[llvm] d76d8e5 - [AMDGPU][NFC] Update GISel memory-legalizer-atomic-fence test (#72829)

via llvm-commits llvm-commits at lists.llvm.org
Mon Nov 27 16:45:00 PST 2023


Author: Pierre van Houtryve
Date: 2023-11-23T09:09:05+01:00
New Revision: d76d8e541dab6d65b2490957707341352eac8320

URL: https://github.com/llvm/llvm-project/commit/d76d8e541dab6d65b2490957707341352eac8320
DIFF: https://github.com/llvm/llvm-project/commit/d76d8e541dab6d65b2490957707341352eac8320.diff

LOG: [AMDGPU][NFC] Update GISel memory-legalizer-atomic-fence test (#72829)

Test needs to be moved to MIR checks and use
stop-after=si-memory-legalizer to avoid being optimized out in a future
patch.

Added: 
    

Modified: 
    llvm/test/CodeGen/AMDGPU/GlobalISel/memory-legalizer-atomic-fence.ll

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/memory-legalizer-atomic-fence.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/memory-legalizer-atomic-fence.ll
index 0fbfe4cb6f35f13..601cc791c04141a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/memory-legalizer-atomic-fence.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/memory-legalizer-atomic-fence.ll
@@ -1,720 +1,1510 @@
-; RUN: llc -global-isel -mtriple=amdgcn-amd- -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN,GFX6,GFX68 %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd- -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN,GFX8,GFX68 %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN,GFX8,GFX68 %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN,GFX10,GFX10WGP %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+cumode -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN,GFX10,GFX10CU %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN,GFX10,GFX10WGP %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+cumode -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN,GFX10,GFX10CU %s
-
-; FUNC-LABEL: {{^}}system_one_as_acquire:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0){{$}}
-; GFX6-NEXT:  buffer_wbinvl1{{$}}
-; GFX8:       s_waitcnt vmcnt(0){{$}}
-; GFX8-NEXT:  buffer_wbinvl1_vol{{$}}
-; GFX10:      s_waitcnt vmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel system_one_as_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -global-isel -stop-after=si-memory-legalizer -mtriple=amdgcn-amd- -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX6 %s
+; RUN: llc -global-isel -stop-after=si-memory-legalizer -mtriple=amdgcn-amd- -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX8 %s
+; RUN: llc -global-isel -stop-after=si-memory-legalizer -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX8 %s
+; RUN: llc -global-isel -stop-after=si-memory-legalizer -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX10WGP %s
+; RUN: llc -global-isel -stop-after=si-memory-legalizer -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+cumode -verify-machineinstrs < %s | FileCheck -check-prefix=GFX10CU %s
+; RUN: llc -global-isel -stop-after=si-memory-legalizer -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX11WGP %s
+; RUN: llc -global-isel -stop-after=si-memory-legalizer -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+cumode -verify-machineinstrs < %s | FileCheck -check-prefix=GFX11CU %s
+
+; Note: we use MIR test checks + stop after legalizer to prevent
+; tests from being optimized out.
+
 define amdgpu_kernel void @system_one_as_acquire() {
+  ; GFX6-LABEL: name: system_one_as_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 3952
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: system_one_as_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 3952
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: system_one_as_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: system_one_as_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 16240
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: system_one_as_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: system_one_as_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 1015
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("one-as") acquire
   ret void
 }
 
-; FUNC-LABEL: {{^}}system_one_as_release:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_waitcnt vmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel system_one_as_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @system_one_as_release() {
+  ; GFX6-LABEL: name: system_one_as_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 3952
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: system_one_as_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 3952
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: system_one_as_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: system_one_as_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 16240
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: system_one_as_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: system_one_as_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 1015
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("one-as") release
   ret void
 }
 
-; FUNC-LABEL: {{^}}system_one_as_acq_rel:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_waitcnt vmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX6:       buffer_wbinvl1{{$}}
-; GFX8:       buffer_wbinvl1_vol{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel system_one_as_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @system_one_as_acq_rel() {
+  ; GFX6-LABEL: name: system_one_as_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 3952
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: system_one_as_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 3952
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: system_one_as_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: system_one_as_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 16240
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: system_one_as_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: system_one_as_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 1015
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("one-as") acq_rel
   ret void
 }
 
-; FUNC-LABEL: {{^}}system_one_as_seq_cst:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_waitcnt vmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX6:       buffer_wbinvl1{{$}}
-; GFX8:       buffer_wbinvl1_vol{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel system_one_as_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @system_one_as_seq_cst() {
+  ; GFX6-LABEL: name: system_one_as_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 3952
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: system_one_as_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 3952
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: system_one_as_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: system_one_as_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 16240
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: system_one_as_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: system_one_as_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 1015
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("one-as") seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}singlethread_one_as_acquire:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel singlethread_one_as_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @singlethread_one_as_acquire() {
+  ; GFX6-LABEL: name: singlethread_one_as_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: singlethread_one_as_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: singlethread_one_as_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: singlethread_one_as_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: singlethread_one_as_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: singlethread_one_as_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("singlethread-one-as") acquire
   ret void
 }
 
-; FUNC-LABEL: {{^}}singlethread_one_as_release:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel singlethread_one_as_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @singlethread_one_as_release() {
+  ; GFX6-LABEL: name: singlethread_one_as_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: singlethread_one_as_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: singlethread_one_as_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: singlethread_one_as_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: singlethread_one_as_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: singlethread_one_as_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("singlethread-one-as") release
   ret void
 }
 
-; FUNC-LABEL: {{^}}singlethread_one_as_acq_rel:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel singlethread_one_as_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @singlethread_one_as_acq_rel() {
+  ; GFX6-LABEL: name: singlethread_one_as_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: singlethread_one_as_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: singlethread_one_as_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: singlethread_one_as_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: singlethread_one_as_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: singlethread_one_as_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("singlethread-one-as") acq_rel
   ret void
 }
 
-; FUNC-LABEL: {{^}}singlethread_one_as_seq_cst:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel singlethread_one_as_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @singlethread_one_as_seq_cst() {
+  ; GFX6-LABEL: name: singlethread_one_as_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: singlethread_one_as_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: singlethread_one_as_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: singlethread_one_as_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: singlethread_one_as_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: singlethread_one_as_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("singlethread-one-as") seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}agent_one_as_acquire:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0){{$}}
-; GFX6-NEXT:  buffer_wbinvl1{{$}}
-; GFX8:       s_waitcnt vmcnt(0){{$}}
-; GFX8-NEXT:  buffer_wbinvl1_vol{{$}}
-; GFX10:      s_waitcnt vmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel agent_one_as_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @agent_one_as_acquire() {
+  ; GFX6-LABEL: name: agent_one_as_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 3952
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: agent_one_as_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 3952
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: agent_one_as_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: agent_one_as_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 16240
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: agent_one_as_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: agent_one_as_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 1015
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("agent-one-as") acquire
   ret void
 }
 
-; FUNC-LABEL: {{^}}agent_one_as_release:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_waitcnt vmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel agent_one_as_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @agent_one_as_release() {
+  ; GFX6-LABEL: name: agent_one_as_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 3952
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: agent_one_as_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 3952
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: agent_one_as_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: agent_one_as_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 16240
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: agent_one_as_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: agent_one_as_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 1015
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("agent-one-as") release
   ret void
 }
 
-; FUNC-LABEL: {{^}}agent_one_as_acq_rel:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_waitcnt vmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX6:       buffer_wbinvl1{{$}}
-; GFX8:       buffer_wbinvl1_vol{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel agent_one_as_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @agent_one_as_acq_rel() {
+  ; GFX6-LABEL: name: agent_one_as_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 3952
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: agent_one_as_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 3952
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: agent_one_as_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: agent_one_as_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 16240
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: agent_one_as_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: agent_one_as_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 1015
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("agent-one-as") acq_rel
   ret void
 }
 
-; FUNC-LABEL: {{^}}agent_one_as_seq_cst:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_waitcnt vmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX6:       buffer_wbinvl1{{$}}
-; GFX8:       buffer_wbinvl1_vol{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel agent_one_as_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @agent_one_as_seq_cst() {
+  ; GFX6-LABEL: name: agent_one_as_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 3952
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: agent_one_as_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 3952
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: agent_one_as_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: agent_one_as_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 16240
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: agent_one_as_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: agent_one_as_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 1015
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("agent-one-as") seq_cst
   ret void
 }
 
-; FUNC-LABEL:    {{^}}workgroup_one_as_acquire:
-; GCN:           %bb.0
-; GFX68-NOT:     s_waitcnt vmcnt(0){{$}}
-; GFX10WGP:      s_waitcnt vmcnt(0){{$}}
-; GFX10WGP-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10CU-NOT:   s_waitcnt vmcnt(0){{$}}
-; GFX10CU-NOT:   s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10WGP-NEXT: buffer_gl0_inv{{$}}
-; GFX10CU-NOT:   buffer_gl0_inv{{$}}
-; GCN-NOT:       ATOMIC_FENCE
-; GCN:           s_endpgm
-; GFX10:         .amdhsa_kernel workgroup_one_as_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @workgroup_one_as_acquire() {
+  ; GFX6-LABEL: name: workgroup_one_as_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: workgroup_one_as_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: workgroup_one_as_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: workgroup_one_as_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: workgroup_one_as_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: workgroup_one_as_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("workgroup-one-as") acquire
   ret void
 }
 
-; FUNC-LABEL:    {{^}}workgroup_one_as_release:
-; GCN:           %bb.0
-; GFX68-NOT:     s_waitcnt vmcnt(0){{$}}
-; GFX10WGP:      s_waitcnt vmcnt(0){{$}}
-; GFX10WGP-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10CU-NOT:   s_waitcnt vmcnt(0){{$}}
-; GFX10CU-NOT:   s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10-NOT:     buffer_gl0_inv
-; GCN-NOT:       ATOMIC_FENCE
-; GCN:           s_endpgm
-; GFX10:         .amdhsa_kernel workgroup_one_as_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @workgroup_one_as_release() {
+  ; GFX6-LABEL: name: workgroup_one_as_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: workgroup_one_as_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: workgroup_one_as_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: workgroup_one_as_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: workgroup_one_as_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: workgroup_one_as_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("workgroup-one-as") release
   ret void
 }
 
-; FUNC-LABEL:    {{^}}workgroup_one_as_acq_rel:
-; GCN:           %bb.0
-; GFX68-NOT:     s_waitcnt vmcnt(0){{$}}
-; GFX10WGP:      s_waitcnt vmcnt(0){{$}}
-; GFX10WGP-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10WGP-NEXT: buffer_gl0_inv{{$}}
-; GFX10CU-NOT:   s_waitcnt vmcnt(0){{$}}
-; GFX10CU-NOT:   s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10CU-NOT:   buffer_gl0_inv{{$}}
-; GCN-NOT:       ATOMIC_FENCE
-; GCN:           s_endpgm
-; GFX10:         .amdhsa_kernel workgroup_one_as_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @workgroup_one_as_acq_rel() {
+  ; GFX6-LABEL: name: workgroup_one_as_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: workgroup_one_as_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: workgroup_one_as_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: workgroup_one_as_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: workgroup_one_as_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: workgroup_one_as_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("workgroup-one-as") acq_rel
   ret void
 }
 
-; FUNC-LABEL:    {{^}}workgroup_one_as_seq_cst:
-; GCN:           %bb.0
-; GFX68-NOT:     s_waitcnt vmcnt(0){{$}}
-; GFX10WGP:      s_waitcnt vmcnt(0){{$}}
-; GFX10WGP-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10WGP-NEXT: buffer_gl0_inv{{$}}
-; GFX10CU-NOT:   s_waitcnt vmcnt(0){{$}}
-; GFX10CU-NOT:   s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10CU-NOT:   buffer_gl0_inv{{$}}
-; GCN-NOT:       ATOMIC_FENCE
-; GCN:           s_endpgm
-; GFX10:         .amdhsa_kernel workgroup_one_as_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @workgroup_one_as_seq_cst() {
+  ; GFX6-LABEL: name: workgroup_one_as_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: workgroup_one_as_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: workgroup_one_as_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 16240
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: workgroup_one_as_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: workgroup_one_as_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 1015
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: workgroup_one_as_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("workgroup-one-as") seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}wavefront_one_as_acquire:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel wavefront_one_as_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @wavefront_one_as_acquire() {
+  ; GFX6-LABEL: name: wavefront_one_as_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: wavefront_one_as_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: wavefront_one_as_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: wavefront_one_as_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: wavefront_one_as_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: wavefront_one_as_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("wavefront-one-as") acquire
   ret void
 }
 
-; FUNC-LABEL: {{^}}wavefront_one_as_release:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel wavefront_one_as_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @wavefront_one_as_release() {
+  ; GFX6-LABEL: name: wavefront_one_as_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: wavefront_one_as_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: wavefront_one_as_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: wavefront_one_as_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: wavefront_one_as_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: wavefront_one_as_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("wavefront-one-as") release
   ret void
 }
 
-; FUNC-LABEL: {{^}}wavefront_one_as_acq_rel:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel wavefront_one_as_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @wavefront_one_as_acq_rel() {
+  ; GFX6-LABEL: name: wavefront_one_as_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: wavefront_one_as_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: wavefront_one_as_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: wavefront_one_as_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: wavefront_one_as_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: wavefront_one_as_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("wavefront-one-as") acq_rel
   ret void
 }
 
-; FUNC-LABEL: {{^}}wavefront_one_as_seq_cst:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel wavefront_one_as_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @wavefront_one_as_seq_cst() {
+  ; GFX6-LABEL: name: wavefront_one_as_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: wavefront_one_as_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: wavefront_one_as_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: wavefront_one_as_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: wavefront_one_as_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: wavefront_one_as_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("wavefront-one-as") seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}system_acquire:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX6-NEXT:  buffer_wbinvl1{{$}}
-; GFX8:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX8-NEXT:  buffer_wbinvl1_vol{{$}}
-; GFX10:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel system_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @system_acquire() {
+  ; GFX6-LABEL: name: system_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 112
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: system_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 112
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: system_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: system_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 112
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: system_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: system_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 7
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence acquire
   ret void
 }
 
-; FUNC-LABEL: {{^}}system_release:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX8:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel system_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @system_release() {
+  ; GFX6-LABEL: name: system_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 112
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: system_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 112
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: system_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: system_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 112
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: system_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: system_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 7
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence release
   ret void
 }
 
-; FUNC-LABEL: {{^}}system_acq_rel:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX8:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX6:       buffer_wbinvl1{{$}}
-; GFX8:       buffer_wbinvl1_vol{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel system_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @system_acq_rel() {
+  ; GFX6-LABEL: name: system_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 112
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: system_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 112
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: system_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: system_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 112
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: system_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: system_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 7
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence acq_rel
   ret void
 }
 
-; FUNC-LABEL: {{^}}system_seq_cst:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX8:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX6:       buffer_wbinvl1{{$}}
-; GFX8:       buffer_wbinvl1_vol{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel system_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @system_seq_cst() {
+  ; GFX6-LABEL: name: system_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 112
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: system_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 112
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: system_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: system_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 112
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: system_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: system_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 7
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}singlethread_acquire:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel singlethread_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @singlethread_acquire() {
+  ; GFX6-LABEL: name: singlethread_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: singlethread_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: singlethread_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: singlethread_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: singlethread_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: singlethread_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("singlethread") acquire
   ret void
 }
 
-; FUNC-LABEL: {{^}}singlethread_release:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel singlethread_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @singlethread_release() {
+  ; GFX6-LABEL: name: singlethread_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: singlethread_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: singlethread_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: singlethread_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: singlethread_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: singlethread_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("singlethread") release
   ret void
 }
 
-; FUNC-LABEL: {{^}}singlethread_acq_rel:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel singlethread_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @singlethread_acq_rel() {
+  ; GFX6-LABEL: name: singlethread_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: singlethread_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: singlethread_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: singlethread_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: singlethread_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: singlethread_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("singlethread") acq_rel
   ret void
 }
 
-; FUNC-LABEL: {{^}}singlethread_seq_cst:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel singlethread_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @singlethread_seq_cst() {
+  ; GFX6-LABEL: name: singlethread_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: singlethread_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: singlethread_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: singlethread_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: singlethread_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: singlethread_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("singlethread") seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}agent_acquire:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX6-NEXT:  buffer_wbinvl1{{$}}
-; GFX8:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX8-NEXT:  buffer_wbinvl1_vol{{$}}
-; GFX10:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel agent_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @agent_acquire() {
+  ; GFX6-LABEL: name: agent_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 112
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: agent_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 112
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: agent_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: agent_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 112
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: agent_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: agent_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 7
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("agent") acquire
   ret void
 }
 
-; FUNC-LABEL: {{^}}agent_release:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX8:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel agent_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @agent_release() {
+  ; GFX6-LABEL: name: agent_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 112
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: agent_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 112
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: agent_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: agent_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 112
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: agent_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: agent_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 7
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("agent") release
   ret void
 }
 
-; FUNC-LABEL: {{^}}agent_acq_rel:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX8:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX6:       buffer_wbinvl1{{$}}
-; GFX8:       buffer_wbinvl1_vol{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel agent_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @agent_acq_rel() {
+  ; GFX6-LABEL: name: agent_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 112
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: agent_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 112
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: agent_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: agent_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 112
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: agent_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: agent_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 7
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("agent") acq_rel
   ret void
 }
 
-; FUNC-LABEL: {{^}}agent_seq_cst:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GFX6:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX8:       s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX6:       buffer_wbinvl1{{$}}
-; GFX8:       buffer_wbinvl1_vol{{$}}
-; GFX10-NEXT: buffer_gl0_inv{{$}}
-; GFX10-NEXT: buffer_gl1_inv{{$}}
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel agent_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @agent_seq_cst() {
+  ; GFX6-LABEL: name: agent_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 112
+  ; GFX6-NEXT:   BUFFER_WBINVL1 implicit $exec
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: agent_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 112
+  ; GFX8-NEXT:   BUFFER_WBINVL1_VOL implicit $exec
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: agent_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: agent_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 112
+  ; GFX10CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: agent_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: agent_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 7
+  ; GFX11CU-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11CU-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11CU-NEXT:   BUFFER_GL1_INV implicit $exec
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("agent") seq_cst
   ret void
 }
 
-; FUNC-LABEL:    {{^}}workgroup_acquire:
-; GCN:           %bb.0
-; GFX68-NOT:     s_waitcnt vmcnt(0){{$}}
-; GFX10WGP:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10WGP-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10CU-NOT:   s_waitcnt vmcnt(0){{$}}
-; GFX10CU-NOT:   s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10WGP-NEXT: buffer_gl0_inv{{$}}
-; GFX10CU-NOT:   buffer_gl0_inv{{$}}
-; GCN-NOT:       ATOMIC_FENCE
-; GCN:           s_endpgm
-; GFX10:         .amdhsa_kernel workgroup_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @workgroup_acquire() {
+  ; GFX6-LABEL: name: workgroup_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 127
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: workgroup_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 127
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: workgroup_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: workgroup_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 49279
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: workgroup_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: workgroup_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 64519
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("workgroup") acquire
   ret void
 }
 
-; FUNC-LABEL:    {{^}}workgroup_release:
-; GCN:           %bb.0
-; GFX68-NOT:     s_waitcnt vmcnt(0){{$}}
-; GFX10WGP:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10WGP-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10CU-NOT:   s_waitcnt vmcnt(0){{$}}
-; GFX10CU-NOT:   s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10-NOT:     buffer_gl0_inv
-; GCN-NOT:       ATOMIC_FENCE
-; GCN:           s_endpgm
-; GFX10:         .amdhsa_kernel workgroup_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @workgroup_release() {
+  ; GFX6-LABEL: name: workgroup_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 127
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: workgroup_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 127
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: workgroup_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: workgroup_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 49279
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: workgroup_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: workgroup_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 64519
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("workgroup") release
   ret void
 }
 
-; FUNC-LABEL:    {{^}}workgroup_acq_rel:
-; GCN:           %bb.0
-; GFX68-NOT:     s_waitcnt vmcnt(0){{$}}
-; GFX10WGP:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10WGP-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10WGP-NEXT: buffer_gl0_inv{{$}}
-; GFX10CU-NOT:   s_waitcnt vmcnt(0){{$}}
-; GFX10CU-NOT:   s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10CU-NOT:   buffer_gl0_inv{{$}}
-; GCN-NOT:       ATOMIC_FENCE
-; GCN:           s_endpgm
-; GFX10:         .amdhsa_kernel workgroup_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @workgroup_acq_rel() {
+  ; GFX6-LABEL: name: workgroup_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 127
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: workgroup_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 127
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: workgroup_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: workgroup_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 49279
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: workgroup_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: workgroup_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 64519
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("workgroup") acq_rel
   ret void
 }
 
-; FUNC-LABEL:    {{^}}workgroup_seq_cst:
-; GCN:           %bb.0
-; GFX68-NOT:     s_waitcnt vmcnt(0){{$}}
-; GFX10WGP:      s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
-; GFX10WGP-NEXT: s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10WGP-NEXT: buffer_gl0_inv{{$}}
-; GFX10CU-NOT:   s_waitcnt vmcnt(0){{$}}
-; GFX10CU-NOT:   s_waitcnt_vscnt null, 0x0{{$}}
-; GFX10CU-NOT:   buffer_gl0_inv{{$}}
-; GCN-NOT:       ATOMIC_FENCE
-; GCN:           s_endpgm
-; GFX10:         .amdhsa_kernel workgroup_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @workgroup_seq_cst() {
+  ; GFX6-LABEL: name: workgroup_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_WAITCNT 127
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: workgroup_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_WAITCNT 127
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: workgroup_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_WAITCNT 112
+  ; GFX10WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX10WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: workgroup_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_WAITCNT 49279
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: workgroup_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_WAITCNT 7
+  ; GFX11WGP-NEXT:   S_WAITCNT_VSCNT undef $sgpr_null, 0
+  ; GFX11WGP-NEXT:   BUFFER_GL0_INV implicit $exec
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: workgroup_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_WAITCNT 64519
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("workgroup") seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}wavefront_acquire:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel wavefront_acquire
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @wavefront_acquire() {
+  ; GFX6-LABEL: name: wavefront_acquire
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: wavefront_acquire
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: wavefront_acquire
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: wavefront_acquire
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: wavefront_acquire
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: wavefront_acquire
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("wavefront") acquire
   ret void
 }
 
-; FUNC-LABEL: {{^}}wavefront_release:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel wavefront_release
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @wavefront_release() {
+  ; GFX6-LABEL: name: wavefront_release
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: wavefront_release
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: wavefront_release
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: wavefront_release
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: wavefront_release
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: wavefront_release
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("wavefront") release
   ret void
 }
 
-; FUNC-LABEL: {{^}}wavefront_acq_rel:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel wavefront_acq_rel
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @wavefront_acq_rel() {
+  ; GFX6-LABEL: name: wavefront_acq_rel
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: wavefront_acq_rel
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: wavefront_acq_rel
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: wavefront_acq_rel
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: wavefront_acq_rel
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: wavefront_acq_rel
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("wavefront") acq_rel
   ret void
 }
 
-; FUNC-LABEL: {{^}}wavefront_seq_cst:
-; GCN:        %bb.0
-; GCN-NOT:    ATOMIC_FENCE
-; GCN:        s_endpgm
-; GFX10:         .amdhsa_kernel wavefront_seq_cst
-; GFX10WGP-NOT:  .amdhsa_workgroup_processor_mode 0
-; GFX10CU:       .amdhsa_workgroup_processor_mode 0
-; GFX10-NOT:     .amdhsa_memory_ordered 0
 define amdgpu_kernel void @wavefront_seq_cst() {
+  ; GFX6-LABEL: name: wavefront_seq_cst
+  ; GFX6: bb.0.entry:
+  ; GFX6-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX8-LABEL: name: wavefront_seq_cst
+  ; GFX8: bb.0.entry:
+  ; GFX8-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10WGP-LABEL: name: wavefront_seq_cst
+  ; GFX10WGP: bb.0.entry:
+  ; GFX10WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX10CU-LABEL: name: wavefront_seq_cst
+  ; GFX10CU: bb.0.entry:
+  ; GFX10CU-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11WGP-LABEL: name: wavefront_seq_cst
+  ; GFX11WGP: bb.0.entry:
+  ; GFX11WGP-NEXT:   S_ENDPGM 0
+  ;
+  ; GFX11CU-LABEL: name: wavefront_seq_cst
+  ; GFX11CU: bb.0.entry:
+  ; GFX11CU-NEXT:   S_ENDPGM 0
 entry:
   fence syncscope("wavefront") seq_cst
   ret void


        


More information about the llvm-commits mailing list