From 64fa2f45135e24136a5e261f88cf6d7f182a3c89 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Tue, 12 Apr 2016 14:05:11 +0000
Subject: [PATCH] AMDGPU: Implement i64 global atomics

llvm-svn: 266075
---
 llvm/lib/Target/AMDGPU/CIInstructions.td       |  10 +
 llvm/lib/Target/AMDGPU/SIInstructions.td       |  48 +-
 llvm/test/CodeGen/AMDGPU/global_atomics_i64.ll | 842 +++++++++++++++++++++++++
 3 files changed, 886 insertions(+), 14 deletions(-)
 create mode 100644 llvm/test/CodeGen/AMDGPU/global_atomics_i64.ll

diff --git a/llvm/lib/Target/AMDGPU/CIInstructions.td b/llvm/lib/Target/AMDGPU/CIInstructions.td
index 9958bd4..6adc013 100644
--- a/llvm/lib/Target/AMDGPU/CIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/CIInstructions.td
@@ -346,8 +346,18 @@ def : FlatAtomicPat <FLAT_ATOMIC_SWAP_RTN, atomic_swap_global, i32>;
 def : FlatAtomicPat <FLAT_ATOMIC_CMPSWAP_RTN, atomic_cmp_swap_global, i32, v2i32>;
 def : FlatAtomicPat <FLAT_ATOMIC_XOR_RTN, atomic_xor_global, i32>;
 
+def : FlatAtomicPat <FLAT_ATOMIC_ADD_X2_RTN, atomic_add_global, i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_SUB_X2_RTN, atomic_sub_global, i64>;
 def : FlatAtomicPat <FLAT_ATOMIC_INC_X2_RTN, atomic_inc_global, i64>;
 def : FlatAtomicPat <FLAT_ATOMIC_DEC_X2_RTN, atomic_dec_global, i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_AND_X2_RTN, atomic_and_global, i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_SMAX_X2_RTN, atomic_max_global, i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_UMAX_X2_RTN, atomic_umax_global, i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_SMIN_X2_RTN, atomic_min_global, i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_UMIN_X2_RTN, atomic_umin_global, i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_OR_X2_RTN, atomic_or_global, i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_SWAP_X2_RTN, atomic_swap_global, i64>;
 def : FlatAtomicPat <FLAT_ATOMIC_CMPSWAP_X2_RTN, atomic_cmp_swap_global, i64, v2i64>;
+def : FlatAtomicPat <FLAT_ATOMIC_XOR_X2_RTN, atomic_xor_global, i64>;
 
 } // End Predicates = [isCIVI]
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index d794fca..f6244d4 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -1049,23 +1049,43 @@ defm BUFFER_ATOMIC_DEC : MUBUF_Atomic <
   mubuf<0x3d, 0x4c>, "buffer_atomic_dec", VGPR_32, i32, atomic_dec_global
 >;
 
-//def BUFFER_ATOMIC_FCMPSWAP : MUBUF_ <mubuf<0x3e>, "buffer_atomic_fcmpswap", []>; // isn't on VI
-//def BUFFER_ATOMIC_FMIN : MUBUF_ <mubuf<0x3f>, "buffer_atomic_fmin", []>; // isn't on VI
-//def BUFFER_ATOMIC_FMAX : MUBUF_ <mubuf<0x40>, "buffer_atomic_fmax", []>; // isn't on VI
-//def BUFFER_ATOMIC_SWAP_X2 : MUBUF_X2 <mubuf<0x50, 0x60>, "buffer_atomic_swap_x2", []>;
+//def BUFFER_ATOMIC_FCMPSWAP : MUBUF_Atomic <mubuf<0x3e>, "buffer_atomic_fcmpswap", []>; // isn't on VI
+//def BUFFER_ATOMIC_FMIN : MUBUF_Atomic <mubuf<0x3f>, "buffer_atomic_fmin", []>; // isn't on VI
+//def BUFFER_ATOMIC_FMAX : MUBUF_Atomic <mubuf<0x40>, "buffer_atomic_fmax", []>; // isn't on VI
+defm BUFFER_ATOMIC_SWAP_X2 : MUBUF_Atomic <
+  mubuf<0x50, 0x60>, "buffer_atomic_swap_x2", VReg_64, i64, atomic_swap_global
+>;
 defm BUFFER_ATOMIC_CMPSWAP_X2 : MUBUF_Atomic <
   mubuf<0x51, 0x61>, "buffer_atomic_cmpswap_x2", VReg_128, v2i64, null_frag
 >;
-//def BUFFER_ATOMIC_ADD_X2 : MUBUF_X2 <mubuf<0x52, 0x62>, "buffer_atomic_add_x2", []>;
-//def BUFFER_ATOMIC_SUB_X2 : MUBUF_X2 <mubuf<0x53, 0x63>, "buffer_atomic_sub_x2", []>;
-//def BUFFER_ATOMIC_RSUB_X2 : MUBUF_X2 <mubuf<0x54>, "buffer_atomic_rsub_x2", []>; // isn't on CI & VI
-//def BUFFER_ATOMIC_SMIN_X2 : MUBUF_X2 <mubuf<0x55, 0x64>, "buffer_atomic_smin_x2", []>;
-//def BUFFER_ATOMIC_UMIN_X2 : MUBUF_X2 <mubuf<0x56, 0x65>, "buffer_atomic_umin_x2", []>;
-//def BUFFER_ATOMIC_SMAX_X2 : MUBUF_X2 <mubuf<0x57, 0x66>, "buffer_atomic_smax_x2", []>;
-//def BUFFER_ATOMIC_UMAX_X2 : MUBUF_X2 <mubuf<0x58, 0x67>, "buffer_atomic_umax_x2", []>;
-//def BUFFER_ATOMIC_AND_X2 : MUBUF_X2 <mubuf<0x59, 0x68>, "buffer_atomic_and_x2", []>;
-//def BUFFER_ATOMIC_OR_X2 : MUBUF_X2 <mubuf<0x5a, 0x69>, "buffer_atomic_or_x2", []>;
-//def BUFFER_ATOMIC_XOR_X2 : MUBUF_X2 <mubuf<0x5b, 0x6a>, "buffer_atomic_xor_x2", []>;
+defm BUFFER_ATOMIC_ADD_X2 : MUBUF_Atomic <
+  mubuf<0x52, 0x62>, "buffer_atomic_add_x2", VReg_64, i64, atomic_add_global
+>;
+defm BUFFER_ATOMIC_SUB_X2 : MUBUF_Atomic <
+  mubuf<0x53, 0x63>, "buffer_atomic_sub_x2", VReg_64, i64, atomic_sub_global
+>;
+//defm BUFFER_ATOMIC_RSUB_X2 : MUBUF_Atomic <mubuf<0x54>, "buffer_atomic_rsub_x2", []>; // isn't on CI & VI
+defm BUFFER_ATOMIC_SMIN_X2 : MUBUF_Atomic <
+  mubuf<0x55, 0x64>, "buffer_atomic_smin_x2", VReg_64, i64, atomic_min_global
+>;
+defm BUFFER_ATOMIC_UMIN_X2 : MUBUF_Atomic <
+  mubuf<0x56, 0x65>, "buffer_atomic_umin_x2", VReg_64, i64, atomic_umin_global
+>;
+defm BUFFER_ATOMIC_SMAX_X2 : MUBUF_Atomic <
+  mubuf<0x57, 0x66>, "buffer_atomic_smax_x2", VReg_64, i64, atomic_max_global
+>;
+defm BUFFER_ATOMIC_UMAX_X2 : MUBUF_Atomic <
+  mubuf<0x58, 0x67>, "buffer_atomic_umax_x2", VReg_64, i64, atomic_umax_global
+>;
+defm BUFFER_ATOMIC_AND_X2 : MUBUF_Atomic <
+  mubuf<0x59, 0x68>, "buffer_atomic_and_x2", VReg_64, i64, atomic_and_global
+>;
+defm BUFFER_ATOMIC_OR_X2 : MUBUF_Atomic <
+  mubuf<0x5a, 0x69>, "buffer_atomic_or_x2", VReg_64, i64, atomic_or_global
+>;
+defm BUFFER_ATOMIC_XOR_X2 : MUBUF_Atomic <
+  mubuf<0x5b, 0x6a>, "buffer_atomic_xor_x2", VReg_64, i64, atomic_xor_global
+>;
 defm BUFFER_ATOMIC_INC_X2 : MUBUF_Atomic <
   mubuf<0x5c, 0x6b>, "buffer_atomic_inc_x2", VReg_64, i64, atomic_inc_global
 >;
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_i64.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_i64.ll
new file mode 100644
index 0000000..3eabec4
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_i64.ll
@@ -0,0 +1,842 @@
+; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
+
+
+; GCN-LABEL: {{^}}atomic_add_i64_offset:
+; GCN: buffer_atomic_add_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_add_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile add i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_ret_offset:
+; GCN: buffer_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_add_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile add i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_addr64_offset:
+; CI: buffer_atomic_add_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_add_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}}{{$}}
+define void @atomic_add_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile add i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_ret_addr64_offset:
+; CI: buffer_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_add_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile add i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64:
+; GCN: buffer_atomic_add_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_add_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile add i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_ret:
+; GCN: buffer_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_add_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile add i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_addr64:
+; CI: buffer_atomic_add_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_add_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_add_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile add i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_add_i64_ret_addr64:
+; CI: buffer_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_add_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_add_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile add i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_offset:
+; GCN: buffer_atomic_and_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_and_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile and i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_ret_offset:
+; GCN: buffer_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_and_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile and i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_addr64_offset:
+; CI: buffer_atomic_and_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_and_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_and_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile and i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_ret_addr64_offset:
+; CI: buffer_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_and_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile and i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64:
+; GCN: buffer_atomic_and_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_and_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile and i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_ret:
+; GCN: buffer_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_and_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile and i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_addr64:
+; CI: buffer_atomic_and_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_and_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_and_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile and i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_and_i64_ret_addr64:
+; CI: buffer_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_and_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_and_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile and i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_offset:
+; GCN: buffer_atomic_sub_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_sub_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile sub i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_ret_offset:
+; GCN: buffer_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_sub_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile sub i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_addr64_offset:
+; CI: buffer_atomic_sub_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_sub_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_sub_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile sub i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_ret_addr64_offset:
+; CI: buffer_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_sub_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile sub i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64:
+; GCN: buffer_atomic_sub_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_sub_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile sub i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_ret:
+; GCN: buffer_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_sub_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile sub i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_addr64:
+; CI: buffer_atomic_sub_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_sub_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_sub_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile sub i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_sub_i64_ret_addr64:
+; CI: buffer_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_sub_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_sub_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile sub i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_offset:
+; GCN: buffer_atomic_smax_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_max_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile max i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_ret_offset:
+; GCN: buffer_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_max_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile max i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_addr64_offset:
+; CI: buffer_atomic_smax_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_smax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_max_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile max i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_ret_addr64_offset:
+; CI: buffer_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_max_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile max i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64:
+; GCN: buffer_atomic_smax_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_max_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile max i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_ret:
+; GCN: buffer_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_max_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile max i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_addr64:
+; CI: buffer_atomic_smax_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_smax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_max_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile max i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_max_i64_ret_addr64:
+; CI: buffer_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_smax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_max_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile max i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_offset:
+; GCN: buffer_atomic_umax_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_umax_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile umax i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_ret_offset:
+; GCN: buffer_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_umax_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile umax i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_addr64_offset:
+; CI: buffer_atomic_umax_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_umax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umax_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile umax i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_ret_addr64_offset:
+; CI: buffer_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_umax_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile umax i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64:
+; GCN: buffer_atomic_umax_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_umax_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile umax i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_ret:
+; GCN: buffer_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_umax_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile umax i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_addr64:
+; CI: buffer_atomic_umax_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_umax_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umax_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile umax i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umax_i64_ret_addr64:
+; CI: buffer_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_umax_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_umax_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile umax i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_offset:
+; GCN: buffer_atomic_smin_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_min_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile min i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_ret_offset:
+; GCN: buffer_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_min_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile min i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_addr64_offset:
+; CI: buffer_atomic_smin_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_smin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_min_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile min i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_ret_addr64_offset:
+; CI: buffer_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_min_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile min i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64:
+; GCN: buffer_atomic_smin_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_min_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile min i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_ret:
+; GCN: buffer_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_min_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile min i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_addr64:
+; CI: buffer_atomic_smin_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_smin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_min_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile min i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_min_i64_ret_addr64:
+; CI: buffer_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_smin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_min_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile min i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_offset:
+; GCN: buffer_atomic_umin_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_umin_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile umin i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_ret_offset:
+; GCN: buffer_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_umin_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile umin i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_addr64_offset:
+; CI: buffer_atomic_umin_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_umin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umin_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile umin i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_ret_addr64_offset:
+; CI: buffer_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_umin_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile umin i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64:
+; GCN: buffer_atomic_umin_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_umin_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile umin i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_ret:
+; CI: buffer_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_umin_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile umin i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_addr64:
+; CI: buffer_atomic_umin_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_umin_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_umin_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile umin i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_umin_i64_ret_addr64:
+; CI: buffer_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_umin_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_umin_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile umin i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_offset:
+; GCN: buffer_atomic_or_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_or_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile or i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_ret_offset:
+; GCN: buffer_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_or_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile or i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_addr64_offset:
+; CI: buffer_atomic_or_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_or_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_or_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile or i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_ret_addr64_offset:
+; CI: buffer_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_or_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile or i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64:
+; GCN: buffer_atomic_or_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_or_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile or i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_ret:
+; GCN: buffer_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_or_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile or i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_addr64:
+; CI: buffer_atomic_or_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_or_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_or_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile or i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_or_i64_ret_addr64:
+; CI: buffer_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_or_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_or_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile or i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_offset:
+; GCN: buffer_atomic_swap_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_xchg_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile xchg i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_ret_offset:
+; GCN: buffer_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_xchg_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile xchg i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_addr64_offset:
+; CI: buffer_atomic_swap_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+define void @atomic_xchg_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile xchg i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_ret_addr64_offset:
+; CI: buffer_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_xchg_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile xchg i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64:
+; GCN: buffer_atomic_swap_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_xchg_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile xchg i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_ret:
+; GCN: buffer_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_xchg_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile xchg i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_addr64:
+; CI: buffer_atomic_swap_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_swap_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xchg_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile xchg i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xchg_i64_ret_addr64:
+; CI: buffer_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_swap_x2 [[RET:v\[[0-9]+:[0-9]+\]]],  v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_xchg_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile xchg i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_offset:
+; GCN: buffer_atomic_xor_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32{{$}}
+define void @atomic_xor_i64_offset(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile xor i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_ret_offset:
+; GCN: buffer_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:32 glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_xor_i64_ret_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %gep = getelementptr i64, i64 addrspace(1)* %out, i64 4
+  %tmp0  = atomicrmw volatile xor i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_addr64_offset:
+; CI: buffer_atomic_xor_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32{{$}}
+; VI: flat_atomic_xor_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xor_i64_addr64_offset(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile xor i64 addrspace(1)* %gep, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_ret_addr64_offset:
+; CI: buffer_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 offset:32 glc{{$}}
+; VI: flat_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_xor_i64_ret_addr64_offset(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %gep = getelementptr i64, i64 addrspace(1)* %ptr, i64 4
+  %tmp0  = atomicrmw volatile xor i64 addrspace(1)* %gep, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64:
+; GCN: buffer_atomic_xor_x2 v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+}}:{{[0-9]+}}], 0{{$}}
+define void @atomic_xor_i64(i64 addrspace(1)* %out, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile xor i64 addrspace(1)* %out, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_ret:
+; GCN: buffer_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+}}:{{[0-9]+}}], 0 glc
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_xor_i64_ret(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in) {
+entry:
+  %tmp0  = atomicrmw volatile xor i64 addrspace(1)* %out, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_addr64:
+; CI: buffer_atomic_xor_x2 v{{\[[0-9]+:[0-9]+\]}}, v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64{{$}}
+; VI: flat_atomic_xor_x2 v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]$}}
+define void @atomic_xor_i64_addr64(i64 addrspace(1)* %out, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile xor i64 addrspace(1)* %ptr, i64 %in seq_cst
+  ret void
+}
+
+; GCN-LABEL: {{^}}atomic_xor_i64_ret_addr64:
+; CI: buffer_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+}}:{{[0-9]+}}], s[{{[0-9]+}}:{{[0-9]+}}], 0 addr64 glc{{$}}
+; VI: flat_atomic_xor_x2 [[RET:v\[[0-9]+:[0-9]+\]]], v[{{[0-9]+:[0-9]+}}], v{{\[[0-9]+:[0-9]+\]}} glc{{$}}
+; GCN: buffer_store_dwordx2 [[RET]]
+define void @atomic_xor_i64_ret_addr64(i64 addrspace(1)* %out, i64 addrspace(1)* %out2, i64 %in, i64 %index) {
+entry:
+  %ptr = getelementptr i64, i64 addrspace(1)* %out, i64 %index
+  %tmp0  = atomicrmw volatile xor i64 addrspace(1)* %ptr, i64 %in seq_cst
+  store i64 %tmp0, i64 addrspace(1)* %out2
+  ret void
+}
-- 
2.7.4