[AMDGPU] Simplify test and extend to gfx9 and gfx10

[foad]

Summary

This is in preparation for adding more test cases for D69661 and other
bug fixes in the same area.

[arsenm]

llvm/test/CodeGen/AMDGPU/vccz-corrupt-bug-workaround.mir

60

Losing tracksRegLiveness seem like a problem, since this is a case where you're using reg liveness (along with the loss of the liveins lists)

[foad]

llvm/test/CodeGen/AMDGPU/vccz-corrupt-bug-workaround.mir

60

I can put tracksRegLiveness back in of course, but are you saying the insert waitcnts pass depends on liveness? I wasn't aware of that.

[arsenm]

llvm/test/CodeGen/AMDGPU/vccz-corrupt-bug-workaround.mir

60

It has to check if vccz is live out for example. Not tracking liveness at any point is just insane

[foad]

Track liveness.

[arsenm]

llvm/test/CodeGen/AMDGPU/vccz-corrupt-bug-workaround.mir

15

vcc doesn't need to be live in here

[merge_guards_bot]

Unit tests: pass. 62213 tests passed, 0 failed and 815 were skipped.

clang-tidy: pass.

clang-format: pass.

Build artifacts: diff.json, clang-tidy.txt, clang-format.patch, CMakeCache.txt, console-log.txt, test-results.xml

Pre-merge checks is in beta. Report issue. Please join beta or enable it for your project.

llvm/test/CodeGen/AMDGPU/vccz-corrupt-bug-workaround.mir

-# RUN: llc -run-pass si-insert-waitcnts -march=amdgcn -mcpu=tahiti -o - %s | FileCheck %s
---- |
-
-  define amdgpu_kernel void @vccz_corrupt_workaround(float %cond, i32 addrspace(1)* %out) #0 {
-  entry:
-    %cmp0 = fcmp oeq float %cond, 0.000000e+00
-    br i1 %cmp0, label %if, label %else, !structurizecfg.uniform !0, !amdgpu.uniform !0
-
-  else:                                             ; preds = %entry
-    store volatile i32 100, i32 addrspace(1)* undef
-    br label %done, !structurizecfg.uniform !0
-
-  if:                                               ; preds = %entry
-    store volatile i32 9, i32 addrspace(1)* undef
-    br label %done, !structurizecfg.uniform !0
-
-  done:                                             ; preds = %if, %else
-    %value = phi i32 [ 0, %if ], [ 1, %else ]
-    store i32 %value, i32 addrspace(1)* %out
-    ret void
-  }
-
-  define amdgpu_kernel void @vccz_corrupt_undef_vcc(float %cond, i32 addrspace(1)* %out) #0 {
-  entry:
-    br i1 undef, label %if, label %else, !structurizecfg.uniform !0, !amdgpu.uniform !0
-
-  else:                                             ; preds = %entry
-    store volatile i32 100, i32 addrspace(1)* undef
-    br label %done, !structurizecfg.uniform !0
-
-  if:                                               ; preds = %entry
-    store volatile i32 9, i32 addrspace(1)* undef
-    br label %done, !structurizecfg.uniform !0
-
-  done:                                             ; preds = %if, %else
-    %value = phi i32 [ 0, %if ], [ 1, %else ]
-    store i32 %value, i32 addrspace(1)* %out
-    ret void
-  }
-
-  attributes #0 = { nounwind }
-  attributes #1 = { readnone }
-
-  !0 = !{}
-
-...
+# RUN: llc -run-pass=si-insert-waitcnts -march=amdgcn -mcpu=tahiti -o - %s | FileCheck %s -check-prefixes=CHECK,SI
+# RUN: llc -run-pass=si-insert-waitcnts -march=amdgcn -mcpu=gfx900 -o - %s | FileCheck %s -check-prefixes=CHECK,GFX9
+# RUN: llc -run-pass=si-insert-waitcnts -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -o - %s | FileCheck %s -check-prefixes=CHECK,GFX10
 ---
 # CHECK-LABEL: name: vccz_corrupt_workaround
 # CHECK: $vcc = V_CMP_EQ_F32
-# CHECK-NEXT: S_WAITCNT 127
-# CHECK-NEXT: $vcc = S_MOV_B64 $vcc
+# SI-NEXT: S_WAITCNT 127
+# SI-NEXT: $vcc = S_MOV_B64 $vcc
 # CHECK-NEXT: S_CBRANCH_VCCZ %bb.2, implicit killed $vcc
 
-name:            vccz_corrupt_workaround
-alignment:       1
-exposesReturnsTwice: false
-legalized:       false
-regBankSelected: false
-selected:        false
+name: vccz_corrupt_workaround
 tracksRegLiveness: true
-liveins:
-  - { reg: '$sgpr0_sgpr1' }
-frameInfo:
-  isFrameAddressTaken: false
-  isReturnAddressTaken: false
-  hasStackMap:     false
-  hasPatchPoint:   false
-  stackSize:       0
-  offsetAdjustment: 0
-  maxAlignment:    0
-  adjustsStack:    false
-  hasCalls:        false
-  maxCallFrameSize: 0
-  hasOpaqueSPAdjustment: false
-  hasVAStart:      false
-  hasMustTailInVarArgFunc: false
-body:             |
-  bb.0.entry:
+body: |
+  bb.0:
     liveins: $sgpr0_sgpr1, $vcc
 
-    $sgpr0_sgpr1 = S_LOAD_DWORDX2_IMM killed $sgpr0_sgpr1, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
+    $sgpr0_sgpr1 = S_LOAD_DWORDX2_IMM killed $sgpr0_sgpr1, 11, 0, 0
     $sgpr7 = S_MOV_B32 61440
     $sgpr6 = S_MOV_B32 -1
     $vcc = V_CMP_EQ_F32_e64 0, 0, 0, undef $sgpr2, 0, implicit $exec
     S_CBRANCH_VCCZ %bb.1, implicit killed $vcc
 
-  bb.2.if:
+  bb.2:
     liveins: $sgpr6, $sgpr7, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
 
     $vgpr0 = V_MOV_B32_e32 9, implicit $exec
-    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (volatile store 4 into `i32 addrspace(1)* undef`)
+    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec
     $vgpr0 = V_MOV_B32_e32 0, implicit $exec
     S_BRANCH %bb.3
 
-  bb.1.else:
+  bb.1:
     liveins: $sgpr6, $sgpr7, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
 
     $vgpr0 = V_MOV_B32_e32 100, implicit $exec
-    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (volatile store 4 into `i32 addrspace(1)* undef`)
+    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec
     $vgpr0 = V_MOV_B32_e32 1, implicit $exec
 
-  bb.3.done:
+  bb.3:
     liveins: $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
 
     $sgpr3 = S_MOV_B32 61440
     $sgpr2 = S_MOV_B32 -1
-    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (store 4 into %ir.out)
+    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, 0, 0, implicit $exec
     S_ENDPGM 0
 
 ...
 ---
 # CHECK-LABEL: name: vccz_corrupt_undef_vcc
-# CHECK: S_WAITCNT 3855
+# CHECK: BUFFER_STORE_DWORD_OFFSET
+# SI-NEXT: S_WAITCNT 3855
 # CHECK-NEXT: $vgpr0 = V_MOV_B32_e32
 
-name:            vccz_corrupt_undef_vcc
-alignment:       1
-exposesReturnsTwice: false
-legalized:       false
-regBankSelected: false
-selected:        false
+name: vccz_corrupt_undef_vcc
 tracksRegLiveness: true
-liveins:
-  - { reg: '$sgpr0_sgpr1' }
-frameInfo:
-  isFrameAddressTaken: false
-  isReturnAddressTaken: false
-  hasStackMap:     false
-  hasPatchPoint:   false
-  stackSize:       0
-  offsetAdjustment: 0
-  maxAlignment:    0
-  adjustsStack:    false
-  hasCalls:        false
-  maxCallFrameSize: 0
-  hasOpaqueSPAdjustment: false
-  hasVAStart:      false
-  hasMustTailInVarArgFunc: false
-body:             |
-  bb.0.entry:
+body: |
+  bb.0:
     liveins: $sgpr0_sgpr1
 
-    $sgpr0_sgpr1 = S_LOAD_DWORDX2_IMM killed $sgpr0_sgpr1, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
+    $sgpr0_sgpr1 = S_LOAD_DWORDX2_IMM killed $sgpr0_sgpr1, 11, 0, 0
     $sgpr7 = S_MOV_B32 61440
     $sgpr6 = S_MOV_B32 -1
     S_CBRANCH_VCCZ %bb.1, implicit undef $vcc
 
-  bb.2.if:
+  bb.2:
     liveins: $sgpr6, $sgpr7, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
 
     $vgpr0 = V_MOV_B32_e32 9, implicit $exec
-    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (volatile store 4 into `i32 addrspace(1)* undef`)
+    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec
     $vgpr0 = V_MOV_B32_e32 0, implicit $exec
     S_BRANCH %bb.3
 
-  bb.1.else:
+  bb.1:
     liveins: $sgpr6, $sgpr7, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
 
     $vgpr0 = V_MOV_B32_e32 100, implicit $exec
-    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (volatile store 4 into `i32 addrspace(1)* undef`)
+    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec
     $vgpr0 = V_MOV_B32_e32 1, implicit $exec
 
-  bb.3.done:
+  bb.3:
     liveins: $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
 
     $sgpr3 = S_MOV_B32 61440
     $sgpr2 = S_MOV_B32 -1
-    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (store 4 into %ir.out)
+    BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, 0, 0, implicit $exec
     S_ENDPGM 0
 
 ...