[llvm-branch-commits] [llvm] AMDGPU: Start selecting buffer fat pointer atomicrmw fmin/fmax (PR #95593)

Mon Jun 17 23:41:04 PDT 2024

https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/95593

>From 12f8d7fec05c29628b8ff50a8ca6b6f3b5b36726 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Thu, 13 Jun 2024 21:34:53 +0200
Subject: [PATCH] AMDGPU: Start selecting buffer fat pointer atomicrmw
 fmin/fmax

---
 llvm/lib/Target/AMDGPU/SIISelLowering.cpp     |    3 +-
 .../buffer-fat-pointer-atomicrmw-fmax.ll      | 1170 ++---------------
 .../buffer-fat-pointer-atomicrmw-fmin.ll      | 1170 ++---------------
 3 files changed, 284 insertions(+), 2059 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index f280f43a32a15..5e616dc9c0bf3 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -16084,7 +16084,8 @@ SITargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *RMW) const {
         return AtomicExpansionKind::None;
       if (Subtarget->hasAtomicFMinFMaxF64FlatInsts() && Ty->isDoubleTy())
         return AtomicExpansionKind::None;
-    } else if (AMDGPU::isExtendedGlobalAddrSpace(AS)) {
+    } else if (AMDGPU::isExtendedGlobalAddrSpace(AS) ||
+               AS == AMDGPUAS::BUFFER_FAT_POINTER) {
       if (Subtarget->hasAtomicFMinFMaxF32GlobalInsts() && Ty->isFloatTy())
         return AtomicExpansionKind::None;
       if (Subtarget->hasAtomicFMinFMaxF64GlobalInsts() && Ty->isDoubleTy())
diff --git a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll
index fb068e35fc597..4b5cdd2ed32ef 100644
--- a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll
+++ b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll
@@ -21,32 +21,11 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset(ptr addrspace(7)
 ; GFX12-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
-; GFX12-NEXT:    s_addk_co_i32 s4, 0x400
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    v_dual_mov_b32 v3, s4 :: v_dual_max_num_f32 v2, v1, v1
-; GFX12-NEXT:    buffer_load_b32 v0, v0, s[0:3], null offen offset:1024
-; GFX12-NEXT:    s_mov_b32 s4, 0
-; GFX12-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX12-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_mov_b32_e32 v5, v0
+; GFX12-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX12-NEXT:    s_wait_storecnt 0x0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-NEXT:    v_max_num_f32_e32 v0, v5, v5
-; GFX12-NEXT:    v_max_num_f32_e32 v4, v0, v2
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5
-; GFX12-NEXT:    buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], null offen th:TH_ATOMIC_RETURN
+; GFX12-NEXT:    buffer_atomic_max_num_f32 v0, v1, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-NEXT:    global_inv scope:SCOPE_DEV
-; GFX12-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX12-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX12-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX12-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX12-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset:
@@ -81,64 +60,23 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset(ptr addrspace(7)
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
-; GFX11-NEXT:    s_addk_i32 s4, 0x400
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    v_dual_mov_b32 v3, s4 :: v_dual_max_f32 v2, v1, v1
-; GFX11-NEXT:    buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024
-; GFX11-NEXT:    s_mov_b32 s4, 0
-; GFX11-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_mov_b32_e32 v5, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_max_f32_e32 v0, v5, v5
-; GFX11-NEXT:    v_max_f32_e32 v4, v0, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5
-; GFX11-NEXT:    buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], 0 offen glc
+; GFX11-NEXT:    buffer_atomic_max_f32 v0, v1, s[0:3], 0 offen offset:1024 glc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    buffer_gl1_inv
 ; GFX11-NEXT:    buffer_gl0_inv
-; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX11-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX11-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX11-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v1, v0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s8
-; GFX10-NEXT:    s_addk_i32 s8, 0x400
-; GFX10-NEXT:    v_mov_b32_e32 v3, s8
-; GFX10-NEXT:    v_max_f32_e32 v2, v1, v1
-; GFX10-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v5, v0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s8
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f32_e32 v0, v5, v5
-; GFX10-NEXT:    v_max_f32_e32 v4, v0, v2
-; GFX10-NEXT:    v_mov_b32_e32 v0, v4
-; GFX10-NEXT:    v_mov_b32_e32 v1, v5
-; GFX10-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX10-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset:
@@ -230,60 +168,19 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset(ptr addrspace(7)
 ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v1, v0
-; GFX7-NEXT:    v_mov_b32_e32 v0, s8
-; GFX7-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; GFX7-NEXT:    v_mov_b32_e32 v3, s10
-; GFX7-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v5, v0
-; GFX7-NEXT:    v_mul_f32_e32 v0, 1.0, v5
-; GFX7-NEXT:    v_max_f32_e32 v4, v0, v2
-; GFX7-NEXT:    v_mov_b32_e32 v0, v4
-; GFX7-NEXT:    v_mov_b32_e32 v1, v5
-; GFX7-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s8
+; GFX7-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v5
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v1, v0
-; GFX6-NEXT:    v_mov_b32_e32 v0, s8
-; GFX6-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s10
-; GFX6-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v5, v0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_mul_f32_e32 v0, 1.0, v5
-; GFX6-NEXT:    v_max_f32_e32 v4, v0, v2
-; GFX6-NEXT:    v_mov_b32_e32 v0, v4
-; GFX6-NEXT:    v_mov_b32_e32 v1, v5
-; GFX6-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX6-NEXT:    v_mov_b32_e32 v1, s8
+; GFX6-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v5
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr float, ptr addrspace(7) %ptr, i32 256
@@ -299,31 +196,11 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset(ptr addrspace(7)
 ; GFX12-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_dual_mov_b32 v1, s4 :: v_dual_max_num_f32 v2, v0, v0
-; GFX12-NEXT:    s_addk_co_i32 s4, 0x400
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    v_mov_b32_e32 v3, s4
-; GFX12-NEXT:    buffer_load_b32 v1, v1, s[0:3], null offen offset:1024
-; GFX12-NEXT:    s_mov_b32 s4, 0
-; GFX12-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX12-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_max_num_f32_e32 v0, v1, v1
+; GFX12-NEXT:    v_mov_b32_e32 v1, s4
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    buffer_atomic_max_num_f32 v0, v1, s[0:3], null offen offset:1024
 ; GFX12-NEXT:    s_wait_storecnt 0x0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-NEXT:    v_max_num_f32_e32 v0, v0, v2
-; GFX12-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v4, v0
-; GFX12-NEXT:    buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], null offen th:TH_ATOMIC_RETURN
-; GFX12-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-NEXT:    global_inv scope:SCOPE_DEV
-; GFX12-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v1
-; GFX12-NEXT:    v_mov_b32_e32 v1, v4
-; GFX12-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX12-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX12-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX12-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset:
@@ -357,62 +234,23 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset(ptr addrspace(7)
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v1, s4 :: v_dual_max_f32 v2, v0, v0
-; GFX11-NEXT:    s_addk_i32 s4, 0x400
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    v_mov_b32_e32 v3, s4
-; GFX11-NEXT:    buffer_load_b32 v1, v1, s[0:3], 0 offen offset:1024
-; GFX11-NEXT:    s_mov_b32 s4, 0
-; GFX11-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_max_f32_e32 v0, v1, v1
+; GFX11-NEXT:    v_mov_b32_e32 v1, s4
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    buffer_atomic_max_f32 v0, v1, s[0:3], 0 offen offset:1024
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_max_f32_e32 v0, v0, v2
-; GFX11-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v4, v0
-; GFX11-NEXT:    buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], 0 offen glc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    buffer_gl1_inv
 ; GFX11-NEXT:    buffer_gl0_inv
-; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v1
-; GFX11-NEXT:    v_mov_b32_e32 v1, v4
-; GFX11-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX11-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX11-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s8
-; GFX10-NEXT:    s_addk_i32 s8, 0x400
-; GFX10-NEXT:    v_max_f32_e32 v2, v0, v0
-; GFX10-NEXT:    v_mov_b32_e32 v3, s8
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:    buffer_load_dword v1, v1, s[4:7], 0 offen offset:1024
-; GFX10-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_max_f32_e32 v0, v1, v1
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f32_e32 v0, v0, v2
-; GFX10-NEXT:    v_mov_b32_e32 v5, v1
-; GFX10-NEXT:    v_mov_b32_e32 v4, v0
-; GFX10-NEXT:    buffer_atomic_cmpswap v[4:5], v3, s[4:7], 0 offen glc
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v1
-; GFX10-NEXT:    v_mov_b32_e32 v1, v4
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset:
@@ -502,57 +340,18 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset(ptr addrspace(7)
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s8
-; GFX7-NEXT:    buffer_load_dword v1, v1, s[4:7], 0 offen offset:1024
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mul_f32_e32 v2, 1.0, v0
-; GFX7-NEXT:    v_mov_b32_e32 v3, s10
-; GFX7-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mul_f32_e32 v0, 1.0, v1
-; GFX7-NEXT:    v_max_f32_e32 v0, v0, v2
-; GFX7-NEXT:    v_mov_b32_e32 v5, v1
-; GFX7-NEXT:    v_mov_b32_e32 v4, v0
-; GFX7-NEXT:    buffer_atomic_cmpswap v[4:5], v3, s[4:7], 0 offen glc
+; GFX7-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v1
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    v_mov_b32_e32 v1, v4
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s8
-; GFX6-NEXT:    buffer_load_dword v1, v1, s[4:7], 0 offen offset:1024
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v0
-; GFX6-NEXT:    v_mov_b32_e32 v3, s10
-; GFX6-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_mul_f32_e32 v0, 1.0, v1
-; GFX6-NEXT:    v_max_f32_e32 v0, v0, v2
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v5, v1
-; GFX6-NEXT:    v_mov_b32_e32 v4, v0
-; GFX6-NEXT:    buffer_atomic_cmpswap v[4:5], v3, s[4:7], 0 offen glc
+; GFX6-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v1
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    v_mov_b32_e32 v1, v4
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr float, ptr addrspace(7) %ptr, i32 256
@@ -568,8 +367,8 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall(ptr ad
 ; GFX12-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_add_nc_u32_e32 v7, 0x400, v4
 ; GFX12-NEXT:    s_mov_b32 s1, exec_lo
+; GFX12-NEXT:    s_wait_storecnt 0x0
 ; GFX12-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX12-NEXT:    v_readfirstlane_b32 s4, v0
 ; GFX12-NEXT:    v_readfirstlane_b32 s5, v1
@@ -581,54 +380,17 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall(ptr ad
 ; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX12-NEXT:    s_and_b32 s0, vcc_lo, s0
 ; GFX12-NEXT:    s_and_saveexec_b32 s0, s0
-; GFX12-NEXT:    buffer_load_b32 v6, v4, s[4:7], null offen offset:1024
+; GFX12-NEXT:    s_wait_loadcnt 0x0
+; GFX12-NEXT:    buffer_atomic_max_num_f32 v5, v4, s[4:7], null offen offset:1024 th:TH_ATOMIC_RETURN
+; GFX12-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX12-NEXT:    ; implicit-def: $vgpr4
 ; GFX12-NEXT:    s_xor_b32 exec_lo, exec_lo, s0
 ; GFX12-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX12-NEXT:  ; %bb.2:
 ; GFX12-NEXT:    s_mov_b32 exec_lo, s1
-; GFX12-NEXT:    v_max_num_f32_e32 v8, v5, v5
-; GFX12-NEXT:    s_mov_b32 s1, 0
-; GFX12-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX12-NEXT:    ; =>This Loop Header: Depth=1
-; GFX12-NEXT:    ; Child Loop BB2_4 Depth 2
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_max_num_f32_e32 v4, v6, v6
-; GFX12-NEXT:    s_mov_b32 s2, exec_lo
-; GFX12-NEXT:    s_wait_storecnt 0x0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-NEXT:    v_max_num_f32_e32 v5, v4, v8
-; GFX12-NEXT:    v_mov_b32_e32 v4, v5
-; GFX12-NEXT:    v_mov_b32_e32 v5, v6
-; GFX12-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX12-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX12-NEXT:    v_readfirstlane_b32 s4, v0
-; GFX12-NEXT:    v_readfirstlane_b32 s5, v1
-; GFX12-NEXT:    v_readfirstlane_b32 s6, v2
-; GFX12-NEXT:    v_readfirstlane_b32 s7, v3
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX12-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[0:1]
-; GFX12-NEXT:    v_cmp_eq_u64_e64 s0, s[6:7], v[2:3]
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_b32 s0, vcc_lo, s0
-; GFX12-NEXT:    s_and_saveexec_b32 s0, s0
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], null offen th:TH_ATOMIC_RETURN
-; GFX12-NEXT:    s_xor_b32 exec_lo, exec_lo, s0
-; GFX12-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX12-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX12-NEXT:    s_mov_b32 exec_lo, s2
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v6
-; GFX12-NEXT:    v_mov_b32_e32 v6, v4
+; GFX12-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX12-NEXT:    global_inv scope:SCOPE_DEV
-; GFX12-NEXT:    s_or_b32 s1, vcc_lo, s1
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s1
-; GFX12-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX12-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s1
-; GFX12-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX12-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall:
@@ -695,9 +457,8 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall(ptr ad
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_add_nc_u32_e32 v7, 0x400, v4
-; GFX11-NEXT:    s_mov_b32 s1, 0
-; GFX11-NEXT:    s_mov_b32 s2, exec_lo
+; GFX11-NEXT:    s_mov_b32 s1, exec_lo
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX11-NEXT:    v_readfirstlane_b32 s4, v0
 ; GFX11-NEXT:    v_readfirstlane_b32 s5, v1
@@ -709,63 +470,25 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall(ptr ad
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_and_b32 s0, vcc_lo, s0
 ; GFX11-NEXT:    s_and_saveexec_b32 s0, s0
-; GFX11-NEXT:    buffer_load_b32 v6, v4, s[4:7], 0 offen offset:1024
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    buffer_atomic_max_f32 v5, v4, s[4:7], 0 offen offset:1024 glc
+; GFX11-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX11-NEXT:    ; implicit-def: $vgpr4
 ; GFX11-NEXT:    s_xor_b32 exec_lo, exec_lo, s0
 ; GFX11-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX11-NEXT:  ; %bb.2:
-; GFX11-NEXT:    s_mov_b32 exec_lo, s2
-; GFX11-NEXT:    v_max_f32_e32 v8, v5, v5
-; GFX11-NEXT:    .p2align 6
-; GFX11-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX11-NEXT:    ; =>This Loop Header: Depth=1
-; GFX11-NEXT:    ; Child Loop BB2_4 Depth 2
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_max_f32_e32 v4, v6, v6
-; GFX11-NEXT:    s_mov_b32 s2, exec_lo
-; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_max_f32_e32 v5, v4, v8
-; GFX11-NEXT:    v_mov_b32_e32 v4, v5
-; GFX11-NEXT:    v_mov_b32_e32 v5, v6
-; GFX11-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX11-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX11-NEXT:    v_readfirstlane_b32 s4, v0
-; GFX11-NEXT:    v_readfirstlane_b32 s5, v1
-; GFX11-NEXT:    v_readfirstlane_b32 s6, v2
-; GFX11-NEXT:    v_readfirstlane_b32 s7, v3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[0:1]
-; GFX11-NEXT:    v_cmp_eq_u64_e64 s0, s[6:7], v[2:3]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_b32 s0, vcc_lo, s0
-; GFX11-NEXT:    s_and_saveexec_b32 s0, s0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], 0 offen glc
-; GFX11-NEXT:    s_xor_b32 exec_lo, exec_lo, s0
-; GFX11-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX11-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX11-NEXT:    s_mov_b32 exec_lo, s2
+; GFX11-NEXT:    s_mov_b32 exec_lo, s1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v6
-; GFX11-NEXT:    v_mov_b32_e32 v6, v4
+; GFX11-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX11-NEXT:    buffer_gl1_inv
 ; GFX11-NEXT:    buffer_gl0_inv
-; GFX11-NEXT:    s_or_b32 s1, vcc_lo, s1
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s1
-; GFX11-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX11-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s1
-; GFX11-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_add_nc_u32_e32 v7, 0x400, v4
-; GFX10-NEXT:    s_mov_b32 s5, 0
-; GFX10-NEXT:    s_mov_b32 s6, exec_lo
+; GFX10-NEXT:    s_mov_b32 s5, exec_lo
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX10-NEXT:    v_readfirstlane_b32 s8, v0
 ; GFX10-NEXT:    v_readfirstlane_b32 s9, v1
@@ -775,53 +498,19 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall(ptr ad
 ; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[2:3]
 ; GFX10-NEXT:    s_and_b32 s4, vcc_lo, s4
 ; GFX10-NEXT:    s_and_saveexec_b32 s4, s4
-; GFX10-NEXT:    buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    buffer_atomic_fmax v5, v4, s[8:11], 0 offen offset:1024 glc
+; GFX10-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX10-NEXT:    ; implicit-def: $vgpr4
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
 ; GFX10-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX10-NEXT:  ; %bb.2:
-; GFX10-NEXT:    s_mov_b32 exec_lo, s6
-; GFX10-NEXT:    v_max_f32_e32 v8, v5, v5
-; GFX10-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Loop Header: Depth=1
-; GFX10-NEXT:    ; Child Loop BB2_4 Depth 2
+; GFX10-NEXT:    s_mov_b32 exec_lo, s5
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_max_f32_e32 v4, v6, v6
-; GFX10-NEXT:    s_mov_b32 s6, exec_lo
-; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f32_e32 v5, v4, v8
-; GFX10-NEXT:    v_mov_b32_e32 v4, v5
-; GFX10-NEXT:    v_mov_b32_e32 v5, v6
-; GFX10-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX10-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX10-NEXT:    v_readfirstlane_b32 s8, v0
-; GFX10-NEXT:    v_readfirstlane_b32 s9, v1
-; GFX10-NEXT:    v_readfirstlane_b32 s10, v2
-; GFX10-NEXT:    v_readfirstlane_b32 s11, v3
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[0:1]
-; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[2:3]
-; GFX10-NEXT:    s_and_b32 s4, vcc_lo, s4
-; GFX10-NEXT:    s_and_saveexec_b32 s4, s4
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
-; GFX10-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX10-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX10-NEXT:    s_mov_b32 exec_lo, s6
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v6
-; GFX10-NEXT:    v_mov_b32_e32 v6, v4
+; GFX10-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    s_or_b32 s5, vcc_lo, s5
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s5
-; GFX10-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX10-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s5
-; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall:
@@ -1006,7 +695,6 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall(ptr ad
 ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_add_i32_e32 v7, vcc, 0x400, v4
 ; GFX7-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX7-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX7-NEXT:    v_readfirstlane_b32 s8, v0
@@ -1017,55 +705,22 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall(ptr ad
 ; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX7-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX7-NEXT:    buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_atomic_fmax v5, v4, s[8:11], 0 offen offset:1024 glc
+; GFX7-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX7-NEXT:    ; implicit-def: $vgpr4
 ; GFX7-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX7-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX7-NEXT:  ; %bb.2:
 ; GFX7-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX7-NEXT:    s_mov_b64 s[6:7], 0
-; GFX7-NEXT:    v_mul_f32_e32 v8, 1.0, v5
-; GFX7-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Loop Header: Depth=1
-; GFX7-NEXT:    ; Child Loop BB2_4 Depth 2
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mul_f32_e32 v4, 1.0, v6
-; GFX7-NEXT:    v_max_f32_e32 v5, v4, v8
-; GFX7-NEXT:    v_mov_b32_e32 v4, v5
-; GFX7-NEXT:    s_mov_b64 s[12:13], exec
-; GFX7-NEXT:    v_mov_b32_e32 v5, v6
-; GFX7-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX7-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX7-NEXT:    v_readfirstlane_b32 s8, v0
-; GFX7-NEXT:    v_readfirstlane_b32 s9, v1
-; GFX7-NEXT:    v_readfirstlane_b32 s10, v2
-; GFX7-NEXT:    v_readfirstlane_b32 s11, v3
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
-; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
-; GFX7-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc
-; GFX7-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX7-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX7-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX7-NEXT:    s_mov_b64 exec, s[12:13]
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v6
-; GFX7-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX7-NEXT:    v_mov_b32_e32 v6, v4
+; GFX7-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX7-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX7-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[6:7]
-; GFX7-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_add_i32_e32 v7, vcc, 0x400, v4
 ; GFX6-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX6-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX6-NEXT:    v_readfirstlane_b32 s8, v0
@@ -1076,49 +731,17 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall(ptr ad
 ; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX6-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX6-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX6-NEXT:    buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024
+; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
+; GFX6-NEXT:    buffer_atomic_fmax v5, v4, s[8:11], 0 offen offset:1024 glc
+; GFX6-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX6-NEXT:    ; implicit-def: $vgpr4
 ; GFX6-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX6-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX6-NEXT:  ; %bb.2:
 ; GFX6-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX6-NEXT:    s_mov_b64 s[6:7], 0
-; GFX6-NEXT:    v_mul_f32_e32 v8, 1.0, v5
-; GFX6-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Loop Header: Depth=1
-; GFX6-NEXT:    ; Child Loop BB2_4 Depth 2
-; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
-; GFX6-NEXT:    v_mul_f32_e32 v4, 1.0, v6
-; GFX6-NEXT:    v_max_f32_e32 v5, v4, v8
-; GFX6-NEXT:    v_mov_b32_e32 v4, v5
-; GFX6-NEXT:    s_mov_b64 s[12:13], exec
-; GFX6-NEXT:    v_mov_b32_e32 v5, v6
-; GFX6-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX6-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX6-NEXT:    v_readfirstlane_b32 s8, v0
-; GFX6-NEXT:    v_readfirstlane_b32 s9, v1
-; GFX6-NEXT:    v_readfirstlane_b32 s10, v2
-; GFX6-NEXT:    v_readfirstlane_b32 s11, v3
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
-; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
-; GFX6-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX6-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
-; GFX6-NEXT:    buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc
-; GFX6-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX6-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX6-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX6-NEXT:    s_mov_b64 exec, s[12:13]
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v6
-; GFX6-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX6-NEXT:    v_mov_b32_e32 v6, v4
+; GFX6-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX6-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX6-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[6:7]
-; GFX6-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr float, ptr addrspace(7) %ptr, i32 256
@@ -1172,32 +795,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset(ptr addrspace(7)
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset:
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX940-NEXT:    v_mov_b32_e32 v2, v0
-; GFX940-NEXT:    v_mov_b32_e32 v0, s4
-; GFX940-NEXT:    v_mov_b32_e32 v3, v1
-; GFX940-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[0:3], 0 offen offset:2048
-; GFX940-NEXT:    s_add_i32 s6, s4, 0x800
-; GFX940-NEXT:    s_mov_b64 s[4:5], 0
-; GFX940-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX940-NEXT:    v_mov_b32_e32 v6, s6
-; GFX940-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX940-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    v_mov_b64_e32 v[10:11], v[0:1]
-; GFX940-NEXT:    v_max_f64 v[0:1], v[10:11], v[10:11]
-; GFX940-NEXT:    v_max_f64 v[8:9], v[0:1], v[4:5]
-; GFX940-NEXT:    v_mov_b64_e32 v[0:1], v[8:9]
-; GFX940-NEXT:    v_mov_b64_e32 v[2:3], v[10:11]
+; GFX940-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX940-NEXT:    buffer_wbl2 sc1
-; GFX940-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[0:3], 0 offen sc0
+; GFX940-NEXT:    buffer_atomic_max_f64 v[0:1], v2, s[0:3], 0 offen offset:2048 sc0
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    buffer_inv sc1
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[10:11]
-; GFX940-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
-; GFX940-NEXT:    s_andn2_b64 exec, exec, s[4:5]
-; GFX940-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX940-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX940-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset:
@@ -1238,66 +840,21 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset(ptr addrspace(7)
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s8
-; GFX10-NEXT:    v_mov_b32_e32 v3, v1
-; GFX10-NEXT:    s_addk_i32 s8, 0x800
-; GFX10-NEXT:    v_mov_b32_e32 v6, s8
-; GFX10-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[4:7], 0 offen offset:2048
-; GFX10-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v10, v1
-; GFX10-NEXT:    v_mov_b32_e32 v9, v0
+; GFX10-NEXT:    v_mov_b32_e32 v2, s8
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f64 v[0:1], v[9:10], v[9:10]
-; GFX10-NEXT:    v_max_f64 v[7:8], v[0:1], v[4:5]
-; GFX10-NEXT:    v_mov_b32_e32 v0, v7
-; GFX10-NEXT:    v_mov_b32_e32 v1, v8
-; GFX10-NEXT:    v_mov_b32_e32 v2, v9
-; GFX10-NEXT:    v_mov_b32_e32 v3, v10
-; GFX10-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[4:7], 0 offen glc
+; GFX10-NEXT:    buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen offset:2048 glc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[9:10]
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX90A-NEXT:    v_mov_b32_e32 v2, v0
-; GFX90A-NEXT:    v_mov_b32_e32 v0, s8
-; GFX90A-NEXT:    v_mov_b32_e32 v3, v1
-; GFX90A-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[4:7], 0 offen offset:2048
-; GFX90A-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX90A-NEXT:    s_mov_b64 s[8:9], 0
-; GFX90A-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX90A-NEXT:    v_mov_b32_e32 v6, s10
-; GFX90A-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_pk_mov_b32 v[10:11], v[0:1], v[0:1] op_sel:[0,1]
-; GFX90A-NEXT:    v_max_f64 v[0:1], v[10:11], v[10:11]
-; GFX90A-NEXT:    v_max_f64 v[8:9], v[0:1], v[4:5]
-; GFX90A-NEXT:    v_pk_mov_b32 v[0:1], v[8:9], v[8:9] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[2:3], v[10:11], v[10:11] op_sel:[0,1]
-; GFX90A-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[4:7], 0 offen glc
+; GFX90A-NEXT:    v_mov_b32_e32 v2, s8
+; GFX90A-NEXT:    buffer_atomic_max_f64 v[0:1], v2, s[4:7], 0 offen offset:2048 glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    buffer_wbinvl1
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[10:11]
-; GFX90A-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX90A-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX90A-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX90A-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset:
@@ -1364,73 +921,24 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset(ptr addrspace(7)
 ; GFX8-NEXT:    s_cbranch_execnz .LBB3_1
 ; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
 ; GFX8-NEXT:    s_or_b64 exec, exec, s[8:9]
-; GFX8-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset:
-; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v2, v0
-; GFX7-NEXT:    v_mov_b32_e32 v0, s8
-; GFX7-NEXT:    v_mov_b32_e32 v3, v1
-; GFX7-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[4:7], 0 offen offset:2048
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX7-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mov_b32_e32 v6, s10
-; GFX7-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v10, v1
-; GFX7-NEXT:    v_mov_b32_e32 v9, v0
-; GFX7-NEXT:    v_max_f64 v[0:1], v[9:10], v[9:10]
-; GFX7-NEXT:    v_max_f64 v[7:8], v[0:1], v[4:5]
-; GFX7-NEXT:    v_mov_b32_e32 v0, v7
-; GFX7-NEXT:    v_mov_b32_e32 v1, v8
-; GFX7-NEXT:    v_mov_b32_e32 v2, v9
-; GFX7-NEXT:    v_mov_b32_e32 v3, v10
-; GFX7-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[4:7], 0 offen glc
+; GFX8-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v2, s8
+; GFX7-NEXT:    buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen offset:2048 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10]
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v2, v0
-; GFX6-NEXT:    v_mov_b32_e32 v0, s8
-; GFX6-NEXT:    v_mov_b32_e32 v3, v1
-; GFX6-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[4:7], 0 offen offset:2048
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX6-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mov_b32_e32 v6, s10
-; GFX6-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v10, v1
-; GFX6-NEXT:    v_mov_b32_e32 v9, v0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_max_f64 v[0:1], v[9:10], v[9:10]
-; GFX6-NEXT:    v_max_f64 v[7:8], v[0:1], v[4:5]
-; GFX6-NEXT:    v_mov_b32_e32 v0, v7
-; GFX6-NEXT:    v_mov_b32_e32 v1, v8
-; GFX6-NEXT:    v_mov_b32_e32 v2, v9
-; GFX6-NEXT:    v_mov_b32_e32 v3, v10
-; GFX6-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[4:7], 0 offen glc
+; GFX6-NEXT:    v_mov_b32_e32 v2, s8
+; GFX6-NEXT:    buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen offset:2048 glc
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10]
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr double, ptr addrspace(7) %ptr, i32 256
@@ -1479,29 +987,10 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset(ptr addrspace(7)
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX940-NEXT:    v_mov_b32_e32 v2, s4
-; GFX940-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[0:3], 0 offen offset:2048
-; GFX940-NEXT:    s_add_i32 s6, s4, 0x800
-; GFX940-NEXT:    s_mov_b64 s[4:5], 0
-; GFX940-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX940-NEXT:    v_mov_b32_e32 v6, s6
-; GFX940-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX940-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
-; GFX940-NEXT:    v_max_f64 v[0:1], v[0:1], v[4:5]
-; GFX940-NEXT:    v_mov_b64_e32 v[10:11], v[2:3]
-; GFX940-NEXT:    v_mov_b64_e32 v[8:9], v[0:1]
 ; GFX940-NEXT:    buffer_wbl2 sc1
-; GFX940-NEXT:    buffer_atomic_cmpswap_x2 v[8:11], v6, s[0:3], 0 offen sc0
+; GFX940-NEXT:    buffer_atomic_max_f64 v[0:1], v2, s[0:3], 0 offen offset:2048
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    buffer_inv sc1
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[2:3]
-; GFX940-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
-; GFX940-NEXT:    v_mov_b64_e32 v[2:3], v[8:9]
-; GFX940-NEXT:    s_andn2_b64 exec, exec, s[4:5]
-; GFX940-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX940-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX940-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset:
@@ -1541,61 +1030,20 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset(ptr addrspace(7)
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s8
-; GFX10-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX10-NEXT:    s_addk_i32 s8, 0x800
-; GFX10-NEXT:    v_mov_b32_e32 v6, s8
-; GFX10-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[4:7], 0 offen offset:2048
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f64 v[0:1], v[0:1], v[4:5]
-; GFX10-NEXT:    v_mov_b32_e32 v10, v3
-; GFX10-NEXT:    v_mov_b32_e32 v9, v2
-; GFX10-NEXT:    v_mov_b32_e32 v8, v1
-; GFX10-NEXT:    v_mov_b32_e32 v7, v0
-; GFX10-NEXT:    buffer_atomic_cmpswap_x2 v[7:10], v6, s[4:7], 0 offen glc
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen offset:2048
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[2:3]
-; GFX10-NEXT:    v_mov_b32_e32 v2, v7
-; GFX10-NEXT:    v_mov_b32_e32 v3, v8
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX90A-NEXT:    v_mov_b32_e32 v2, s8
-; GFX90A-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[4:7], 0 offen offset:2048
-; GFX90A-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX90A-NEXT:    s_mov_b64 s[8:9], 0
-; GFX90A-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX90A-NEXT:    v_mov_b32_e32 v6, s10
-; GFX90A-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
-; GFX90A-NEXT:    v_max_f64 v[0:1], v[0:1], v[4:5]
-; GFX90A-NEXT:    v_pk_mov_b32 v[10:11], v[2:3], v[2:3] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[8:9], v[0:1], v[0:1] op_sel:[0,1]
-; GFX90A-NEXT:    buffer_atomic_cmpswap_x2 v[8:11], v6, s[4:7], 0 offen glc
+; GFX90A-NEXT:    buffer_atomic_max_f64 v[0:1], v2, s[4:7], 0 offen offset:2048
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    buffer_wbinvl1
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[2:3]
-; GFX90A-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX90A-NEXT:    v_pk_mov_b32 v[2:3], v[8:9], v[8:9] op_sel:[0,1]
-; GFX90A-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX90A-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX90A-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset:
@@ -1664,63 +1112,18 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset(ptr addrspace(7)
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s8
-; GFX7-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[4:7], 0 offen offset:2048
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX7-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mov_b32_e32 v6, s10
-; GFX7-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
-; GFX7-NEXT:    v_max_f64 v[0:1], v[0:1], v[4:5]
-; GFX7-NEXT:    v_mov_b32_e32 v10, v3
-; GFX7-NEXT:    v_mov_b32_e32 v9, v2
-; GFX7-NEXT:    v_mov_b32_e32 v8, v1
-; GFX7-NEXT:    v_mov_b32_e32 v7, v0
-; GFX7-NEXT:    buffer_atomic_cmpswap_x2 v[7:10], v6, s[4:7], 0 offen glc
+; GFX7-NEXT:    buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen offset:2048
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[2:3]
-; GFX7-NEXT:    v_mov_b32_e32 v2, v7
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    v_mov_b32_e32 v3, v8
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s8
-; GFX6-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[4:7], 0 offen offset:2048
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX6-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mov_b32_e32 v6, s10
-; GFX6-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
-; GFX6-NEXT:    v_max_f64 v[0:1], v[0:1], v[4:5]
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v10, v3
-; GFX6-NEXT:    v_mov_b32_e32 v9, v2
-; GFX6-NEXT:    v_mov_b32_e32 v8, v1
-; GFX6-NEXT:    v_mov_b32_e32 v7, v0
-; GFX6-NEXT:    buffer_atomic_cmpswap_x2 v[7:10], v6, s[4:7], 0 offen glc
+; GFX6-NEXT:    buffer_atomic_fmax_x2 v[0:1], v2, s[4:7], 0 offen offset:2048
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[2:3]
-; GFX6-NEXT:    v_mov_b32_e32 v2, v7
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    v_mov_b32_e32 v3, v8
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr double, ptr addrspace(7) %ptr, i32 256
@@ -1804,68 +1207,32 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall(ptr a
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall:
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX940-NEXT:    v_mov_b32_e32 v11, v6
-; GFX940-NEXT:    v_mov_b32_e32 v10, v5
-; GFX940-NEXT:    v_mov_b32_e32 v9, v3
-; GFX940-NEXT:    v_mov_b32_e32 v8, v2
-; GFX940-NEXT:    v_mov_b32_e32 v7, v1
-; GFX940-NEXT:    v_mov_b32_e32 v6, v0
-; GFX940-NEXT:    v_add_u32_e32 v14, 0x800, v4
+; GFX940-NEXT:    v_mov_b32_e32 v7, v6
+; GFX940-NEXT:    v_mov_b32_e32 v6, v5
 ; GFX940-NEXT:    s_mov_b64 s[2:3], exec
+; GFX940-NEXT:    buffer_wbl2 sc1
 ; GFX940-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX940-NEXT:    v_readfirstlane_b32 s4, v6
-; GFX940-NEXT:    v_readfirstlane_b32 s5, v7
-; GFX940-NEXT:    v_readfirstlane_b32 s6, v8
-; GFX940-NEXT:    v_readfirstlane_b32 s7, v9
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, s[4:5], v[6:7]
+; GFX940-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX940-NEXT:    v_readfirstlane_b32 s5, v1
+; GFX940-NEXT:    v_readfirstlane_b32 s6, v2
+; GFX940-NEXT:    v_readfirstlane_b32 s7, v3
+; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, s[4:5], v[0:1]
 ; GFX940-NEXT:    s_nop 0
-; GFX940-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[6:7], v[8:9]
+; GFX940-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[6:7], v[2:3]
 ; GFX940-NEXT:    s_and_b64 s[0:1], vcc, s[0:1]
 ; GFX940-NEXT:    s_and_saveexec_b64 s[0:1], s[0:1]
-; GFX940-NEXT:    buffer_load_dwordx2 v[12:13], v4, s[4:7], 0 offen offset:2048
+; GFX940-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NEXT:    buffer_atomic_max_f64 v[6:7], v4, s[4:7], 0 offen offset:2048 sc0
+; GFX940-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX940-NEXT:    ; implicit-def: $vgpr4
 ; GFX940-NEXT:    s_xor_b64 exec, exec, s[0:1]
 ; GFX940-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX940-NEXT:  ; %bb.2:
 ; GFX940-NEXT:    s_mov_b64 exec, s[2:3]
-; GFX940-NEXT:    s_mov_b64 s[2:3], 0
-; GFX940-NEXT:    v_max_f64 v[4:5], v[10:11], v[10:11]
-; GFX940-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX940-NEXT:    ; =>This Loop Header: Depth=1
-; GFX940-NEXT:    ; Child Loop BB5_4 Depth 2
-; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    v_max_f64 v[0:1], v[12:13], v[12:13]
-; GFX940-NEXT:    v_max_f64 v[10:11], v[0:1], v[4:5]
-; GFX940-NEXT:    v_mov_b64_e32 v[0:1], v[10:11]
-; GFX940-NEXT:    s_mov_b64 s[8:9], exec
-; GFX940-NEXT:    v_mov_b64_e32 v[2:3], v[12:13]
-; GFX940-NEXT:    buffer_wbl2 sc1
-; GFX940-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX940-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX940-NEXT:    v_readfirstlane_b32 s4, v6
-; GFX940-NEXT:    v_readfirstlane_b32 s5, v7
-; GFX940-NEXT:    v_readfirstlane_b32 s6, v8
-; GFX940-NEXT:    v_readfirstlane_b32 s7, v9
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, s[4:5], v[6:7]
-; GFX940-NEXT:    s_nop 0
-; GFX940-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[6:7], v[8:9]
-; GFX940-NEXT:    s_and_b64 s[0:1], vcc, s[0:1]
-; GFX940-NEXT:    s_and_saveexec_b64 s[0:1], s[0:1]
-; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v14, s[4:7], 0 offen sc0
-; GFX940-NEXT:    s_xor_b64 exec, exec, s[0:1]
-; GFX940-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX940-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX940-NEXT:    s_mov_b64 exec, s[8:9]
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[12:13]
-; GFX940-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX940-NEXT:    v_mov_b64_e32 v[12:13], v[0:1]
+; GFX940-NEXT:    v_mov_b32_e32 v0, v6
+; GFX940-NEXT:    v_mov_b32_e32 v1, v7
 ; GFX940-NEXT:    buffer_inv sc1
-; GFX940-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX940-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX940-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX940-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall:
@@ -1941,136 +1308,60 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall(ptr a
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v8, v3
-; GFX10-NEXT:    v_mov_b32_e32 v7, v2
-; GFX10-NEXT:    v_mov_b32_e32 v10, v1
-; GFX10-NEXT:    v_mov_b32_e32 v9, v0
-; GFX10-NEXT:    v_add_nc_u32_e32 v15, 0x800, v4
-; GFX10-NEXT:    s_mov_b32 s5, 0
-; GFX10-NEXT:    s_mov_b32 s6, exec_lo
+; GFX10-NEXT:    s_mov_b32 s5, exec_lo
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX10-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX10-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX10-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[9:10]
-; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[7:8]
+; GFX10-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX10-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX10-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX10-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[0:1]
+; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[2:3]
 ; GFX10-NEXT:    s_and_b32 s4, vcc_lo, s4
 ; GFX10-NEXT:    s_and_saveexec_b32 s4, s4
-; GFX10-NEXT:    buffer_load_dwordx2 v[13:14], v4, s[8:11], 0 offen offset:2048
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    buffer_atomic_fmax_x2 v[5:6], v4, s[8:11], 0 offen offset:2048 glc
+; GFX10-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX10-NEXT:    ; implicit-def: $vgpr4
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
 ; GFX10-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX10-NEXT:  ; %bb.2:
-; GFX10-NEXT:    s_mov_b32 exec_lo, s6
-; GFX10-NEXT:    v_max_f64 v[4:5], v[5:6], v[5:6]
-; GFX10-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Loop Header: Depth=1
-; GFX10-NEXT:    ; Child Loop BB5_4 Depth 2
+; GFX10-NEXT:    s_mov_b32 exec_lo, s5
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_max_f64 v[0:1], v[13:14], v[13:14]
-; GFX10-NEXT:    s_mov_b32 s6, exec_lo
-; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f64 v[11:12], v[0:1], v[4:5]
-; GFX10-NEXT:    v_mov_b32_e32 v0, v11
-; GFX10-NEXT:    v_mov_b32_e32 v1, v12
-; GFX10-NEXT:    v_mov_b32_e32 v2, v13
-; GFX10-NEXT:    v_mov_b32_e32 v3, v14
-; GFX10-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX10-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX10-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX10-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX10-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX10-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[9:10]
-; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[7:8]
-; GFX10-NEXT:    s_and_b32 s4, vcc_lo, s4
-; GFX10-NEXT:    s_and_saveexec_b32 s4, s4
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v15, s[8:11], 0 offen glc
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
-; GFX10-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX10-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX10-NEXT:    s_mov_b32 exec_lo, s6
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[13:14]
-; GFX10-NEXT:    v_mov_b32_e32 v14, v1
-; GFX10-NEXT:    v_mov_b32_e32 v13, v0
+; GFX10-NEXT:    v_mov_b32_e32 v0, v5
+; GFX10-NEXT:    v_mov_b32_e32 v1, v6
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    s_or_b32 s5, vcc_lo, s5
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s5
-; GFX10-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX10-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s5
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX90A-NEXT:    v_mov_b32_e32 v11, v6
-; GFX90A-NEXT:    v_mov_b32_e32 v10, v5
-; GFX90A-NEXT:    v_mov_b32_e32 v9, v3
-; GFX90A-NEXT:    v_mov_b32_e32 v8, v2
-; GFX90A-NEXT:    v_mov_b32_e32 v7, v1
-; GFX90A-NEXT:    v_mov_b32_e32 v6, v0
-; GFX90A-NEXT:    v_add_u32_e32 v14, 0x800, v4
+; GFX90A-NEXT:    v_mov_b32_e32 v7, v6
+; GFX90A-NEXT:    v_mov_b32_e32 v6, v5
 ; GFX90A-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX90A-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT:    v_readfirstlane_b32 s8, v6
-; GFX90A-NEXT:    v_readfirstlane_b32 s9, v7
-; GFX90A-NEXT:    v_readfirstlane_b32 s10, v8
-; GFX90A-NEXT:    v_readfirstlane_b32 s11, v9
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[6:7]
-; GFX90A-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[8:9]
+; GFX90A-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX90A-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX90A-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX90A-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
+; GFX90A-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX90A-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX90A-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX90A-NEXT:    s_nop 0
-; GFX90A-NEXT:    buffer_load_dwordx2 v[12:13], v4, s[8:11], 0 offen offset:2048
+; GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GFX90A-NEXT:    buffer_atomic_max_f64 v[6:7], v4, s[8:11], 0 offen offset:2048 glc
+; GFX90A-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX90A-NEXT:    ; implicit-def: $vgpr4
 ; GFX90A-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX90A-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX90A-NEXT:  ; %bb.2:
 ; GFX90A-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX90A-NEXT:    s_mov_b64 s[6:7], 0
-; GFX90A-NEXT:    v_max_f64 v[4:5], v[10:11], v[10:11]
-; GFX90A-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX90A-NEXT:    ; =>This Loop Header: Depth=1
-; GFX90A-NEXT:    ; Child Loop BB5_4 Depth 2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_max_f64 v[0:1], v[12:13], v[12:13]
-; GFX90A-NEXT:    v_max_f64 v[10:11], v[0:1], v[4:5]
-; GFX90A-NEXT:    v_pk_mov_b32 v[0:1], v[10:11], v[10:11] op_sel:[0,1]
-; GFX90A-NEXT:    s_mov_b64 s[12:13], exec
-; GFX90A-NEXT:    v_pk_mov_b32 v[2:3], v[12:13], v[12:13] op_sel:[0,1]
-; GFX90A-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX90A-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX90A-NEXT:    v_readfirstlane_b32 s8, v6
-; GFX90A-NEXT:    v_readfirstlane_b32 s9, v7
-; GFX90A-NEXT:    v_readfirstlane_b32 s10, v8
-; GFX90A-NEXT:    v_readfirstlane_b32 s11, v9
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[6:7]
-; GFX90A-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[8:9]
-; GFX90A-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX90A-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v14, s[8:11], 0 offen glc
-; GFX90A-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX90A-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX90A-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX90A-NEXT:    s_mov_b64 exec, s[12:13]
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[12:13]
-; GFX90A-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX90A-NEXT:    v_pk_mov_b32 v[12:13], v[0:1], v[0:1] op_sel:[0,1]
+; GFX90A-NEXT:    v_mov_b32_e32 v0, v6
+; GFX90A-NEXT:    v_mov_b32_e32 v1, v7
 ; GFX90A-NEXT:    buffer_wbinvl1
-; GFX90A-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX90A-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX90A-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX90A-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall:
@@ -2208,131 +1499,55 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall(ptr a
 ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v8, v3
-; GFX7-NEXT:    v_mov_b32_e32 v7, v2
-; GFX7-NEXT:    v_mov_b32_e32 v10, v1
-; GFX7-NEXT:    v_mov_b32_e32 v9, v0
-; GFX7-NEXT:    v_add_i32_e32 v15, vcc, 0x800, v4
 ; GFX7-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX7-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX7-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX7-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX7-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[9:10]
-; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[7:8]
+; GFX7-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX7-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX7-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX7-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
+; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX7-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX7-NEXT:    buffer_load_dwordx2 v[13:14], v4, s[8:11], 0 offen offset:2048
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_atomic_fmax_x2 v[5:6], v4, s[8:11], 0 offen offset:2048 glc
+; GFX7-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX7-NEXT:    ; implicit-def: $vgpr4
 ; GFX7-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX7-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX7-NEXT:  ; %bb.2:
 ; GFX7-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX7-NEXT:    v_max_f64 v[4:5], v[5:6], v[5:6]
-; GFX7-NEXT:    s_mov_b64 s[6:7], 0
-; GFX7-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Loop Header: Depth=1
-; GFX7-NEXT:    ; Child Loop BB5_4 Depth 2
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_max_f64 v[0:1], v[13:14], v[13:14]
-; GFX7-NEXT:    s_mov_b64 s[12:13], exec
-; GFX7-NEXT:    v_max_f64 v[11:12], v[0:1], v[4:5]
-; GFX7-NEXT:    v_mov_b32_e32 v0, v11
-; GFX7-NEXT:    v_mov_b32_e32 v1, v12
-; GFX7-NEXT:    v_mov_b32_e32 v2, v13
-; GFX7-NEXT:    v_mov_b32_e32 v3, v14
-; GFX7-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX7-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX7-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX7-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX7-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX7-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[9:10]
-; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[7:8]
-; GFX7-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v15, s[8:11], 0 offen glc
-; GFX7-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX7-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX7-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX7-NEXT:    s_mov_b64 exec, s[12:13]
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[13:14]
-; GFX7-NEXT:    v_mov_b32_e32 v14, v1
-; GFX7-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX7-NEXT:    v_mov_b32_e32 v13, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, v5
+; GFX7-NEXT:    v_mov_b32_e32 v1, v6
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX7-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX7-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v8, v3
-; GFX6-NEXT:    v_mov_b32_e32 v7, v2
-; GFX6-NEXT:    v_mov_b32_e32 v10, v1
-; GFX6-NEXT:    v_mov_b32_e32 v9, v0
-; GFX6-NEXT:    v_add_i32_e32 v15, vcc, 0x800, v4
 ; GFX6-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX6-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX6-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX6-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX6-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[9:10]
-; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[7:8]
+; GFX6-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX6-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX6-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
+; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX6-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX6-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX6-NEXT:    buffer_load_dwordx2 v[13:14], v4, s[8:11], 0 offen offset:2048
+; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
+; GFX6-NEXT:    buffer_atomic_fmax_x2 v[5:6], v4, s[8:11], 0 offen offset:2048 glc
+; GFX6-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX6-NEXT:    ; implicit-def: $vgpr4
 ; GFX6-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX6-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX6-NEXT:  ; %bb.2:
 ; GFX6-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX6-NEXT:    v_max_f64 v[4:5], v[5:6], v[5:6]
-; GFX6-NEXT:    s_mov_b64 s[6:7], 0
-; GFX6-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Loop Header: Depth=1
-; GFX6-NEXT:    ; Child Loop BB5_4 Depth 2
-; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
-; GFX6-NEXT:    v_max_f64 v[0:1], v[13:14], v[13:14]
-; GFX6-NEXT:    s_mov_b64 s[12:13], exec
-; GFX6-NEXT:    v_max_f64 v[11:12], v[0:1], v[4:5]
-; GFX6-NEXT:    v_mov_b32_e32 v0, v11
-; GFX6-NEXT:    v_mov_b32_e32 v1, v12
-; GFX6-NEXT:    v_mov_b32_e32 v2, v13
-; GFX6-NEXT:    v_mov_b32_e32 v3, v14
-; GFX6-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX6-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX6-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX6-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX6-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX6-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[9:10]
-; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[7:8]
-; GFX6-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX6-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
-; GFX6-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v15, s[8:11], 0 offen glc
-; GFX6-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX6-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX6-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX6-NEXT:    s_mov_b64 exec, s[12:13]
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[13:14]
-; GFX6-NEXT:    v_mov_b32_e32 v14, v1
-; GFX6-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX6-NEXT:    v_mov_b32_e32 v13, v0
+; GFX6-NEXT:    v_mov_b32_e32 v0, v5
+; GFX6-NEXT:    v_mov_b32_e32 v1, v6
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX6-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX6-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr double, ptr addrspace(7) %ptr, i32 256
@@ -7992,32 +7207,11 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset(ptr addrspace(7)
 ; GFX12-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
-; GFX12-NEXT:    s_addk_co_i32 s4, 0x400
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    v_dual_mov_b32 v3, s4 :: v_dual_max_num_f32 v2, v1, v1
-; GFX12-NEXT:    buffer_load_b32 v0, v0, s[0:3], null offen offset:1024
-; GFX12-NEXT:    s_mov_b32 s4, 0
-; GFX12-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX12-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_mov_b32_e32 v5, v0
+; GFX12-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX12-NEXT:    s_wait_storecnt 0x0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-NEXT:    v_max_num_f32_e32 v0, v5, v5
-; GFX12-NEXT:    v_max_num_f32_e32 v4, v0, v2
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5
-; GFX12-NEXT:    buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], null offen th:TH_ATOMIC_RETURN
+; GFX12-NEXT:    buffer_atomic_max_num_f32 v0, v1, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-NEXT:    global_inv scope:SCOPE_SYS
-; GFX12-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX12-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX12-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX12-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX12-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: buffer_fat_ptr_system_atomic_fmax_ret_f32__offset:
@@ -8052,64 +7246,23 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset(ptr addrspace(7)
 ; GFX11-LABEL: buffer_fat_ptr_system_atomic_fmax_ret_f32__offset:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
-; GFX11-NEXT:    s_addk_i32 s4, 0x400
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    v_dual_mov_b32 v3, s4 :: v_dual_max_f32 v2, v1, v1
-; GFX11-NEXT:    buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024
-; GFX11-NEXT:    s_mov_b32 s4, 0
-; GFX11-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_mov_b32_e32 v5, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_max_f32_e32 v0, v5, v5
-; GFX11-NEXT:    v_max_f32_e32 v4, v0, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5
-; GFX11-NEXT:    buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], 0 offen glc
+; GFX11-NEXT:    buffer_atomic_max_f32 v0, v1, s[0:3], 0 offen offset:1024 glc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    buffer_gl1_inv
 ; GFX11-NEXT:    buffer_gl0_inv
-; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX11-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX11-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX11-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: buffer_fat_ptr_system_atomic_fmax_ret_f32__offset:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v1, v0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s8
-; GFX10-NEXT:    s_addk_i32 s8, 0x400
-; GFX10-NEXT:    v_mov_b32_e32 v3, s8
-; GFX10-NEXT:    v_max_f32_e32 v2, v1, v1
-; GFX10-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v5, v0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s8
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f32_e32 v0, v5, v5
-; GFX10-NEXT:    v_max_f32_e32 v4, v0, v2
-; GFX10-NEXT:    v_mov_b32_e32 v0, v4
-; GFX10-NEXT:    v_mov_b32_e32 v1, v5
-; GFX10-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX10-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_system_atomic_fmax_ret_f32__offset:
@@ -8203,60 +7356,19 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset(ptr addrspace(7)
 ; GFX7-LABEL: buffer_fat_ptr_system_atomic_fmax_ret_f32__offset:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v1, v0
-; GFX7-NEXT:    v_mov_b32_e32 v0, s8
-; GFX7-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; GFX7-NEXT:    v_mov_b32_e32 v3, s10
-; GFX7-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v5, v0
-; GFX7-NEXT:    v_mul_f32_e32 v0, 1.0, v5
-; GFX7-NEXT:    v_max_f32_e32 v4, v0, v2
-; GFX7-NEXT:    v_mov_b32_e32 v0, v4
-; GFX7-NEXT:    v_mov_b32_e32 v1, v5
-; GFX7-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s8
+; GFX7-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v5
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_system_atomic_fmax_ret_f32__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v1, v0
-; GFX6-NEXT:    v_mov_b32_e32 v0, s8
-; GFX6-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s10
-; GFX6-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v5, v0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_mul_f32_e32 v0, 1.0, v5
-; GFX6-NEXT:    v_max_f32_e32 v4, v0, v2
-; GFX6-NEXT:    v_mov_b32_e32 v0, v4
-; GFX6-NEXT:    v_mov_b32_e32 v1, v5
-; GFX6-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX6-NEXT:    v_mov_b32_e32 v1, s8
+; GFX6-NEXT:    buffer_atomic_fmax v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v5
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr float, ptr addrspace(7) %ptr, i32 256
diff --git a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmin.ll b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmin.ll
index 89289c15dcae5..4f43ae95533f3 100644
--- a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmin.ll
+++ b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmin.ll
@@ -21,32 +21,11 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset(ptr addrspace(7)
 ; GFX12-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
-; GFX12-NEXT:    s_addk_co_i32 s4, 0x400
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    v_dual_mov_b32 v3, s4 :: v_dual_max_num_f32 v2, v1, v1
-; GFX12-NEXT:    buffer_load_b32 v0, v0, s[0:3], null offen offset:1024
-; GFX12-NEXT:    s_mov_b32 s4, 0
-; GFX12-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX12-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_mov_b32_e32 v5, v0
+; GFX12-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX12-NEXT:    s_wait_storecnt 0x0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-NEXT:    v_max_num_f32_e32 v0, v5, v5
-; GFX12-NEXT:    v_min_num_f32_e32 v4, v0, v2
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5
-; GFX12-NEXT:    buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], null offen th:TH_ATOMIC_RETURN
+; GFX12-NEXT:    buffer_atomic_min_num_f32 v0, v1, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-NEXT:    global_inv scope:SCOPE_DEV
-; GFX12-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX12-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX12-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX12-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX12-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset:
@@ -81,64 +60,23 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset(ptr addrspace(7)
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
-; GFX11-NEXT:    s_addk_i32 s4, 0x400
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    v_dual_mov_b32 v3, s4 :: v_dual_max_f32 v2, v1, v1
-; GFX11-NEXT:    buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024
-; GFX11-NEXT:    s_mov_b32 s4, 0
-; GFX11-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_mov_b32_e32 v5, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_max_f32_e32 v0, v5, v5
-; GFX11-NEXT:    v_min_f32_e32 v4, v0, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5
-; GFX11-NEXT:    buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], 0 offen glc
+; GFX11-NEXT:    buffer_atomic_min_f32 v0, v1, s[0:3], 0 offen offset:1024 glc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    buffer_gl1_inv
 ; GFX11-NEXT:    buffer_gl0_inv
-; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX11-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX11-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX11-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v1, v0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s8
-; GFX10-NEXT:    s_addk_i32 s8, 0x400
-; GFX10-NEXT:    v_mov_b32_e32 v3, s8
-; GFX10-NEXT:    v_max_f32_e32 v2, v1, v1
-; GFX10-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v5, v0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s8
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f32_e32 v0, v5, v5
-; GFX10-NEXT:    v_min_f32_e32 v4, v0, v2
-; GFX10-NEXT:    v_mov_b32_e32 v0, v4
-; GFX10-NEXT:    v_mov_b32_e32 v1, v5
-; GFX10-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX10-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset:
@@ -230,60 +168,19 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset(ptr addrspace(7)
 ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v1, v0
-; GFX7-NEXT:    v_mov_b32_e32 v0, s8
-; GFX7-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; GFX7-NEXT:    v_mov_b32_e32 v3, s10
-; GFX7-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v5, v0
-; GFX7-NEXT:    v_mul_f32_e32 v0, 1.0, v5
-; GFX7-NEXT:    v_min_f32_e32 v4, v0, v2
-; GFX7-NEXT:    v_mov_b32_e32 v0, v4
-; GFX7-NEXT:    v_mov_b32_e32 v1, v5
-; GFX7-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s8
+; GFX7-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v5
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v1, v0
-; GFX6-NEXT:    v_mov_b32_e32 v0, s8
-; GFX6-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s10
-; GFX6-NEXT:  .LBB0_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v5, v0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_mul_f32_e32 v0, 1.0, v5
-; GFX6-NEXT:    v_min_f32_e32 v4, v0, v2
-; GFX6-NEXT:    v_mov_b32_e32 v0, v4
-; GFX6-NEXT:    v_mov_b32_e32 v1, v5
-; GFX6-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX6-NEXT:    v_mov_b32_e32 v1, s8
+; GFX6-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v5
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB0_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr float, ptr addrspace(7) %ptr, i32 256
@@ -299,31 +196,11 @@ define void @buffer_fat_ptr_agent_atomic_fmin_noret_f32__offset(ptr addrspace(7)
 ; GFX12-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_dual_mov_b32 v1, s4 :: v_dual_max_num_f32 v2, v0, v0
-; GFX12-NEXT:    s_addk_co_i32 s4, 0x400
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    v_mov_b32_e32 v3, s4
-; GFX12-NEXT:    buffer_load_b32 v1, v1, s[0:3], null offen offset:1024
-; GFX12-NEXT:    s_mov_b32 s4, 0
-; GFX12-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX12-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_max_num_f32_e32 v0, v1, v1
+; GFX12-NEXT:    v_mov_b32_e32 v1, s4
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    buffer_atomic_min_num_f32 v0, v1, s[0:3], null offen offset:1024
 ; GFX12-NEXT:    s_wait_storecnt 0x0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-NEXT:    v_min_num_f32_e32 v0, v0, v2
-; GFX12-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v4, v0
-; GFX12-NEXT:    buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], null offen th:TH_ATOMIC_RETURN
-; GFX12-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-NEXT:    global_inv scope:SCOPE_DEV
-; GFX12-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v1
-; GFX12-NEXT:    v_mov_b32_e32 v1, v4
-; GFX12-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX12-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX12-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX12-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f32__offset:
@@ -357,62 +234,23 @@ define void @buffer_fat_ptr_agent_atomic_fmin_noret_f32__offset(ptr addrspace(7)
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f32__offset:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v1, s4 :: v_dual_max_f32 v2, v0, v0
-; GFX11-NEXT:    s_addk_i32 s4, 0x400
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    v_mov_b32_e32 v3, s4
-; GFX11-NEXT:    buffer_load_b32 v1, v1, s[0:3], 0 offen offset:1024
-; GFX11-NEXT:    s_mov_b32 s4, 0
-; GFX11-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_max_f32_e32 v0, v1, v1
+; GFX11-NEXT:    v_mov_b32_e32 v1, s4
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    buffer_atomic_min_f32 v0, v1, s[0:3], 0 offen offset:1024
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_min_f32_e32 v0, v0, v2
-; GFX11-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v4, v0
-; GFX11-NEXT:    buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], 0 offen glc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    buffer_gl1_inv
 ; GFX11-NEXT:    buffer_gl0_inv
-; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v1
-; GFX11-NEXT:    v_mov_b32_e32 v1, v4
-; GFX11-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX11-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX11-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f32__offset:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s8
-; GFX10-NEXT:    s_addk_i32 s8, 0x400
-; GFX10-NEXT:    v_max_f32_e32 v2, v0, v0
-; GFX10-NEXT:    v_mov_b32_e32 v3, s8
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:    buffer_load_dword v1, v1, s[4:7], 0 offen offset:1024
-; GFX10-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_max_f32_e32 v0, v1, v1
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_min_f32_e32 v0, v0, v2
-; GFX10-NEXT:    v_mov_b32_e32 v5, v1
-; GFX10-NEXT:    v_mov_b32_e32 v4, v0
-; GFX10-NEXT:    buffer_atomic_cmpswap v[4:5], v3, s[4:7], 0 offen glc
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v1
-; GFX10-NEXT:    v_mov_b32_e32 v1, v4
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f32__offset:
@@ -502,57 +340,18 @@ define void @buffer_fat_ptr_agent_atomic_fmin_noret_f32__offset(ptr addrspace(7)
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s8
-; GFX7-NEXT:    buffer_load_dword v1, v1, s[4:7], 0 offen offset:1024
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mul_f32_e32 v2, 1.0, v0
-; GFX7-NEXT:    v_mov_b32_e32 v3, s10
-; GFX7-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mul_f32_e32 v0, 1.0, v1
-; GFX7-NEXT:    v_min_f32_e32 v0, v0, v2
-; GFX7-NEXT:    v_mov_b32_e32 v5, v1
-; GFX7-NEXT:    v_mov_b32_e32 v4, v0
-; GFX7-NEXT:    buffer_atomic_cmpswap v[4:5], v3, s[4:7], 0 offen glc
+; GFX7-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v1
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    v_mov_b32_e32 v1, v4
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f32__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s8
-; GFX6-NEXT:    buffer_load_dword v1, v1, s[4:7], 0 offen offset:1024
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v0
-; GFX6-NEXT:    v_mov_b32_e32 v3, s10
-; GFX6-NEXT:  .LBB1_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_mul_f32_e32 v0, 1.0, v1
-; GFX6-NEXT:    v_min_f32_e32 v0, v0, v2
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v5, v1
-; GFX6-NEXT:    v_mov_b32_e32 v4, v0
-; GFX6-NEXT:    buffer_atomic_cmpswap v[4:5], v3, s[4:7], 0 offen glc
+; GFX6-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v1
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    v_mov_b32_e32 v1, v4
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB1_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr float, ptr addrspace(7) %ptr, i32 256
@@ -568,8 +367,8 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall(ptr ad
 ; GFX12-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_add_nc_u32_e32 v7, 0x400, v4
 ; GFX12-NEXT:    s_mov_b32 s1, exec_lo
+; GFX12-NEXT:    s_wait_storecnt 0x0
 ; GFX12-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX12-NEXT:    v_readfirstlane_b32 s4, v0
 ; GFX12-NEXT:    v_readfirstlane_b32 s5, v1
@@ -581,54 +380,17 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall(ptr ad
 ; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX12-NEXT:    s_and_b32 s0, vcc_lo, s0
 ; GFX12-NEXT:    s_and_saveexec_b32 s0, s0
-; GFX12-NEXT:    buffer_load_b32 v6, v4, s[4:7], null offen offset:1024
+; GFX12-NEXT:    s_wait_loadcnt 0x0
+; GFX12-NEXT:    buffer_atomic_min_num_f32 v5, v4, s[4:7], null offen offset:1024 th:TH_ATOMIC_RETURN
+; GFX12-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX12-NEXT:    ; implicit-def: $vgpr4
 ; GFX12-NEXT:    s_xor_b32 exec_lo, exec_lo, s0
 ; GFX12-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX12-NEXT:  ; %bb.2:
 ; GFX12-NEXT:    s_mov_b32 exec_lo, s1
-; GFX12-NEXT:    v_max_num_f32_e32 v8, v5, v5
-; GFX12-NEXT:    s_mov_b32 s1, 0
-; GFX12-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX12-NEXT:    ; =>This Loop Header: Depth=1
-; GFX12-NEXT:    ; Child Loop BB2_4 Depth 2
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_max_num_f32_e32 v4, v6, v6
-; GFX12-NEXT:    s_mov_b32 s2, exec_lo
-; GFX12-NEXT:    s_wait_storecnt 0x0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-NEXT:    v_min_num_f32_e32 v5, v4, v8
-; GFX12-NEXT:    v_mov_b32_e32 v4, v5
-; GFX12-NEXT:    v_mov_b32_e32 v5, v6
-; GFX12-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX12-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX12-NEXT:    v_readfirstlane_b32 s4, v0
-; GFX12-NEXT:    v_readfirstlane_b32 s5, v1
-; GFX12-NEXT:    v_readfirstlane_b32 s6, v2
-; GFX12-NEXT:    v_readfirstlane_b32 s7, v3
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX12-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[0:1]
-; GFX12-NEXT:    v_cmp_eq_u64_e64 s0, s[6:7], v[2:3]
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_b32 s0, vcc_lo, s0
-; GFX12-NEXT:    s_and_saveexec_b32 s0, s0
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], null offen th:TH_ATOMIC_RETURN
-; GFX12-NEXT:    s_xor_b32 exec_lo, exec_lo, s0
-; GFX12-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX12-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX12-NEXT:    s_mov_b32 exec_lo, s2
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v6
-; GFX12-NEXT:    v_mov_b32_e32 v6, v4
+; GFX12-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX12-NEXT:    global_inv scope:SCOPE_DEV
-; GFX12-NEXT:    s_or_b32 s1, vcc_lo, s1
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s1
-; GFX12-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX12-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s1
-; GFX12-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX12-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall:
@@ -695,9 +457,8 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall(ptr ad
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_add_nc_u32_e32 v7, 0x400, v4
-; GFX11-NEXT:    s_mov_b32 s1, 0
-; GFX11-NEXT:    s_mov_b32 s2, exec_lo
+; GFX11-NEXT:    s_mov_b32 s1, exec_lo
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX11-NEXT:    v_readfirstlane_b32 s4, v0
 ; GFX11-NEXT:    v_readfirstlane_b32 s5, v1
@@ -709,63 +470,25 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall(ptr ad
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_and_b32 s0, vcc_lo, s0
 ; GFX11-NEXT:    s_and_saveexec_b32 s0, s0
-; GFX11-NEXT:    buffer_load_b32 v6, v4, s[4:7], 0 offen offset:1024
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    buffer_atomic_min_f32 v5, v4, s[4:7], 0 offen offset:1024 glc
+; GFX11-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX11-NEXT:    ; implicit-def: $vgpr4
 ; GFX11-NEXT:    s_xor_b32 exec_lo, exec_lo, s0
 ; GFX11-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX11-NEXT:  ; %bb.2:
-; GFX11-NEXT:    s_mov_b32 exec_lo, s2
-; GFX11-NEXT:    v_max_f32_e32 v8, v5, v5
-; GFX11-NEXT:    .p2align 6
-; GFX11-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX11-NEXT:    ; =>This Loop Header: Depth=1
-; GFX11-NEXT:    ; Child Loop BB2_4 Depth 2
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_max_f32_e32 v4, v6, v6
-; GFX11-NEXT:    s_mov_b32 s2, exec_lo
-; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_min_f32_e32 v5, v4, v8
-; GFX11-NEXT:    v_mov_b32_e32 v4, v5
-; GFX11-NEXT:    v_mov_b32_e32 v5, v6
-; GFX11-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX11-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX11-NEXT:    v_readfirstlane_b32 s4, v0
-; GFX11-NEXT:    v_readfirstlane_b32 s5, v1
-; GFX11-NEXT:    v_readfirstlane_b32 s6, v2
-; GFX11-NEXT:    v_readfirstlane_b32 s7, v3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[0:1]
-; GFX11-NEXT:    v_cmp_eq_u64_e64 s0, s[6:7], v[2:3]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_b32 s0, vcc_lo, s0
-; GFX11-NEXT:    s_and_saveexec_b32 s0, s0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], 0 offen glc
-; GFX11-NEXT:    s_xor_b32 exec_lo, exec_lo, s0
-; GFX11-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX11-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX11-NEXT:    s_mov_b32 exec_lo, s2
+; GFX11-NEXT:    s_mov_b32 exec_lo, s1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v6
-; GFX11-NEXT:    v_mov_b32_e32 v6, v4
+; GFX11-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX11-NEXT:    buffer_gl1_inv
 ; GFX11-NEXT:    buffer_gl0_inv
-; GFX11-NEXT:    s_or_b32 s1, vcc_lo, s1
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s1
-; GFX11-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX11-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s1
-; GFX11-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_add_nc_u32_e32 v7, 0x400, v4
-; GFX10-NEXT:    s_mov_b32 s5, 0
-; GFX10-NEXT:    s_mov_b32 s6, exec_lo
+; GFX10-NEXT:    s_mov_b32 s5, exec_lo
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX10-NEXT:    v_readfirstlane_b32 s8, v0
 ; GFX10-NEXT:    v_readfirstlane_b32 s9, v1
@@ -775,53 +498,19 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall(ptr ad
 ; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[2:3]
 ; GFX10-NEXT:    s_and_b32 s4, vcc_lo, s4
 ; GFX10-NEXT:    s_and_saveexec_b32 s4, s4
-; GFX10-NEXT:    buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    buffer_atomic_fmin v5, v4, s[8:11], 0 offen offset:1024 glc
+; GFX10-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX10-NEXT:    ; implicit-def: $vgpr4
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
 ; GFX10-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX10-NEXT:  ; %bb.2:
-; GFX10-NEXT:    s_mov_b32 exec_lo, s6
-; GFX10-NEXT:    v_max_f32_e32 v8, v5, v5
-; GFX10-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Loop Header: Depth=1
-; GFX10-NEXT:    ; Child Loop BB2_4 Depth 2
+; GFX10-NEXT:    s_mov_b32 exec_lo, s5
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_max_f32_e32 v4, v6, v6
-; GFX10-NEXT:    s_mov_b32 s6, exec_lo
-; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_min_f32_e32 v5, v4, v8
-; GFX10-NEXT:    v_mov_b32_e32 v4, v5
-; GFX10-NEXT:    v_mov_b32_e32 v5, v6
-; GFX10-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX10-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX10-NEXT:    v_readfirstlane_b32 s8, v0
-; GFX10-NEXT:    v_readfirstlane_b32 s9, v1
-; GFX10-NEXT:    v_readfirstlane_b32 s10, v2
-; GFX10-NEXT:    v_readfirstlane_b32 s11, v3
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[0:1]
-; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[2:3]
-; GFX10-NEXT:    s_and_b32 s4, vcc_lo, s4
-; GFX10-NEXT:    s_and_saveexec_b32 s4, s4
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
-; GFX10-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX10-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX10-NEXT:    s_mov_b32 exec_lo, s6
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v4, v6
-; GFX10-NEXT:    v_mov_b32_e32 v6, v4
+; GFX10-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    s_or_b32 s5, vcc_lo, s5
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s5
-; GFX10-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX10-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s5
-; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall:
@@ -1006,7 +695,6 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall(ptr ad
 ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_add_i32_e32 v7, vcc, 0x400, v4
 ; GFX7-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX7-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX7-NEXT:    v_readfirstlane_b32 s8, v0
@@ -1017,55 +705,22 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall(ptr ad
 ; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX7-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX7-NEXT:    buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_atomic_fmin v5, v4, s[8:11], 0 offen offset:1024 glc
+; GFX7-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX7-NEXT:    ; implicit-def: $vgpr4
 ; GFX7-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX7-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX7-NEXT:  ; %bb.2:
 ; GFX7-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX7-NEXT:    s_mov_b64 s[6:7], 0
-; GFX7-NEXT:    v_mul_f32_e32 v8, 1.0, v5
-; GFX7-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Loop Header: Depth=1
-; GFX7-NEXT:    ; Child Loop BB2_4 Depth 2
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mul_f32_e32 v4, 1.0, v6
-; GFX7-NEXT:    v_min_f32_e32 v5, v4, v8
-; GFX7-NEXT:    v_mov_b32_e32 v4, v5
-; GFX7-NEXT:    s_mov_b64 s[12:13], exec
-; GFX7-NEXT:    v_mov_b32_e32 v5, v6
-; GFX7-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX7-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX7-NEXT:    v_readfirstlane_b32 s8, v0
-; GFX7-NEXT:    v_readfirstlane_b32 s9, v1
-; GFX7-NEXT:    v_readfirstlane_b32 s10, v2
-; GFX7-NEXT:    v_readfirstlane_b32 s11, v3
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
-; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
-; GFX7-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc
-; GFX7-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX7-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX7-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX7-NEXT:    s_mov_b64 exec, s[12:13]
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v6
-; GFX7-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX7-NEXT:    v_mov_b32_e32 v6, v4
+; GFX7-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX7-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX7-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[6:7]
-; GFX7-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_add_i32_e32 v7, vcc, 0x400, v4
 ; GFX6-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX6-NEXT:  .LBB2_1: ; =>This Inner Loop Header: Depth=1
 ; GFX6-NEXT:    v_readfirstlane_b32 s8, v0
@@ -1076,49 +731,17 @@ define float @buffer_fat_ptr_agent_atomic_fmin_ret_f32__offset__waterfall(ptr ad
 ; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX6-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX6-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX6-NEXT:    buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024
+; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
+; GFX6-NEXT:    buffer_atomic_fmin v5, v4, s[8:11], 0 offen offset:1024 glc
+; GFX6-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX6-NEXT:    ; implicit-def: $vgpr4
 ; GFX6-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX6-NEXT:    s_cbranch_execnz .LBB2_1
 ; GFX6-NEXT:  ; %bb.2:
 ; GFX6-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX6-NEXT:    s_mov_b64 s[6:7], 0
-; GFX6-NEXT:    v_mul_f32_e32 v8, 1.0, v5
-; GFX6-NEXT:  .LBB2_3: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Loop Header: Depth=1
-; GFX6-NEXT:    ; Child Loop BB2_4 Depth 2
-; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
-; GFX6-NEXT:    v_mul_f32_e32 v4, 1.0, v6
-; GFX6-NEXT:    v_min_f32_e32 v5, v4, v8
-; GFX6-NEXT:    v_mov_b32_e32 v4, v5
-; GFX6-NEXT:    s_mov_b64 s[12:13], exec
-; GFX6-NEXT:    v_mov_b32_e32 v5, v6
-; GFX6-NEXT:  .LBB2_4: ; Parent Loop BB2_3 Depth=1
-; GFX6-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX6-NEXT:    v_readfirstlane_b32 s8, v0
-; GFX6-NEXT:    v_readfirstlane_b32 s9, v1
-; GFX6-NEXT:    v_readfirstlane_b32 s10, v2
-; GFX6-NEXT:    v_readfirstlane_b32 s11, v3
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
-; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
-; GFX6-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX6-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
-; GFX6-NEXT:    buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc
-; GFX6-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX6-NEXT:    s_cbranch_execnz .LBB2_4
-; GFX6-NEXT:  ; %bb.5: ; in Loop: Header=BB2_3 Depth=1
-; GFX6-NEXT:    s_mov_b64 exec, s[12:13]
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v6
-; GFX6-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX6-NEXT:    v_mov_b32_e32 v6, v4
+; GFX6-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX6-NEXT:    s_cbranch_execnz .LBB2_3
-; GFX6-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[6:7]
-; GFX6-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr float, ptr addrspace(7) %ptr, i32 256
@@ -1172,32 +795,11 @@ define double @buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset(ptr addrspace(7)
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset:
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX940-NEXT:    v_mov_b32_e32 v2, v0
-; GFX940-NEXT:    v_mov_b32_e32 v0, s4
-; GFX940-NEXT:    v_mov_b32_e32 v3, v1
-; GFX940-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[0:3], 0 offen offset:2048
-; GFX940-NEXT:    s_add_i32 s6, s4, 0x800
-; GFX940-NEXT:    s_mov_b64 s[4:5], 0
-; GFX940-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX940-NEXT:    v_mov_b32_e32 v6, s6
-; GFX940-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX940-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    v_mov_b64_e32 v[10:11], v[0:1]
-; GFX940-NEXT:    v_max_f64 v[0:1], v[10:11], v[10:11]
-; GFX940-NEXT:    v_min_f64 v[8:9], v[0:1], v[4:5]
-; GFX940-NEXT:    v_mov_b64_e32 v[0:1], v[8:9]
-; GFX940-NEXT:    v_mov_b64_e32 v[2:3], v[10:11]
+; GFX940-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX940-NEXT:    buffer_wbl2 sc1
-; GFX940-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[0:3], 0 offen sc0
+; GFX940-NEXT:    buffer_atomic_min_f64 v[0:1], v2, s[0:3], 0 offen offset:2048 sc0
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    buffer_inv sc1
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[10:11]
-; GFX940-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
-; GFX940-NEXT:    s_andn2_b64 exec, exec, s[4:5]
-; GFX940-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX940-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX940-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset:
@@ -1238,66 +840,21 @@ define double @buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset(ptr addrspace(7)
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s8
-; GFX10-NEXT:    v_mov_b32_e32 v3, v1
-; GFX10-NEXT:    s_addk_i32 s8, 0x800
-; GFX10-NEXT:    v_mov_b32_e32 v6, s8
-; GFX10-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[4:7], 0 offen offset:2048
-; GFX10-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v10, v1
-; GFX10-NEXT:    v_mov_b32_e32 v9, v0
+; GFX10-NEXT:    v_mov_b32_e32 v2, s8
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f64 v[0:1], v[9:10], v[9:10]
-; GFX10-NEXT:    v_min_f64 v[7:8], v[0:1], v[4:5]
-; GFX10-NEXT:    v_mov_b32_e32 v0, v7
-; GFX10-NEXT:    v_mov_b32_e32 v1, v8
-; GFX10-NEXT:    v_mov_b32_e32 v2, v9
-; GFX10-NEXT:    v_mov_b32_e32 v3, v10
-; GFX10-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[4:7], 0 offen glc
+; GFX10-NEXT:    buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen offset:2048 glc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[9:10]
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX90A-NEXT:    v_mov_b32_e32 v2, v0
-; GFX90A-NEXT:    v_mov_b32_e32 v0, s8
-; GFX90A-NEXT:    v_mov_b32_e32 v3, v1
-; GFX90A-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[4:7], 0 offen offset:2048
-; GFX90A-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX90A-NEXT:    s_mov_b64 s[8:9], 0
-; GFX90A-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX90A-NEXT:    v_mov_b32_e32 v6, s10
-; GFX90A-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_pk_mov_b32 v[10:11], v[0:1], v[0:1] op_sel:[0,1]
-; GFX90A-NEXT:    v_max_f64 v[0:1], v[10:11], v[10:11]
-; GFX90A-NEXT:    v_min_f64 v[8:9], v[0:1], v[4:5]
-; GFX90A-NEXT:    v_pk_mov_b32 v[0:1], v[8:9], v[8:9] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[2:3], v[10:11], v[10:11] op_sel:[0,1]
-; GFX90A-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[4:7], 0 offen glc
+; GFX90A-NEXT:    v_mov_b32_e32 v2, s8
+; GFX90A-NEXT:    buffer_atomic_min_f64 v[0:1], v2, s[4:7], 0 offen offset:2048 glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    buffer_wbinvl1
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[10:11]
-; GFX90A-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX90A-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX90A-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX90A-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset:
@@ -1364,73 +921,24 @@ define double @buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset(ptr addrspace(7)
 ; GFX8-NEXT:    s_cbranch_execnz .LBB3_1
 ; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
 ; GFX8-NEXT:    s_or_b64 exec, exec, s[8:9]
-; GFX8-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset:
-; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v2, v0
-; GFX7-NEXT:    v_mov_b32_e32 v0, s8
-; GFX7-NEXT:    v_mov_b32_e32 v3, v1
-; GFX7-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[4:7], 0 offen offset:2048
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX7-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mov_b32_e32 v6, s10
-; GFX7-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v10, v1
-; GFX7-NEXT:    v_mov_b32_e32 v9, v0
-; GFX7-NEXT:    v_max_f64 v[0:1], v[9:10], v[9:10]
-; GFX7-NEXT:    v_min_f64 v[7:8], v[0:1], v[4:5]
-; GFX7-NEXT:    v_mov_b32_e32 v0, v7
-; GFX7-NEXT:    v_mov_b32_e32 v1, v8
-; GFX7-NEXT:    v_mov_b32_e32 v2, v9
-; GFX7-NEXT:    v_mov_b32_e32 v3, v10
-; GFX7-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[4:7], 0 offen glc
+; GFX8-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v2, s8
+; GFX7-NEXT:    buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen offset:2048 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10]
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v2, v0
-; GFX6-NEXT:    v_mov_b32_e32 v0, s8
-; GFX6-NEXT:    v_mov_b32_e32 v3, v1
-; GFX6-NEXT:    buffer_load_dwordx2 v[0:1], v0, s[4:7], 0 offen offset:2048
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX6-NEXT:    v_max_f64 v[4:5], v[2:3], v[2:3]
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mov_b32_e32 v6, s10
-; GFX6-NEXT:  .LBB3_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v10, v1
-; GFX6-NEXT:    v_mov_b32_e32 v9, v0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_max_f64 v[0:1], v[9:10], v[9:10]
-; GFX6-NEXT:    v_min_f64 v[7:8], v[0:1], v[4:5]
-; GFX6-NEXT:    v_mov_b32_e32 v0, v7
-; GFX6-NEXT:    v_mov_b32_e32 v1, v8
-; GFX6-NEXT:    v_mov_b32_e32 v2, v9
-; GFX6-NEXT:    v_mov_b32_e32 v3, v10
-; GFX6-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v6, s[4:7], 0 offen glc
+; GFX6-NEXT:    v_mov_b32_e32 v2, s8
+; GFX6-NEXT:    buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen offset:2048 glc
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10]
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB3_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr double, ptr addrspace(7) %ptr, i32 256
@@ -1479,29 +987,10 @@ define void @buffer_fat_ptr_agent_atomic_fmin_noret_f64__offset(ptr addrspace(7)
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX940-NEXT:    v_mov_b32_e32 v2, s4
-; GFX940-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[0:3], 0 offen offset:2048
-; GFX940-NEXT:    s_add_i32 s6, s4, 0x800
-; GFX940-NEXT:    s_mov_b64 s[4:5], 0
-; GFX940-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX940-NEXT:    v_mov_b32_e32 v6, s6
-; GFX940-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX940-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
-; GFX940-NEXT:    v_min_f64 v[0:1], v[0:1], v[4:5]
-; GFX940-NEXT:    v_mov_b64_e32 v[10:11], v[2:3]
-; GFX940-NEXT:    v_mov_b64_e32 v[8:9], v[0:1]
 ; GFX940-NEXT:    buffer_wbl2 sc1
-; GFX940-NEXT:    buffer_atomic_cmpswap_x2 v[8:11], v6, s[0:3], 0 offen sc0
+; GFX940-NEXT:    buffer_atomic_min_f64 v[0:1], v2, s[0:3], 0 offen offset:2048
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    buffer_inv sc1
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[2:3]
-; GFX940-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
-; GFX940-NEXT:    v_mov_b64_e32 v[2:3], v[8:9]
-; GFX940-NEXT:    s_andn2_b64 exec, exec, s[4:5]
-; GFX940-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX940-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX940-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f64__offset:
@@ -1541,61 +1030,20 @@ define void @buffer_fat_ptr_agent_atomic_fmin_noret_f64__offset(ptr addrspace(7)
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s8
-; GFX10-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX10-NEXT:    s_addk_i32 s8, 0x800
-; GFX10-NEXT:    v_mov_b32_e32 v6, s8
-; GFX10-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[4:7], 0 offen offset:2048
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_min_f64 v[0:1], v[0:1], v[4:5]
-; GFX10-NEXT:    v_mov_b32_e32 v10, v3
-; GFX10-NEXT:    v_mov_b32_e32 v9, v2
-; GFX10-NEXT:    v_mov_b32_e32 v8, v1
-; GFX10-NEXT:    v_mov_b32_e32 v7, v0
-; GFX10-NEXT:    buffer_atomic_cmpswap_x2 v[7:10], v6, s[4:7], 0 offen glc
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen offset:2048
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[2:3]
-; GFX10-NEXT:    v_mov_b32_e32 v2, v7
-; GFX10-NEXT:    v_mov_b32_e32 v3, v8
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f64__offset:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX90A-NEXT:    v_mov_b32_e32 v2, s8
-; GFX90A-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[4:7], 0 offen offset:2048
-; GFX90A-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX90A-NEXT:    s_mov_b64 s[8:9], 0
-; GFX90A-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX90A-NEXT:    v_mov_b32_e32 v6, s10
-; GFX90A-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX90A-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
-; GFX90A-NEXT:    v_min_f64 v[0:1], v[0:1], v[4:5]
-; GFX90A-NEXT:    v_pk_mov_b32 v[10:11], v[2:3], v[2:3] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[8:9], v[0:1], v[0:1] op_sel:[0,1]
-; GFX90A-NEXT:    buffer_atomic_cmpswap_x2 v[8:11], v6, s[4:7], 0 offen glc
+; GFX90A-NEXT:    buffer_atomic_min_f64 v[0:1], v2, s[4:7], 0 offen offset:2048
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
 ; GFX90A-NEXT:    buffer_wbinvl1
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[2:3]
-; GFX90A-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX90A-NEXT:    v_pk_mov_b32 v[2:3], v[8:9], v[8:9] op_sel:[0,1]
-; GFX90A-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX90A-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX90A-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX90A-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f64__offset:
@@ -1664,63 +1112,18 @@ define void @buffer_fat_ptr_agent_atomic_fmin_noret_f64__offset(ptr addrspace(7)
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s8
-; GFX7-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[4:7], 0 offen offset:2048
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX7-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mov_b32_e32 v6, s10
-; GFX7-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
-; GFX7-NEXT:    v_min_f64 v[0:1], v[0:1], v[4:5]
-; GFX7-NEXT:    v_mov_b32_e32 v10, v3
-; GFX7-NEXT:    v_mov_b32_e32 v9, v2
-; GFX7-NEXT:    v_mov_b32_e32 v8, v1
-; GFX7-NEXT:    v_mov_b32_e32 v7, v0
-; GFX7-NEXT:    buffer_atomic_cmpswap_x2 v[7:10], v6, s[4:7], 0 offen glc
+; GFX7-NEXT:    buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen offset:2048
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[2:3]
-; GFX7-NEXT:    v_mov_b32_e32 v2, v7
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    v_mov_b32_e32 v3, v8
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmin_noret_f64__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s8
-; GFX6-NEXT:    buffer_load_dwordx2 v[2:3], v2, s[4:7], 0 offen offset:2048
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x800
-; GFX6-NEXT:    v_max_f64 v[4:5], v[0:1], v[0:1]
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mov_b32_e32 v6, s10
-; GFX6-NEXT:  .LBB4_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_max_f64 v[0:1], v[2:3], v[2:3]
-; GFX6-NEXT:    v_min_f64 v[0:1], v[0:1], v[4:5]
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v10, v3
-; GFX6-NEXT:    v_mov_b32_e32 v9, v2
-; GFX6-NEXT:    v_mov_b32_e32 v8, v1
-; GFX6-NEXT:    v_mov_b32_e32 v7, v0
-; GFX6-NEXT:    buffer_atomic_cmpswap_x2 v[7:10], v6, s[4:7], 0 offen glc
+; GFX6-NEXT:    buffer_atomic_fmin_x2 v[0:1], v2, s[4:7], 0 offen offset:2048
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[2:3]
-; GFX6-NEXT:    v_mov_b32_e32 v2, v7
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    v_mov_b32_e32 v3, v8
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB4_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr double, ptr addrspace(7) %ptr, i32 256
@@ -1804,68 +1207,32 @@ define double @buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall(ptr a
 ; GFX940-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall:
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX940-NEXT:    v_mov_b32_e32 v11, v6
-; GFX940-NEXT:    v_mov_b32_e32 v10, v5
-; GFX940-NEXT:    v_mov_b32_e32 v9, v3
-; GFX940-NEXT:    v_mov_b32_e32 v8, v2
-; GFX940-NEXT:    v_mov_b32_e32 v7, v1
-; GFX940-NEXT:    v_mov_b32_e32 v6, v0
-; GFX940-NEXT:    v_add_u32_e32 v14, 0x800, v4
+; GFX940-NEXT:    v_mov_b32_e32 v7, v6
+; GFX940-NEXT:    v_mov_b32_e32 v6, v5
 ; GFX940-NEXT:    s_mov_b64 s[2:3], exec
+; GFX940-NEXT:    buffer_wbl2 sc1
 ; GFX940-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX940-NEXT:    v_readfirstlane_b32 s4, v6
-; GFX940-NEXT:    v_readfirstlane_b32 s5, v7
-; GFX940-NEXT:    v_readfirstlane_b32 s6, v8
-; GFX940-NEXT:    v_readfirstlane_b32 s7, v9
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, s[4:5], v[6:7]
+; GFX940-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX940-NEXT:    v_readfirstlane_b32 s5, v1
+; GFX940-NEXT:    v_readfirstlane_b32 s6, v2
+; GFX940-NEXT:    v_readfirstlane_b32 s7, v3
+; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, s[4:5], v[0:1]
 ; GFX940-NEXT:    s_nop 0
-; GFX940-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[6:7], v[8:9]
+; GFX940-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[6:7], v[2:3]
 ; GFX940-NEXT:    s_and_b64 s[0:1], vcc, s[0:1]
 ; GFX940-NEXT:    s_and_saveexec_b64 s[0:1], s[0:1]
-; GFX940-NEXT:    buffer_load_dwordx2 v[12:13], v4, s[4:7], 0 offen offset:2048
+; GFX940-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NEXT:    buffer_atomic_min_f64 v[6:7], v4, s[4:7], 0 offen offset:2048 sc0
+; GFX940-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX940-NEXT:    ; implicit-def: $vgpr4
 ; GFX940-NEXT:    s_xor_b64 exec, exec, s[0:1]
 ; GFX940-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX940-NEXT:  ; %bb.2:
 ; GFX940-NEXT:    s_mov_b64 exec, s[2:3]
-; GFX940-NEXT:    s_mov_b64 s[2:3], 0
-; GFX940-NEXT:    v_max_f64 v[4:5], v[10:11], v[10:11]
-; GFX940-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX940-NEXT:    ; =>This Loop Header: Depth=1
-; GFX940-NEXT:    ; Child Loop BB5_4 Depth 2
-; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    v_max_f64 v[0:1], v[12:13], v[12:13]
-; GFX940-NEXT:    v_min_f64 v[10:11], v[0:1], v[4:5]
-; GFX940-NEXT:    v_mov_b64_e32 v[0:1], v[10:11]
-; GFX940-NEXT:    s_mov_b64 s[8:9], exec
-; GFX940-NEXT:    v_mov_b64_e32 v[2:3], v[12:13]
-; GFX940-NEXT:    buffer_wbl2 sc1
-; GFX940-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX940-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX940-NEXT:    v_readfirstlane_b32 s4, v6
-; GFX940-NEXT:    v_readfirstlane_b32 s5, v7
-; GFX940-NEXT:    v_readfirstlane_b32 s6, v8
-; GFX940-NEXT:    v_readfirstlane_b32 s7, v9
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, s[4:5], v[6:7]
-; GFX940-NEXT:    s_nop 0
-; GFX940-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[6:7], v[8:9]
-; GFX940-NEXT:    s_and_b64 s[0:1], vcc, s[0:1]
-; GFX940-NEXT:    s_and_saveexec_b64 s[0:1], s[0:1]
-; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v14, s[4:7], 0 offen sc0
-; GFX940-NEXT:    s_xor_b64 exec, exec, s[0:1]
-; GFX940-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX940-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX940-NEXT:    s_mov_b64 exec, s[8:9]
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[12:13]
-; GFX940-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
-; GFX940-NEXT:    v_mov_b64_e32 v[12:13], v[0:1]
+; GFX940-NEXT:    v_mov_b32_e32 v0, v6
+; GFX940-NEXT:    v_mov_b32_e32 v1, v7
 ; GFX940-NEXT:    buffer_inv sc1
-; GFX940-NEXT:    s_andn2_b64 exec, exec, s[2:3]
-; GFX940-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX940-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX940-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall:
@@ -1941,136 +1308,60 @@ define double @buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall(ptr a
 ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v8, v3
-; GFX10-NEXT:    v_mov_b32_e32 v7, v2
-; GFX10-NEXT:    v_mov_b32_e32 v10, v1
-; GFX10-NEXT:    v_mov_b32_e32 v9, v0
-; GFX10-NEXT:    v_add_nc_u32_e32 v15, 0x800, v4
-; GFX10-NEXT:    s_mov_b32 s5, 0
-; GFX10-NEXT:    s_mov_b32 s6, exec_lo
+; GFX10-NEXT:    s_mov_b32 s5, exec_lo
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX10-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX10-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX10-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[9:10]
-; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[7:8]
+; GFX10-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX10-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX10-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX10-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[0:1]
+; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[2:3]
 ; GFX10-NEXT:    s_and_b32 s4, vcc_lo, s4
 ; GFX10-NEXT:    s_and_saveexec_b32 s4, s4
-; GFX10-NEXT:    buffer_load_dwordx2 v[13:14], v4, s[8:11], 0 offen offset:2048
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    buffer_atomic_fmin_x2 v[5:6], v4, s[8:11], 0 offen offset:2048 glc
+; GFX10-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX10-NEXT:    ; implicit-def: $vgpr4
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
 ; GFX10-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX10-NEXT:  ; %bb.2:
-; GFX10-NEXT:    s_mov_b32 exec_lo, s6
-; GFX10-NEXT:    v_max_f64 v[4:5], v[5:6], v[5:6]
-; GFX10-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Loop Header: Depth=1
-; GFX10-NEXT:    ; Child Loop BB5_4 Depth 2
+; GFX10-NEXT:    s_mov_b32 exec_lo, s5
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_max_f64 v[0:1], v[13:14], v[13:14]
-; GFX10-NEXT:    s_mov_b32 s6, exec_lo
-; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_min_f64 v[11:12], v[0:1], v[4:5]
-; GFX10-NEXT:    v_mov_b32_e32 v0, v11
-; GFX10-NEXT:    v_mov_b32_e32 v1, v12
-; GFX10-NEXT:    v_mov_b32_e32 v2, v13
-; GFX10-NEXT:    v_mov_b32_e32 v3, v14
-; GFX10-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX10-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX10-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX10-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX10-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX10-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[9:10]
-; GFX10-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[7:8]
-; GFX10-NEXT:    s_and_b32 s4, vcc_lo, s4
-; GFX10-NEXT:    s_and_saveexec_b32 s4, s4
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v15, s[8:11], 0 offen glc
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
-; GFX10-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX10-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX10-NEXT:    s_mov_b32 exec_lo, s6
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[13:14]
-; GFX10-NEXT:    v_mov_b32_e32 v14, v1
-; GFX10-NEXT:    v_mov_b32_e32 v13, v0
+; GFX10-NEXT:    v_mov_b32_e32 v0, v5
+; GFX10-NEXT:    v_mov_b32_e32 v1, v6
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    s_or_b32 s5, vcc_lo, s5
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s5
-; GFX10-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX10-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s5
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall:
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX90A-NEXT:    v_mov_b32_e32 v11, v6
-; GFX90A-NEXT:    v_mov_b32_e32 v10, v5
-; GFX90A-NEXT:    v_mov_b32_e32 v9, v3
-; GFX90A-NEXT:    v_mov_b32_e32 v8, v2
-; GFX90A-NEXT:    v_mov_b32_e32 v7, v1
-; GFX90A-NEXT:    v_mov_b32_e32 v6, v0
-; GFX90A-NEXT:    v_add_u32_e32 v14, 0x800, v4
+; GFX90A-NEXT:    v_mov_b32_e32 v7, v6
+; GFX90A-NEXT:    v_mov_b32_e32 v6, v5
 ; GFX90A-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX90A-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX90A-NEXT:    v_readfirstlane_b32 s8, v6
-; GFX90A-NEXT:    v_readfirstlane_b32 s9, v7
-; GFX90A-NEXT:    v_readfirstlane_b32 s10, v8
-; GFX90A-NEXT:    v_readfirstlane_b32 s11, v9
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[6:7]
-; GFX90A-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[8:9]
+; GFX90A-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX90A-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX90A-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX90A-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
+; GFX90A-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX90A-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX90A-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX90A-NEXT:    s_nop 0
-; GFX90A-NEXT:    buffer_load_dwordx2 v[12:13], v4, s[8:11], 0 offen offset:2048
+; GFX90A-NEXT:    s_waitcnt vmcnt(0)
+; GFX90A-NEXT:    buffer_atomic_min_f64 v[6:7], v4, s[8:11], 0 offen offset:2048 glc
+; GFX90A-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX90A-NEXT:    ; implicit-def: $vgpr4
 ; GFX90A-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX90A-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX90A-NEXT:  ; %bb.2:
 ; GFX90A-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX90A-NEXT:    s_mov_b64 s[6:7], 0
-; GFX90A-NEXT:    v_max_f64 v[4:5], v[10:11], v[10:11]
-; GFX90A-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX90A-NEXT:    ; =>This Loop Header: Depth=1
-; GFX90A-NEXT:    ; Child Loop BB5_4 Depth 2
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_max_f64 v[0:1], v[12:13], v[12:13]
-; GFX90A-NEXT:    v_min_f64 v[10:11], v[0:1], v[4:5]
-; GFX90A-NEXT:    v_pk_mov_b32 v[0:1], v[10:11], v[10:11] op_sel:[0,1]
-; GFX90A-NEXT:    s_mov_b64 s[12:13], exec
-; GFX90A-NEXT:    v_pk_mov_b32 v[2:3], v[12:13], v[12:13] op_sel:[0,1]
-; GFX90A-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX90A-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX90A-NEXT:    v_readfirstlane_b32 s8, v6
-; GFX90A-NEXT:    v_readfirstlane_b32 s9, v7
-; GFX90A-NEXT:    v_readfirstlane_b32 s10, v8
-; GFX90A-NEXT:    v_readfirstlane_b32 s11, v9
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[6:7]
-; GFX90A-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[8:9]
-; GFX90A-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX90A-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v14, s[8:11], 0 offen glc
-; GFX90A-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX90A-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX90A-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX90A-NEXT:    s_mov_b64 exec, s[12:13]
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[12:13]
-; GFX90A-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX90A-NEXT:    v_pk_mov_b32 v[12:13], v[0:1], v[0:1] op_sel:[0,1]
+; GFX90A-NEXT:    v_mov_b32_e32 v0, v6
+; GFX90A-NEXT:    v_mov_b32_e32 v1, v7
 ; GFX90A-NEXT:    buffer_wbinvl1
-; GFX90A-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX90A-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX90A-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX90A-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GFX90A-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall:
@@ -2208,131 +1499,55 @@ define double @buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall(ptr a
 ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v8, v3
-; GFX7-NEXT:    v_mov_b32_e32 v7, v2
-; GFX7-NEXT:    v_mov_b32_e32 v10, v1
-; GFX7-NEXT:    v_mov_b32_e32 v9, v0
-; GFX7-NEXT:    v_add_i32_e32 v15, vcc, 0x800, v4
 ; GFX7-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX7-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX7-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX7-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX7-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[9:10]
-; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[7:8]
+; GFX7-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX7-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX7-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX7-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
+; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX7-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX7-NEXT:    buffer_load_dwordx2 v[13:14], v4, s[8:11], 0 offen offset:2048
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_atomic_fmin_x2 v[5:6], v4, s[8:11], 0 offen offset:2048 glc
+; GFX7-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX7-NEXT:    ; implicit-def: $vgpr4
 ; GFX7-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX7-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX7-NEXT:  ; %bb.2:
 ; GFX7-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX7-NEXT:    v_max_f64 v[4:5], v[5:6], v[5:6]
-; GFX7-NEXT:    s_mov_b64 s[6:7], 0
-; GFX7-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Loop Header: Depth=1
-; GFX7-NEXT:    ; Child Loop BB5_4 Depth 2
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_max_f64 v[0:1], v[13:14], v[13:14]
-; GFX7-NEXT:    s_mov_b64 s[12:13], exec
-; GFX7-NEXT:    v_min_f64 v[11:12], v[0:1], v[4:5]
-; GFX7-NEXT:    v_mov_b32_e32 v0, v11
-; GFX7-NEXT:    v_mov_b32_e32 v1, v12
-; GFX7-NEXT:    v_mov_b32_e32 v2, v13
-; GFX7-NEXT:    v_mov_b32_e32 v3, v14
-; GFX7-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX7-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX7-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX7-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX7-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX7-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[9:10]
-; GFX7-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[7:8]
-; GFX7-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v15, s[8:11], 0 offen glc
-; GFX7-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX7-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX7-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX7-NEXT:    s_mov_b64 exec, s[12:13]
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[13:14]
-; GFX7-NEXT:    v_mov_b32_e32 v14, v1
-; GFX7-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX7-NEXT:    v_mov_b32_e32 v13, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, v5
+; GFX7-NEXT:    v_mov_b32_e32 v1, v6
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX7-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX7-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmin_ret_f64__offset__waterfall:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v8, v3
-; GFX6-NEXT:    v_mov_b32_e32 v7, v2
-; GFX6-NEXT:    v_mov_b32_e32 v10, v1
-; GFX6-NEXT:    v_mov_b32_e32 v9, v0
-; GFX6-NEXT:    v_add_i32_e32 v15, vcc, 0x800, v4
 ; GFX6-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX6-NEXT:  .LBB5_1: ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX6-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX6-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX6-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[9:10]
-; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[7:8]
+; GFX6-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX6-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX6-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
+; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3]
 ; GFX6-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
 ; GFX6-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX6-NEXT:    buffer_load_dwordx2 v[13:14], v4, s[8:11], 0 offen offset:2048
+; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
+; GFX6-NEXT:    buffer_atomic_fmin_x2 v[5:6], v4, s[8:11], 0 offen offset:2048 glc
+; GFX6-NEXT:    ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
 ; GFX6-NEXT:    ; implicit-def: $vgpr4
 ; GFX6-NEXT:    s_xor_b64 exec, exec, s[4:5]
 ; GFX6-NEXT:    s_cbranch_execnz .LBB5_1
 ; GFX6-NEXT:  ; %bb.2:
 ; GFX6-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX6-NEXT:    v_max_f64 v[4:5], v[5:6], v[5:6]
-; GFX6-NEXT:    s_mov_b64 s[6:7], 0
-; GFX6-NEXT:  .LBB5_3: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Loop Header: Depth=1
-; GFX6-NEXT:    ; Child Loop BB5_4 Depth 2
-; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
-; GFX6-NEXT:    v_max_f64 v[0:1], v[13:14], v[13:14]
-; GFX6-NEXT:    s_mov_b64 s[12:13], exec
-; GFX6-NEXT:    v_min_f64 v[11:12], v[0:1], v[4:5]
-; GFX6-NEXT:    v_mov_b32_e32 v0, v11
-; GFX6-NEXT:    v_mov_b32_e32 v1, v12
-; GFX6-NEXT:    v_mov_b32_e32 v2, v13
-; GFX6-NEXT:    v_mov_b32_e32 v3, v14
-; GFX6-NEXT:  .LBB5_4: ; Parent Loop BB5_3 Depth=1
-; GFX6-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX6-NEXT:    v_readfirstlane_b32 s8, v9
-; GFX6-NEXT:    v_readfirstlane_b32 s9, v10
-; GFX6-NEXT:    v_readfirstlane_b32 s10, v7
-; GFX6-NEXT:    v_readfirstlane_b32 s11, v8
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[9:10]
-; GFX6-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[10:11], v[7:8]
-; GFX6-NEXT:    s_and_b64 s[4:5], vcc, s[4:5]
-; GFX6-NEXT:    s_and_saveexec_b64 s[4:5], s[4:5]
-; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
-; GFX6-NEXT:    buffer_atomic_cmpswap_x2 v[0:3], v15, s[8:11], 0 offen glc
-; GFX6-NEXT:    s_xor_b64 exec, exec, s[4:5]
-; GFX6-NEXT:    s_cbranch_execnz .LBB5_4
-; GFX6-NEXT:  ; %bb.5: ; in Loop: Header=BB5_3 Depth=1
-; GFX6-NEXT:    s_mov_b64 exec, s[12:13]
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[13:14]
-; GFX6-NEXT:    v_mov_b32_e32 v14, v1
-; GFX6-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
-; GFX6-NEXT:    v_mov_b32_e32 v13, v0
+; GFX6-NEXT:    v_mov_b32_e32 v0, v5
+; GFX6-NEXT:    v_mov_b32_e32 v1, v6
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[6:7]
-; GFX6-NEXT:    s_cbranch_execnz .LBB5_3
-; GFX6-NEXT:  ; %bb.6: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr double, ptr addrspace(7) %ptr, i32 256
@@ -7992,32 +7207,11 @@ define float @buffer_fat_ptr_system_atomic_fmin_ret_f32__offset(ptr addrspace(7)
 ; GFX12-NEXT:    s_wait_samplecnt 0x0
 ; GFX12-NEXT:    s_wait_bvhcnt 0x0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
-; GFX12-NEXT:    s_addk_co_i32 s4, 0x400
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    v_dual_mov_b32 v3, s4 :: v_dual_max_num_f32 v2, v1, v1
-; GFX12-NEXT:    buffer_load_b32 v0, v0, s[0:3], null offen offset:1024
-; GFX12-NEXT:    s_mov_b32 s4, 0
-; GFX12-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX12-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_mov_b32_e32 v5, v0
+; GFX12-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX12-NEXT:    s_wait_storecnt 0x0
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX12-NEXT:    v_max_num_f32_e32 v0, v5, v5
-; GFX12-NEXT:    v_min_num_f32_e32 v4, v0, v2
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5
-; GFX12-NEXT:    buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], null offen th:TH_ATOMIC_RETURN
+; GFX12-NEXT:    buffer_atomic_min_num_f32 v0, v1, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-NEXT:    global_inv scope:SCOPE_SYS
-; GFX12-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX12-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX12-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX12-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX12-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: buffer_fat_ptr_system_atomic_fmin_ret_f32__offset:
@@ -8052,64 +7246,23 @@ define float @buffer_fat_ptr_system_atomic_fmin_ret_f32__offset(ptr addrspace(7)
 ; GFX11-LABEL: buffer_fat_ptr_system_atomic_fmin_ret_f32__offset:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
-; GFX11-NEXT:    s_addk_i32 s4, 0x400
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    v_dual_mov_b32 v3, s4 :: v_dual_max_f32 v2, v1, v1
-; GFX11-NEXT:    buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024
-; GFX11-NEXT:    s_mov_b32 s4, 0
-; GFX11-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_mov_b32_e32 v5, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_max_f32_e32 v0, v5, v5
-; GFX11-NEXT:    v_min_f32_e32 v4, v0, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5
-; GFX11-NEXT:    buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], 0 offen glc
+; GFX11-NEXT:    buffer_atomic_min_f32 v0, v1, s[0:3], 0 offen offset:1024 glc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    buffer_gl1_inv
 ; GFX11-NEXT:    buffer_gl0_inv
-; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX11-NEXT:    s_or_b32 s4, vcc_lo, s4
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s4
-; GFX11-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX11-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: buffer_fat_ptr_system_atomic_fmin_ret_f32__offset:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v1, v0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s8
-; GFX10-NEXT:    s_addk_i32 s8, 0x400
-; GFX10-NEXT:    v_mov_b32_e32 v3, s8
-; GFX10-NEXT:    v_max_f32_e32 v2, v1, v1
-; GFX10-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX10-NEXT:    s_mov_b32 s8, 0
-; GFX10-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v5, v0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s8
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_max_f32_e32 v0, v5, v5
-; GFX10-NEXT:    v_min_f32_e32 v4, v0, v2
-; GFX10-NEXT:    v_mov_b32_e32 v0, v4
-; GFX10-NEXT:    v_mov_b32_e32 v1, v5
-; GFX10-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX10-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    buffer_gl1_inv
 ; GFX10-NEXT:    buffer_gl0_inv
-; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v0, v5
-; GFX10-NEXT:    s_or_b32 s8, vcc_lo, s8
-; GFX10-NEXT:    s_andn2_b32 exec_lo, exec_lo, s8
-; GFX10-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX10-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX10-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX90A-LABEL: buffer_fat_ptr_system_atomic_fmin_ret_f32__offset:
@@ -8203,60 +7356,19 @@ define float @buffer_fat_ptr_system_atomic_fmin_ret_f32__offset(ptr addrspace(7)
 ; GFX7-LABEL: buffer_fat_ptr_system_atomic_fmin_ret_f32__offset:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v1, v0
-; GFX7-NEXT:    v_mov_b32_e32 v0, s8
-; GFX7-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX7-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX7-NEXT:    s_mov_b64 s[8:9], 0
-; GFX7-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; GFX7-NEXT:    v_mov_b32_e32 v3, s10
-; GFX7-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_mov_b32_e32 v5, v0
-; GFX7-NEXT:    v_mul_f32_e32 v0, 1.0, v5
-; GFX7-NEXT:    v_min_f32_e32 v4, v0, v2
-; GFX7-NEXT:    v_mov_b32_e32 v0, v4
-; GFX7-NEXT:    v_mov_b32_e32 v1, v5
-; GFX7-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s8
+; GFX7-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    buffer_wbinvl1
-; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v5
-; GFX7-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX7-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX7-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX7-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX6-LABEL: buffer_fat_ptr_system_atomic_fmin_ret_f32__offset:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v1, v0
-; GFX6-NEXT:    v_mov_b32_e32 v0, s8
-; GFX6-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen offset:1024
-; GFX6-NEXT:    s_add_i32 s10, s8, 0x400
-; GFX6-NEXT:    s_mov_b64 s[8:9], 0
-; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s10
-; GFX6-NEXT:  .LBB18_1: ; %atomicrmw.start
-; GFX6-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_mov_b32_e32 v5, v0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_mul_f32_e32 v0, 1.0, v5
-; GFX6-NEXT:    v_min_f32_e32 v4, v0, v2
-; GFX6-NEXT:    v_mov_b32_e32 v0, v4
-; GFX6-NEXT:    v_mov_b32_e32 v1, v5
-; GFX6-NEXT:    buffer_atomic_cmpswap v[0:1], v3, s[4:7], 0 offen glc
+; GFX6-NEXT:    v_mov_b32_e32 v1, s8
+; GFX6-NEXT:    buffer_atomic_fmin v0, v1, s[4:7], 0 offen offset:1024 glc
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_wbinvl1
-; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, v0, v5
-; GFX6-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GFX6-NEXT:    s_andn2_b64 exec, exec, s[8:9]
-; GFX6-NEXT:    s_cbranch_execnz .LBB18_1
-; GFX6-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX6-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr float, ptr addrspace(7) %ptr, i32 256

