[llvm] 689babd - [SLP] Don't try to vectorize allocas

[Philip Reames via llvm-commits]

Author: Philip Reames
Date: 2022-03-02T10:08:43-08:00
New Revision: 689babdf687f98b6790c66e628bd057af3bc87d8

URL: https://github.com/llvm/llvm-project/commit/689babdf687f98b6790c66e628bd057af3bc87d8
DIFF: https://github.com/llvm/llvm-project/commit/689babdf687f98b6790c66e628bd057af3bc87d8.diff

LOG: [SLP] Don't try to vectorize allocas

While a collection of allocas are technically vectorizeable - by forming a wider alloca - this was not a transform SLP actually knows how to do.  Instead, we were forming a bundle with missing dependencies, and then relying on the scheduling code to preserve program order if multiple instructions were scheduleable at once.  I haven't been able to write a test case, but I'm 99% sure this was wrong in some edge case.

The unknown op case was flowing down the shufflevector path.  This did result in some splat handling being lost with this change, but the same lack of splat handling is visible in a whole bunch of simple examples for the gather path.  I didn't consider this interesting to fix given how narrow the splat of allocas case is.

Added: 
    

Modified: 
    llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
    llvm/test/Transforms/SLPVectorizer/X86/store_alloca.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 73c7cbccf72ff..ddcd0c3a496c4 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -3910,6 +3910,16 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
     return;
   }
 
+  // Avoid attempting to schedule allocas; there are unmodeled dependencies
+  // for "static" alloca status and for reordering with stacksave calls.
+  for (Value *V : VL) {
+    if (isa<AllocaInst>(V)) {
+      LLVM_DEBUG(dbgs() << "SLP: Gathering due to alloca.\n");
+      newTreeEntry(VL, None /*not vectorized*/, S, UserTreeIdx);
+      return;
+    }
+  }
+
   if (StoreInst *SI = dyn_cast<StoreInst>(S.OpValue))
     if (SI->getValueOperand()->getType()->isVectorTy()) {
       LLVM_DEBUG(dbgs() << "SLP: Gathering due to store vector type.\n");

diff  --git a/llvm/test/Transforms/SLPVectorizer/X86/store_alloca.ll b/llvm/test/Transforms/SLPVectorizer/X86/store_alloca.ll
index f46149bf36925..944a7028c80da 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/store_alloca.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/store_alloca.ll
@@ -32,13 +32,13 @@ define void @ham() #1 {
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8** [[VAR32]] to <4 x i8*>*
 ; CHECK-NEXT:    store <4 x i8*> [[SHUFFLE]], <4 x i8*>* [[TMP2]], align 4
 ; CHECK-NEXT:    [[VAR36:%.*]] = getelementptr inbounds [12 x i8*], [12 x i8*]* [[VAR12]], i32 0, i32 4
+; CHECK-NEXT:    store i8* [[VAR4]], i8** [[VAR36]], align 4
 ; CHECK-NEXT:    [[VAR37:%.*]] = getelementptr inbounds [12 x i8*], [12 x i8*]* [[VAR12]], i32 0, i32 5
+; CHECK-NEXT:    store i8* [[VAR5]], i8** [[VAR37]], align 4
 ; CHECK-NEXT:    [[VAR38:%.*]] = getelementptr inbounds [12 x i8*], [12 x i8*]* [[VAR12]], i32 0, i32 6
+; CHECK-NEXT:    store i8* [[VAR5]], i8** [[VAR38]], align 4
 ; CHECK-NEXT:    [[VAR39:%.*]] = getelementptr inbounds [12 x i8*], [12 x i8*]* [[VAR12]], i32 0, i32 7
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x i8*> [[TMP1]], i8* [[VAR5]], i32 1
-; CHECK-NEXT:    [[SHUFFLE1:%.*]] = shufflevector <4 x i8*> [[TMP3]], <4 x i8*> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 1>
-; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i8** [[VAR36]] to <4 x i8*>*
-; CHECK-NEXT:    store <4 x i8*> [[SHUFFLE1]], <4 x i8*>* [[TMP4]], align 4
+; CHECK-NEXT:    store i8* [[VAR5]], i8** [[VAR39]], align 4
 ; CHECK-NEXT:    ret void
 ;
   %var2 = alloca i8
@@ -78,14 +78,13 @@ define void @spam() #1 {
 ; CHECK-NEXT:    [[VAR5:%.*]] = alloca i8, align 1
 ; CHECK-NEXT:    [[VAR12:%.*]] = alloca [12 x i8*], align 4
 ; CHECK-NEXT:    [[VAR36:%.*]] = getelementptr inbounds [12 x i8*], [12 x i8*]* [[VAR12]], i32 0, i32 4
+; CHECK-NEXT:    store i8* [[VAR4]], i8** [[VAR36]], align 4
 ; CHECK-NEXT:    [[VAR37:%.*]] = getelementptr inbounds [12 x i8*], [12 x i8*]* [[VAR12]], i32 0, i32 5
+; CHECK-NEXT:    store i8* [[VAR5]], i8** [[VAR37]], align 4
 ; CHECK-NEXT:    [[VAR38:%.*]] = getelementptr inbounds [12 x i8*], [12 x i8*]* [[VAR12]], i32 0, i32 6
+; CHECK-NEXT:    store i8* [[VAR5]], i8** [[VAR38]], align 4
 ; CHECK-NEXT:    [[VAR39:%.*]] = getelementptr inbounds [12 x i8*], [12 x i8*]* [[VAR12]], i32 0, i32 7
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x i8*> poison, i8* [[VAR4]], i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x i8*> [[TMP1]], i8* [[VAR5]], i32 1
-; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i8*> [[TMP2]], <4 x i8*> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 1>
-; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i8** [[VAR36]] to <4 x i8*>*
-; CHECK-NEXT:    store <4 x i8*> [[SHUFFLE]], <4 x i8*>* [[TMP3]], align 4
+; CHECK-NEXT:    store i8* [[VAR5]], i8** [[VAR39]], align 4
 ; CHECK-NEXT:    ret void
 ;
   %var4 = alloca i8