[llvm] 85eb44e - [SLP]Fix number of operands for the split node
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Tue Apr 15 13:33:47 PDT 2025
Author: Alexey Bataev
Date: 2025-04-15T13:33:36-07:00
New Revision: 85eb44e304e0a0a7da78448ceee60fdfec235edb
URL: https://github.com/llvm/llvm-project/commit/85eb44e304e0a0a7da78448ceee60fdfec235edb
DIFF: https://github.com/llvm/llvm-project/commit/85eb44e304e0a0a7da78448ceee60fdfec235edb.diff
LOG: [SLP]Fix number of operands for the split node
FOr the split node number of operands should be requested via
getNumOperands() function, even if the main op is CallInst.
Added:
llvm/test/Transforms/SLPVectorizer/X86/split-node-num-operands.ll
Modified:
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 234cd340ebc13..b174f0f03fca6 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -7577,6 +7577,8 @@ void BoUpSLP::reorderBottomToTop(bool IgnoreReorder) {
return Res.takeVector();
};
auto GetNumOperands = [](const TreeEntry *TE) {
+ if (TE->State == TreeEntry::SplitVectorize)
+ return TE->getNumOperands();
if (auto *CI = dyn_cast<CallInst>(TE->getMainOp()); CI)
return CI->arg_size();
return TE->getNumOperands();
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/split-node-num-operands.ll b/llvm/test/Transforms/SLPVectorizer/X86/split-node-num-operands.ll
new file mode 100644
index 0000000000000..5aa4dba2b8a1b
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/split-node-num-operands.ll
@@ -0,0 +1,121 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu -mattr=+avx -slp-threshold=-1000 < %s | FileCheck %s
+
+define i64 @Foo(ptr align 8 dereferenceable(344) %0, i64 %1) {
+; CHECK-LABEL: define i64 @Foo(
+; CHECK-SAME: ptr align 8 dereferenceable(344) [[TMP0:%.*]], i64 [[TMP1:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP0]], i64 104
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[TMP0]], i64 112
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP0]], i64 24
+; CHECK-NEXT: [[TMP6:%.*]] = load i64, ptr [[TMP3]], align 8
+; CHECK-NEXT: [[TMP7:%.*]] = load i64, ptr [[TMP4]], align 8
+; CHECK-NEXT: [[TMP8:%.*]] = load i64, ptr [[TMP5]], align 8
+; CHECK-NEXT: [[TMP9:%.*]] = load i64, ptr [[TMP0]], align 8
+; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x i64> poison, i64 [[TMP6]], i32 0
+; CHECK-NEXT: [[TMP11:%.*]] = insertelement <2 x i64> [[TMP10]], i64 [[TMP9]], i32 1
+; CHECK-NEXT: [[TMP12:%.*]] = insertelement <2 x i64> poison, i64 [[TMP7]], i32 0
+; CHECK-NEXT: [[TMP13:%.*]] = insertelement <2 x i64> [[TMP12]], i64 [[TMP8]], i32 1
+; CHECK-NEXT: [[TMP14:%.*]] = insertelement <2 x i64> poison, i64 0, i32 0
+; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x i64> <i64 0, i64 poison>, i64 [[TMP1]], i32 1
+; CHECK-NEXT: br label %[[BB16:.*]]
+; CHECK: [[BB16]]:
+; CHECK-NEXT: [[TMP17:%.*]] = phi <2 x i64> [ [[TMP11]], [[TMP2:%.*]] ], [ zeroinitializer, %[[TMP25:.*]] ]
+; CHECK-NEXT: [[TMP18:%.*]] = phi <2 x i64> [ [[TMP13]], [[TMP2]] ], [ [[TMP29:%.*]], %[[TMP25]] ]
+; CHECK-NEXT: switch i32 0, label %[[BB19:.*]] [
+; CHECK-NEXT: i32 0, label %[[TMP25]]
+; CHECK-NEXT: ]
+; CHECK: [[BB19]]:
+; CHECK-NEXT: [[TMP20:%.*]] = shufflevector <2 x i64> [[TMP18]], <2 x i64> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i64> [[TMP20]], i64 0, i32 1
+; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x i64> [[TMP21]], i64 0, i32 2
+; CHECK-NEXT: [[TMP23:%.*]] = shufflevector <4 x i64> [[TMP22]], <4 x i64> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
+; CHECK-NEXT: [[TMP24:%.*]] = shufflevector <2 x i64> [[TMP14]], <2 x i64> [[TMP18]], <2 x i32> <i32 0, i32 2>
+; CHECK-NEXT: br label %[[TMP25]]
+; CHECK: [[TMP25]]:
+; CHECK-NEXT: [[TMP26:%.*]] = phi <2 x i64> [ [[TMP17]], %[[BB19]] ], [ zeroinitializer, %[[BB16]] ]
+; CHECK-NEXT: [[TMP27:%.*]] = phi <4 x i64> [ [[TMP23]], %[[BB19]] ], [ zeroinitializer, %[[BB16]] ]
+; CHECK-NEXT: [[TMP28:%.*]] = phi <2 x i64> [ [[TMP24]], %[[BB19]] ], [ [[TMP15]], %[[BB16]] ]
+; CHECK-NEXT: [[TMP29]] = shufflevector <2 x i64> [[TMP18]], <2 x i64> <i64 0, i64 poison>, <2 x i32> <i32 2, i32 1>
+; CHECK-NEXT: br i1 false, label %[[DOTLOOPEXIT206:.*]], label %[[BB16]]
+; CHECK: [[_LOOPEXIT206:.*:]]
+; CHECK-NEXT: switch i32 0, label %[[BB32:.*]] [
+; CHECK-NEXT: i32 0, [[DOTCONT174:label %.*]]
+; CHECK-NEXT: i32 1, label %[[BB30:.*]]
+; CHECK-NEXT: ]
+; CHECK: [[BB30]]:
+; CHECK-NEXT: [[TMP31:%.*]] = shufflevector <4 x i64> [[TMP27]], <4 x i64> <i64 0, i64 0, i64 poison, i64 0>, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
+; CHECK-NEXT: br [[DOTCONT174]]
+; CHECK: [[BB32]]:
+; CHECK-NEXT: [[TMP33:%.*]] = insertelement <4 x i64> [[TMP27]], i64 0, i32 1
+; CHECK-NEXT: [[TMP34:%.*]] = insertelement <4 x i64> [[TMP33]], i64 0, i32 2
+; CHECK-NEXT: [[TMP35:%.*]] = shufflevector <4 x i64> [[TMP34]], <4 x i64> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
+; CHECK-NEXT: [[TMP36:%.*]] = insertelement <2 x i64> [[TMP28]], i64 0, i32 0
+; CHECK-NEXT: br [[DOTCONT174]]
+; CHECK: [[_CONT174:.*:]]
+; CHECK-NEXT: [[TMP37:%.*]] = phi <2 x i64> [ [[TMP26]], %[[BB32]] ], [ zeroinitializer, %[[BB30]] ], [ [[TMP26]], %[[DOTLOOPEXIT206]] ]
+; CHECK-NEXT: [[TMP38:%.*]] = phi <4 x i64> [ [[TMP35]], %[[BB32]] ], [ [[TMP31]], %[[BB30]] ], [ [[TMP27]], %[[DOTLOOPEXIT206]] ]
+; CHECK-NEXT: [[TMP39:%.*]] = phi <2 x i64> [ [[TMP36]], %[[BB32]] ], [ zeroinitializer, %[[BB30]] ], [ [[TMP28]], %[[DOTLOOPEXIT206]] ]
+; CHECK-NEXT: ret i64 0
+;
+ %3 = getelementptr i8, ptr %0, i64 104
+ %4 = getelementptr i8, ptr %0, i64 112
+ %5 = getelementptr i8, ptr %0, i64 24
+ %6 = load i64, ptr %3, align 8
+ %7 = load i64, ptr %4, align 8
+ %8 = load i64, ptr %5, align 8
+ %9 = load i64, ptr %0, align 8
+ br label %10
+
+10:
+ %11 = phi i64 [ %9, %2 ], [ 0, %18 ]
+ %12 = phi i64 [ %8, %2 ], [ %12, %18 ]
+ %13 = phi i64 [ %7, %2 ], [ 0, %18 ]
+ %14 = phi i64 [ %6, %2 ], [ 0, %18 ]
+ switch i32 0, label %15 [
+ i32 0, label %18
+ ]
+
+15:
+ %16 = tail call i64 @llvm.umin.i64(i64 0, i64 0)
+ %17 = tail call i64 @llvm.umax.i64(i64 0, i64 0)
+ br label %18
+
+18:
+ %19 = phi i64 [ %17, %15 ], [ 0, %10 ]
+ %20 = phi i64 [ %16, %15 ], [ 0, %10 ]
+ %21 = phi i64 [ %11, %15 ], [ 0, %10 ]
+ %22 = phi i64 [ %12, %15 ], [ 0, %10 ]
+ %23 = phi i64 [ %13, %15 ], [ %1, %10 ]
+ %24 = phi i64 [ %14, %15 ], [ 0, %10 ]
+ br i1 false, label %.loopexit206, label %10
+
+.loopexit206:
+ switch i32 0, label %26 [
+ i32 0, label %.cont174
+ i32 1, label %25
+ ]
+
+25:
+ br label %.cont174
+
+26:
+ %27 = tail call i64 @llvm.umin.i64(i64 0, i64 0)
+ %28 = tail call i64 @llvm.umax.i64(i64 0, i64 0)
+ br label %.cont174
+
+.cont174:
+ %.sroa.139.1 = phi i64 [ %28, %26 ], [ %19, %25 ], [ %19, %.loopexit206 ]
+ %.sroa.133.1 = phi i64 [ %27, %26 ], [ 0, %25 ], [ %20, %.loopexit206 ]
+ %.sroa.81.1 = phi i64 [ %23, %26 ], [ 0, %25 ], [ %23, %.loopexit206 ]
+ %.sroa.75.1 = phi i64 [ %24, %26 ], [ 0, %25 ], [ %24, %.loopexit206 ]
+ %.sroa.21.1 = phi i64 [ %21, %26 ], [ 0, %25 ], [ %21, %.loopexit206 ]
+ %.sroa.15.1 = phi i64 [ %22, %26 ], [ 0, %25 ], [ %22, %.loopexit206 ]
+ %29 = phi i64 [ %28, %26 ], [ 0, %25 ], [ %19, %.loopexit206 ]
+ %30 = phi i64 [ %27, %26 ], [ 0, %25 ], [ %20, %.loopexit206 ]
+ ret i64 0
+}
+
+declare i64 @llvm.umax.i64(i64, i64)
+
+declare i64 @llvm.umin.i64(i64, i64)
+
More information about the llvm-commits
mailing list