AMDGPU: Fix si-fix-sgpr-copies asserting on VReg_1 phi #128903
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
We should just need to leave these alone for the si-i1-copies to deal with it. Note the IR test did not fail previously, and only establishes a baseline result for a pending patch which hit the failure.
|
@llvm/pr-subscribers-backend-amdgpu Author: Matt Arsenault (arsenm) ChangesWe should just need to leave these alone for the si-i1-copies Full diff: https://github.com/llvm/llvm-project/pull/128903.diff 3 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp b/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
index a99fd25477553..52d00485385c2 100644
--- a/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
@@ -669,6 +669,9 @@ bool SIFixSGPRCopies::run(MachineFunction &MF) {
if (!MO.isReg() || !MO.getReg().isVirtual())
continue;
const TargetRegisterClass *SrcRC = MRI->getRegClass(MO.getReg());
+ if (SrcRC == &AMDGPU::VReg_1RegClass)
+ continue;
+
if (TRI->hasVectorRegisters(SrcRC)) {
const TargetRegisterClass *DestRC =
TRI->getEquivalentSGPRClass(SrcRC);
diff --git a/llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll b/llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll
index 757458363284c..0a420396f52a9 100644
--- a/llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll
+++ b/llvm/test/CodeGen/AMDGPU/divergent-branch-uniform-condition.ll
@@ -121,6 +121,68 @@ endloop: ; preds = %if1, %Flow2
ret void
}
+define amdgpu_ps void @i1_copy_assert(i1 %v4) {
+; ISA-LABEL: i1_copy_assert:
+; ISA: ; %bb.0: ; %start
+; ISA-NEXT: v_and_b32_e32 v0, 1, v0
+; ISA-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
+; ISA-NEXT: s_mov_b32 s8, 0
+; ISA-NEXT: s_mov_b64 s[0:1], 0
+; ISA-NEXT: ; implicit-def: $sgpr4_sgpr5
+; ISA-NEXT: ; implicit-def: $sgpr2_sgpr3
+; ISA-NEXT: s_branch .LBB1_3
+; ISA-NEXT: .LBB1_1: ; %endif1
+; ISA-NEXT: ; in Loop: Header=BB1_3 Depth=1
+; ISA-NEXT: s_andn2_b64 s[4:5], s[4:5], exec
+; ISA-NEXT: s_and_b64 s[8:9], vcc, exec
+; ISA-NEXT: s_mov_b64 s[6:7], 0
+; ISA-NEXT: s_or_b64 s[4:5], s[4:5], s[8:9]
+; ISA-NEXT: .LBB1_2: ; %Flow
+; ISA-NEXT: ; in Loop: Header=BB1_3 Depth=1
+; ISA-NEXT: s_and_b64 s[8:9], exec, s[4:5]
+; ISA-NEXT: s_or_b64 s[0:1], s[8:9], s[0:1]
+; ISA-NEXT: s_andn2_b64 s[2:3], s[2:3], exec
+; ISA-NEXT: s_and_b64 s[6:7], s[6:7], exec
+; ISA-NEXT: s_mov_b32 s8, 1
+; ISA-NEXT: s_or_b64 s[2:3], s[2:3], s[6:7]
+; ISA-NEXT: s_andn2_b64 exec, exec, s[0:1]
+; ISA-NEXT: s_cbranch_execz .LBB1_5
+; ISA-NEXT: .LBB1_3: ; %loop
+; ISA-NEXT: ; =>This Inner Loop Header: Depth=1
+; ISA-NEXT: s_or_b64 s[4:5], s[4:5], exec
+; ISA-NEXT: s_cmp_lg_u32 s8, 0
+; ISA-NEXT: s_cbranch_scc1 .LBB1_1
+; ISA-NEXT: ; %bb.4: ; in Loop: Header=BB1_3 Depth=1
+; ISA-NEXT: s_mov_b64 s[6:7], -1
+; ISA-NEXT: s_branch .LBB1_2
+; ISA-NEXT: .LBB1_5: ; %Flow2
+; ISA-NEXT: s_or_b64 exec, exec, s[0:1]
+; ISA-NEXT: v_mov_b32_e32 v0, 0
+; ISA-NEXT: v_cndmask_b32_e64 v1, 0, 1.0, s[2:3]
+; ISA-NEXT: exp mrt0 off, off, off, off
+; ISA-NEXT: s_endpgm
+start:
+ br label %loop
+
+loop: ; preds = %Flow, %start
+ %v1 = phi i32 [ 0, %start ], [ 1, %Flow ]
+ %v2 = icmp ugt i32 %v1, 0
+ br i1 %v2, label %endif1, label %Flow
+
+Flow2: ; preds = %Flow
+ %spec.select = select i1 %i1, float 1.000000e+00, float 0.000000e+00
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float %spec.select, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i1 false, i1 false)
+ ret void
+
+endif1: ; preds = %loop
+ br label %Flow
+
+Flow: ; preds = %endif1, %loop
+ %i = phi i1 [ %v4, %endif1 ], [ true, %loop ]
+ %i1 = phi i1 [ false, %endif1 ], [ true, %loop ]
+ br i1 %i, label %Flow2, label %loop
+}
+
; Function Attrs: nounwind readnone speculatable willreturn
declare float @llvm.sqrt.f32(float) #0
diff --git a/llvm/test/CodeGen/AMDGPU/i1-divergent-phi-fix-sgpr-copies-assert.mir b/llvm/test/CodeGen/AMDGPU/i1-divergent-phi-fix-sgpr-copies-assert.mir
new file mode 100644
index 0000000000000..1b5999c689178
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/i1-divergent-phi-fix-sgpr-copies-assert.mir
@@ -0,0 +1,97 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=si-fix-sgpr-copies,si-i1-copies -o - %s | FileCheck %s
+
+# Make sure SIFixSGPRCopies does not assert on a phi with vreg_1
+# inputs.
+
+---
+name: i1_copy_assert
+tracksRegLiveness: true
+body: |
+ ; CHECK-LABEL: name: i1_copy_assert
+ ; CHECK: bb.0:
+ ; CHECK-NEXT: successors: %bb.1(0x80000000)
+ ; CHECK-NEXT: liveins: $vgpr0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; CHECK-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 killed [[COPY]], 1, implicit $exec
+ ; CHECK-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
+ ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.1:
+ ; CHECK-NEXT: successors: %bb.3(0x40000000), %bb.4(0x40000000)
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF1]], %bb.0, %14, %bb.4
+ ; CHECK-NEXT: [[PHI1:%[0-9]+]]:sreg_64 = PHI [[DEF]], %bb.0, %8, %bb.4
+ ; CHECK-NEXT: [[S_MOV_B64_1:%[0-9]+]]:sreg_64 = S_MOV_B64 -1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sreg_64 = COPY [[S_MOV_B64_1]]
+ ; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
+ ; CHECK-NEXT: [[DEF2:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[S_OR_B64_:%[0-9]+]]:sreg_64 = S_OR_B64 [[PHI]], $exec, implicit-def $scc
+ ; CHECK-NEXT: S_CMP_LG_U32 [[DEF2]], killed [[S_MOV_B32_]], implicit-def $scc
+ ; CHECK-NEXT: S_CBRANCH_SCC1 %bb.3, implicit $scc
+ ; CHECK-NEXT: S_BRANCH %bb.4
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.2:
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sreg_64_xexec = COPY %8
+ ; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
+ ; CHECK-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, [[V_MOV_B32_e32_]], [[COPY2]], implicit $exec
+ ; CHECK-NEXT: S_ENDPGM 0, implicit killed [[V_CNDMASK_B32_e64_]]
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.3:
+ ; CHECK-NEXT: successors: %bb.4(0x80000000)
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[S_MOV_B64_2:%[0-9]+]]:sreg_64 = S_MOV_B64 0
+ ; CHECK-NEXT: [[S_ANDN2_B64_:%[0-9]+]]:sreg_64 = S_ANDN2_B64 [[S_OR_B64_]], $exec, implicit-def $scc
+ ; CHECK-NEXT: [[S_AND_B64_:%[0-9]+]]:sreg_64 = S_AND_B64 [[V_CMP_EQ_U32_e64_]], $exec, implicit-def $scc
+ ; CHECK-NEXT: [[S_OR_B64_1:%[0-9]+]]:sreg_64 = S_OR_B64 [[S_ANDN2_B64_]], [[S_AND_B64_]], implicit-def $scc
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.4:
+ ; CHECK-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000)
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[PHI2:%[0-9]+]]:sreg_64 = PHI [[S_OR_B64_]], %bb.1, [[S_OR_B64_1]], %bb.3
+ ; CHECK-NEXT: [[PHI3:%[0-9]+]]:sreg_64 = PHI [[COPY1]], %bb.1, [[S_MOV_B64_2]], %bb.3
+ ; CHECK-NEXT: [[COPY3:%[0-9]+]]:sreg_64 = COPY [[PHI2]]
+ ; CHECK-NEXT: [[DEF3:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[S_ANDN2_B64_1:%[0-9]+]]:sreg_64 = S_ANDN2_B64 [[PHI1]], $exec, implicit-def $scc
+ ; CHECK-NEXT: [[S_AND_B64_1:%[0-9]+]]:sreg_64 = S_AND_B64 [[PHI3]], $exec, implicit-def $scc
+ ; CHECK-NEXT: [[S_OR_B64_2:%[0-9]+]]:sreg_64 = S_OR_B64 [[S_ANDN2_B64_1]], [[S_AND_B64_1]], implicit-def $scc
+ ; CHECK-NEXT: SI_LOOP [[DEF3]], %bb.1, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+ ; CHECK-NEXT: S_BRANCH %bb.2
+ bb.0:
+ liveins: $vgpr0
+
+ %0:vgpr_32 = COPY $vgpr0
+ %1:sreg_64 = V_CMP_EQ_U32_e64 killed %0, 1, implicit $exec
+ %2:sreg_64 = S_MOV_B64 0
+ %3:vreg_1 = COPY %1
+
+ bb.1:
+ %4:sreg_64 = S_MOV_B64 -1
+ %5:vreg_1 = COPY %4
+ %6:sreg_32 = S_MOV_B32 0
+ %7:sreg_32 = IMPLICIT_DEF
+ S_CMP_LG_U32 %7, killed %6, implicit-def $scc
+ S_CBRANCH_SCC1 %bb.3, implicit $scc
+ S_BRANCH %bb.4
+
+ bb.2:
+ %8:vreg_1 = PHI %9, %bb.4
+ %10:sreg_64_xexec = COPY %8
+ %11:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
+ %12:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, %11, %10, implicit $exec
+ S_ENDPGM 0, implicit killed %12
+
+ bb.3:
+ %13:sreg_64 = S_MOV_B64 0
+
+ bb.4:
+ %14:vreg_1 = PHI %5, %bb.1, %3, %bb.3
+ %9:sreg_64 = PHI %5, %bb.1, %13, %bb.3
+ %15:sreg_64 = COPY %14
+ %16:sreg_64 = IMPLICIT_DEF
+ SI_LOOP %16, %bb.1, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+ S_BRANCH %bb.2
+
+...
|
cdevadas
reviewed
Feb 26, 2025
| @@ -0,0 +1,97 @@ | |||
| # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5 | |||
| # RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=si-fix-sgpr-copies,si-i1-copies -o - %s | FileCheck %s | |||
There was a problem hiding this comment.
Wouldn't this case be relevant to compute as well? Asking because of the triple with mesa.
There was a problem hiding this comment.
The triple does not imply the use case, and it does not matter
rampitec
approved these changes
Feb 26, 2025
arsenm
deleted the
users/arsenm/amdgpu/si-fix-sgpr-copies-assert-vreg1-phi
branch
joaosaffran
pushed a commit
to joaosaffran/llvm-project
that referenced
this pull request
Mar 3, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We should just need to leave these alone for the si-i1-copies
to deal with it. Note the IR test did not fail previously, and
only establishes a baseline result for a pending patch which
hit the failure.