[llvm] [SLP] Don't create zero-count items in SameValuesCounter (PR #107042)

[Yingwei Zheng via llvm-commits]

https://github.com/dtcxzyw created https://github.com/llvm/llvm-project/pull/107042

Fix https://github.com/llvm/llvm-project/issues/107037.


>From 9c4fabd10f645a86a14e625ad8dbf4964ad562d9 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng <dtcxzyw2333 at gmail.com>
Date: Tue, 3 Sep 2024 10:53:19 +0800
Subject: [PATCH] [SLP] Don't create zero-count items in SameValuesCounter

---
 .../Transforms/Vectorize/SLPVectorizer.cpp    |  3 +-
 .../Transforms/SLPVectorizer/X86/pr107037.ll  | 47 +++++++++++++++++++
 2 files changed, 49 insertions(+), 1 deletion(-)
 create mode 100644 llvm/test/Transforms/SLPVectorizer/X86/pr107037.ll

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 58137cd6f543f2..9bf586ebefd184 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -17962,7 +17962,7 @@ class HorizontalReduction {
           }
           Value *OrigV = TrackedToOrig.find(RdxVal)->second;
           unsigned NumOps =
-              VectorizedVals.lookup(RdxVal) + SameValuesCounter[OrigV];
+              VectorizedVals.lookup(RdxVal) + SameValuesCounter.lookup(OrigV);
           if (NumOps != ReducedValsToOps.find(OrigV)->second.size())
             LocalExternallyUsedValues[RdxVal];
         }
@@ -18104,6 +18104,7 @@ class HorizontalReduction {
       }
       if (OptReusedScalars && !AnyVectorized) {
         for (const std::pair<Value *, unsigned> &P : SameValuesCounter) {
+          assert(P.second && "Expected at least one occurrence.");
           Value *RedVal = emitScaleForReusedOps(P.first, Builder, P.second);
           VectorizedTree = GetNewVectorizedTree(VectorizedTree, RedVal);
           Value *OrigV = TrackedToOrig.find(P.first)->second;
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/pr107037.ll b/llvm/test/Transforms/SLPVectorizer/X86/pr107037.ll
new file mode 100644
index 00000000000000..9fa68e46c02d76
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/pr107037.ll
@@ -0,0 +1,47 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu -S < %s | FileCheck %s
+
+define i8 @pr107037() {
+; CHECK-LABEL: define i8 @pr107037() {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[TMP0:%.*]] = trunc i32 0 to i8
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 0 to i8
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc i32 0 to i8
+; CHECK-NEXT:    [[TMP3:%.*]] = trunc i32 0 to i8
+; CHECK-NEXT:    [[TMP4:%.*]] = call i8 @llvm.vector.reduce.or.v8i8(<8 x i8> zeroinitializer)
+; CHECK-NEXT:    [[OP_RDX:%.*]] = or i8 [[TMP4]], [[TMP0]]
+; CHECK-NEXT:    [[OP_RDX1:%.*]] = or i8 [[OP_RDX]], [[TMP2]]
+; CHECK-NEXT:    [[OP_RDX2:%.*]] = or i8 [[OP_RDX1]], [[TMP0]]
+; CHECK-NEXT:    [[OP_RDX3:%.*]] = or i8 [[OP_RDX2]], [[TMP1]]
+; CHECK-NEXT:    [[OP_RDX4:%.*]] = or i8 [[OP_RDX3]], [[TMP3]]
+; CHECK-NEXT:    ret i8 [[OP_RDX4]]
+;
+entry:
+  %0 = trunc i32 0 to i8
+  %1 = add i8 %0, 0
+  %2 = add i8 %0, 0
+  %3 = add i8 %0, 0
+  %4 = add i8 %0, 0
+  %5 = trunc i32 0 to i8
+  %6 = or i8 %5, %0
+  %7 = or i8 %6, %2
+  %8 = or i8 %7, %3
+  %9 = or i8 %8, %0
+  %10 = or i8 %9, %4
+  %conv223.us.us.us.i.us.us.22.i.us.4 = or i8 %10, %1
+  %11 = trunc i32 0 to i8
+  %12 = add i8 %11, 0
+  %conv223.us.us.us.peel.i.i.us.5 = or i8 %conv223.us.us.us.i.us.us.22.i.us.4, %12
+  %13 = add i8 %11, 0
+  %14 = add i8 %11, 0
+  %15 = add i8 %11, 0
+  %16 = trunc i32 0 to i8
+  %17 = or i8 %13, %16
+  %18 = or i8 %17, %14
+  %19 = or i8 %18, %11
+  %20 = or i8 %19, %15
+  %conv223.us.us.us.i.us.us.22.i.us.5 = or i8 %20, %conv223.us.us.us.peel.i.i.us.5
+  %21 = trunc i32 0 to i8
+  %conv223.us.us.us.i.us.us.22.i.us.6 = or i8 %21, %conv223.us.us.us.i.us.us.22.i.us.5
+  ret i8 %conv223.us.us.us.i.us.us.22.i.us.6
+}