[llvm] 4dede75 - [SLP]Transform nodes before building externally used values

[Alexey Bataev via llvm-commits]

Author: Alexey Bataev
Date: 2024-10-02T06:01:25-07:00
New Revision: 4dede756f25b153a81ea883ef84064f5a1eaaec2

URL: https://github.com/llvm/llvm-project/commit/4dede756f25b153a81ea883ef84064f5a1eaaec2
DIFF: https://github.com/llvm/llvm-project/commit/4dede756f25b153a81ea883ef84064f5a1eaaec2.diff

LOG: [SLP]Transform nodes before building externally used values

transformNodes function may create new vector nodes, so the reduced
values might be vectorized later. Need to build the list of the
externally used values after the transformNodes() function call to avoid
compiler crash.

Fixe #110787

Added: 
    llvm/test/Transforms/SLPVectorizer/X86/reduced-val-vectorized-in-transform.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 15e798bd6c98f9..ce1e5753c0c81e 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -18848,6 +18848,7 @@ class HorizontalReduction {
             ++SameValuesCounter.try_emplace(OrigV).first->second;
           }
         }
+        V.transformNodes();
         SmallPtrSet<Value *, 4> VLScalars(VL.begin(), VL.end());
         // Gather externally used values.
         SmallPtrSet<Value *, 4> Visited;
@@ -18877,7 +18878,6 @@ class HorizontalReduction {
         for (Value *RdxVal : VL)
           if (RequiredExtract.contains(RdxVal))
             LocalExternallyUsedValues[RdxVal];
-        V.transformNodes();
         V.buildExternalUses(LocalExternallyUsedValues);
 
         V.computeMinimumValueSizes();

diff  --git a/llvm/test/Transforms/SLPVectorizer/X86/reduced-val-vectorized-in-transform.ll b/llvm/test/Transforms/SLPVectorizer/X86/reduced-val-vectorized-in-transform.ll
new file mode 100644
index 00000000000000..81f3bf99f3fd89
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/reduced-val-vectorized-in-transform.ll
@@ -0,0 +1,47 @@
+; 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 -slp-threshold=-10 < %s | FileCheck %s
+
+define i32 @test(i1 %cond) {
+; CHECK-LABEL: define i32 @test(
+; CHECK-SAME: i1 [[COND:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    br label %[[BB:.*]]
+; CHECK:       [[BB]]:
+; CHECK-NEXT:    [[P1:%.*]] = phi i32 [ [[OR92:%.*]], %[[BB]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = phi <2 x i32> [ [[TMP8:%.*]], %[[BB]] ], [ zeroinitializer, %[[ENTRY]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = or i32 1, 0
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> [[TMP0]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> <i32 poison, i32 poison, i32 0, i32 0>, <4 x i32> <i32 poison, i32 1, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x i32> [[TMP3]], i32 [[P1]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = or <4 x i32> zeroinitializer, [[TMP4]]
+; CHECK-NEXT:    [[OR92]] = or i32 1, 0
+; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.vector.reduce.xor.v4i32(<4 x i32> [[TMP5]])
+; CHECK-NEXT:    [[OP_RDX:%.*]] = xor i32 [[TMP6]], [[OR92]]
+; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x i32> poison, i32 [[OP_RDX]], i32 0
+; CHECK-NEXT:    [[TMP8]] = insertelement <2 x i32> [[TMP7]], i32 [[TMP1]], i32 1
+; CHECK-NEXT:    br i1 [[COND]], label %[[EXIT:.*]], label %[[BB]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    ret i32 [[OP_RDX]]
+;
+entry:
+  br label %bb
+
+bb:
+  %p1 = phi i32 [ %or92, %bb ], [ 0, %entry ]
+  %p2 = phi i32 [ %0, %bb ], [ 0, %entry ]
+  %p3 = phi i32 [ %4, %bb ], [ 0, %entry ]
+  %0 = or i32 1, 0
+  %or8.i = or i32 0, 0
+  %or9.i = or i32 0, 0
+  %or91 = or i32 %p1, 0
+  %or12.i = or i32 %p2, 0
+  %or92 = or i32 1, 0
+  %1 = xor i32 %or91, %or12.i
+  %2 = xor i32 %1, %or9.i
+  %3 = xor i32 %2, %or8.i
+  %4 = xor i32 %3, %or92
+  br i1 %cond, label %exit, label %bb
+
+exit:
+  ret i32 %4
+}