[llvm] [VPlan] Fix crash when unrolling in-loop reduction chains (PR #129840)
Luke Lau via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 4 22:44:10 PST 2025
https://github.com/lukel97 created https://github.com/llvm/llvm-project/pull/129840
If an in-loop reduction is chained e.g.
WIDEN-REDUCTION-PHI ir<%rdx> = phi ir<0>, ir<%add2>
REDUCE ir<%add1> = ir<%rdx> + reduce.add (ir<%x>)
REDUCE ir<%add2> = ir<%add1> + reduce.add (ir<%y>)
When we try to unroll the second add reduction, we crash because we currently expect the first operand to be a VPReductionPHIRecipe, when in fact it's the previous chain. This relaxes the cast to a dyn_cast, so we end up unrolling to:
WIDEN-REDUCTION-PHI ir<%rdx> = phi ir<0>, ir<%add2>
WIDEN-REDUCTION-PHI ir<%rdx>.1 = phi ir<0>, ir<%add2>.1, ir<1>
WIDEN-REDUCTION-PHI ir<%rdx>.2 = phi ir<0>, ir<%add2>.2, ir<2>
WIDEN-REDUCTION-PHI ir<%rdx>.3 = phi ir<0>, ir<%add2>.3, ir<3>
REDUCE ir<%add1> = ir<%rdx> + reduce.add (ir<%x>)
REDUCE ir<%add1>.1 = ir<%rdx>.1 + reduce.add (ir<%x>.1)
REDUCE ir<%add1>.2 = ir<%rdx>.2 + reduce.add (ir<%x>.2)
REDUCE ir<%add1>.3 = ir<%rdx>.3 + reduce.add (ir<%x>.3)
REDUCE ir<%add2> = ir<%add1> + reduce.add (ir<%y>)
REDUCE ir<%add2>.1 = ir<%add1>.1 + reduce.add (ir<%y>.1)
REDUCE ir<%add2>.2 = ir<%add1>.2 + reduce.add (ir<%y>.2)
REDUCE ir<%add2>.3 = ir<%add1>.3 + reduce.add (ir<%y>.3)
This fixes a crash when building 525.x264_r from SPEC CPU 2017 on AArch64 with -mllvm -prefer-inloop-reductions
>From af4d8f1575e66af384946e221ae3da7e6f3a5362 Mon Sep 17 00:00:00 2001
From: Luke Lau <luke at igalia.com>
Date: Wed, 5 Mar 2025 14:37:21 +0800
Subject: [PATCH] [VPlan] Fix crash when unrolling in-loop reduction chains
If an in-loop reduction is chained e.g.
WIDEN-REDUCTION-PHI ir<%rdx> = phi ir<0>, ir<%add2>
REDUCE ir<%add1> = ir<%rdx> + reduce.add (ir<%x>)
REDUCE ir<%add2> = ir<%add1> + reduce.add (ir<%y>)
When we try to unroll the second add reduction, we crash because we currently expect the first operand to be a VPReductionPHIRecipe, when in fact it's the previous chain. This relaxes the cast to a dyn_cast, so we end up unrolling to:
WIDEN-REDUCTION-PHI ir<%rdx> = phi ir<0>, ir<%add2>
WIDEN-REDUCTION-PHI ir<%rdx>.1 = phi ir<0>, ir<%add2>.1, ir<1>
WIDEN-REDUCTION-PHI ir<%rdx>.2 = phi ir<0>, ir<%add2>.2, ir<2>
WIDEN-REDUCTION-PHI ir<%rdx>.3 = phi ir<0>, ir<%add2>.3, ir<3>
REDUCE ir<%add1> = ir<%rdx> + reduce.add (ir<%x>)
REDUCE ir<%add1>.1 = ir<%rdx>.1 + reduce.add (ir<%x>.1)
REDUCE ir<%add1>.2 = ir<%rdx>.2 + reduce.add (ir<%x>.2)
REDUCE ir<%add1>.3 = ir<%rdx>.3 + reduce.add (ir<%x>.3)
REDUCE ir<%add2> = ir<%add1> + reduce.add (ir<%y>)
REDUCE ir<%add2>.1 = ir<%add1>.1 + reduce.add (ir<%y>.1)
REDUCE ir<%add2>.2 = ir<%add1>.2 + reduce.add (ir<%y>.2)
REDUCE ir<%add2>.3 = ir<%add1>.3 + reduce.add (ir<%y>.3)
This fixes a crash when building 525.x264_r from SPEC CPU 2017 on AArch64 with -mllvm -prefer-inloop-reductions
---
llvm/lib/Transforms/Vectorize/VPlanUnroll.cpp | 4 +-
.../LoopVectorize/reduction-inloop-uf4.ll | 90 ++++++++++++++++++-
2 files changed, 88 insertions(+), 6 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanUnroll.cpp b/llvm/lib/Transforms/Vectorize/VPlanUnroll.cpp
index 89e372d6b46cf..dd2e666e8d150 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanUnroll.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanUnroll.cpp
@@ -295,8 +295,8 @@ void UnrollState::unrollRecipeByUF(VPRecipeBase &R) {
continue;
}
if (auto *Red = dyn_cast<VPReductionRecipe>(&R)) {
- auto *Phi = cast<VPReductionPHIRecipe>(R.getOperand(0));
- if (Phi->isOrdered()) {
+ if (auto *Phi = dyn_cast<VPReductionPHIRecipe>(R.getOperand(0));
+ Phi && Phi->isOrdered()) {
auto &Parts = VPV2Parts[Phi];
if (Part == 1) {
Parts.clear();
diff --git a/llvm/test/Transforms/LoopVectorize/reduction-inloop-uf4.ll b/llvm/test/Transforms/LoopVectorize/reduction-inloop-uf4.ll
index 81163c45abe95..3393ea81992da 100644
--- a/llvm/test/Transforms/LoopVectorize/reduction-inloop-uf4.ll
+++ b/llvm/test/Transforms/LoopVectorize/reduction-inloop-uf4.ll
@@ -66,6 +66,88 @@ entry:
ret i32 %sum.0.lcssa
}
+define i64 @reduction_sum_double(ptr noalias %p, ptr noalias %q) {
+; CHECK-LABEL: @reduction_sum_double(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI2:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i64, ptr [[P:%.*]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i64, ptr [[Q:%.*]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i64 32
+; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP0]], i64 64
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[TMP0]], i64 96
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP0]], align 8
+; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP2]], align 8
+; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <4 x i64>, ptr [[TMP3]], align 8
+; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <4 x i64>, ptr [[TMP4]], align 8
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP1]], i64 32
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[TMP1]], i64 64
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP1]], i64 96
+; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <4 x i64>, ptr [[TMP1]], align 8
+; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <4 x i64>, ptr [[TMP5]], align 8
+; CHECK-NEXT: [[WIDE_LOAD9:%.*]] = load <4 x i64>, ptr [[TMP6]], align 8
+; CHECK-NEXT: [[WIDE_LOAD10:%.*]] = load <4 x i64>, ptr [[TMP7]], align 8
+; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD]])
+; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[TMP8]], [[VEC_PHI]]
+; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD4]])
+; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[TMP10]], [[VEC_PHI1]]
+; CHECK-NEXT: [[TMP12:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD5]])
+; CHECK-NEXT: [[TMP13:%.*]] = add i64 [[TMP12]], [[VEC_PHI2]]
+; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD6]])
+; CHECK-NEXT: [[TMP15:%.*]] = add i64 [[TMP14]], [[VEC_PHI3]]
+; CHECK-NEXT: [[TMP16:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD7]])
+; CHECK-NEXT: [[TMP17]] = add i64 [[TMP16]], [[TMP9]]
+; CHECK-NEXT: [[TMP18:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD8]])
+; CHECK-NEXT: [[TMP19]] = add i64 [[TMP18]], [[TMP11]]
+; CHECK-NEXT: [[TMP20:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD9]])
+; CHECK-NEXT: [[TMP21]] = add i64 [[TMP20]], [[TMP13]]
+; CHECK-NEXT: [[TMP22:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD10]])
+; CHECK-NEXT: [[TMP23]] = add i64 [[TMP22]], [[TMP15]]
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
+; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
+; CHECK-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK: middle.block:
+; CHECK-NEXT: [[BIN_RDX:%.*]] = add i64 [[TMP19]], [[TMP17]]
+; CHECK-NEXT: [[BIN_RDX11:%.*]] = add i64 [[TMP21]], [[BIN_RDX]]
+; CHECK-NEXT: [[BIN_RDX12:%.*]] = add i64 [[TMP23]], [[BIN_RDX11]]
+; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: br label [[LOOP:%.*]]
+; CHECK: loop:
+; CHECK-NEXT: br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK: exit:
+; CHECK-NEXT: [[ADD2_LCSSA:%.*]] = phi i64 [ poison, [[LOOP]] ], [ [[BIN_RDX12]], [[MIDDLE_BLOCK]] ]
+; CHECK-NEXT: ret i64 [[ADD2_LCSSA]]
+;
+entry:
+ br label %loop
+loop:
+ %iv = phi i64 [0, %entry], [%iv.next, %loop]
+ %rdx = phi i64 [0, %entry], [%add2, %loop]
+
+ %p.gep = getelementptr i64, ptr %p, i64 %iv
+ %q.gep = getelementptr i64, ptr %q, i64 %iv
+
+ %x = load i64, ptr %p.gep
+ %y = load i64, ptr %q.gep
+
+ %add1 = add i64 %rdx, %x
+ %add2 = add i64 %add1, %y
+
+ %iv.next = add i64 %iv, 1
+ %done = icmp eq i64 %iv.next, 256
+ br i1 %done, label %exit, label %loop
+exit:
+ ret i64 %add2
+}
+
define i32 @predicated(ptr noalias nocapture %A) {
; CHECK-LABEL: @predicated(
; CHECK-NEXT: entry:
@@ -260,7 +342,7 @@ define i32 @predicated(ptr noalias nocapture %A) {
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 16)
; CHECK-NEXT: [[TMP111:%.*]] = icmp eq i64 [[INDEX_NEXT]], 272
-; CHECK-NEXT: br i1 [[TMP111]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP111]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[BIN_RDX:%.*]] = add i32 [[TMP104]], [[TMP101]]
; CHECK-NEXT: [[BIN_RDX37:%.*]] = add i32 [[TMP107]], [[BIN_RDX]]
@@ -269,7 +351,7 @@ define i32 @predicated(ptr noalias nocapture %A) {
; CHECK: scalar.ph:
; CHECK-NEXT: br label [[DOTLR_PH:%.*]]
; CHECK: .lr.ph:
-; CHECK-NEXT: br i1 poison, label [[DOT_CRIT_EDGE]], label [[DOTLR_PH]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK-NEXT: br i1 poison, label [[DOT_CRIT_EDGE]], label [[DOTLR_PH]], !llvm.loop [[LOOP7:![0-9]+]]
; CHECK: ._crit_edge:
; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ poison, [[DOTLR_PH]] ], [ [[BIN_RDX38]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: ret i32 [[SUM_0_LCSSA]]
@@ -499,7 +581,7 @@ define i32 @cond_rdx_pred(i32 %cond, ptr noalias %a, i64 %N) {
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 16
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 16)
; CHECK-NEXT: [[TMP119:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP119]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP119]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[BIN_RDX:%.*]] = mul i32 [[TMP112]], [[TMP109]]
; CHECK-NEXT: [[BIN_RDX39:%.*]] = mul i32 [[TMP115]], [[BIN_RDX]]
@@ -512,7 +594,7 @@ define i32 @cond_rdx_pred(i32 %cond, ptr noalias %a, i64 %N) {
; CHECK: if.then:
; CHECK-NEXT: br label [[FOR_INC]]
; CHECK: for.inc:
-; CHECK-NEXT: br i1 poison, label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
+; CHECK-NEXT: br i1 poison, label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
; CHECK: for.end:
; CHECK-NEXT: [[RES_LCSSA:%.*]] = phi i32 [ poison, [[FOR_INC]] ], [ [[BIN_RDX40]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: ret i32 [[RES_LCSSA]]
More information about the llvm-commits
mailing list