[llvm] 48ebe42 - [SLPVectorizer] Make aliasing check more precise

[Nikita Popov via llvm-commits]

Author: Nikita Popov
Date: 2021-08-31T22:35:30+02:00
New Revision: 48ebe427c9c516c406f6fc1e4883bafaf5f4f992

URL: https://github.com/llvm/llvm-project/commit/48ebe427c9c516c406f6fc1e4883bafaf5f4f992
DIFF: https://github.com/llvm/llvm-project/commit/48ebe427c9c516c406f6fc1e4883bafaf5f4f992.diff

LOG: [SLPVectorizer] Make aliasing check more precise

SLPVectorizer currently uses AA::isNoAlias() to determine whether
two locations alias. This does not work if one of the instructions
is a call. Instead, we should check getModRefInfo(), which
determines whether an arbitrary instruction modifies or references
a given location.

Among other things, this prevents @llvm.experimental.noalias.scope.decl()
and other inaccessiblmemonly intrinsics from interfering with SLP
vectorization.

Differential Revision: https://reviews.llvm.org/D109012

Added: 
    

Modified: 
    llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
    llvm/test/Transforms/SLPVectorizer/int_sideeffect.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 1520a70c81f29..a96bfa69a232c 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -1965,12 +1965,9 @@ class BoUpSLP {
     if (result.hasValue()) {
       return result.getValue();
     }
-    MemoryLocation Loc2 = getLocation(Inst2, AA);
     bool aliased = true;
-    if (Loc1.Ptr && Loc2.Ptr && isSimple(Inst1) && isSimple(Inst2)) {
-      // Do the alias check.
-      aliased = !AA->isNoAlias(Loc1, Loc2);
-    }
+    if (Loc1.Ptr && isSimple(Inst1))
+      aliased = isModOrRefSet(AA->getModRefInfo(Inst2, Loc1));
     // Store the result in the cache.
     result = aliased;
     return aliased;

diff  --git a/llvm/test/Transforms/SLPVectorizer/int_sideeffect.ll b/llvm/test/Transforms/SLPVectorizer/int_sideeffect.ll
index 74925e219ce89..b54d67fb82011 100644
--- a/llvm/test/Transforms/SLPVectorizer/int_sideeffect.ll
+++ b/llvm/test/Transforms/SLPVectorizer/int_sideeffect.ll
@@ -44,16 +44,12 @@ define void @test_inaccessiblememonly(float* %p) {
 ; CHECK-NEXT:    [[P1:%.*]] = getelementptr float, float* [[P]], i64 1
 ; CHECK-NEXT:    [[P2:%.*]] = getelementptr float, float* [[P]], i64 2
 ; CHECK-NEXT:    [[P3:%.*]] = getelementptr float, float* [[P]], i64 3
-; CHECK-NEXT:    [[L0:%.*]] = load float, float* [[P0]], align 4
-; CHECK-NEXT:    [[L1:%.*]] = load float, float* [[P1]], align 4
-; CHECK-NEXT:    [[L2:%.*]] = load float, float* [[P2]], align 4
 ; CHECK-NEXT:    call void @foo() #[[ATTR1:[0-9]+]]
-; CHECK-NEXT:    [[L3:%.*]] = load float, float* [[P3]], align 4
-; CHECK-NEXT:    store float [[L0]], float* [[P0]], align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast float* [[P0]] to <4 x float>*
+; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* [[TMP1]], align 4
 ; CHECK-NEXT:    call void @foo() #[[ATTR1]]
-; CHECK-NEXT:    store float [[L1]], float* [[P1]], align 4
-; CHECK-NEXT:    store float [[L2]], float* [[P2]], align 4
-; CHECK-NEXT:    store float [[L3]], float* [[P3]], align 4
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast float* [[P0]] to <4 x float>*
+; CHECK-NEXT:    store <4 x float> [[TMP2]], <4 x float>* [[TMP3]], align 4
 ; CHECK-NEXT:    ret void
 ;
   %p0 = getelementptr float, float* %p, i64 0