llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-wavefront.ll - rust-lang/llvm-project - Git at Google

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX7 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-WGP %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+cumode -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-CU %s
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx700 -amdgcn-skip-cache-invalidations -verify-machineinstrs < %s | FileCheck --check-prefixes=SKIP-CACHE-INV %s

 define amdgpu_kernel void @flat_wavefront_unordered_load(
 ; GFX7-LABEL: flat_wavefront_unordered_load:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[2:3], v0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_unordered_load:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_unordered_load:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %in, i32* %out) {
 entry:
   %val = load atomic i32, i32* %in syncscope("wavefront") unordered, align 4
   store i32 %val, i32* %out
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_monotonic_load(
 ; GFX7-LABEL: flat_wavefront_monotonic_load:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[2:3], v0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_monotonic_load:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_monotonic_load:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %in, i32* %out) {
 entry:
   %val = load atomic i32, i32* %in syncscope("wavefront") monotonic, align 4
   store i32 %val, i32* %out
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acquire_load(
 ; GFX7-LABEL: flat_wavefront_acquire_load:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[2:3], v0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acquire_load:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acquire_load:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %in, i32* %out) {
 entry:
   %val = load atomic i32, i32* %in syncscope("wavefront") acquire, align 4
   store i32 %val, i32* %out
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_load(
 ; GFX7-LABEL: flat_wavefront_seq_cst_load:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[2:3], v0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_load:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_load:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %in, i32* %out) {
 entry:
   %val = load atomic i32, i32* %in syncscope("wavefront") seq_cst, align 4
   store i32 %val, i32* %out
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_unordered_store(
 ; GFX7-LABEL: flat_wavefront_unordered_store:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_unordered_store:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_unordered_store:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32 %in, i32* %out) {
 entry:
   store atomic i32 %in, i32* %out syncscope("wavefront") unordered, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_monotonic_store(
 ; GFX7-LABEL: flat_wavefront_monotonic_store:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_monotonic_store:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_monotonic_store:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32 %in, i32* %out) {
 entry:
   store atomic i32 %in, i32* %out syncscope("wavefront") monotonic, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_release_store(
 ; GFX7-LABEL: flat_wavefront_release_store:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_release_store:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_release_store:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32 %in, i32* %out) {
 entry:
   store atomic i32 %in, i32* %out syncscope("wavefront") release, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_store(
 ; GFX7-LABEL: flat_wavefront_seq_cst_store:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_store:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_store:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32 %in, i32* %out) {
 entry:
   store atomic i32 %in, i32* %out syncscope("wavefront") seq_cst, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_monotonic_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_monotonic_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_monotonic_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_monotonic_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront") monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acquire_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_acquire_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acquire_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acquire_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront") acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_release_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_release_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_release_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_release_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront") release
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acq_rel_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_acq_rel_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acq_rel_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acq_rel_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront") acq_rel
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_seq_cst_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront") seq_cst
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acquire_ret_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_acquire_ret_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acquire_ret_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acquire_ret_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront") acquire
   store i32 %val, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acq_rel_ret_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_acq_rel_ret_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acq_rel_ret_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acq_rel_ret_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront") acq_rel
   store i32 %val, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_ret_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_seq_cst_ret_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_ret_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_ret_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront") seq_cst
   store i32 %val, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_monotonic_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_monotonic_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_monotonic_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_monotonic_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") monotonic monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acquire_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_acquire_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acquire_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acquire_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") acquire monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_release_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_release_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_release_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_release_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") release monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acq_rel_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_acq_rel_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acq_rel_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acq_rel_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_seq_cst_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acquire_acquire_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_acquire_acquire_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acquire_acquire_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acquire_acquire_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") acquire acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_release_acquire_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_release_acquire_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_release_acquire_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_release_acquire_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") release acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acq_rel_acquire_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_acq_rel_acquire_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acq_rel_acquire_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acq_rel_acquire_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_acquire_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_seq_cst_acquire_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_acquire_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_acquire_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_seq_cst_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_seq_cst_seq_cst_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_seq_cst_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_seq_cst_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst seq_cst
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acquire_monotonic_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_acquire_monotonic_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acquire_monotonic_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acquire_monotonic_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acq_rel_monotonic_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_monotonic_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acquire_acquire_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_acquire_acquire_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acquire_acquire_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acquire_acquire_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_release_acquire_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_release_acquire_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_release_acquire_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_release_acquire_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_acq_rel_acquire_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_acquire_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_seq_cst_seq_cst_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_unordered_load(
 ; GFX7-LABEL: flat_wavefront_one_as_unordered_load:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[2:3], v0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_unordered_load:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_unordered_load:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %in, i32* %out) {
 entry:
   %val = load atomic i32, i32* %in syncscope("wavefront-one-as") unordered, align 4
   store i32 %val, i32* %out
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_monotonic_load(
 ; GFX7-LABEL: flat_wavefront_one_as_monotonic_load:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[2:3], v0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_monotonic_load:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_monotonic_load:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %in, i32* %out) {
 entry:
   %val = load atomic i32, i32* %in syncscope("wavefront-one-as") monotonic, align 4
   store i32 %val, i32* %out
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acquire_load(
 ; GFX7-LABEL: flat_wavefront_one_as_acquire_load:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[2:3], v0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acquire_load:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acquire_load:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %in, i32* %out) {
 entry:
   %val = load atomic i32, i32* %in syncscope("wavefront-one-as") acquire, align 4
   store i32 %val, i32* %out
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_load(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_load:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[2:3], v0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_load:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_load:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %in, i32* %out) {
 entry:
   %val = load atomic i32, i32* %in syncscope("wavefront-one-as") seq_cst, align 4
   store i32 %val, i32* %out
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_unordered_store(
 ; GFX7-LABEL: flat_wavefront_one_as_unordered_store:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_unordered_store:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_unordered_store:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32 %in, i32* %out) {
 entry:
   store atomic i32 %in, i32* %out syncscope("wavefront-one-as") unordered, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_monotonic_store(
 ; GFX7-LABEL: flat_wavefront_one_as_monotonic_store:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_monotonic_store:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_monotonic_store:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32 %in, i32* %out) {
 entry:
   store atomic i32 %in, i32* %out syncscope("wavefront-one-as") monotonic, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_release_store(
 ; GFX7-LABEL: flat_wavefront_one_as_release_store:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_release_store:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_release_store:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32 %in, i32* %out) {
 entry:
   store atomic i32 %in, i32* %out syncscope("wavefront-one-as") release, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_store(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_store:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_store:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_store:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32 %in, i32* %out) {
 entry:
   store atomic i32 %in, i32* %out syncscope("wavefront-one-as") seq_cst, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_monotonic_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_one_as_monotonic_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_monotonic_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_monotonic_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront-one-as") monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acquire_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_one_as_acquire_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acquire_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acquire_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront-one-as") acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_release_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_one_as_release_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_release_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_release_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront-one-as") release
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acq_rel_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_one_as_acq_rel_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acq_rel_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acq_rel_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront-one-as") acq_rel
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront-one-as") seq_cst
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acquire_ret_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_one_as_acquire_ret_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acquire_ret_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acquire_ret_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront-one-as") acquire
   store i32 %val, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acq_rel_ret_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_one_as_acq_rel_ret_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acq_rel_ret_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acq_rel_ret_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront-one-as") acq_rel
   store i32 %val, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_ret_atomicrmw(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_ret_atomicrmw:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_ret_atomicrmw:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_ret_atomicrmw:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in) {
 entry:
   %val = atomicrmw volatile xchg i32* %out, i32 %in syncscope("wavefront-one-as") seq_cst
   store i32 %val, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_monotonic_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") monotonic monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acquire_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_release_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_release_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_release_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_release_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") release monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acq_rel_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_monotonic_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst monotonic
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acquire_acquire_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_acquire_acquire_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acquire_acquire_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acquire_acquire_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_release_acquire_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_release_acquire_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_release_acquire_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_release_acquire_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") release acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acq_rel_acquire_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_acquire_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst acquire
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_seq_cst_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s0, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s0, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst seq_cst
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acquire_monotonic_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst monotonic
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acquire_acquire_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acquire acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_release_acquire_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") release acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_acq_rel_acquire_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") acq_rel acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_acquire_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst acquire
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }

 define amdgpu_kernel void @flat_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg(
 ; GFX7-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    flat_store_dword v[0:1], v2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-WGP-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX10-WGP:       ; %bb.0: ; %entry
 ; GFX10-WGP-NEXT:    s_clause 0x1
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-WGP-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-WGP-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-WGP-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-WGP-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-WGP-NEXT:    s_endpgm
 ;
 ; GFX10-CU-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX10-CU:       ; %bb.0: ; %entry
 ; GFX10-CU-NEXT:    s_clause 0x1
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX10-CU-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x8
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_add_u32 s4, s0, 16
 ; GFX10-CU-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
     i32* %out, i32 %in, i32 %old) {
 entry:
   %gep = getelementptr i32, i32* %out, i32 4
   %val = cmpxchg volatile i32* %gep, i32 %old, i32 %in syncscope("wavefront-one-as") seq_cst seq_cst
   %val0 = extractvalue { i32, i1 } %val, 0
   store i32 %val0, i32* %out, align 4
   ret void
 }