[LAA] Use getTypeSizeInBits for stride type size comparison

[wwei]

Summary

SCEV type conversion functions, such as getNoopOrSignExtend, will use getTypeSizeInBits to check the conversion type size.
This patch attempts to fix some assertion failures when stride type is undef or the size of alloc type does not match the bits type size.

[wwei]

ping

llvm/lib/Analysis/LoopAccessAnalysis.cpp

static bool isSafeDependenceDistance(const DataLayout &DL, ScalarEvolution &SE,
   // also guarantees that distance >= VF.
   //
   const uint64_t ByteStride = Stride * TypeByteSize;
   const SCEV *Step = SE.getConstant(BackedgeTakenCount.getType(), ByteStride);
   const SCEV *Product = SE.getMulExpr(&BackedgeTakenCount, Step);
 
   const SCEV *CastedDist = &Dist;
   const SCEV *CastedProduct = Product;
-  uint64_t DistTypeSize = DL.getTypeAllocSize(Dist.getType());
-  uint64_t ProductTypeSize = DL.getTypeAllocSize(Product->getType());
+  uint64_t DistTypeSize = DL.getTypeSizeInBits(Dist.getType());
+  uint64_t ProductTypeSize = DL.getTypeSizeInBits(Product->getType());
 
   // The dependence distance can be positive/negative, so we sign extend Dist;
   // The multiplication of the absolute stride in bytes and the
   // backedgeTakenCount is non-negative, so we zero extend Product.
   if (DistTypeSize > ProductTypeSize)
     CastedProduct = SE.getZeroExtendExpr(Product, Dist.getType());
   else
     CastedDist = SE.getNoopOrSignExtend(&Dist, Product->getType());
void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
 
   const SCEV *StrideExpr = PSE->getSCEV(Stride);
   const SCEV *BETakenCount = PSE->getBackedgeTakenCount();
 
   // Match the types so we can compare the stride and the BETakenCount.
   // The Stride can be positive/negative, so we sign extend Stride;
   // The backedgeTakenCount is non-negative, so we zero extend BETakenCount.
   const DataLayout &DL = TheLoop->getHeader()->getModule()->getDataLayout();
-  uint64_t StrideTypeSize = DL.getTypeAllocSize(StrideExpr->getType());
-  uint64_t BETypeSize = DL.getTypeAllocSize(BETakenCount->getType());
+  uint64_t StrideTypeSize = DL.getTypeSizeInBits(StrideExpr->getType());
+  uint64_t BETypeSize = DL.getTypeSizeInBits(BETakenCount->getType());
   const SCEV *CastedStride = StrideExpr;
   const SCEV *CastedBECount = BETakenCount;
   ScalarEvolution *SE = PSE->getSE();
   if (BETypeSize >= StrideTypeSize)
     CastedStride = SE->getNoopOrSignExtend(StrideExpr, BETakenCount->getType());
   else
     CastedBECount = SE->getZeroExtendExpr(BETakenCount, StrideExpr->getType());
   const SCEV *StrideMinusBETaken = SE->getMinusSCEV(CastedStride, CastedBECount);

llvm/test/Transforms/LoopLoadElim/undef-stride.ll

+; RUN: opt -enable-mem-access-versioning --loop-load-elim < %s -S | FileCheck %s
+; REQUIRES: asserts
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+
+; Make sure it does not crash with assert.
+define void @f([2 x [4 x i32]]* nocapture readonly %g) {
+
+for.cond.us.preheader:
+  br label %for.cond.us
+
+for.cond.us:                                      ; preds = %for.cond.cleanup.us-lcssa.us.us, %for.cond.us.preheader
+  %h.0.us = phi i8 [ %conv14.us, %for.cond.cleanup.us-lcssa.us.us ], [ 0, %for.cond.us.preheader ]
+  %idxprom.us = zext i8 %h.0.us to i64
+  %arrayidx4.us = getelementptr inbounds [2 x [4 x i32]], [2 x [4 x i32]]* %g, i64 0, i64 0, i64 %idxprom.us
+  br i1 undef, label %for.body.us.us.lr.ph, label %for.cond.cleanup.us-lcssa.us.us
+
+for.body.us.us.lr.ph:                             ; preds = %for.cond.us
+  br label %for.body.us.us
+
+for.body.us.us:                                   ; preds = %for.body.us.us, %for.body.us.us.lr.ph
+  %0 = load i32, i32* %arrayidx4.us, align 4
+  br i1 undef, label %for.body.us.us, label %for.cond1.us.us.for.cond.cleanup.us-lcssa.us.us_crit_edge
+
+for.cond1.us.us.for.cond.cleanup.us-lcssa.us.us_crit_edge: ; preds = %for.body.us.us
+  br label %for.cond.cleanup.us-lcssa.us.us
+
+for.cond.cleanup.us-lcssa.us.us:                  ; preds = %for.cond1.us.us.for.cond.cleanup.us-lcssa.us.us_crit_edge, %for.cond.us
+  %conv14.us = add i8 %h.0.us, undef
+  br label %for.cond.us
+}
+