Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[AMDGPU][NFC] Update GISel memory-legalizer-atomic-fence test #72829

Merged
merged 1 commit into from
Nov 23, 2023
Merged

[AMDGPU][NFC] Update GISel memory-legalizer-atomic-fence test #72829

merged 1 commit into from
Nov 23, 2023

Conversation

Pierre-vh
Copy link
Contributor

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

@Pierre-vh
[AMDGPU][NFC] Update GISel memory-legalizer-atomic-fence test
1ef3f6c 
Test needs to be moved to MIR checks and use stop-after=si-memory-legalizer to avoid being optimized out in a future patch.
@llvmbot
Copy link
Collaborator

llvmbot commented Nov 20, 2023

@llvm/pr-subscribers-llvm-globalisel

Author: Pierre van Houtryve (Pierre-vh)

Changes

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

Patch is 66.63 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/72829.diff

1 Files Affected:

  • (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/memory-legalizer-atomic-fence.ll (+1272-482)
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...
[truncated]

@llvmbot
Copy link
Collaborator

llvmbot commented Nov 20, 2023

@llvm/pr-subscribers-backend-amdgpu

Author: Pierre van Houtryve (Pierre-vh)

Changes

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

Patch is 66.63 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/72829.diff

1 Files Affected:

  • (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/memory-legalizer-atomic-fence.ll (+1272-482)
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...
[truncated]

@Pierre-vh Pierre-vh mentioned this pull request Nov 20, 2023
Merged
rovka
rovka approved these changes Nov 20, 2023
View reviewed changes
Copy link
Collaborator

@rovka rovka left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@Pierre-vh Pierre-vh merged commit d76d8e5 into llvm:main Nov 23, 2023
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@rovka rovka rovka approved these changes

@jayfoad jayfoad Awaiting requested review from jayfoad

@arsenm arsenm Awaiting requested review from arsenm

Assignees
No one assigned
Labels
backend:AMDGPU llvm:globalisel
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
3 participants
@Pierre-vh @llvmbot @rovka