5
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v4
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v2
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[36:39], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[4:5]
+; GFX7LESS-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB12_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s38, -1
+; GFX9-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-NEXT: s_add_u32 s36, s36, s9
+; GFX9-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-NEXT: s_mov_b32 s14, s8
+; GFX9-NEXT: s_add_u32 s8, s2, 44
+; GFX9-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-NEXT: s_getpc_b64 s[2:3]
+; GFX9-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-NEXT: s_mov_b32 s12, s6
+; GFX9-NEXT: s_mov_b32 s13, s7
+; GFX9-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT: s_mov_b32 s32, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-NEXT: s_cbranch_execnz .LBB12_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s38, -1
+; GFX1064-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-NEXT: s_mov_b32 s14, s8
+; GFX1064-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s6
+; GFX1064-NEXT: s_mov_b32 s13, s7
+; GFX1064-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-NEXT: s_mov_b32 s32, 0
+; GFX1064-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s38, -1
+; GFX1032-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-NEXT: s_mov_b32 s14, s8
+; GFX1032-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s6
+; GFX1032-NEXT: s_mov_b32 s13, s7
+; GFX1032-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-NEXT: s_mov_b32 s32, 0
+; GFX1032-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-NEXT: s_mov_b32 s0, 0
+; GFX1032-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s14, s8
+; GFX1164-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s32, 0
+; GFX1164-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-NEXT: s_mov_b32 s0, 0
+; GFX1132-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s38, -1
+; GFX9-DPP-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX9-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-DPP-NEXT: s_mov_b32 s14, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-DPP-NEXT: s_getpc_b64 s[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b32 s12, s6
+; GFX9-DPP-NEXT: s_mov_b32 s13, s7
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b32 s32, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1064-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1032-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1032-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_one_as_scope_unsafe_structfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1132-DPP-NEXT: .LBB12_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB12_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value() strictfp
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %divValue syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(ptr addrspace(1) %ptr) #2{
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s12, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s14, -1
+; GFX7LESS-NEXT: s_mov_b32 s15, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s12, s12, s3
+; GFX7LESS-NEXT: s_addc_u32 s13, s13, 0
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], exec
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s2, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s3, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB13_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7LESS-NEXT: s_bcnt1_i32_b64 s6, s[2:3]
+; GFX7LESS-NEXT: s_mov_b32 s7, 0x43300000
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], 0
+; GFX7LESS-NEXT: s_mov_b32 s3, 0xf000
+; GFX7LESS-NEXT: v_add_f64 v[0:1], s[6:7], v[0:1]
+; GFX7LESS-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s8
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s9
+; GFX7LESS-NEXT: s_mov_b32 s2, -1
+; GFX7LESS-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v2
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v1
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v0
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[0:3], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[2:3]
+; GFX7LESS-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[4:5]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB13_2
+; GFX7LESS-NEXT: .LBB13_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s10, -1
+; GFX9-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9-NEXT: s_add_u32 s8, s8, s3
+; GFX9-NEXT: s_mov_b64 s[2:3], exec
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX9-NEXT: s_addc_u32 s9, s9, 0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB13_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-NEXT: s_mov_b32 s3, 0x43300000
+; GFX9-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-NEXT: s_cbranch_execnz .LBB13_2
+; GFX9-NEXT: .LBB13_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s10, -1
+; GFX1064-NEXT: s_mov_b32 s11, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s8, s8, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-NEXT: s_addc_u32 s9, s9, 0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB13_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-NEXT: s_mov_b32 s3, 0x43300000
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-NEXT: v_add_f64 v[0:1], 0xc3300000, s[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1064-NEXT: .LBB13_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s10, -1
+; GFX1032-NEXT: s_mov_b32 s11, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s8, s8, s3
+; GFX1032-NEXT: s_mov_b32 s3, exec_lo
+; GFX1032-NEXT: s_addc_u32 s9, s9, 0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
+; GFX1032-NEXT: s_mov_b32 s2, 0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB13_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_bcnt1_i32_b32 s4, s3
+; GFX1032-NEXT: s_mov_b32 s5, 0x43300000
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-NEXT: v_add_f64 v[0:1], 0xc3300000, s[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1032-NEXT: .LBB13_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_bcnt1_i32_b64 s2, exec
+; GFX1164-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-NEXT: v_mov_b32_e32 v1, s2
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1164-NEXT: scratch_store_b32 off, v1, off
+; GFX1164-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-NEXT: s_cbranch_execz .LBB13_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1164-NEXT: .LBB13_3:
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_bcnt1_i32_b32 s2, exec_lo
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: v_dual_mov_b32 v0, 0x43300000 :: v_dual_mov_b32 v1, s2
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-NEXT: s_mov_b32 s2, 0
+; GFX1132-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1132-NEXT: scratch_store_b32 off, v1, off
+; GFX1132-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-NEXT: s_cbranch_execz .LBB13_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-NEXT: v_mov_b32_e32 v6, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1132-NEXT: .LBB13_3:
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s10, -1
+; GFX9-DPP-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX9-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-DPP-NEXT: s_mov_b32 s3, 0x43300000
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX9-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX9-DPP-NEXT: .LBB13_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s10, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s11, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-DPP-NEXT: s_mov_b32 s3, 0x43300000
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1064-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[2:3]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1064-DPP-NEXT: .LBB13_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s10, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s11, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s8, s8, s3
+; GFX1032-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s9, 0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s4, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_bcnt1_i32_b32 s4, s3
+; GFX1032-DPP-NEXT: s_mov_b32 s5, 0x43300000
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s4
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s5
+; GFX1032-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1032-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s2
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1032-DPP-NEXT: .LBB13_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_bcnt1_i32_b64 s2, exec
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s2
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v1, off
+; GFX1164-DPP-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[2:3], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s2
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s3
+; GFX1164-DPP-NEXT: s_mov_b64 s[2:3], 0
+; GFX1164-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, v1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[2:3], vcc, s[2:3]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[2:3]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1164-DPP-NEXT: .LBB13_3:
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_bcnt1_i32_b32 s2, exec_lo
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v0, 0x43300000 :: v_dual_mov_b32 v1, s2
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s2, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s3, exec_lo
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v0, off offset:4
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v1, off
+; GFX1132-DPP-NEXT: scratch_load_b64 v[0:1], off, off
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB13_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[4:5], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT: v_mul_f64 v[4:5], 4.0, v[0:1]
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s4 :: v_dual_mov_b32 v3, s5
+; GFX1132-DPP-NEXT: .LBB13_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], v[2:3], -v[4:5]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[0:1], v6, v[0:3], s[0:1] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[2:3]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0
+; GFX1132-DPP-NEXT: s_or_b32 s2, vcc_lo, s2
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s2
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB13_2
+; GFX1132-DPP-NEXT: .LBB13_3:
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double 4.0 syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe(ptr addrspace(1) %ptr) #0 {
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s32, 0
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s9
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s14, s8
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s39, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s38, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s2, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s3, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[2:3]
+; GFX7LESS-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v31, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s12, s6
+; GFX7LESS-NEXT: s_mov_b32 s13, s7
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[4:5], off, s[36:39], 0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], 0
+; GFX7LESS-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v5
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v4
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v2
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[36:39], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[4:5]
+; GFX7LESS-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB14_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s38, -1
+; GFX9-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-NEXT: s_add_u32 s36, s36, s9
+; GFX9-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-NEXT: s_mov_b32 s14, s8
+; GFX9-NEXT: s_add_u32 s8, s2, 44
+; GFX9-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-NEXT: s_getpc_b64 s[2:3]
+; GFX9-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-NEXT: s_mov_b32 s12, s6
+; GFX9-NEXT: s_mov_b32 s13, s7
+; GFX9-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT: s_mov_b32 s32, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-NEXT: s_cbranch_execnz .LBB14_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s38, -1
+; GFX1064-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-NEXT: s_mov_b32 s14, s8
+; GFX1064-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s6
+; GFX1064-NEXT: s_mov_b32 s13, s7
+; GFX1064-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-NEXT: s_mov_b32 s32, 0
+; GFX1064-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s38, -1
+; GFX1032-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-NEXT: s_mov_b32 s14, s8
+; GFX1032-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s6
+; GFX1032-NEXT: s_mov_b32 s13, s7
+; GFX1032-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-NEXT: s_mov_b32 s32, 0
+; GFX1032-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-NEXT: s_mov_b32 s0, 0
+; GFX1032-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s14, s8
+; GFX1164-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s32, 0
+; GFX1164-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-NEXT: s_mov_b32 s0, 0
+; GFX1132-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s38, -1
+; GFX9-DPP-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX9-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-DPP-NEXT: s_mov_b32 s14, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-DPP-NEXT: s_getpc_b64 s[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s2, s2, div.double.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s3, s3, div.double.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b32 s12, s6
+; GFX9-DPP-NEXT: s_mov_b32 s13, s7
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b32 s32, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1064-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1032-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1032-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-DPP-NEXT: s_add_u32 s4, s4, div.double.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s5, s5, div.double.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1132-DPP-NEXT: .LBB14_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB14_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.double.value()
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %divValue syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp(ptr addrspace(1) %ptr) #1 {
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_mov_b32 s32, 0
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s9
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s14, s8
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s39, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s38, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s2, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s3, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[2:3]
+; GFX7LESS-NEXT: s_add_u32 s2, s2, div.float.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s3, s3, div.float.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v31, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s12, s6
+; GFX7LESS-NEXT: s_mov_b32 s13, s7
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[4:5], off, s[36:39], 0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], 0
+; GFX7LESS-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v9, v5
+; GFX7LESS-NEXT: v_mov_b32_e32 v8, v4
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, v3
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, v2
+; GFX7LESS-NEXT: buffer_atomic_cmpswap_x2 v[6:9], off, s[36:39], 0 glc
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_cmp_eq_u64_e32 vcc, v[6:7], v[4:5]
+; GFX7LESS-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, v6
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, v7
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB15_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s38, -1
+; GFX9-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-NEXT: s_add_u32 s36, s36, s9
+; GFX9-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-NEXT: s_mov_b32 s14, s8
+; GFX9-NEXT: s_add_u32 s8, s2, 44
+; GFX9-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-NEXT: s_getpc_b64 s[2:3]
+; GFX9-NEXT: s_add_u32 s2, s2, div.float.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s3, s3, div.float.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-NEXT: s_mov_b32 s12, s6
+; GFX9-NEXT: s_mov_b32 s13, s7
+; GFX9-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT: s_mov_b32 s32, 0
+; GFX9-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX9-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-NEXT: s_cbranch_execnz .LBB15_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s38, -1
+; GFX1064-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-NEXT: s_mov_b32 s14, s8
+; GFX1064-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s6
+; GFX1064-NEXT: s_mov_b32 s13, s7
+; GFX1064-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-NEXT: s_mov_b32 s32, 0
+; GFX1064-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1064-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s38, -1
+; GFX1032-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-NEXT: s_mov_b32 s14, s8
+; GFX1032-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s6
+; GFX1032-NEXT: s_mov_b32 s13, s7
+; GFX1032-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-NEXT: s_mov_b32 s32, 0
+; GFX1032-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-NEXT: s_mov_b32 s0, 0
+; GFX1032-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1032-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b32 s14, s8
+; GFX1164-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s32, 0
+; GFX1164-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1164-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1132-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-NEXT: s_mov_b32 s0, 0
+; GFX1132-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1132-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s38, -1
+; GFX9-DPP-NEXT: s_mov_b32 s39, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX9-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX9-DPP-NEXT: s_mov_b32 s14, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX9-DPP-NEXT: s_getpc_b64 s[2:3]
+; GFX9-DPP-NEXT: s_add_u32 s2, s2, div.float.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s3, s3, div.float.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[16:17], s[2:3], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX9-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b32 s12, s6
+; GFX9-DPP-NEXT: s_mov_b32 s13, s7
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b32 s32, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX9-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX9-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s39, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1064-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1064-DPP-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1064-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1064-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1064-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1064-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s38, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s39, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s36, s36, s9
+; GFX1032-DPP-NEXT: s_addc_u32 s37, s37, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1032-DPP-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[34:35], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1032-DPP-NEXT: v_or3_b32 v31, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[4:5], v40, s[34:35]
+; GFX1032-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1032-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1032-DPP-NEXT: global_atomic_cmpswap_x2 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1032-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s0
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1164-DPP-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[16:17], s[4:5], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX1164-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], 0
+; GFX1164-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1164-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, v3
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, v2
+; GFX1164-DPP-NEXT: s_or_b64 s[0:1], vcc, s[0:1]
+; GFX1164-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[0:1]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_agent_scope_unsafe_structfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_add_u32 s8, s2, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s3, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX1132-DPP-NEXT: s_getpc_b64 s[4:5]
+; GFX1132-DPP-NEXT: s_add_u32 s4, s4, div.float.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s5, s5, div.float.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[34:35], s[2:3], 0x24
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, 0 :: v_dual_mov_b32 v31, v0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1132-DPP-NEXT: global_load_b64 v[4:5], v40, s[34:35]
+; GFX1132-DPP-NEXT: s_mov_b32 s0, 0
+; GFX1132-DPP-NEXT: .LBB15_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[2:3], v[4:5], -v[0:1]
+; GFX1132-DPP-NEXT: global_atomic_cmpswap_b64 v[2:3], v40, v[2:5], s[34:35] glc
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, v3 :: v_dual_mov_b32 v4, v2
+; GFX1132-DPP-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB15_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.float.value() strictfp
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %divValue syncscope("agent") monotonic
+ ret void
+}
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp(ptr addrspace(1) %ptr) #2 {
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s42, -1
+; GFX7LESS-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s40, s40, s3
+; GFX7LESS-NEXT: s_addc_u32 s41, s41, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s2
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], exec
+; GFX7LESS-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s0, 0
+; GFX7LESS-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s1, v0
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7LESS-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX7LESS-NEXT: s_cbranch_execz .LBB16_3
+; GFX7LESS-NEXT: ; %bb.1:
+; GFX7LESS-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x9
+; GFX7LESS-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX7LESS-NEXT: s_mov_b32 s1, 0x43300000
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_load_dwordx2 s[2:3], s[36:37], 0x0
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX7LESS-NEXT: v_add_f64 v[0:1], s[0:1], v[0:1]
+; GFX7LESS-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], 0
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, s2
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, s3
+; GFX7LESS-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[0:1], -v[41:42]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s34, 44
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:8
+; GFX7LESS-NEXT: s_addc_u32 s9, s35, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(2)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b32 s12, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v40
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s36
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s37
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[40:43], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[40:43], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB16_2
+; GFX7LESS-NEXT: .LBB16_3:
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s42, -1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-NEXT: s_mov_b64 s[0:1], exec
+; GFX9-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 0
+; GFX9-NEXT: s_add_u32 s40, s40, s3
+; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v0
+; GFX9-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-NEXT: s_mov_b32 s33, s2
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-NEXT: s_cbranch_execz .LBB16_3
+; GFX9-NEXT: ; %bb.1:
+; GFX9-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX9-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-NEXT: s_mov_b32 s1, 0x43300000
+; GFX9-NEXT: v_add_f64 v[0:1], s[0:1], v[0:1]
+; GFX9-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX9-NEXT: s_add_u32 s8, s34, 44
+; GFX9-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-NEXT: s_mov_b32 s12, s33
+; GFX9-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-NEXT: s_cbranch_execnz .LBB16_2
+; GFX9-NEXT: .LBB16_3:
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s42, -1
+; GFX1064-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-NEXT: s_mov_b32 s33, s2
+; GFX1064-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-NEXT: s_cbranch_execz .LBB16_3
+; GFX1064-NEXT: ; %bb.1:
+; GFX1064-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1064-NEXT: s_mov_b32 s1, 0x43300000
+; GFX1064-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-NEXT: v_add_f64 v[0:1], 0xc3300000, s[0:1]
+; GFX1064-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-NEXT: s_mov_b32 s12, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1064-NEXT: .LBB16_3:
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s33, s2
+; GFX1032-NEXT: s_mov_b32 s2, exec_lo
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1032-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s42, -1
+; GFX1032-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-NEXT: s_mov_b32 s38, 0
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-NEXT: s_cbranch_execz .LBB16_3
+; GFX1032-NEXT: ; %bb.1:
+; GFX1032-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1032-NEXT: s_mov_b32 s1, 0x43300000
+; GFX1032-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-NEXT: v_add_f64 v[0:1], 0xc3300000, s[0:1]
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-NEXT: s_mov_b32 s12, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1032-NEXT: .LBB16_3:
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-NEXT: s_bcnt1_i32_b64 s0, exec
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b32 off, v0, off offset:20
+; GFX1164-NEXT: scratch_store_b32 off, v1, off offset:16
+; GFX1164-NEXT: scratch_load_b64 v[0:1], off, off offset:16
+; GFX1164-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-NEXT: s_cbranch_execz .LBB16_3
+; GFX1164-NEXT: ; %bb.1:
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-NEXT: s_mov_b32 s33, s2
+; GFX1164-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b32 s12, s33
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1164-NEXT: .LBB16_3:
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-NEXT: s_bcnt1_i32_b32 s0, exec_lo
+; GFX1132-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1132-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b32 off, v0, off offset:20
+; GFX1132-NEXT: scratch_store_b32 off, v1, off offset:16
+; GFX1132-NEXT: scratch_load_b64 v[0:1], off, off offset:16
+; GFX1132-NEXT: s_mov_b32 s38, 0
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-NEXT: s_cbranch_execz .LBB16_3
+; GFX1132-NEXT: ; %bb.1:
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b32 s12, s33
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1132-NEXT: .LBB16_3:
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s42, -1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX9-DPP-NEXT: s_mov_b32 s43, 0xe00000
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 0
+; GFX9-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX9-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v0
+; GFX9-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX9-DPP-NEXT: s_mov_b32 s33, s2
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: s_and_saveexec_b64 s[2:3], vcc
+; GFX9-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX9-DPP-NEXT: ; %bb.1:
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT: s_bcnt1_i32_b64 s0, s[0:1]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0xc3300000
+; GFX9-DPP-NEXT: s_mov_b32 s1, 0x43300000
+; GFX9-DPP-NEXT: v_add_f64 v[0:1], s[0:1], v[0:1]
+; GFX9-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX9-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX9-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX9-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: s_mov_b32 s12, s33
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX9-DPP-NEXT: .LBB16_3:
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s43, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], exec
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1064-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mbcnt_hi_u32_b32 v0, s3, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT: s_and_saveexec_b64 s[0:1], vcc
+; GFX1064-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX1064-DPP-NEXT: ; %bb.1:
+; GFX1064-DPP-NEXT: s_bcnt1_i32_b64 s0, s[2:3]
+; GFX1064-DPP-NEXT: s_mov_b32 s1, 0x43300000
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1064-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[0:1]
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1064-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1064-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[38:39]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1064-DPP-NEXT: .LBB16_3:
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1032-DPP-NEXT: s_mov_b32 s2, exec_lo
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: v_mbcnt_lo_u32_b32 v0, s2, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s42, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s43, 0x31c16000
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT: s_add_u32 s40, s40, s3
+; GFX1032-DPP-NEXT: s_addc_u32 s41, s41, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1032-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: s_and_saveexec_b32 s0, vcc_lo
+; GFX1032-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX1032-DPP-NEXT: ; %bb.1:
+; GFX1032-DPP-NEXT: s_bcnt1_i32_b32 s0, s2
+; GFX1032-DPP-NEXT: s_mov_b32 s1, 0x43300000
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[36:37], s[34:35], 0x24
+; GFX1032-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, s[0:1]
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[0:1], s[36:37], 0x0
+; GFX1032-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1032-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[40:41]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[42:43]
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[40:43], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[40:43], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[40:43], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[40:43], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s38
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1032-DPP-NEXT: .LBB16_3:
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1164-DPP-NEXT: s_bcnt1_i32_b64 s0, exec
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v0, off offset:20
+; GFX1164-DPP-NEXT: scratch_store_b32 off, v1, off offset:16
+; GFX1164-DPP-NEXT: scratch_load_b64 v[0:1], off, off offset:16
+; GFX1164-DPP-NEXT: v_mbcnt_hi_u32_b32 v2, exec_hi, v2
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: s_mov_b64 s[0:1], exec
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1164-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX1164-DPP-NEXT: ; %bb.1:
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1164-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s2
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s1
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, s0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v40
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s36
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s37
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[38:39], vcc, s[38:39]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[38:39]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1164-DPP-NEXT: .LBB16_3:
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_uni_value_defalut_scope_strictfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[0:1]
+; GFX1132-DPP-NEXT: s_bcnt1_i32_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v0, 0x43300000
+; GFX1132-DPP-NEXT: v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v0, off offset:20
+; GFX1132-DPP-NEXT: scratch_store_b32 off, v1, off offset:16
+; GFX1132-DPP-NEXT: scratch_load_b64 v[0:1], off, off offset:16
+; GFX1132-DPP-NEXT: s_mov_b32 s38, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s0, exec_lo
+; GFX1132-DPP-NEXT: v_cmpx_eq_u32_e32 0, v2
+; GFX1132-DPP-NEXT: s_cbranch_execz .LBB16_3
+; GFX1132-DPP-NEXT: ; %bb.1:
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[0:1], 0xc3300000, v[0:1]
+; GFX1132-DPP-NEXT: s_load_b64 s[36:37], s[34:35], 0x24
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[36:37], 0x0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_mul_f64 v[41:42], 4.0, v[0:1]
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB16_2: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1132-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[41:42]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v40 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s33
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s36
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s37 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s38, vcc_lo, s38
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s38
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB16_2
+; GFX1132-DPP-NEXT: .LBB16_3:
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double 4.0 monotonic, align 4
+ ret void
+}
+
+define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp(ptr addrspace(1) %ptr) #2 {
+; GFX7LESS-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX7LESS: ; %bb.0:
+; GFX7LESS-NEXT: s_movk_i32 s32, 0x800
+; GFX7LESS-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX7LESS-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX7LESS-NEXT: s_mov_b32 s50, -1
+; GFX7LESS-NEXT: s_mov_b32 s51, 0xe8f000
+; GFX7LESS-NEXT: s_add_u32 s48, s48, s9
+; GFX7LESS-NEXT: s_addc_u32 s49, s49, 0
+; GFX7LESS-NEXT: s_mov_b32 s33, s8
+; GFX7LESS-NEXT: s_mov_b32 s40, s7
+; GFX7LESS-NEXT: s_mov_b32 s41, s6
+; GFX7LESS-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX7LESS-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX7LESS-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX7LESS-NEXT: s_load_dwordx2 s[44:45], s[2:3], 0x9
+; GFX7LESS-NEXT: s_mov_b32 s47, 0xf000
+; GFX7LESS-NEXT: s_mov_b32 s46, -1
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX7LESS-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX7LESS-NEXT: v_or_b32_e32 v42, v0, v2
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v42
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_mov_b32_e32 v40, v0
+; GFX7LESS-NEXT: v_mov_b32_e32 v41, v1
+; GFX7LESS-NEXT: buffer_load_dwordx2 v[0:1], off, s[44:47], 0
+; GFX7LESS-NEXT: s_mov_b64 s[42:43], 0
+; GFX7LESS-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX7LESS-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX7LESS-NEXT: s_waitcnt vmcnt(0)
+; GFX7LESS-NEXT: v_add_f64 v[2:3], v[0:1], -v[40:41]
+; GFX7LESS-NEXT: buffer_store_dword v1, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: buffer_store_dword v0, off, s[48:51], 0
+; GFX7LESS-NEXT: s_add_u32 s8, s36, 44
+; GFX7LESS-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:12
+; GFX7LESS-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:8
+; GFX7LESS-NEXT: s_addc_u32 s9, s37, 0
+; GFX7LESS-NEXT: s_getpc_b64 s[0:1]
+; GFX7LESS-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX7LESS-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX7LESS-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX7LESS-NEXT: s_waitcnt expcnt(2)
+; GFX7LESS-NEXT: v_mov_b32_e32 v0, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v1, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v4, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v5, 8
+; GFX7LESS-NEXT: v_mov_b32_e32 v6, 0
+; GFX7LESS-NEXT: v_mov_b32_e32 v7, 0
+; GFX7LESS-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX7LESS-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX7LESS-NEXT: s_mov_b32 s12, s41
+; GFX7LESS-NEXT: s_mov_b32 s13, s40
+; GFX7LESS-NEXT: s_mov_b32 s14, s33
+; GFX7LESS-NEXT: v_mov_b32_e32 v31, v42
+; GFX7LESS-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX7LESS-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX7LESS-NEXT: s_waitcnt expcnt(0)
+; GFX7LESS-NEXT: v_mov_b32_e32 v2, s44
+; GFX7LESS-NEXT: v_mov_b32_e32 v3, s45
+; GFX7LESS-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7LESS-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX7LESS-NEXT: v_and_b32_e32 v2, 1, v0
+; GFX7LESS-NEXT: buffer_load_dword v0, off, s[48:51], 0
+; GFX7LESS-NEXT: buffer_load_dword v1, off, s[48:51], 0 offset:4
+; GFX7LESS-NEXT: v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7LESS-NEXT: s_or_b64 s[42:43], vcc, s[42:43]
+; GFX7LESS-NEXT: s_andn2_b64 exec, exec, s[42:43]
+; GFX7LESS-NEXT: s_cbranch_execnz .LBB17_1
+; GFX7LESS-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX7LESS-NEXT: s_endpgm
+;
+; GFX9-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-NEXT: s_mov_b32 s50, -1
+; GFX9-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-NEXT: s_add_u32 s48, s48, s9
+; GFX9-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-NEXT: s_mov_b32 s33, s8
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX9-NEXT: s_mov_b32 s40, s7
+; GFX9-NEXT: s_mov_b32 s41, s6
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: s_movk_i32 s32, 0x800
+; GFX9-NEXT: v_mov_b32_e32 v43, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: v_mov_b32_e32 v41, v1
+; GFX9-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX9-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX9-NEXT: s_add_u32 s8, s36, 44
+; GFX9-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-NEXT: s_getpc_b64 s[0:1]
+; GFX9-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX9-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX9-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX9-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-NEXT: s_mov_b32 s12, s41
+; GFX9-NEXT: s_mov_b32 s13, s40
+; GFX9-NEXT: s_mov_b32 s14, s33
+; GFX9-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX9-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX9-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-NEXT: s_cbranch_execnz .LBB17_1
+; GFX9-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-NEXT: s_endpgm
+;
+; GFX1064-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1064: ; %bb.0:
+; GFX1064-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-NEXT: s_mov_b32 s50, -1
+; GFX1064-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-NEXT: s_mov_b32 s33, s8
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1064-NEXT: s_mov_b32 s40, s7
+; GFX1064-NEXT: s_mov_b32 s41, s6
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-NEXT: v_mov_b32_e32 v43, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: v_mov_b32_e32 v41, v1
+; GFX1064-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1064-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1064-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1064-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1064-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-NEXT: s_mov_b32 s12, s41
+; GFX1064-NEXT: s_mov_b32 s13, s40
+; GFX1064-NEXT: s_mov_b32 s14, s33
+; GFX1064-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1064-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1064-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-NEXT: s_clause 0x1
+; GFX1064-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1064-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1064-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-NEXT: s_endpgm
+;
+; GFX1032-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1032: ; %bb.0:
+; GFX1032-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-NEXT: s_mov_b32 s50, -1
+; GFX1032-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-NEXT: s_mov_b32 s33, s8
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1032-NEXT: s_mov_b32 s40, s7
+; GFX1032-NEXT: s_mov_b32 s41, s6
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-NEXT: v_mov_b32_e32 v43, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: v_mov_b32_e32 v41, v1
+; GFX1032-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1032-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-NEXT: s_mov_b32 s44, 0
+; GFX1032-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1032-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1032-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1032-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-NEXT: s_mov_b32 s12, s41
+; GFX1032-NEXT: s_mov_b32 s13, s40
+; GFX1032-NEXT: s_mov_b32 s14, s33
+; GFX1032-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1032-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1032-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-NEXT: s_clause 0x1
+; GFX1032-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1032-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1032-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-NEXT: s_endpgm
+;
+; GFX1164-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1164: ; %bb.0:
+; GFX1164-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-NEXT: s_mov_b32 s33, s8
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s6
+; GFX1164-NEXT: s_mov_b32 s13, s7
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_mov_b32 s32, 32
+; GFX1164-NEXT: v_mov_b32_e32 v42, v0
+; GFX1164-NEXT: s_mov_b32 s40, s7
+; GFX1164-NEXT: s_mov_b32 s41, s6
+; GFX1164-NEXT: v_mov_b32_e32 v43, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: v_mov_b32_e32 v41, v1
+; GFX1164-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1164-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-NEXT: .p2align 6
+; GFX1164-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1164-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1164-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-NEXT: v_mov_b32_e32 v31, v42
+; GFX1164-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-NEXT: s_mov_b32 s12, s41
+; GFX1164-NEXT: s_mov_b32 s13, s40
+; GFX1164-NEXT: s_mov_b32 s14, s33
+; GFX1164-NEXT: s_clause 0x1
+; GFX1164-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1164-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-NEXT: s_endpgm
+;
+; GFX1132-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1132: ; %bb.0:
+; GFX1132-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1132-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-NEXT: s_mov_b32 s40, s14
+; GFX1132-NEXT: s_mov_b32 s41, s13
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s13
+; GFX1132-NEXT: s_mov_b32 s13, s14
+; GFX1132-NEXT: s_mov_b32 s14, s15
+; GFX1132-NEXT: s_mov_b32 s32, 32
+; GFX1132-NEXT: s_mov_b32 s33, s15
+; GFX1132-NEXT: v_dual_mov_b32 v42, v0 :: v_dual_mov_b32 v43, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, v1
+; GFX1132-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1132-NEXT: s_mov_b32 s44, 0
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-NEXT: .p2align 6
+; GFX1132-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1132-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1132-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-NEXT: v_dual_mov_b32 v31, v42 :: v_dual_mov_b32 v0, 8
+; GFX1132-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-NEXT: s_mov_b32 s12, s41
+; GFX1132-NEXT: s_mov_b32 s13, s40
+; GFX1132-NEXT: s_mov_b32 s14, s33
+; GFX1132-NEXT: s_clause 0x1
+; GFX1132-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s42
+; GFX1132-NEXT: v_dual_mov_b32 v3, s43 :: v_dual_mov_b32 v4, 0
+; GFX1132-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1132-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-NEXT: s_endpgm
+;
+; GFX9-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX9-DPP: ; %bb.0:
+; GFX9-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX9-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX9-DPP-NEXT: s_mov_b32 s50, -1
+; GFX9-DPP-NEXT: s_mov_b32 s51, 0xe00000
+; GFX9-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX9-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX9-DPP-NEXT: s_mov_b32 s33, s8
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_mov_b32 s40, s7
+; GFX9-DPP-NEXT: s_mov_b32 s41, s6
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX9-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX9-DPP-NEXT: s_mov_b64 s[34:35], s[4:5]
+; GFX9-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX9-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX9-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX9-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX9-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX9-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX9-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX9-DPP-NEXT: s_add_u32 s8, s36, 44
+; GFX9-DPP-NEXT: s_addc_u32 s9, s37, 0
+; GFX9-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX9-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX9-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX9-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX9-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX9-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX9-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX9-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX9-DPP-NEXT: s_mov_b64 s[10:11], s[34:35]
+; GFX9-DPP-NEXT: s_mov_b32 s12, s41
+; GFX9-DPP-NEXT: s_mov_b32 s13, s40
+; GFX9-DPP-NEXT: s_mov_b32 s14, s33
+; GFX9-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX9-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX9-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX9-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX9-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX9-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX9-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX9-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX9-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX9-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX9-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX9-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX9-DPP-NEXT: s_endpgm
+;
+; GFX1064-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1064-DPP: ; %bb.0:
+; GFX1064-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1064-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1064-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1064-DPP-NEXT: s_mov_b32 s51, 0x31e16000
+; GFX1064-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1064-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1064-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1064-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1064-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1064-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1064-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-DPP-NEXT: s_movk_i32 s32, 0x800
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1064-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1064-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1064-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1064-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1064-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1064-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1064-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1064-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1064-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1064-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1064-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1064-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1064-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1064-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1064-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1064-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1064-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1064-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1064-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1064-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1064-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1064-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1064-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1064-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1064-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1064-DPP-NEXT: s_clause 0x1
+; GFX1064-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1064-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1064-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1064-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1064-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1064-DPP-NEXT: s_andn2_b64 exec, exec, s[44:45]
+; GFX1064-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1064-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1064-DPP-NEXT: s_endpgm
+;
+; GFX1032-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1032-DPP: ; %bb.0:
+; GFX1032-DPP-NEXT: s_mov_b32 s48, SCRATCH_RSRC_DWORD0
+; GFX1032-DPP-NEXT: s_mov_b32 s49, SCRATCH_RSRC_DWORD1
+; GFX1032-DPP-NEXT: s_mov_b32 s50, -1
+; GFX1032-DPP-NEXT: s_mov_b32 s51, 0x31c16000
+; GFX1032-DPP-NEXT: s_add_u32 s48, s48, s9
+; GFX1032-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1032-DPP-NEXT: s_addc_u32 s49, s49, 0
+; GFX1032-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1032-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[42:43], s[2:3], 0x24
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v2, 20, v2
+; GFX1032-DPP-NEXT: v_lshlrev_b32_e32 v1, 10, v1
+; GFX1032-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: v_or3_b32 v42, v0, v1, v2
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-DPP-NEXT: s_movk_i32 s32, 0x400
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1032-DPP-NEXT: global_load_dwordx2 v[1:2], v43, s[42:43]
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1032-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1032-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1032-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1032-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1032-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1032-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1032-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1032-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1032-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1032-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1032-DPP-NEXT: buffer_store_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: buffer_store_dword v1, off, s[48:51], 0
+; GFX1032-DPP-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1032-DPP-NEXT: s_mov_b64 s[0:1], s[48:49]
+; GFX1032-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1032-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1032-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1032-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1032-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1032-DPP-NEXT: s_mov_b64 s[2:3], s[50:51]
+; GFX1032-DPP-NEXT: buffer_store_dword v4, off, s[48:51], 0 offset:12
+; GFX1032-DPP-NEXT: buffer_store_dword v3, off, s[48:51], 0 offset:8
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1032-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1032-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1032-DPP-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX1032-DPP-NEXT: s_clause 0x1
+; GFX1032-DPP-NEXT: buffer_load_dword v1, off, s[48:51], 0
+; GFX1032-DPP-NEXT: buffer_load_dword v2, off, s[48:51], 0 offset:4
+; GFX1032-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1032-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1032-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1032-DPP-NEXT: s_andn2_b32 exec_lo, exec_lo, s44
+; GFX1032-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1032-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1032-DPP-NEXT: s_endpgm
+;
+; GFX1164-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1164-DPP: ; %bb.0:
+; GFX1164-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1164-DPP-NEXT: s_mov_b32 s33, s8
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s6
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v42, v0
+; GFX1164-DPP-NEXT: s_mov_b32 s40, s7
+; GFX1164-DPP-NEXT: s_mov_b32 s41, s6
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v43, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v41, v1
+; GFX1164-DPP-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v40, v0
+; GFX1164-DPP-NEXT: s_mov_b64 s[44:45], 0
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1164-DPP-NEXT: .p2align 6
+; GFX1164-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1164-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1164-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1164-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1164-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1164-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1164-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1164-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1164-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v31, v42
+; GFX1164-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v0, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v5, 8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v6, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1164-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1164-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1164-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1164-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1164-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1164-DPP-NEXT: s_clause 0x1
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1164-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v1, 0
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v2, s42
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v3, s43
+; GFX1164-DPP-NEXT: v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1164-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1164-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1164-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1164-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1164-DPP-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX1164-DPP-NEXT: s_or_b64 s[44:45], vcc, s[44:45]
+; GFX1164-DPP-NEXT: s_and_not1_b64 exec, exec, s[44:45]
+; GFX1164-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1164-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1164-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1164-DPP-NEXT: s_endpgm
+;
+; GFX1132-DPP-LABEL: global_atomic_fsub_double_uni_address_div_value_defalut_scope_strictfp:
+; GFX1132-DPP: ; %bb.0:
+; GFX1132-DPP-NEXT: s_mov_b64 s[34:35], s[2:3]
+; GFX1132-DPP-NEXT: s_mov_b64 s[38:39], s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, div.float.value at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, div.float.value at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v31, v0
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: s_load_b64 s[42:43], s[2:3], 0x24
+; GFX1132-DPP-NEXT: s_mov_b64 s[36:37], s[4:5]
+; GFX1132-DPP-NEXT: s_mov_b32 s40, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s41, s13
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s13
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s14
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s15
+; GFX1132-DPP-NEXT: s_mov_b32 s32, 32
+; GFX1132-DPP-NEXT: s_mov_b32 s33, s15
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v42, v0 :: v_dual_mov_b32 v43, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v40, v0 :: v_dual_mov_b32 v41, v1
+; GFX1132-DPP-NEXT: global_load_b64 v[1:2], v43, s[42:43]
+; GFX1132-DPP-NEXT: s_mov_b32 s44, 0
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x1
+; GFX1132-DPP-NEXT: .p2align 6
+; GFX1132-DPP-NEXT: .LBB17_1: ; %atomicrmw.start
+; GFX1132-DPP-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1132-DPP-NEXT: s_waitcnt vmcnt(0)
+; GFX1132-DPP-NEXT: v_add_f64 v[3:4], v[1:2], -v[40:41]
+; GFX1132-DPP-NEXT: s_add_u32 s8, s34, 44
+; GFX1132-DPP-NEXT: s_addc_u32 s9, s35, 0
+; GFX1132-DPP-NEXT: s_getpc_b64 s[0:1]
+; GFX1132-DPP-NEXT: s_add_u32 s0, s0, __atomic_compare_exchange at gotpcrel32@lo+4
+; GFX1132-DPP-NEXT: s_addc_u32 s1, s1, __atomic_compare_exchange at gotpcrel32@hi+12
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v31, v42 :: v_dual_mov_b32 v0, 8
+; GFX1132-DPP-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v5, 8 :: v_dual_mov_b32 v6, 0
+; GFX1132-DPP-NEXT: v_mov_b32_e32 v7, 0
+; GFX1132-DPP-NEXT: s_mov_b64 s[4:5], s[38:39]
+; GFX1132-DPP-NEXT: s_mov_b64 s[10:11], s[36:37]
+; GFX1132-DPP-NEXT: s_mov_b32 s12, s41
+; GFX1132-DPP-NEXT: s_mov_b32 s13, s40
+; GFX1132-DPP-NEXT: s_mov_b32 s14, s33
+; GFX1132-DPP-NEXT: s_clause 0x1
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[1:2], off
+; GFX1132-DPP-NEXT: scratch_store_b64 off, v[3:4], off offset:8
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v2, s42
+; GFX1132-DPP-NEXT: v_dual_mov_b32 v3, s43 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1132-DPP-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GFX1132-DPP-NEXT: scratch_load_b64 v[1:2], off, off
+; GFX1132-DPP-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1132-DPP-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX1132-DPP-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GFX1132-DPP-NEXT: s_or_b32 s44, vcc_lo, s44
+; GFX1132-DPP-NEXT: s_and_not1_b32 exec_lo, exec_lo, s44
+; GFX1132-DPP-NEXT: s_cbranch_execnz .LBB17_1
+; GFX1132-DPP-NEXT: ; %bb.2: ; %atomicrmw.end
+; GFX1132-DPP-NEXT: s_set_inst_prefetch_distance 0x2
+; GFX1132-DPP-NEXT: s_endpgm
+ %divValue = call double @div.float.value() strictfp
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %divValue monotonic, align 4
+ ret void
+}
+
attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #1 = { strictfp "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #2 = { strictfp}
>From 3115963fa76eccdab63e90fa9a767a9df502da54 Mon Sep 17 00:00:00 2001
From: Pravin Jagtap <Pravin.Jagtap at amd.com>
Date: Mon, 18 Mar 2024 00:39:23 -0400
Subject: [PATCH 2/2] added ret and no-ret opt tests for floating point atomic
ops with double type.
---
.../Target/AMDGPU/AMDGPUAtomicOptimizer.cpp | 6 +-
.../AMDGPU/global_atomic_optimizer_fp_rtn.ll | 706 ++++++++++++++++++
.../global_atomics_optimizer_fp_no_rtn.ll | 537 +++++++++++++
3 files changed, 1247 insertions(+), 2 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
index a6013176676bde..dbb3de76b4ddae 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
@@ -921,8 +921,10 @@ void AMDGPUAtomicOptimizerImpl::optimizeAtomic(Instruction &I,
Value *BroadcastI = nullptr;
if (TyBitWidth == 64) {
- Value *const ExtractLo = B.CreateTrunc(PHI, Int32Ty);
- Value *const ExtractHi = B.CreateTrunc(B.CreateLShr(PHI, 32), Int32Ty);
+ Value *CastedPhi = B.CreateBitCast(PHI, IntNTy);
+ Value *const ExtractLo = B.CreateTrunc(CastedPhi, Int32Ty);
+ Value *const ExtractHi =
+ B.CreateTrunc(B.CreateLShr(CastedPhi, 32), Int32Ty);
CallInst *const ReadFirstLaneLo =
B.CreateIntrinsic(Intrinsic::amdgcn_readfirstlane, {}, ExtractLo);
CallInst *const ReadFirstLaneHi =
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomic_optimizer_fp_rtn.ll b/llvm/test/CodeGen/AMDGPU/global_atomic_optimizer_fp_rtn.ll
index e3fada3459a07f..4ab662b6eb26d6 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomic_optimizer_fp_rtn.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomic_optimizer_fp_rtn.ll
@@ -1165,6 +1165,712 @@ define amdgpu_ps float @global_atomic_fadd_div_address_div_value_system_scope_st
ret float %result
}
+define amdgpu_ps double @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double inreg %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP30:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]])
+; IR-ITERATIVE-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = uitofp i32 [[TMP10]] to double
+; IR-ITERATIVE-NEXT: [[TMP12:%.*]] = fmul double [[VAL:%.*]], [[TMP11]]
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-ITERATIVE: 14:
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 4
+; IR-ITERATIVE-NEXT: br label [[TMP16]]
+; IR-ITERATIVE: 16:
+; IR-ITERATIVE-NEXT: [[TMP17:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP15]], [[TMP14]] ]
+; IR-ITERATIVE-NEXT: [[TMP18:%.*]] = bitcast double [[TMP17]] to i64
+; IR-ITERATIVE-NEXT: [[TMP19:%.*]] = trunc i64 [[TMP18]] to i32
+; IR-ITERATIVE-NEXT: [[TMP20:%.*]] = lshr i64 [[TMP18]], 32
+; IR-ITERATIVE-NEXT: [[TMP21:%.*]] = trunc i64 [[TMP20]] to i32
+; IR-ITERATIVE-NEXT: [[TMP22:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP19]])
+; IR-ITERATIVE-NEXT: [[TMP23:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP21]])
+; IR-ITERATIVE-NEXT: [[TMP24:%.*]] = insertelement <2 x i32> poison, i32 [[TMP22]], i32 0
+; IR-ITERATIVE-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP24]], i32 [[TMP23]], i32 1
+; IR-ITERATIVE-NEXT: [[TMP26:%.*]] = bitcast <2 x i32> [[TMP25]] to double
+; IR-ITERATIVE-NEXT: [[TMP27:%.*]] = uitofp i32 [[TMP8]] to double
+; IR-ITERATIVE-NEXT: [[TMP28:%.*]] = fmul double [[VAL]], [[TMP27]]
+; IR-ITERATIVE-NEXT: [[TMP29:%.*]] = fadd double [[TMP26]], [[TMP28]]
+; IR-ITERATIVE-NEXT: br label [[TMP30]]
+; IR-ITERATIVE: 30:
+; IR-ITERATIVE-NEXT: [[TMP31:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP29]], [[TMP16]] ]
+; IR-ITERATIVE-NEXT: ret double [[TMP31]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP30:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
+; IR-DPP-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]])
+; IR-DPP-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-DPP-NEXT: [[TMP11:%.*]] = uitofp i32 [[TMP10]] to double
+; IR-DPP-NEXT: [[TMP12:%.*]] = fmul double [[VAL:%.*]], [[TMP11]]
+; IR-DPP-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-DPP: 14:
+; IR-DPP-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 4
+; IR-DPP-NEXT: br label [[TMP16]]
+; IR-DPP: 16:
+; IR-DPP-NEXT: [[TMP17:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP15]], [[TMP14]] ]
+; IR-DPP-NEXT: [[TMP18:%.*]] = bitcast double [[TMP17]] to i64
+; IR-DPP-NEXT: [[TMP19:%.*]] = trunc i64 [[TMP18]] to i32
+; IR-DPP-NEXT: [[TMP20:%.*]] = lshr i64 [[TMP18]], 32
+; IR-DPP-NEXT: [[TMP21:%.*]] = trunc i64 [[TMP20]] to i32
+; IR-DPP-NEXT: [[TMP22:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP19]])
+; IR-DPP-NEXT: [[TMP23:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP21]])
+; IR-DPP-NEXT: [[TMP24:%.*]] = insertelement <2 x i32> poison, i32 [[TMP22]], i32 0
+; IR-DPP-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP24]], i32 [[TMP23]], i32 1
+; IR-DPP-NEXT: [[TMP26:%.*]] = bitcast <2 x i32> [[TMP25]] to double
+; IR-DPP-NEXT: [[TMP27:%.*]] = uitofp i32 [[TMP8]] to double
+; IR-DPP-NEXT: [[TMP28:%.*]] = fmul double [[VAL]], [[TMP27]]
+; IR-DPP-NEXT: [[TMP29:%.*]] = fadd double [[TMP26]], [[TMP28]]
+; IR-DPP-NEXT: br label [[TMP30]]
+; IR-DPP: 30:
+; IR-DPP-NEXT: [[TMP31:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP29]], [[TMP16]] ]
+; IR-DPP-NEXT: ret double [[TMP31]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_uni_address_div_value_scope_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_div_value_scope_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_div_value_scope_agent_scope_unsafe(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp(ptr addrspace(1) inreg %ptr, double inreg %val) #1 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP30:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-ITERATIVE: 14:
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("one-as") monotonic, align 8
+; IR-ITERATIVE-NEXT: br label [[TMP16]]
+; IR-ITERATIVE: 16:
+; IR-ITERATIVE-NEXT: [[TMP17:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP15]], [[TMP14]] ]
+; IR-ITERATIVE-NEXT: [[TMP18:%.*]] = bitcast double [[TMP17]] to i64
+; IR-ITERATIVE-NEXT: [[TMP19:%.*]] = trunc i64 [[TMP18]] to i32
+; IR-ITERATIVE-NEXT: [[TMP20:%.*]] = lshr i64 [[TMP18]], 32
+; IR-ITERATIVE-NEXT: [[TMP21:%.*]] = trunc i64 [[TMP20]] to i32
+; IR-ITERATIVE-NEXT: [[TMP22:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP19]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP23:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP21]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP24:%.*]] = insertelement <2 x i32> poison, i32 [[TMP22]], i32 0
+; IR-ITERATIVE-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP24]], i32 [[TMP23]], i32 1
+; IR-ITERATIVE-NEXT: [[TMP26:%.*]] = bitcast <2 x i32> [[TMP25]] to double
+; IR-ITERATIVE-NEXT: [[TMP27:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP8]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP28:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL]], double [[TMP27]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP29:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[TMP26]], double [[TMP28]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br label [[TMP30]]
+; IR-ITERATIVE: 30:
+; IR-ITERATIVE-NEXT: [[TMP31:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP29]], [[TMP16]] ]
+; IR-ITERATIVE-NEXT: ret double [[TMP31]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP30:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-DPP-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-DPP: 14:
+; IR-DPP-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("one-as") monotonic, align 8
+; IR-DPP-NEXT: br label [[TMP16]]
+; IR-DPP: 16:
+; IR-DPP-NEXT: [[TMP17:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP15]], [[TMP14]] ]
+; IR-DPP-NEXT: [[TMP18:%.*]] = bitcast double [[TMP17]] to i64
+; IR-DPP-NEXT: [[TMP19:%.*]] = trunc i64 [[TMP18]] to i32
+; IR-DPP-NEXT: [[TMP20:%.*]] = lshr i64 [[TMP18]], 32
+; IR-DPP-NEXT: [[TMP21:%.*]] = trunc i64 [[TMP20]] to i32
+; IR-DPP-NEXT: [[TMP22:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP19]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP23:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP21]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP24:%.*]] = insertelement <2 x i32> poison, i32 [[TMP22]], i32 0
+; IR-DPP-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP24]], i32 [[TMP23]], i32 1
+; IR-DPP-NEXT: [[TMP26:%.*]] = bitcast <2 x i32> [[TMP25]] to double
+; IR-DPP-NEXT: [[TMP27:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP8]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP28:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL]], double [[TMP27]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP29:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[TMP26]], double [[TMP28]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: br label [[TMP30]]
+; IR-DPP: 30:
+; IR-DPP-NEXT: [[TMP31:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP29]], [[TMP16]] ]
+; IR-DPP-NEXT: ret double [[TMP31]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp(ptr addrspace(1) inreg %ptr, double %val) #1 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(ptr addrspace(1) inreg %ptr, double inreg %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP30:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-ITERATIVE: 14:
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: br label [[TMP16]]
+; IR-ITERATIVE: 16:
+; IR-ITERATIVE-NEXT: [[TMP17:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP15]], [[TMP14]] ]
+; IR-ITERATIVE-NEXT: [[TMP18:%.*]] = bitcast double [[TMP17]] to i64
+; IR-ITERATIVE-NEXT: [[TMP19:%.*]] = trunc i64 [[TMP18]] to i32
+; IR-ITERATIVE-NEXT: [[TMP20:%.*]] = lshr i64 [[TMP18]], 32
+; IR-ITERATIVE-NEXT: [[TMP21:%.*]] = trunc i64 [[TMP20]] to i32
+; IR-ITERATIVE-NEXT: [[TMP22:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP19]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP23:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP21]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP24:%.*]] = insertelement <2 x i32> poison, i32 [[TMP22]], i32 0
+; IR-ITERATIVE-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP24]], i32 [[TMP23]], i32 1
+; IR-ITERATIVE-NEXT: [[TMP26:%.*]] = bitcast <2 x i32> [[TMP25]] to double
+; IR-ITERATIVE-NEXT: [[TMP27:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP8]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP28:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL]], double [[TMP27]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP29:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[TMP26]], double [[TMP28]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br label [[TMP30]]
+; IR-ITERATIVE: 30:
+; IR-ITERATIVE-NEXT: [[TMP31:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP29]], [[TMP16]] ]
+; IR-ITERATIVE-NEXT: ret double [[TMP31]]
+;
+; IR-DPP-LABEL: @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP30:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-DPP-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-DPP: 14:
+; IR-DPP-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: br label [[TMP16]]
+; IR-DPP: 16:
+; IR-DPP-NEXT: [[TMP17:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP15]], [[TMP14]] ]
+; IR-DPP-NEXT: [[TMP18:%.*]] = bitcast double [[TMP17]] to i64
+; IR-DPP-NEXT: [[TMP19:%.*]] = trunc i64 [[TMP18]] to i32
+; IR-DPP-NEXT: [[TMP20:%.*]] = lshr i64 [[TMP18]], 32
+; IR-DPP-NEXT: [[TMP21:%.*]] = trunc i64 [[TMP20]] to i32
+; IR-DPP-NEXT: [[TMP22:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP19]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP23:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP21]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP24:%.*]] = insertelement <2 x i32> poison, i32 [[TMP22]], i32 0
+; IR-DPP-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP24]], i32 [[TMP23]], i32 1
+; IR-DPP-NEXT: [[TMP26:%.*]] = bitcast <2 x i32> [[TMP25]] to double
+; IR-DPP-NEXT: [[TMP27:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP8]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP28:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL]], double [[TMP27]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP29:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[TMP26]], double [[TMP28]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: br label [[TMP30]]
+; IR-DPP: 30:
+; IR-DPP-NEXT: [[TMP31:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP29]], [[TMP16]] ]
+; IR-DPP-NEXT: ret double [[TMP31]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fsub_double_uni_address_div_value_agent_scope_strictfp(ptr addrspace(1) inreg %ptr, double %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_uni_address_div_value_agent_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fsub_double_uni_address_div_value_agent_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double inreg %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP26:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
+; IR-ITERATIVE: 10:
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: br label [[TMP12]]
+; IR-ITERATIVE: 12:
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP11]], [[TMP10]] ]
+; IR-ITERATIVE-NEXT: [[TMP14:%.*]] = bitcast double [[TMP13]] to i64
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
+; IR-ITERATIVE-NEXT: [[TMP16:%.*]] = lshr i64 [[TMP14]], 32
+; IR-ITERATIVE-NEXT: [[TMP17:%.*]] = trunc i64 [[TMP16]] to i32
+; IR-ITERATIVE-NEXT: [[TMP18:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP15]])
+; IR-ITERATIVE-NEXT: [[TMP19:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP17]])
+; IR-ITERATIVE-NEXT: [[TMP20:%.*]] = insertelement <2 x i32> poison, i32 [[TMP18]], i32 0
+; IR-ITERATIVE-NEXT: [[TMP21:%.*]] = insertelement <2 x i32> [[TMP20]], i32 [[TMP19]], i32 1
+; IR-ITERATIVE-NEXT: [[TMP22:%.*]] = bitcast <2 x i32> [[TMP21]] to double
+; IR-ITERATIVE-NEXT: [[TMP23:%.*]] = uitofp i32 [[TMP8]] to double
+; IR-ITERATIVE-NEXT: [[TMP24:%.*]] = select i1 [[TMP9]], double 0x7FF0000000000000, double [[VAL]]
+; IR-ITERATIVE-NEXT: [[TMP25:%.*]] = call double @llvm.minnum.f64(double [[TMP22]], double [[TMP24]])
+; IR-ITERATIVE-NEXT: br label [[TMP26]]
+; IR-ITERATIVE: 26:
+; IR-ITERATIVE-NEXT: [[TMP27:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP25]], [[TMP12]] ]
+; IR-ITERATIVE-NEXT: ret double [[TMP27]]
+;
+; IR-DPP-LABEL: @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP26:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
+; IR-DPP-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
+; IR-DPP: 10:
+; IR-DPP-NEXT: [[TMP11:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: br label [[TMP12]]
+; IR-DPP: 12:
+; IR-DPP-NEXT: [[TMP13:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP11]], [[TMP10]] ]
+; IR-DPP-NEXT: [[TMP14:%.*]] = bitcast double [[TMP13]] to i64
+; IR-DPP-NEXT: [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
+; IR-DPP-NEXT: [[TMP16:%.*]] = lshr i64 [[TMP14]], 32
+; IR-DPP-NEXT: [[TMP17:%.*]] = trunc i64 [[TMP16]] to i32
+; IR-DPP-NEXT: [[TMP18:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP15]])
+; IR-DPP-NEXT: [[TMP19:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP17]])
+; IR-DPP-NEXT: [[TMP20:%.*]] = insertelement <2 x i32> poison, i32 [[TMP18]], i32 0
+; IR-DPP-NEXT: [[TMP21:%.*]] = insertelement <2 x i32> [[TMP20]], i32 [[TMP19]], i32 1
+; IR-DPP-NEXT: [[TMP22:%.*]] = bitcast <2 x i32> [[TMP21]] to double
+; IR-DPP-NEXT: [[TMP23:%.*]] = uitofp i32 [[TMP8]] to double
+; IR-DPP-NEXT: [[TMP24:%.*]] = select i1 [[TMP9]], double 0x7FF0000000000000, double [[VAL]]
+; IR-DPP-NEXT: [[TMP25:%.*]] = call double @llvm.minnum.f64(double [[TMP22]], double [[TMP24]])
+; IR-DPP-NEXT: br label [[TMP26]]
+; IR-DPP: 26:
+; IR-DPP-NEXT: [[TMP27:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP25]], [[TMP12]] ]
+; IR-DPP-NEXT: ret double [[TMP27]]
+;
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic__fmax_double_uni_address_uni_value_agent_scope_unsafe_structfp(ptr addrspace(1) inreg %ptr, double inreg %val) #1{
+; IR-ITERATIVE-LABEL: @global_atomic__fmax_double_uni_address_uni_value_agent_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP26:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
+; IR-ITERATIVE: 10:
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: br label [[TMP12]]
+; IR-ITERATIVE: 12:
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP11]], [[TMP10]] ]
+; IR-ITERATIVE-NEXT: [[TMP14:%.*]] = bitcast double [[TMP13]] to i64
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
+; IR-ITERATIVE-NEXT: [[TMP16:%.*]] = lshr i64 [[TMP14]], 32
+; IR-ITERATIVE-NEXT: [[TMP17:%.*]] = trunc i64 [[TMP16]] to i32
+; IR-ITERATIVE-NEXT: [[TMP18:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP15]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP19:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP17]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP20:%.*]] = insertelement <2 x i32> poison, i32 [[TMP18]], i32 0
+; IR-ITERATIVE-NEXT: [[TMP21:%.*]] = insertelement <2 x i32> [[TMP20]], i32 [[TMP19]], i32 1
+; IR-ITERATIVE-NEXT: [[TMP22:%.*]] = bitcast <2 x i32> [[TMP21]] to double
+; IR-ITERATIVE-NEXT: [[TMP23:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP8]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP24:%.*]] = select i1 [[TMP9]], double 0xFFF0000000000000, double [[VAL]]
+; IR-ITERATIVE-NEXT: [[TMP25:%.*]] = call double @llvm.experimental.constrained.maxnum.f64(double [[TMP22]], double [[TMP24]], metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br label [[TMP26]]
+; IR-ITERATIVE: 26:
+; IR-ITERATIVE-NEXT: [[TMP27:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP25]], [[TMP12]] ]
+; IR-ITERATIVE-NEXT: ret double [[TMP27]]
+;
+; IR-DPP-LABEL: @global_atomic__fmax_double_uni_address_uni_value_agent_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP26:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
+; IR-DPP: 10:
+; IR-DPP-NEXT: [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: br label [[TMP12]]
+; IR-DPP: 12:
+; IR-DPP-NEXT: [[TMP13:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP11]], [[TMP10]] ]
+; IR-DPP-NEXT: [[TMP14:%.*]] = bitcast double [[TMP13]] to i64
+; IR-DPP-NEXT: [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
+; IR-DPP-NEXT: [[TMP16:%.*]] = lshr i64 [[TMP14]], 32
+; IR-DPP-NEXT: [[TMP17:%.*]] = trunc i64 [[TMP16]] to i32
+; IR-DPP-NEXT: [[TMP18:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP15]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP19:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP17]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP20:%.*]] = insertelement <2 x i32> poison, i32 [[TMP18]], i32 0
+; IR-DPP-NEXT: [[TMP21:%.*]] = insertelement <2 x i32> [[TMP20]], i32 [[TMP19]], i32 1
+; IR-DPP-NEXT: [[TMP22:%.*]] = bitcast <2 x i32> [[TMP21]] to double
+; IR-DPP-NEXT: [[TMP23:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP8]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP24:%.*]] = select i1 [[TMP9]], double 0xFFF0000000000000, double [[VAL]]
+; IR-DPP-NEXT: [[TMP25:%.*]] = call double @llvm.experimental.constrained.maxnum.f64(double [[TMP22]], double [[TMP24]], metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: br label [[TMP26]]
+; IR-DPP: 26:
+; IR-DPP-NEXT: [[TMP27:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP25]], [[TMP12]] ]
+; IR-DPP-NEXT: ret double [[TMP27]]
+;
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic__fmax_double_uni_address_div_value_agent_scope_unsafe_structfp(ptr addrspace(1) inreg %ptr, double %val) #1{
+; IR-ITERATIVE-LABEL: @global_atomic__fmax_double_uni_address_div_value_agent_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic__fmax_double_uni_address_div_value_agent_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(ptr addrspace(1) inreg %ptr, double inreg %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP30:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-ITERATIVE: 14:
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] monotonic, align 4
+; IR-ITERATIVE-NEXT: br label [[TMP16]]
+; IR-ITERATIVE: 16:
+; IR-ITERATIVE-NEXT: [[TMP17:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP15]], [[TMP14]] ]
+; IR-ITERATIVE-NEXT: [[TMP18:%.*]] = bitcast double [[TMP17]] to i64
+; IR-ITERATIVE-NEXT: [[TMP19:%.*]] = trunc i64 [[TMP18]] to i32
+; IR-ITERATIVE-NEXT: [[TMP20:%.*]] = lshr i64 [[TMP18]], 32
+; IR-ITERATIVE-NEXT: [[TMP21:%.*]] = trunc i64 [[TMP20]] to i32
+; IR-ITERATIVE-NEXT: [[TMP22:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP19]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP23:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP21]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP24:%.*]] = insertelement <2 x i32> poison, i32 [[TMP22]], i32 0
+; IR-ITERATIVE-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP24]], i32 [[TMP23]], i32 1
+; IR-ITERATIVE-NEXT: [[TMP26:%.*]] = bitcast <2 x i32> [[TMP25]] to double
+; IR-ITERATIVE-NEXT: [[TMP27:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP8]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP28:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL]], double [[TMP27]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP29:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[TMP26]], double [[TMP28]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br label [[TMP30]]
+; IR-ITERATIVE: 30:
+; IR-ITERATIVE-NEXT: [[TMP31:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP29]], [[TMP16]] ]
+; IR-ITERATIVE-NEXT: ret double [[TMP31]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP30:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-DPP-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-DPP: 14:
+; IR-DPP-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] monotonic, align 4
+; IR-DPP-NEXT: br label [[TMP16]]
+; IR-DPP: 16:
+; IR-DPP-NEXT: [[TMP17:%.*]] = phi double [ poison, [[TMP2]] ], [ [[TMP15]], [[TMP14]] ]
+; IR-DPP-NEXT: [[TMP18:%.*]] = bitcast double [[TMP17]] to i64
+; IR-DPP-NEXT: [[TMP19:%.*]] = trunc i64 [[TMP18]] to i32
+; IR-DPP-NEXT: [[TMP20:%.*]] = lshr i64 [[TMP18]], 32
+; IR-DPP-NEXT: [[TMP21:%.*]] = trunc i64 [[TMP20]] to i32
+; IR-DPP-NEXT: [[TMP22:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP19]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP23:%.*]] = call i32 @llvm.amdgcn.readfirstlane(i32 [[TMP21]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP24:%.*]] = insertelement <2 x i32> poison, i32 [[TMP22]], i32 0
+; IR-DPP-NEXT: [[TMP25:%.*]] = insertelement <2 x i32> [[TMP24]], i32 [[TMP23]], i32 1
+; IR-DPP-NEXT: [[TMP26:%.*]] = bitcast <2 x i32> [[TMP25]] to double
+; IR-DPP-NEXT: [[TMP27:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP8]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP28:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL]], double [[TMP27]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP29:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[TMP26]], double [[TMP28]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: br label [[TMP30]]
+; IR-DPP: 30:
+; IR-DPP-NEXT: [[TMP31:%.*]] = phi double [ poison, [[TMP0:%.*]] ], [ [[TMP29]], [[TMP16]] ]
+; IR-DPP-NEXT: ret double [[TMP31]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_uni_address_div_value_system_scope_strictfp(ptr addrspace(1) inreg %ptr, double %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_div_value_system_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_div_value_system_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_div_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr, double inreg %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_uni_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_uni_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_div_address_div_value_agent_scope_unsafe(ptr addrspace(1) %ptr, double %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_div_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_div_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_div_address_uni_value_one_as_scope_unsafe_structfp(ptr addrspace(1) %ptr, double inreg %val) #1 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_uni_value_one_as_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_uni_value_one_as_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_div_address_div_value_one_as_scope_unsafe_structfp(ptr addrspace(1) %ptr, double %val) #1 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_div_value_one_as_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_div_value_one_as_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fsub_double_div_address_uni_value_agent_scope_strictfp(ptr addrspace(1) %ptr, double inreg %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_div_address_uni_value_agent_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fsub_double_div_address_uni_value_agent_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fsub_double_div_address_div_value_agent_scope_strictfp(ptr addrspace(1) %ptr, double %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_div_address_div_value_agent_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fsub_double_div_address_div_value_agent_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fmin_double_div_address_uni_value_agent_scope(ptr addrspace(1) %ptr, double inreg %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fmin_double_div_address_uni_value_agent_scope(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fmin_double_div_address_uni_value_agent_scope(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fmin_double_div_address_div_value_agent_scope(ptr addrspace(1) %ptr, double %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fmin_double_div_address_div_value_agent_scope(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fmin_double_div_address_div_value_agent_scope(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic__fmax_double_div_address_uni_value_agent_scope_unsafe_structfp(ptr addrspace(1) %ptr, double inreg %val) #1{
+; IR-ITERATIVE-LABEL: @global_atomic__fmax_double_div_address_uni_value_agent_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic__fmax_double_div_address_uni_value_agent_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic__fmax_double_div_address_div_value_agent_scope_unsafe_structfp(ptr addrspace(1) %ptr, double %val) #1{
+; IR-ITERATIVE-LABEL: @global_atomic__fmax_double_div_address_div_value_agent_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic__fmax_double_div_address_div_value_agent_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_div_address_uni_value_system_scope_strictfp(ptr addrspace(1) %ptr, double inreg %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_uni_value_system_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_uni_value_system_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
+ ret double %result
+}
+
+define amdgpu_ps double @global_atomic_fadd_double_div_address_div_value_system_scope_strictfp(ptr addrspace(1) %ptr, double %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_div_value_system_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-ITERATIVE-NEXT: ret double [[RESULT]]
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_div_value_system_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-DPP-NEXT: ret double [[RESULT]]
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
+ ret double %result
+}
+
+
+
attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #1 = { strictfp "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #2 = { strictfp }
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_optimizer_fp_no_rtn.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_optimizer_fp_no_rtn.ll
index f87932b9361ed9..25d91e0213f8c5 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_optimizer_fp_no_rtn.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_optimizer_fp_no_rtn.ll
@@ -955,6 +955,543 @@ define amdgpu_ps void @global_atomic_fadd_div_address_div_value_system_scope_str
ret void
}
+define amdgpu_ps void @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double inreg %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]])
+; IR-ITERATIVE-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = uitofp i32 [[TMP10]] to double
+; IR-ITERATIVE-NEXT: [[TMP12:%.*]] = fmul double [[VAL:%.*]], [[TMP11]]
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-ITERATIVE: 14:
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 4
+; IR-ITERATIVE-NEXT: br label [[TMP16]]
+; IR-ITERATIVE: 16:
+; IR-ITERATIVE-NEXT: br label [[TMP17]]
+; IR-ITERATIVE: 17:
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
+; IR-DPP-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]])
+; IR-DPP-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-DPP-NEXT: [[TMP11:%.*]] = uitofp i32 [[TMP10]] to double
+; IR-DPP-NEXT: [[TMP12:%.*]] = fmul double [[VAL:%.*]], [[TMP11]]
+; IR-DPP-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-DPP: 14:
+; IR-DPP-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 4
+; IR-DPP-NEXT: br label [[TMP16]]
+; IR-DPP: 16:
+; IR-DPP-NEXT: br label [[TMP17]]
+; IR-DPP: 17:
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_uni_address_div_value_scope_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_div_value_scope_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_div_value_scope_agent_scope_unsafe(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp(ptr addrspace(1) inreg %ptr, double inreg %val) #1 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-ITERATIVE: 14:
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("one-as") monotonic, align 8
+; IR-ITERATIVE-NEXT: br label [[TMP16]]
+; IR-ITERATIVE: 16:
+; IR-ITERATIVE-NEXT: br label [[TMP17]]
+; IR-ITERATIVE: 17:
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-DPP-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-DPP: 14:
+; IR-DPP-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("one-as") monotonic, align 8
+; IR-DPP-NEXT: br label [[TMP16]]
+; IR-DPP: 16:
+; IR-DPP-NEXT: br label [[TMP17]]
+; IR-DPP: 17:
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp(ptr addrspace(1) inreg %ptr, double %val) #1 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(ptr addrspace(1) inreg %ptr, double inreg %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-ITERATIVE: 14:
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: br label [[TMP16]]
+; IR-ITERATIVE: 16:
+; IR-ITERATIVE-NEXT: br label [[TMP17]]
+; IR-ITERATIVE: 17:
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-DPP-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-DPP: 14:
+; IR-DPP-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: br label [[TMP16]]
+; IR-DPP: 16:
+; IR-DPP-NEXT: br label [[TMP17]]
+; IR-DPP: 17:
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fsub_double_uni_address_div_value_agent_scope_strictfp(ptr addrspace(1) inreg %ptr, double %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_uni_address_div_value_agent_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fsub_double_uni_address_div_value_agent_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double inreg %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
+; IR-ITERATIVE: 10:
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: br label [[TMP12]]
+; IR-ITERATIVE: 12:
+; IR-ITERATIVE-NEXT: br label [[TMP13]]
+; IR-ITERATIVE: 13:
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
+; IR-DPP-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
+; IR-DPP: 10:
+; IR-DPP-NEXT: [[TMP11:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: br label [[TMP12]]
+; IR-DPP: 12:
+; IR-DPP-NEXT: br label [[TMP13]]
+; IR-DPP: 13:
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe_structfp(ptr addrspace(1) inreg %ptr, double inreg %val) #1{
+; IR-ITERATIVE-LABEL: @global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
+; IR-ITERATIVE: 10:
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: br label [[TMP12]]
+; IR-ITERATIVE: 12:
+; IR-ITERATIVE-NEXT: br label [[TMP13]]
+; IR-ITERATIVE: 13:
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
+; IR-DPP: 10:
+; IR-DPP-NEXT: [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: br label [[TMP12]]
+; IR-DPP: 12:
+; IR-DPP-NEXT: br label [[TMP13]]
+; IR-DPP: 13:
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe_structfp(ptr addrspace(1) inreg %ptr, double %val) #1{
+; IR-ITERATIVE-LABEL: @global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(ptr addrspace(1) inreg %ptr, double inreg %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
+; IR-ITERATIVE-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
+; IR-ITERATIVE: 2:
+; IR-ITERATIVE-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-ITERATIVE-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-ITERATIVE-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-ITERATIVE-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-ITERATIVE-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
+; IR-ITERATIVE-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-ITERATIVE-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-ITERATIVE: 14:
+; IR-ITERATIVE-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] monotonic, align 4
+; IR-ITERATIVE-NEXT: br label [[TMP16]]
+; IR-ITERATIVE: 16:
+; IR-ITERATIVE-NEXT: br label [[TMP17]]
+; IR-ITERATIVE: 17:
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(
+; IR-DPP-NEXT: [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
+; IR-DPP-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
+; IR-DPP: 2:
+; IR-DPP-NEXT: [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; IR-DPP-NEXT: [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
+; IR-DPP-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
+; IR-DPP-NEXT: [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
+; IR-DPP-NEXT: [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
+; IR-DPP-NEXT: [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
+; IR-DPP-NEXT: br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
+; IR-DPP: 14:
+; IR-DPP-NEXT: [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] monotonic, align 4
+; IR-DPP-NEXT: br label [[TMP16]]
+; IR-DPP: 16:
+; IR-DPP-NEXT: br label [[TMP17]]
+; IR-DPP: 17:
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_uni_address_div_value_system_scope_strictfp(ptr addrspace(1) inreg %ptr, double %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_div_value_system_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_div_value_system_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_div_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr, double inreg %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_uni_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_uni_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_div_address_div_value_agent_scope_unsafe(ptr addrspace(1) %ptr, double %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_div_value_agent_scope_unsafe(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_div_value_agent_scope_unsafe(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_div_address_uni_value_one_as_scope_unsafe_structfp(ptr addrspace(1) %ptr, double inreg %val) #1 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_uni_value_one_as_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_uni_value_one_as_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_div_address_div_value_one_as_scope_unsafe_structfp(ptr addrspace(1) %ptr, double %val) #1 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_div_value_one_as_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_div_value_one_as_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fsub_double_div_address_uni_value_agent_scope_strictfp(ptr addrspace(1) %ptr, double inreg %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_div_address_uni_value_agent_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fsub_double_div_address_uni_value_agent_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fsub_double_div_address_div_value_agent_scope_strictfp(ptr addrspace(1) %ptr, double %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_div_address_div_value_agent_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fsub_double_div_address_div_value_agent_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fsub ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fmin_double_div_address_uni_value_agent_scope(ptr addrspace(1) %ptr, double inreg %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fmin_double_div_address_uni_value_agent_scope(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fmin_double_div_address_uni_value_agent_scope(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fmin_double_div_address_div_value_agent_scope(ptr addrspace(1) %ptr, double %val) #0 {
+; IR-ITERATIVE-LABEL: @global_atomic_fmin_double_div_address_div_value_agent_scope(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fmin_double_div_address_div_value_agent_scope(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fmax_double_div_address_uni_value_agent_scope_unsafe_structfp(ptr addrspace(1) %ptr, double inreg %val) #1{
+; IR-ITERATIVE-LABEL: @global_atomic_fmax_double_div_address_uni_value_agent_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fmax_double_div_address_uni_value_agent_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fmax_double_div_address_div_value_agent_scope_unsafe_structfp(ptr addrspace(1) %ptr, double %val) #1{
+; IR-ITERATIVE-LABEL: @global_atomic_fmax_double_div_address_div_value_agent_scope_unsafe_structfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fmax_double_div_address_div_value_agent_scope_unsafe_structfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_div_address_uni_value_system_scope_strictfp(ptr addrspace(1) %ptr, double inreg %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_uni_value_system_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_uni_value_system_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
+ ret void
+}
+
+define amdgpu_ps void @global_atomic_fadd_double_div_address_div_value_system_scope_strictfp(ptr addrspace(1) %ptr, double %val) #2 {
+; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_div_address_div_value_system_scope_strictfp(
+; IR-ITERATIVE-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-ITERATIVE-NEXT: ret void
+;
+; IR-DPP-LABEL: @global_atomic_fadd_double_div_address_div_value_system_scope_strictfp(
+; IR-DPP-NEXT: [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
+; IR-DPP-NEXT: ret void
+;
+ %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
+ ret void
+}
+
+
attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #1 = { strictfp "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #2 = { strictfp }
More information about the llvm-commits
mailing list