[llvm] e66b1b7 - [AMDGPU] Split fp min/max atomics test. NFC.
Jay Foad via llvm-commits
llvm-commits at lists.llvm.org
Wed Feb 23 07:04:25 PST 2022
Author: Jay Foad
Date: 2022-02-23T15:00:49Z
New Revision: e66b1b73854e43f7e822fbb3be9e9875e6780761
URL: https://github.com/llvm/llvm-project/commit/e66b1b73854e43f7e822fbb3be9e9875e6780761
DIFF: https://github.com/llvm/llvm-project/commit/e66b1b73854e43f7e822fbb3be9e9875e6780761.diff
LOG: [AMDGPU] Split fp min/max atomics test. NFC.
Split out f32 buffer, f64 buffer and image atomics. This just makes
it easier to test subtargets that only have some of these
instructions.
Differential Revision: https://reviews.llvm.org/D120407
Added:
llvm/test/CodeGen/AMDGPU/fp-min-max-buffer-atomics.ll
llvm/test/CodeGen/AMDGPU/fp-min-max-image-atomics.ll
llvm/test/CodeGen/AMDGPU/fp64-min-max-buffer-atomics.ll
Modified:
Removed:
llvm/test/CodeGen/AMDGPU/fp-min-max-atomics.ll
################################################################################
diff --git a/llvm/test/CodeGen/AMDGPU/fp-min-max-atomics.ll b/llvm/test/CodeGen/AMDGPU/fp-min-max-atomics.ll
deleted file mode 100644
index b27943195857..000000000000
--- a/llvm/test/CodeGen/AMDGPU/fp-min-max-atomics.ll
+++ /dev/null
@@ -1,1244 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=SI
-; RUN: llc < %s -march=amdgcn -mcpu=hawaii -verify-machineinstrs | FileCheck %s -check-prefix=GFX7
-; RUN: llc < %s -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=GFX10
-; RUN: llc < %s -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s -check-prefix=GFX1030
-
-; RUN: llc < %s -global-isel -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=G_SI
-; RUN: llc < %s -global-isel -march=amdgcn -mcpu=hawaii -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX7
-; RUN: llc < %s -global-isel -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX10
-; RUN: llc < %s -global-isel -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX1030
-
-declare double @llvm.amdgcn.raw.buffer.atomic.fmin.f64(double, <4 x i32>, i32, i32, i32 immarg)
-declare double @llvm.amdgcn.raw.buffer.atomic.fmax.f64(double, <4 x i32>, i32, i32, i32 immarg)
-
-declare float @llvm.amdgcn.raw.buffer.atomic.fmin.f32(float, <4 x i32>, i32, i32, i32 immarg)
-declare float @llvm.amdgcn.raw.buffer.atomic.fmax.f32(float, <4 x i32>, i32, i32, i32 immarg)
-
-declare float @llvm.amdgcn.image.atomic.fmin.1d.f32.f32(float, i32, <8 x i32>, i32, i32)
-declare float @llvm.amdgcn.image.atomic.fmax.1d.f32.f32(float, i32, <8 x i32>, i32, i32)
-
-
-define amdgpu_kernel void @raw_buffer_atomic_min_noret_f32(<4 x i32> inreg %rsrc, float %data, i32 %vindex) {
-; SI-LABEL: raw_buffer_atomic_min_noret_f32:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; SI-NEXT: s_waitcnt lgkmcnt(0)
-; SI-NEXT: v_mov_b32_e32 v0, s4
-; SI-NEXT: v_mov_b32_e32 v1, s5
-; SI-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_min_noret_f32:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; GFX7-NEXT: v_mov_b32_e32 v0, s4
-; GFX7-NEXT: v_mov_b32_e32 v1, s5
-; GFX7-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_min_noret_f32:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: s_clause 0x1
-; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-NEXT: v_mov_b32_e32 v0, s2
-; GFX10-NEXT: v_mov_b32_e32 v1, s3
-; GFX10-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 0 offen
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_min_noret_f32:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: s_clause 0x1
-; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_min_noret_f32:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_SI-NEXT: s_waitcnt lgkmcnt(0)
-; G_SI-NEXT: v_mov_b32_e32 v0, s4
-; G_SI-NEXT: v_mov_b32_e32 v1, s5
-; G_SI-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_min_noret_f32:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX7-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_min_noret_f32:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: s_clause 0x1
-; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX10-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 0 offen glc
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_min_noret_f32:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: s_clause 0x1
-; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmin.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
- ret void
-}
-
-define amdgpu_kernel void @raw_buffer_atomic_min_noret_f64(<4 x i32> inreg %rsrc, double %data, i32 %vindex) {
-; SI-LABEL: raw_buffer_atomic_min_noret_f64:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; SI-NEXT: s_load_dword s6, s[0:1], 0xf
-; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; SI-NEXT: s_waitcnt lgkmcnt(0)
-; SI-NEXT: v_mov_b32_e32 v0, s4
-; SI-NEXT: v_mov_b32_e32 v1, s5
-; SI-NEXT: v_mov_b32_e32 v2, s6
-; SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_min_noret_f64:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; GFX7-NEXT: s_load_dword s6, s[0:1], 0xf
-; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; GFX7-NEXT: v_mov_b32_e32 v0, s4
-; GFX7-NEXT: v_mov_b32_e32 v1, s5
-; GFX7-NEXT: v_mov_b32_e32 v2, s6
-; GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_min_noret_f64:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: s_clause 0x2
-; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX10-NEXT: s_load_dword s8, s[0:1], 0x3c
-; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-NEXT: v_mov_b32_e32 v0, s2
-; GFX10-NEXT: v_mov_b32_e32 v1, s3
-; GFX10-NEXT: v_mov_b32_e32 v2, s8
-; GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_min_noret_f64:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: s_clause 0x2
-; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; GFX1030-NEXT: s_load_dword s6, s[0:1], 0x3c
-; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; GFX1030-NEXT: v_mov_b32_e32 v2, s6
-; GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_min_noret_f64:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_SI-NEXT: s_load_dword s6, s[0:1], 0xf
-; G_SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_SI-NEXT: s_waitcnt lgkmcnt(0)
-; G_SI-NEXT: v_mov_b32_e32 v0, s4
-; G_SI-NEXT: v_mov_b32_e32 v1, s5
-; G_SI-NEXT: v_mov_b32_e32 v2, s6
-; G_SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_min_noret_f64:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_GFX7-NEXT: s_load_dword s6, s[0:1], 0xf
-; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX7-NEXT: v_mov_b32_e32 v2, s6
-; G_GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_min_noret_f64:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: s_clause 0x2
-; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX10-NEXT: s_load_dword s8, s[0:1], 0x3c
-; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX10-NEXT: v_mov_b32_e32 v2, s8
-; G_GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen glc
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_min_noret_f64:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: s_clause 0x2
-; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; G_GFX1030-NEXT: s_load_dword s6, s[0:1], 0x3c
-; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX1030-NEXT: v_mov_b32_e32 v2, s6
-; G_GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmin.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
- ret void
-}
-
-define amdgpu_ps void @raw_buffer_atomic_min_rtn_f32(<4 x i32> inreg %rsrc, float %data, i32 %vindex) {
-; SI-LABEL: raw_buffer_atomic_min_rtn_f32:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; SI-NEXT: s_mov_b32 s3, 0xf000
-; SI-NEXT: s_mov_b32 s2, -1
-; SI-NEXT: s_waitcnt vmcnt(0)
-; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_min_rtn_f32:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; GFX7-NEXT: s_mov_b32 s3, 0xf000
-; GFX7-NEXT: s_mov_b32 s2, -1
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_min_rtn_f32:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: global_store_dword v[0:1], v0, off
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_min_rtn_f32:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: global_store_dword v[0:1], v0, off
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_min_rtn_f32:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; G_SI-NEXT: s_mov_b32 s2, -1
-; G_SI-NEXT: s_mov_b32 s3, 0xf000
-; G_SI-NEXT: s_waitcnt vmcnt(0)
-; G_SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_min_rtn_f32:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; G_GFX7-NEXT: s_mov_b32 s2, -1
-; G_GFX7-NEXT: s_mov_b32 s3, 0xf000
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_min_rtn_f32:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: global_store_dword v[0:1], v0, off
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_min_rtn_f32:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: global_store_dword v[0:1], v0, off
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmin.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
- store float %ret, float addrspace(1)* undef
- ret void
-}
-
-define amdgpu_ps void @raw_buffer_atomic_min_rtn_f64(<4 x i32> inreg %rsrc, double %data, i32 %vindex) {
-; SI-LABEL: raw_buffer_atomic_min_rtn_f64:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; SI-NEXT: s_mov_b32 m0, -1
-; SI-NEXT: s_waitcnt vmcnt(0)
-; SI-NEXT: ds_write_b64 v0, v[0:1]
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_min_rtn_f64:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; GFX7-NEXT: s_mov_b32 m0, -1
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: ds_write_b64 v0, v[0:1]
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_min_rtn_f64:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: ds_write_b64 v0, v[0:1]
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_min_rtn_f64:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: ds_write_b64 v0, v[0:1]
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_min_rtn_f64:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_SI-NEXT: s_mov_b32 m0, -1
-; G_SI-NEXT: s_waitcnt vmcnt(0)
-; G_SI-NEXT: ds_write_b64 v0, v[0:1]
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_min_rtn_f64:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX7-NEXT: s_mov_b32 m0, -1
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: ds_write_b64 v0, v[0:1]
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_min_rtn_f64:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: ds_write_b64 v0, v[0:1]
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_min_rtn_f64:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: ds_write_b64 v0, v[0:1]
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmin.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
- store double %ret, double addrspace(3)* undef
- ret void
-}
-
-define amdgpu_kernel void @raw_buffer_atomic_min_rtn_f32_off4_slc(<4 x i32> inreg %rsrc, float %data, i32 %vindex, float addrspace(3)* %out) {
-; SI-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; SI-NEXT: s_load_dword s0, s[0:1], 0xf
-; SI-NEXT: s_mov_b32 m0, -1
-; SI-NEXT: s_waitcnt lgkmcnt(0)
-; SI-NEXT: v_mov_b32_e32 v0, s2
-; SI-NEXT: v_mov_b32_e32 v1, s3
-; SI-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
-; SI-NEXT: v_mov_b32_e32 v1, s0
-; SI-NEXT: s_waitcnt vmcnt(0)
-; SI-NEXT: ds_write_b32 v1, v0
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX7-NEXT: s_load_dword s0, s[0:1], 0xf
-; GFX7-NEXT: s_mov_b32 m0, -1
-; GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; GFX7-NEXT: v_mov_b32_e32 v0, s2
-; GFX7-NEXT: v_mov_b32_e32 v1, s3
-; GFX7-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
-; GFX7-NEXT: v_mov_b32_e32 v1, s0
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: ds_write_b32 v1, v0
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: s_clause 0x1
-; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-NEXT: v_mov_b32_e32 v0, s2
-; GFX10-NEXT: v_mov_b32_e32 v1, s3
-; GFX10-NEXT: s_load_dword s0, s[0:1], 0x3c
-; GFX10-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
-; GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-NEXT: v_mov_b32_e32 v1, s0
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: ds_write_b32 v1, v0
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: s_clause 0x2
-; GFX1030-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX1030-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX1030-NEXT: s_load_dword s0, s[0:1], 0x3c
-; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; GFX1030-NEXT: v_mov_b32_e32 v0, s2
-; GFX1030-NEXT: v_mov_b32_e32 v1, s3
-; GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
-; GFX1030-NEXT: v_mov_b32_e32 v1, s0
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: ds_write_b32 v1, v0
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; G_SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; G_SI-NEXT: s_load_dword s0, s[0:1], 0xf
-; G_SI-NEXT: s_mov_b32 m0, -1
-; G_SI-NEXT: s_waitcnt lgkmcnt(0)
-; G_SI-NEXT: v_mov_b32_e32 v0, s2
-; G_SI-NEXT: v_mov_b32_e32 v1, s3
-; G_SI-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
-; G_SI-NEXT: v_mov_b32_e32 v1, s0
-; G_SI-NEXT: s_waitcnt vmcnt(0)
-; G_SI-NEXT: ds_write_b32 v1, v0
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; G_GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; G_GFX7-NEXT: s_load_dword s0, s[0:1], 0xf
-; G_GFX7-NEXT: s_mov_b32 m0, -1
-; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX7-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX7-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX7-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
-; G_GFX7-NEXT: v_mov_b32_e32 v1, s0
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: ds_write_b32 v1, v0
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: s_clause 0x1
-; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX10-NEXT: s_load_dword s0, s[0:1], 0x3c
-; G_GFX10-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
-; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX10-NEXT: v_mov_b32_e32 v1, s0
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: ds_write_b32 v1, v0
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: s_clause 0x2
-; G_GFX1030-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX1030-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX1030-NEXT: s_load_dword s0, s[0:1], 0x3c
-; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX1030-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, s0
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: ds_write_b32 v1, v0
-; G_GFX1030-NEXT: s_endpgm
-; GFX1010-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
-main_body:
- %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmin.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 4, i32 2)
- store float %ret, float addrspace(3)* %out, align 8
- ret void
-}
-
-define amdgpu_ps void @raw_buffer_atomic_min_rtn_f64_off4_slc(<4 x i32> inreg %rsrc, double %data, i32 %vindex, double addrspace(3)* %out) {
-; SI-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; SI-NEXT: s_mov_b32 m0, -1
-; SI-NEXT: s_waitcnt vmcnt(0)
-; SI-NEXT: ds_write_b64 v3, v[0:1]
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; GFX7-NEXT: s_mov_b32 m0, -1
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: ds_write_b64 v3, v[0:1]
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: ds_write_b64 v3, v[0:1]
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: ds_write_b64 v3, v[0:1]
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; G_SI-NEXT: s_mov_b32 m0, -1
-; G_SI-NEXT: s_waitcnt vmcnt(0)
-; G_SI-NEXT: ds_write_b64 v3, v[0:1]
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; G_GFX7-NEXT: s_mov_b32 m0, -1
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: ds_write_b64 v3, v[0:1]
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: ds_write_b64 v3, v[0:1]
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: ds_write_b64 v3, v[0:1]
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmin.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 4, i32 2)
- store double %ret, double addrspace(3)* %out, align 8
- ret void
-}
-
-define amdgpu_kernel void @raw_buffer_atomic_max_noret_f32(<4 x i32> inreg %rsrc, float %data, i32 %vindex) {
-; SI-LABEL: raw_buffer_atomic_max_noret_f32:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; SI-NEXT: s_waitcnt lgkmcnt(0)
-; SI-NEXT: v_mov_b32_e32 v0, s4
-; SI-NEXT: v_mov_b32_e32 v1, s5
-; SI-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_max_noret_f32:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; GFX7-NEXT: v_mov_b32_e32 v0, s4
-; GFX7-NEXT: v_mov_b32_e32 v1, s5
-; GFX7-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_max_noret_f32:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: s_clause 0x1
-; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-NEXT: v_mov_b32_e32 v0, s2
-; GFX10-NEXT: v_mov_b32_e32 v1, s3
-; GFX10-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 0 offen
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_max_noret_f32:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: s_clause 0x1
-; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_max_noret_f32:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_SI-NEXT: s_waitcnt lgkmcnt(0)
-; G_SI-NEXT: v_mov_b32_e32 v0, s4
-; G_SI-NEXT: v_mov_b32_e32 v1, s5
-; G_SI-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_max_noret_f32:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX7-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_max_noret_f32:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: s_clause 0x1
-; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX10-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 0 offen glc
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_max_noret_f32:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: s_clause 0x1
-; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmax.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
- ret void
-}
-
-define amdgpu_kernel void @raw_buffer_atomic_max_noret_f64(<4 x i32> inreg %rsrc, double %data, i32 %vindex) {
-; SI-LABEL: raw_buffer_atomic_max_noret_f64:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; SI-NEXT: s_load_dword s6, s[0:1], 0xf
-; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; SI-NEXT: s_waitcnt lgkmcnt(0)
-; SI-NEXT: v_mov_b32_e32 v0, s4
-; SI-NEXT: v_mov_b32_e32 v1, s5
-; SI-NEXT: v_mov_b32_e32 v2, s6
-; SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_max_noret_f64:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; GFX7-NEXT: s_load_dword s6, s[0:1], 0xf
-; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; GFX7-NEXT: v_mov_b32_e32 v0, s4
-; GFX7-NEXT: v_mov_b32_e32 v1, s5
-; GFX7-NEXT: v_mov_b32_e32 v2, s6
-; GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_max_noret_f64:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: s_clause 0x2
-; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX10-NEXT: s_load_dword s8, s[0:1], 0x3c
-; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-NEXT: v_mov_b32_e32 v0, s2
-; GFX10-NEXT: v_mov_b32_e32 v1, s3
-; GFX10-NEXT: v_mov_b32_e32 v2, s8
-; GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_max_noret_f64:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: s_clause 0x2
-; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; GFX1030-NEXT: s_load_dword s6, s[0:1], 0x3c
-; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; GFX1030-NEXT: v_mov_b32_e32 v2, s6
-; GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_max_noret_f64:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_SI-NEXT: s_load_dword s6, s[0:1], 0xf
-; G_SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_SI-NEXT: s_waitcnt lgkmcnt(0)
-; G_SI-NEXT: v_mov_b32_e32 v0, s4
-; G_SI-NEXT: v_mov_b32_e32 v1, s5
-; G_SI-NEXT: v_mov_b32_e32 v2, s6
-; G_SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_max_noret_f64:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_GFX7-NEXT: s_load_dword s6, s[0:1], 0xf
-; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX7-NEXT: v_mov_b32_e32 v2, s6
-; G_GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_max_noret_f64:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: s_clause 0x2
-; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX10-NEXT: s_load_dword s8, s[0:1], 0x3c
-; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX10-NEXT: v_mov_b32_e32 v2, s8
-; G_GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen glc
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_max_noret_f64:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: s_clause 0x2
-; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; G_GFX1030-NEXT: s_load_dword s6, s[0:1], 0x3c
-; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX1030-NEXT: v_mov_b32_e32 v2, s6
-; G_GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmax.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
- ret void
-}
-
-define amdgpu_ps void @raw_buffer_atomic_max_rtn_f32(<4 x i32> inreg %rsrc, float %data, i32 %vindex) {
-; SI-LABEL: raw_buffer_atomic_max_rtn_f32:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; SI-NEXT: s_mov_b32 s3, 0xf000
-; SI-NEXT: s_mov_b32 s2, -1
-; SI-NEXT: s_waitcnt vmcnt(0)
-; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_max_rtn_f32:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; GFX7-NEXT: s_mov_b32 s3, 0xf000
-; GFX7-NEXT: s_mov_b32 s2, -1
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_max_rtn_f32:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: global_store_dword v[0:1], v0, off
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_max_rtn_f32:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: global_store_dword v[0:1], v0, off
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_max_rtn_f32:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; G_SI-NEXT: s_mov_b32 s2, -1
-; G_SI-NEXT: s_mov_b32 s3, 0xf000
-; G_SI-NEXT: s_waitcnt vmcnt(0)
-; G_SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_max_rtn_f32:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; G_GFX7-NEXT: s_mov_b32 s2, -1
-; G_GFX7-NEXT: s_mov_b32 s3, 0xf000
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_max_rtn_f32:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: global_store_dword v[0:1], v0, off
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_max_rtn_f32:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: global_store_dword v[0:1], v0, off
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmax.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
- store float %ret, float addrspace(1)* undef
- ret void
-}
-
-define amdgpu_ps void @raw_buffer_atomic_max_rtn_f64(<4 x i32> inreg %rsrc, double %data, i32 %vindex) {
-; SI-LABEL: raw_buffer_atomic_max_rtn_f64:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; SI-NEXT: s_mov_b32 m0, -1
-; SI-NEXT: s_waitcnt vmcnt(0)
-; SI-NEXT: ds_write_b64 v0, v[0:1]
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_max_rtn_f64:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; GFX7-NEXT: s_mov_b32 m0, -1
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: ds_write_b64 v0, v[0:1]
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_max_rtn_f64:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: ds_write_b64 v0, v[0:1]
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_max_rtn_f64:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: ds_write_b64 v0, v[0:1]
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_max_rtn_f64:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_SI-NEXT: s_mov_b32 m0, -1
-; G_SI-NEXT: s_waitcnt vmcnt(0)
-; G_SI-NEXT: ds_write_b64 v0, v[0:1]
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_max_rtn_f64:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX7-NEXT: s_mov_b32 m0, -1
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: ds_write_b64 v0, v[0:1]
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_max_rtn_f64:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: ds_write_b64 v0, v[0:1]
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_max_rtn_f64:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: ds_write_b64 v0, v[0:1]
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmax.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
- store double %ret, double addrspace(3)* undef
- ret void
-}
-
-define amdgpu_kernel void @raw_buffer_atomic_max_rtn_f32_off4_slc(<4 x i32> inreg %rsrc, float %data, i32 %vindex, float addrspace(1)* %out) {
-; SI-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; SI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xf
-; SI-NEXT: s_waitcnt lgkmcnt(0)
-; SI-NEXT: v_mov_b32_e32 v0, s2
-; SI-NEXT: v_mov_b32_e32 v1, s3
-; SI-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
-; SI-NEXT: s_mov_b32 s3, 0xf000
-; SI-NEXT: s_mov_b32 s2, -1
-; SI-NEXT: s_waitcnt vmcnt(0)
-; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xf
-; GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; GFX7-NEXT: v_mov_b32_e32 v0, s2
-; GFX7-NEXT: v_mov_b32_e32 v1, s3
-; GFX7-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
-; GFX7-NEXT: s_mov_b32 s3, 0xf000
-; GFX7-NEXT: s_mov_b32 s2, -1
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: s_clause 0x1
-; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-NEXT: v_mov_b32_e32 v0, s2
-; GFX10-NEXT: v_mov_b32_e32 v1, s3
-; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x3c
-; GFX10-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
-; GFX10-NEXT: v_mov_b32_e32 v1, 0
-; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX10-NEXT: global_store_dword v1, v0, s[0:1]
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: s_clause 0x2
-; GFX1030-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX1030-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX1030-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x3c
-; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; GFX1030-NEXT: v_mov_b32_e32 v0, s2
-; GFX1030-NEXT: v_mov_b32_e32 v1, s3
-; GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
-; GFX1030-NEXT: v_mov_b32_e32 v1, 0
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: global_store_dword v1, v0, s[0:1]
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; G_SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; G_SI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xf
-; G_SI-NEXT: s_waitcnt lgkmcnt(0)
-; G_SI-NEXT: v_mov_b32_e32 v0, s2
-; G_SI-NEXT: v_mov_b32_e32 v1, s3
-; G_SI-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
-; G_SI-NEXT: s_mov_b32 s2, -1
-; G_SI-NEXT: s_mov_b32 s3, 0xf000
-; G_SI-NEXT: s_waitcnt vmcnt(0)
-; G_SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; G_GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; G_GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xf
-; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX7-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX7-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX7-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
-; G_GFX7-NEXT: s_mov_b32 s2, -1
-; G_GFX7-NEXT: s_mov_b32 s3, 0xf000
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: s_clause 0x1
-; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x3c
-; G_GFX10-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
-; G_GFX10-NEXT: v_mov_b32_e32 v1, 0
-; G_GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; G_GFX10-NEXT: global_store_dword v1, v0, s[0:1]
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: s_clause 0x2
-; G_GFX1030-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX1030-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX1030-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x3c
-; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX1030-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, 0
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: global_store_dword v1, v0, s[0:1]
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmax.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 4, i32 2)
- store float %ret, float addrspace(1)* %out, align 8
- ret void
-}
-
-define amdgpu_kernel void @raw_buffer_atomic_max_rtn_f64_off4_slc(<4 x i32> inreg %rsrc, double %data, i32 %vindex, double addrspace(3)* %out) {
-; SI-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; SI-NEXT: s_load_dword s8, s[0:1], 0xf
-; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; SI-NEXT: s_load_dword s0, s[0:1], 0x10
-; SI-NEXT: s_mov_b32 m0, -1
-; SI-NEXT: s_waitcnt lgkmcnt(0)
-; SI-NEXT: v_mov_b32_e32 v0, s2
-; SI-NEXT: v_mov_b32_e32 v1, s3
-; SI-NEXT: v_mov_b32_e32 v2, s8
-; SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 4 offen glc slc
-; SI-NEXT: v_mov_b32_e32 v2, s0
-; SI-NEXT: s_waitcnt vmcnt(0)
-; SI-NEXT: ds_write_b64 v2, v[0:1]
-; SI-NEXT: s_endpgm
-;
-; GFX7-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; GFX7-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0xf
-; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GFX7-NEXT: s_mov_b32 m0, -1
-; GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; GFX7-NEXT: v_mov_b32_e32 v0, s4
-; GFX7-NEXT: v_mov_b32_e32 v1, s5
-; GFX7-NEXT: v_mov_b32_e32 v2, s6
-; GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; GFX7-NEXT: v_mov_b32_e32 v2, s7
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: ds_write_b64 v2, v[0:1]
-; GFX7-NEXT: s_endpgm
-;
-; GFX10-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: s_clause 0x2
-; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; GFX10-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x3c
-; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-NEXT: v_mov_b32_e32 v0, s2
-; GFX10-NEXT: v_mov_b32_e32 v1, s3
-; GFX10-NEXT: v_mov_b32_e32 v2, s8
-; GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 4 offen glc slc
-; GFX10-NEXT: v_mov_b32_e32 v2, s9
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: ds_write_b64 v2, v[0:1]
-; GFX10-NEXT: s_endpgm
-;
-; GFX1030-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: s_clause 0x2
-; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; GFX1030-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x3c
-; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; GFX1030-NEXT: v_mov_b32_e32 v2, s6
-; GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; GFX1030-NEXT: v_mov_b32_e32 v2, s7
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: ds_write_b64 v2, v[0:1]
-; GFX1030-NEXT: s_endpgm
-;
-; G_SI-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
-; G_SI-NEXT: s_load_dword s8, s[0:1], 0xf
-; G_SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; G_SI-NEXT: s_load_dword s0, s[0:1], 0x10
-; G_SI-NEXT: s_mov_b32 m0, -1
-; G_SI-NEXT: s_waitcnt lgkmcnt(0)
-; G_SI-NEXT: v_mov_b32_e32 v0, s2
-; G_SI-NEXT: v_mov_b32_e32 v1, s3
-; G_SI-NEXT: v_mov_b32_e32 v2, s8
-; G_SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 4 offen glc slc
-; G_SI-NEXT: v_mov_b32_e32 v2, s0
-; G_SI-NEXT: s_waitcnt vmcnt(0)
-; G_SI-NEXT: ds_write_b64 v2, v[0:1]
-; G_SI-NEXT: s_endpgm
-;
-; G_GFX7-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
-; G_GFX7-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0xf
-; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
-; G_GFX7-NEXT: s_mov_b32 m0, -1
-; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX7-NEXT: v_mov_b32_e32 v2, s6
-; G_GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; G_GFX7-NEXT: v_mov_b32_e32 v2, s7
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: ds_write_b64 v2, v[0:1]
-; G_GFX7-NEXT: s_endpgm
-;
-; G_GFX10-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: s_clause 0x2
-; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
-; G_GFX10-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x3c
-; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
-; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
-; G_GFX10-NEXT: v_mov_b32_e32 v2, s8
-; G_GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 4 offen glc slc
-; G_GFX10-NEXT: v_mov_b32_e32 v2, s9
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: ds_write_b64 v2, v[0:1]
-; G_GFX10-NEXT: s_endpgm
-;
-; G_GFX1030-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: s_clause 0x2
-; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
-; G_GFX1030-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x3c
-; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
-; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
-; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
-; G_GFX1030-NEXT: v_mov_b32_e32 v2, s6
-; G_GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 4 offen glc slc
-; G_GFX1030-NEXT: v_mov_b32_e32 v2, s7
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: ds_write_b64 v2, v[0:1]
-; G_GFX1030-NEXT: s_endpgm
-main_body:
- %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmax.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 4, i32 2)
- store double %ret, double addrspace(3)* %out, align 8
- ret void
-}
-
-define amdgpu_ps float @atomic_fmin_1d(<8 x i32> inreg %rsrc, float %data, i32 %s) {
-; SI-LABEL: atomic_fmin_1d:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 unorm glc
-; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
-; SI-NEXT: ; return to shader part epilog
-;
-; GFX7-LABEL: atomic_fmin_1d:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 unorm glc
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: ; return to shader part epilog
-;
-; GFX10-LABEL: atomic_fmin_1d:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: ; return to shader part epilog
-;
-; GFX1030-LABEL: atomic_fmin_1d:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: ; return to shader part epilog
-;
-; G_SI-LABEL: atomic_fmin_1d:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 unorm glc
-; G_SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
-; G_SI-NEXT: ; return to shader part epilog
-;
-; G_GFX7-LABEL: atomic_fmin_1d:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 unorm glc
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: ; return to shader part epilog
-;
-; G_GFX10-LABEL: atomic_fmin_1d:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: ; return to shader part epilog
-;
-; G_GFX1030-LABEL: atomic_fmin_1d:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: ; return to shader part epilog
-main_body:
- %v = call float @llvm.amdgcn.image.atomic.fmin.1d.f32.f32(float %data, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
- ret float %v
-}
-
-define amdgpu_ps float @atomic_fmax_1d(<8 x i32> inreg %rsrc, float %data, i32 %s) {
-; SI-LABEL: atomic_fmax_1d:
-; SI: ; %bb.0: ; %main_body
-; SI-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 unorm glc
-; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
-; SI-NEXT: ; return to shader part epilog
-;
-; GFX7-LABEL: atomic_fmax_1d:
-; GFX7: ; %bb.0: ; %main_body
-; GFX7-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 unorm glc
-; GFX7-NEXT: s_waitcnt vmcnt(0)
-; GFX7-NEXT: ; return to shader part epilog
-;
-; GFX10-LABEL: atomic_fmax_1d:
-; GFX10: ; %bb.0: ; %main_body
-; GFX10-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
-; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: ; return to shader part epilog
-;
-; GFX1030-LABEL: atomic_fmax_1d:
-; GFX1030: ; %bb.0: ; %main_body
-; GFX1030-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
-; GFX1030-NEXT: s_waitcnt vmcnt(0)
-; GFX1030-NEXT: ; return to shader part epilog
-;
-; G_SI-LABEL: atomic_fmax_1d:
-; G_SI: ; %bb.0: ; %main_body
-; G_SI-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 unorm glc
-; G_SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
-; G_SI-NEXT: ; return to shader part epilog
-;
-; G_GFX7-LABEL: atomic_fmax_1d:
-; G_GFX7: ; %bb.0: ; %main_body
-; G_GFX7-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 unorm glc
-; G_GFX7-NEXT: s_waitcnt vmcnt(0)
-; G_GFX7-NEXT: ; return to shader part epilog
-;
-; G_GFX10-LABEL: atomic_fmax_1d:
-; G_GFX10: ; %bb.0: ; %main_body
-; G_GFX10-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
-; G_GFX10-NEXT: s_waitcnt vmcnt(0)
-; G_GFX10-NEXT: ; return to shader part epilog
-;
-; G_GFX1030-LABEL: atomic_fmax_1d:
-; G_GFX1030: ; %bb.0: ; %main_body
-; G_GFX1030-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
-; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
-; G_GFX1030-NEXT: ; return to shader part epilog
-main_body:
- %v = call float @llvm.amdgcn.image.atomic.fmax.1d.f32.f32(float %data, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
- ret float %v
-}
-
diff --git a/llvm/test/CodeGen/AMDGPU/fp-min-max-buffer-atomics.ll b/llvm/test/CodeGen/AMDGPU/fp-min-max-buffer-atomics.ll
new file mode 100644
index 000000000000..2e63b837bd76
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/fp-min-max-buffer-atomics.ll
@@ -0,0 +1,587 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=SI
+; RUN: llc < %s -march=amdgcn -mcpu=hawaii -verify-machineinstrs | FileCheck %s -check-prefix=GFX7
+; RUN: llc < %s -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=GFX10
+; RUN: llc < %s -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s -check-prefix=GFX1030
+
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=G_SI
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=hawaii -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX7
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX10
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX1030
+
+declare float @llvm.amdgcn.raw.buffer.atomic.fmin.f32(float, <4 x i32>, i32, i32, i32 immarg)
+declare float @llvm.amdgcn.raw.buffer.atomic.fmax.f32(float, <4 x i32>, i32, i32, i32 immarg)
+
+
+define amdgpu_kernel void @raw_buffer_atomic_min_noret_f32(<4 x i32> inreg %rsrc, float %data, i32 %vindex) {
+; SI-LABEL: raw_buffer_atomic_min_noret_f32:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT: s_waitcnt lgkmcnt(0)
+; SI-NEXT: v_mov_b32_e32 v0, s4
+; SI-NEXT: v_mov_b32_e32 v1, s5
+; SI-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_min_noret_f32:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: v_mov_b32_e32 v1, s5
+; GFX7-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_min_noret_f32:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: s_clause 0x1
+; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 0 offen
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_min_noret_f32:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: s_clause 0x1
+; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_min_noret_f32:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_SI-NEXT: s_waitcnt lgkmcnt(0)
+; G_SI-NEXT: v_mov_b32_e32 v0, s4
+; G_SI-NEXT: v_mov_b32_e32 v1, s5
+; G_SI-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_min_noret_f32:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX7-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_min_noret_f32:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: s_clause 0x1
+; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX10-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 0 offen glc
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_min_noret_f32:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: s_clause 0x1
+; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmin.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
+ ret void
+}
+
+define amdgpu_ps void @raw_buffer_atomic_min_rtn_f32(<4 x i32> inreg %rsrc, float %data, i32 %vindex) {
+; SI-LABEL: raw_buffer_atomic_min_rtn_f32:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; SI-NEXT: s_mov_b32 s3, 0xf000
+; SI-NEXT: s_mov_b32 s2, -1
+; SI-NEXT: s_waitcnt vmcnt(0)
+; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_min_rtn_f32:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; GFX7-NEXT: s_mov_b32 s3, 0xf000
+; GFX7-NEXT: s_mov_b32 s2, -1
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_min_rtn_f32:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: global_store_dword v[0:1], v0, off
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_min_rtn_f32:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: global_store_dword v[0:1], v0, off
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_min_rtn_f32:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; G_SI-NEXT: s_mov_b32 s2, -1
+; G_SI-NEXT: s_mov_b32 s3, 0xf000
+; G_SI-NEXT: s_waitcnt vmcnt(0)
+; G_SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_min_rtn_f32:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; G_GFX7-NEXT: s_mov_b32 s2, -1
+; G_GFX7-NEXT: s_mov_b32 s3, 0xf000
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_min_rtn_f32:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: global_store_dword v[0:1], v0, off
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_min_rtn_f32:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[0:3], 0 offen glc
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: global_store_dword v[0:1], v0, off
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmin.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
+ store float %ret, float addrspace(1)* undef
+ ret void
+}
+
+define amdgpu_kernel void @raw_buffer_atomic_min_rtn_f32_off4_slc(<4 x i32> inreg %rsrc, float %data, i32 %vindex, float addrspace(3)* %out) {
+; SI-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT: s_load_dword s0, s[0:1], 0xf
+; SI-NEXT: s_mov_b32 m0, -1
+; SI-NEXT: s_waitcnt lgkmcnt(0)
+; SI-NEXT: v_mov_b32_e32 v0, s2
+; SI-NEXT: v_mov_b32_e32 v1, s3
+; SI-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
+; SI-NEXT: v_mov_b32_e32 v1, s0
+; SI-NEXT: s_waitcnt vmcnt(0)
+; SI-NEXT: ds_write_b32 v1, v0
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT: s_load_dword s0, s[0:1], 0xf
+; GFX7-NEXT: s_mov_b32 m0, -1
+; GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7-NEXT: v_mov_b32_e32 v0, s2
+; GFX7-NEXT: v_mov_b32_e32 v1, s3
+; GFX7-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
+; GFX7-NEXT: v_mov_b32_e32 v1, s0
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: ds_write_b32 v1, v0
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: s_clause 0x1
+; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: s_load_dword s0, s[0:1], 0x3c
+; GFX10-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: v_mov_b32_e32 v1, s0
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: ds_write_b32 v1, v0
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: s_clause 0x2
+; GFX1030-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX1030-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX1030-NEXT: s_load_dword s0, s[0:1], 0x3c
+; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1030-NEXT: v_mov_b32_e32 v0, s2
+; GFX1030-NEXT: v_mov_b32_e32 v1, s3
+; GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
+; GFX1030-NEXT: v_mov_b32_e32 v1, s0
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: ds_write_b32 v1, v0
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; G_SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; G_SI-NEXT: s_load_dword s0, s[0:1], 0xf
+; G_SI-NEXT: s_mov_b32 m0, -1
+; G_SI-NEXT: s_waitcnt lgkmcnt(0)
+; G_SI-NEXT: v_mov_b32_e32 v0, s2
+; G_SI-NEXT: v_mov_b32_e32 v1, s3
+; G_SI-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
+; G_SI-NEXT: v_mov_b32_e32 v1, s0
+; G_SI-NEXT: s_waitcnt vmcnt(0)
+; G_SI-NEXT: ds_write_b32 v1, v0
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; G_GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; G_GFX7-NEXT: s_load_dword s0, s[0:1], 0xf
+; G_GFX7-NEXT: s_mov_b32 m0, -1
+; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX7-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX7-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX7-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
+; G_GFX7-NEXT: v_mov_b32_e32 v1, s0
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: ds_write_b32 v1, v0
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: s_clause 0x1
+; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX10-NEXT: s_load_dword s0, s[0:1], 0x3c
+; G_GFX10-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
+; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX10-NEXT: v_mov_b32_e32 v1, s0
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: ds_write_b32 v1, v0
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: s_clause 0x2
+; G_GFX1030-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX1030-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX1030-NEXT: s_load_dword s0, s[0:1], 0x3c
+; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX1030-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX1030-NEXT: buffer_atomic_fmin v0, v1, s[4:7], 4 offen glc slc
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, s0
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: ds_write_b32 v1, v0
+; G_GFX1030-NEXT: s_endpgm
+; GFX1010-LABEL: raw_buffer_atomic_min_rtn_f32_off4_slc:
+main_body:
+ %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmin.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 4, i32 2)
+ store float %ret, float addrspace(3)* %out, align 8
+ ret void
+}
+
+define amdgpu_kernel void @raw_buffer_atomic_max_noret_f32(<4 x i32> inreg %rsrc, float %data, i32 %vindex) {
+; SI-LABEL: raw_buffer_atomic_max_noret_f32:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT: s_waitcnt lgkmcnt(0)
+; SI-NEXT: v_mov_b32_e32 v0, s4
+; SI-NEXT: v_mov_b32_e32 v1, s5
+; SI-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_max_noret_f32:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: v_mov_b32_e32 v1, s5
+; GFX7-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_max_noret_f32:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: s_clause 0x1
+; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 0 offen
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_max_noret_f32:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: s_clause 0x1
+; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_max_noret_f32:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_SI-NEXT: s_waitcnt lgkmcnt(0)
+; G_SI-NEXT: v_mov_b32_e32 v0, s4
+; G_SI-NEXT: v_mov_b32_e32 v1, s5
+; G_SI-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_max_noret_f32:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX7-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_max_noret_f32:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: s_clause 0x1
+; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX10-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 0 offen glc
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_max_noret_f32:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: s_clause 0x1
+; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmax.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
+ ret void
+}
+
+define amdgpu_ps void @raw_buffer_atomic_max_rtn_f32(<4 x i32> inreg %rsrc, float %data, i32 %vindex) {
+; SI-LABEL: raw_buffer_atomic_max_rtn_f32:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; SI-NEXT: s_mov_b32 s3, 0xf000
+; SI-NEXT: s_mov_b32 s2, -1
+; SI-NEXT: s_waitcnt vmcnt(0)
+; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_max_rtn_f32:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; GFX7-NEXT: s_mov_b32 s3, 0xf000
+; GFX7-NEXT: s_mov_b32 s2, -1
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_max_rtn_f32:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: global_store_dword v[0:1], v0, off
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_max_rtn_f32:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: global_store_dword v[0:1], v0, off
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_max_rtn_f32:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; G_SI-NEXT: s_mov_b32 s2, -1
+; G_SI-NEXT: s_mov_b32 s3, 0xf000
+; G_SI-NEXT: s_waitcnt vmcnt(0)
+; G_SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_max_rtn_f32:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; G_GFX7-NEXT: s_mov_b32 s2, -1
+; G_GFX7-NEXT: s_mov_b32 s3, 0xf000
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_max_rtn_f32:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: global_store_dword v[0:1], v0, off
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_max_rtn_f32:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[0:3], 0 offen glc
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: global_store_dword v[0:1], v0, off
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmax.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
+ store float %ret, float addrspace(1)* undef
+ ret void
+}
+
+define amdgpu_kernel void @raw_buffer_atomic_max_rtn_f32_off4_slc(<4 x i32> inreg %rsrc, float %data, i32 %vindex, float addrspace(1)* %out) {
+; SI-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xf
+; SI-NEXT: s_waitcnt lgkmcnt(0)
+; SI-NEXT: v_mov_b32_e32 v0, s2
+; SI-NEXT: v_mov_b32_e32 v1, s3
+; SI-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
+; SI-NEXT: s_mov_b32 s3, 0xf000
+; SI-NEXT: s_mov_b32 s2, -1
+; SI-NEXT: s_waitcnt vmcnt(0)
+; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xf
+; GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7-NEXT: v_mov_b32_e32 v0, s2
+; GFX7-NEXT: v_mov_b32_e32 v1, s3
+; GFX7-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
+; GFX7-NEXT: s_mov_b32 s3, 0xf000
+; GFX7-NEXT: s_mov_b32 s2, -1
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: s_clause 0x1
+; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x3c
+; GFX10-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
+; GFX10-NEXT: v_mov_b32_e32 v1, 0
+; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-NEXT: global_store_dword v1, v0, s[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: s_clause 0x2
+; GFX1030-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX1030-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX1030-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x3c
+; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1030-NEXT: v_mov_b32_e32 v0, s2
+; GFX1030-NEXT: v_mov_b32_e32 v1, s3
+; GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
+; GFX1030-NEXT: v_mov_b32_e32 v1, 0
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: global_store_dword v1, v0, s[0:1]
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; G_SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; G_SI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xf
+; G_SI-NEXT: s_waitcnt lgkmcnt(0)
+; G_SI-NEXT: v_mov_b32_e32 v0, s2
+; G_SI-NEXT: v_mov_b32_e32 v1, s3
+; G_SI-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
+; G_SI-NEXT: s_mov_b32 s2, -1
+; G_SI-NEXT: s_mov_b32 s3, 0xf000
+; G_SI-NEXT: s_waitcnt vmcnt(0)
+; G_SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; G_GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; G_GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xf
+; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX7-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX7-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX7-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
+; G_GFX7-NEXT: s_mov_b32 s2, -1
+; G_GFX7-NEXT: s_mov_b32 s3, 0xf000
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: s_clause 0x1
+; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x3c
+; G_GFX10-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
+; G_GFX10-NEXT: v_mov_b32_e32 v1, 0
+; G_GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; G_GFX10-NEXT: global_store_dword v1, v0, s[0:1]
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_max_rtn_f32_off4_slc:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: s_clause 0x2
+; G_GFX1030-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX1030-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX1030-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x3c
+; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX1030-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX1030-NEXT: buffer_atomic_fmax v0, v1, s[4:7], 4 offen glc slc
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, 0
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: global_store_dword v1, v0, s[0:1]
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call float @llvm.amdgcn.raw.buffer.atomic.fmax.f32(float %data, <4 x i32> %rsrc, i32 %vindex, i32 4, i32 2)
+ store float %ret, float addrspace(1)* %out, align 8
+ ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/fp-min-max-image-atomics.ll b/llvm/test/CodeGen/AMDGPU/fp-min-max-image-atomics.ll
new file mode 100644
index 000000000000..7ddeebe1b767
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/fp-min-max-image-atomics.ll
@@ -0,0 +1,120 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=SI
+; RUN: llc < %s -march=amdgcn -mcpu=hawaii -verify-machineinstrs | FileCheck %s -check-prefix=GFX7
+; RUN: llc < %s -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=GFX10
+; RUN: llc < %s -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s -check-prefix=GFX1030
+
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=G_SI
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=hawaii -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX7
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX10
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX1030
+
+declare float @llvm.amdgcn.image.atomic.fmin.1d.f32.f32(float, i32, <8 x i32>, i32, i32)
+declare float @llvm.amdgcn.image.atomic.fmax.1d.f32.f32(float, i32, <8 x i32>, i32, i32)
+
+
+define amdgpu_ps float @atomic_fmin_1d(<8 x i32> inreg %rsrc, float %data, i32 %s) {
+; SI-LABEL: atomic_fmin_1d:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 unorm glc
+; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
+; SI-NEXT: ; return to shader part epilog
+;
+; GFX7-LABEL: atomic_fmin_1d:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 unorm glc
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: ; return to shader part epilog
+;
+; GFX10-LABEL: atomic_fmin_1d:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: ; return to shader part epilog
+;
+; GFX1030-LABEL: atomic_fmin_1d:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: ; return to shader part epilog
+;
+; G_SI-LABEL: atomic_fmin_1d:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 unorm glc
+; G_SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
+; G_SI-NEXT: ; return to shader part epilog
+;
+; G_GFX7-LABEL: atomic_fmin_1d:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 unorm glc
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: ; return to shader part epilog
+;
+; G_GFX10-LABEL: atomic_fmin_1d:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: ; return to shader part epilog
+;
+; G_GFX1030-LABEL: atomic_fmin_1d:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: image_atomic_fmin v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: ; return to shader part epilog
+main_body:
+ %v = call float @llvm.amdgcn.image.atomic.fmin.1d.f32.f32(float %data, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
+ ret float %v
+}
+
+define amdgpu_ps float @atomic_fmax_1d(<8 x i32> inreg %rsrc, float %data, i32 %s) {
+; SI-LABEL: atomic_fmax_1d:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 unorm glc
+; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
+; SI-NEXT: ; return to shader part epilog
+;
+; GFX7-LABEL: atomic_fmax_1d:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 unorm glc
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: ; return to shader part epilog
+;
+; GFX10-LABEL: atomic_fmax_1d:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: ; return to shader part epilog
+;
+; GFX1030-LABEL: atomic_fmax_1d:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: ; return to shader part epilog
+;
+; G_SI-LABEL: atomic_fmax_1d:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 unorm glc
+; G_SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
+; G_SI-NEXT: ; return to shader part epilog
+;
+; G_GFX7-LABEL: atomic_fmax_1d:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 unorm glc
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: ; return to shader part epilog
+;
+; G_GFX10-LABEL: atomic_fmax_1d:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: ; return to shader part epilog
+;
+; G_GFX1030-LABEL: atomic_fmax_1d:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: image_atomic_fmax v0, v1, s[0:7] dmask:0x1 dim:SQ_RSRC_IMG_1D unorm glc
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: ; return to shader part epilog
+main_body:
+ %v = call float @llvm.amdgcn.image.atomic.fmax.1d.f32.f32(float %data, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
+ ret float %v
+}
diff --git a/llvm/test/CodeGen/AMDGPU/fp64-min-max-buffer-atomics.ll b/llvm/test/CodeGen/AMDGPU/fp64-min-max-buffer-atomics.ll
new file mode 100644
index 000000000000..ad30576a4604
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/fp64-min-max-buffer-atomics.ll
@@ -0,0 +1,558 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=SI
+; RUN: llc < %s -march=amdgcn -mcpu=hawaii -verify-machineinstrs | FileCheck %s -check-prefix=GFX7
+; RUN: llc < %s -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=GFX10
+; RUN: llc < %s -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s -check-prefix=GFX1030
+
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=G_SI
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=hawaii -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX7
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX10
+; RUN: llc < %s -global-isel -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s -check-prefix=G_GFX1030
+
+declare double @llvm.amdgcn.raw.buffer.atomic.fmin.f64(double, <4 x i32>, i32, i32, i32 immarg)
+declare double @llvm.amdgcn.raw.buffer.atomic.fmax.f64(double, <4 x i32>, i32, i32, i32 immarg)
+
+
+define amdgpu_kernel void @raw_buffer_atomic_min_noret_f64(<4 x i32> inreg %rsrc, double %data, i32 %vindex) {
+; SI-LABEL: raw_buffer_atomic_min_noret_f64:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; SI-NEXT: s_load_dword s6, s[0:1], 0xf
+; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT: s_waitcnt lgkmcnt(0)
+; SI-NEXT: v_mov_b32_e32 v0, s4
+; SI-NEXT: v_mov_b32_e32 v1, s5
+; SI-NEXT: v_mov_b32_e32 v2, s6
+; SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_min_noret_f64:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT: s_load_dword s6, s[0:1], 0xf
+; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: v_mov_b32_e32 v1, s5
+; GFX7-NEXT: v_mov_b32_e32 v2, s6
+; GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_min_noret_f64:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: s_clause 0x2
+; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX10-NEXT: s_load_dword s8, s[0:1], 0x3c
+; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: v_mov_b32_e32 v2, s8
+; GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_min_noret_f64:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: s_clause 0x2
+; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; GFX1030-NEXT: s_load_dword s6, s[0:1], 0x3c
+; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; GFX1030-NEXT: v_mov_b32_e32 v2, s6
+; GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_min_noret_f64:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_SI-NEXT: s_load_dword s6, s[0:1], 0xf
+; G_SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_SI-NEXT: s_waitcnt lgkmcnt(0)
+; G_SI-NEXT: v_mov_b32_e32 v0, s4
+; G_SI-NEXT: v_mov_b32_e32 v1, s5
+; G_SI-NEXT: v_mov_b32_e32 v2, s6
+; G_SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_min_noret_f64:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_GFX7-NEXT: s_load_dword s6, s[0:1], 0xf
+; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX7-NEXT: v_mov_b32_e32 v2, s6
+; G_GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_min_noret_f64:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: s_clause 0x2
+; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX10-NEXT: s_load_dword s8, s[0:1], 0x3c
+; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX10-NEXT: v_mov_b32_e32 v2, s8
+; G_GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen glc
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_min_noret_f64:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: s_clause 0x2
+; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; G_GFX1030-NEXT: s_load_dword s6, s[0:1], 0x3c
+; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX1030-NEXT: v_mov_b32_e32 v2, s6
+; G_GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmin.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
+ ret void
+}
+
+define amdgpu_ps void @raw_buffer_atomic_min_rtn_f64(<4 x i32> inreg %rsrc, double %data, i32 %vindex) {
+; SI-LABEL: raw_buffer_atomic_min_rtn_f64:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; SI-NEXT: s_mov_b32 m0, -1
+; SI-NEXT: s_waitcnt vmcnt(0)
+; SI-NEXT: ds_write_b64 v0, v[0:1]
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_min_rtn_f64:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; GFX7-NEXT: s_mov_b32 m0, -1
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: ds_write_b64 v0, v[0:1]
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_min_rtn_f64:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: ds_write_b64 v0, v[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_min_rtn_f64:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: ds_write_b64 v0, v[0:1]
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_min_rtn_f64:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_SI-NEXT: s_mov_b32 m0, -1
+; G_SI-NEXT: s_waitcnt vmcnt(0)
+; G_SI-NEXT: ds_write_b64 v0, v[0:1]
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_min_rtn_f64:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX7-NEXT: s_mov_b32 m0, -1
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: ds_write_b64 v0, v[0:1]
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_min_rtn_f64:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: ds_write_b64 v0, v[0:1]
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_min_rtn_f64:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: ds_write_b64 v0, v[0:1]
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmin.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
+ store double %ret, double addrspace(3)* undef
+ ret void
+}
+
+define amdgpu_ps void @raw_buffer_atomic_min_rtn_f64_off4_slc(<4 x i32> inreg %rsrc, double %data, i32 %vindex, double addrspace(3)* %out) {
+; SI-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; SI-NEXT: s_mov_b32 m0, -1
+; SI-NEXT: s_waitcnt vmcnt(0)
+; SI-NEXT: ds_write_b64 v3, v[0:1]
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; GFX7-NEXT: s_mov_b32 m0, -1
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: ds_write_b64 v3, v[0:1]
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: ds_write_b64 v3, v[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: ds_write_b64 v3, v[0:1]
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; G_SI-NEXT: s_mov_b32 m0, -1
+; G_SI-NEXT: s_waitcnt vmcnt(0)
+; G_SI-NEXT: ds_write_b64 v3, v[0:1]
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; G_GFX7-NEXT: s_mov_b32 m0, -1
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: ds_write_b64 v3, v[0:1]
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: ds_write_b64 v3, v[0:1]
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_min_rtn_f64_off4_slc:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: buffer_atomic_fmin_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: ds_write_b64 v3, v[0:1]
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmin.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 4, i32 2)
+ store double %ret, double addrspace(3)* %out, align 8
+ ret void
+}
+
+define amdgpu_kernel void @raw_buffer_atomic_max_noret_f64(<4 x i32> inreg %rsrc, double %data, i32 %vindex) {
+; SI-LABEL: raw_buffer_atomic_max_noret_f64:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; SI-NEXT: s_load_dword s6, s[0:1], 0xf
+; SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT: s_waitcnt lgkmcnt(0)
+; SI-NEXT: v_mov_b32_e32 v0, s4
+; SI-NEXT: v_mov_b32_e32 v1, s5
+; SI-NEXT: v_mov_b32_e32 v2, s6
+; SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_max_noret_f64:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT: s_load_dword s6, s[0:1], 0xf
+; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: v_mov_b32_e32 v1, s5
+; GFX7-NEXT: v_mov_b32_e32 v2, s6
+; GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_max_noret_f64:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: s_clause 0x2
+; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX10-NEXT: s_load_dword s8, s[0:1], 0x3c
+; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: v_mov_b32_e32 v2, s8
+; GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_max_noret_f64:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: s_clause 0x2
+; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; GFX1030-NEXT: s_load_dword s6, s[0:1], 0x3c
+; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; GFX1030-NEXT: v_mov_b32_e32 v2, s6
+; GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_max_noret_f64:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_SI-NEXT: s_load_dword s6, s[0:1], 0xf
+; G_SI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_SI-NEXT: s_waitcnt lgkmcnt(0)
+; G_SI-NEXT: v_mov_b32_e32 v0, s4
+; G_SI-NEXT: v_mov_b32_e32 v1, s5
+; G_SI-NEXT: v_mov_b32_e32 v2, s6
+; G_SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_max_noret_f64:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_GFX7-NEXT: s_load_dword s6, s[0:1], 0xf
+; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX7-NEXT: v_mov_b32_e32 v2, s6
+; G_GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_max_noret_f64:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: s_clause 0x2
+; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX10-NEXT: s_load_dword s8, s[0:1], 0x3c
+; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX10-NEXT: v_mov_b32_e32 v2, s8
+; G_GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen glc
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_max_noret_f64:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: s_clause 0x2
+; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; G_GFX1030-NEXT: s_load_dword s6, s[0:1], 0x3c
+; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX1030-NEXT: v_mov_b32_e32 v2, s6
+; G_GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmax.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
+ ret void
+}
+
+define amdgpu_ps void @raw_buffer_atomic_max_rtn_f64(<4 x i32> inreg %rsrc, double %data, i32 %vindex) {
+; SI-LABEL: raw_buffer_atomic_max_rtn_f64:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; SI-NEXT: s_mov_b32 m0, -1
+; SI-NEXT: s_waitcnt vmcnt(0)
+; SI-NEXT: ds_write_b64 v0, v[0:1]
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_max_rtn_f64:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; GFX7-NEXT: s_mov_b32 m0, -1
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: ds_write_b64 v0, v[0:1]
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_max_rtn_f64:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: ds_write_b64 v0, v[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_max_rtn_f64:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: ds_write_b64 v0, v[0:1]
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_max_rtn_f64:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_SI-NEXT: s_mov_b32 m0, -1
+; G_SI-NEXT: s_waitcnt vmcnt(0)
+; G_SI-NEXT: ds_write_b64 v0, v[0:1]
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_max_rtn_f64:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX7-NEXT: s_mov_b32 m0, -1
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: ds_write_b64 v0, v[0:1]
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_max_rtn_f64:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: ds_write_b64 v0, v[0:1]
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_max_rtn_f64:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 0 offen glc
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: ds_write_b64 v0, v[0:1]
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmax.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0)
+ store double %ret, double addrspace(3)* undef
+ ret void
+}
+
+define amdgpu_kernel void @raw_buffer_atomic_max_rtn_f64_off4_slc(<4 x i32> inreg %rsrc, double %data, i32 %vindex, double addrspace(3)* %out) {
+; SI-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
+; SI: ; %bb.0: ; %main_body
+; SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; SI-NEXT: s_load_dword s8, s[0:1], 0xf
+; SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT: s_load_dword s0, s[0:1], 0x10
+; SI-NEXT: s_mov_b32 m0, -1
+; SI-NEXT: s_waitcnt lgkmcnt(0)
+; SI-NEXT: v_mov_b32_e32 v0, s2
+; SI-NEXT: v_mov_b32_e32 v1, s3
+; SI-NEXT: v_mov_b32_e32 v2, s8
+; SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 4 offen glc slc
+; SI-NEXT: v_mov_b32_e32 v2, s0
+; SI-NEXT: s_waitcnt vmcnt(0)
+; SI-NEXT: ds_write_b64 v2, v[0:1]
+; SI-NEXT: s_endpgm
+;
+; GFX7-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
+; GFX7: ; %bb.0: ; %main_body
+; GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0xf
+; GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX7-NEXT: s_mov_b32 m0, -1
+; GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: v_mov_b32_e32 v1, s5
+; GFX7-NEXT: v_mov_b32_e32 v2, s6
+; GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; GFX7-NEXT: v_mov_b32_e32 v2, s7
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: ds_write_b64 v2, v[0:1]
+; GFX7-NEXT: s_endpgm
+;
+; GFX10-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
+; GFX10: ; %bb.0: ; %main_body
+; GFX10-NEXT: s_clause 0x2
+; GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; GFX10-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x3c
+; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: v_mov_b32_e32 v2, s8
+; GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 4 offen glc slc
+; GFX10-NEXT: v_mov_b32_e32 v2, s9
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: ds_write_b64 v2, v[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX1030-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
+; GFX1030: ; %bb.0: ; %main_body
+; GFX1030-NEXT: s_clause 0x2
+; GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; GFX1030-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x3c
+; GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; GFX1030-NEXT: v_mov_b32_e32 v2, s6
+; GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; GFX1030-NEXT: v_mov_b32_e32 v2, s7
+; GFX1030-NEXT: s_waitcnt vmcnt(0)
+; GFX1030-NEXT: ds_write_b64 v2, v[0:1]
+; GFX1030-NEXT: s_endpgm
+;
+; G_SI-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
+; G_SI: ; %bb.0: ; %main_body
+; G_SI-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd
+; G_SI-NEXT: s_load_dword s8, s[0:1], 0xf
+; G_SI-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
+; G_SI-NEXT: s_load_dword s0, s[0:1], 0x10
+; G_SI-NEXT: s_mov_b32 m0, -1
+; G_SI-NEXT: s_waitcnt lgkmcnt(0)
+; G_SI-NEXT: v_mov_b32_e32 v0, s2
+; G_SI-NEXT: v_mov_b32_e32 v1, s3
+; G_SI-NEXT: v_mov_b32_e32 v2, s8
+; G_SI-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 4 offen glc slc
+; G_SI-NEXT: v_mov_b32_e32 v2, s0
+; G_SI-NEXT: s_waitcnt vmcnt(0)
+; G_SI-NEXT: ds_write_b64 v2, v[0:1]
+; G_SI-NEXT: s_endpgm
+;
+; G_GFX7-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
+; G_GFX7: ; %bb.0: ; %main_body
+; G_GFX7-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
+; G_GFX7-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0xf
+; G_GFX7-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; G_GFX7-NEXT: s_mov_b32 m0, -1
+; G_GFX7-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX7-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX7-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX7-NEXT: v_mov_b32_e32 v2, s6
+; G_GFX7-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; G_GFX7-NEXT: v_mov_b32_e32 v2, s7
+; G_GFX7-NEXT: s_waitcnt vmcnt(0)
+; G_GFX7-NEXT: ds_write_b64 v2, v[0:1]
+; G_GFX7-NEXT: s_endpgm
+;
+; G_GFX10-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
+; G_GFX10: ; %bb.0: ; %main_body
+; G_GFX10-NEXT: s_clause 0x2
+; G_GFX10-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
+; G_GFX10-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x3c
+; G_GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
+; G_GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX10-NEXT: v_mov_b32_e32 v0, s2
+; G_GFX10-NEXT: v_mov_b32_e32 v1, s3
+; G_GFX10-NEXT: v_mov_b32_e32 v2, s8
+; G_GFX10-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 4 offen glc slc
+; G_GFX10-NEXT: v_mov_b32_e32 v2, s9
+; G_GFX10-NEXT: s_waitcnt vmcnt(0)
+; G_GFX10-NEXT: ds_write_b64 v2, v[0:1]
+; G_GFX10-NEXT: s_endpgm
+;
+; G_GFX1030-LABEL: raw_buffer_atomic_max_rtn_f64_off4_slc:
+; G_GFX1030: ; %bb.0: ; %main_body
+; G_GFX1030-NEXT: s_clause 0x2
+; G_GFX1030-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
+; G_GFX1030-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x3c
+; G_GFX1030-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; G_GFX1030-NEXT: s_waitcnt lgkmcnt(0)
+; G_GFX1030-NEXT: v_mov_b32_e32 v0, s4
+; G_GFX1030-NEXT: v_mov_b32_e32 v1, s5
+; G_GFX1030-NEXT: v_mov_b32_e32 v2, s6
+; G_GFX1030-NEXT: buffer_atomic_fmax_x2 v[0:1], v2, s[0:3], 4 offen glc slc
+; G_GFX1030-NEXT: v_mov_b32_e32 v2, s7
+; G_GFX1030-NEXT: s_waitcnt vmcnt(0)
+; G_GFX1030-NEXT: ds_write_b64 v2, v[0:1]
+; G_GFX1030-NEXT: s_endpgm
+main_body:
+ %ret = call double @llvm.amdgcn.raw.buffer.atomic.fmax.f64(double %data, <4 x i32> %rsrc, i32 %vindex, i32 4, i32 2)
+ store double %ret, double addrspace(3)* %out, align 8
+ ret void
+}
More information about the llvm-commits
mailing list