[llvm] 4fc98ca - [NFC][LoopIdiom] Let processLoopStoreOfLoopLoad take StoreSize as SCEV instead of unsigned

[via llvm-commits]

Author: eopXD
Date: 2021-08-21T00:03:28-07:00
New Revision: 4fc98ca6174ad173abc48646b713c91fd50e7182

URL: https://github.com/llvm/llvm-project/commit/4fc98ca6174ad173abc48646b713c91fd50e7182
DIFF: https://github.com/llvm/llvm-project/commit/4fc98ca6174ad173abc48646b713c91fd50e7182.diff

LOG: [NFC][LoopIdiom] Let processLoopStoreOfLoopLoad take StoreSize as SCEV instead of unsigned

Letting it take SCEV allows further modification on the function to optimize
if the StoreSize / Stride is runtime determined.

The plan is to let memcpy / memmove deal with runtime-determined sizes, just
like what D107353 did to memset.

Reviewed By: bmahjour

Differential Revision: https://reviews.llvm.org/D108289

Added: 
    

Modified: 
    llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp b/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
index 804fe1018b0e..f57ee475fcc3 100644
--- a/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
@@ -225,7 +225,7 @@ class LoopIdiomRecognize {
                                bool IsNegStride, bool IsLoopMemset = false);
   bool processLoopStoreOfLoopLoad(StoreInst *SI, const SCEV *BECount);
   bool processLoopStoreOfLoopLoad(Value *DestPtr, Value *SourcePtr,
-                                  unsigned StoreSize, MaybeAlign StoreAlign,
+                                  const SCEV *StoreSize, MaybeAlign StoreAlign,
                                   MaybeAlign LoadAlign, Instruction *TheStore,
                                   Instruction *TheLoad,
                                   const SCEVAddRecExpr *StoreEv,
@@ -858,15 +858,15 @@ bool LoopIdiomRecognize::processLoopMemCpy(MemCpyInst *MCI,
 
   // Check if the stride matches the size of the memcpy. If so, then we know
   // that every byte is touched in the loop.
-  const SCEVConstant *StoreStride =
+  const SCEVConstant *ConstStoreStride =
       dyn_cast<SCEVConstant>(StoreEv->getOperand(1));
-  const SCEVConstant *LoadStride =
+  const SCEVConstant *ConstLoadStride =
       dyn_cast<SCEVConstant>(LoadEv->getOperand(1));
-  if (!StoreStride || !LoadStride)
+  if (!ConstStoreStride || !ConstLoadStride)
     return false;
 
-  APInt StoreStrideValue = StoreStride->getAPInt();
-  APInt LoadStrideValue = LoadStride->getAPInt();
+  APInt StoreStrideValue = ConstStoreStride->getAPInt();
+  APInt LoadStrideValue = ConstLoadStride->getAPInt();
   // Huge stride value - give up
   if (StoreStrideValue.getBitWidth() > 64 || LoadStrideValue.getBitWidth() > 64)
     return false;
@@ -888,9 +888,10 @@ bool LoopIdiomRecognize::processLoopMemCpy(MemCpyInst *MCI,
   if (StoreStrideInt != LoadStrideInt)
     return false;
 
-  return processLoopStoreOfLoopLoad(Dest, Source, (unsigned)SizeInBytes,
-                                    MCI->getDestAlign(), MCI->getSourceAlign(),
-                                    MCI, MCI, StoreEv, LoadEv, BECount);
+  return processLoopStoreOfLoopLoad(
+      Dest, Source, SE->getConstant(Dest->getType(), SizeInBytes),
+      MCI->getDestAlign(), MCI->getSourceAlign(), MCI, MCI, StoreEv, LoadEv,
+      BECount);
 }
 
 /// processLoopMemSet - See if this memset can be promoted to a large memset.
@@ -1242,16 +1243,18 @@ bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(StoreInst *SI,
   // random load we can't handle.
   Value *LoadPtr = LI->getPointerOperand();
   const SCEVAddRecExpr *LoadEv = cast<SCEVAddRecExpr>(SE->getSCEV(LoadPtr));
-  return processLoopStoreOfLoopLoad(StorePtr, LoadPtr, StoreSize,
+
+  const SCEV *StoreSizeSCEV = SE->getConstant(StorePtr->getType(), StoreSize);
+  return processLoopStoreOfLoopLoad(StorePtr, LoadPtr, StoreSizeSCEV,
                                     SI->getAlign(), LI->getAlign(), SI, LI,
                                     StoreEv, LoadEv, BECount);
 }
 
 bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(
-    Value *DestPtr, Value *SourcePtr, unsigned StoreSize, MaybeAlign StoreAlign,
-    MaybeAlign LoadAlign, Instruction *TheStore, Instruction *TheLoad,
-    const SCEVAddRecExpr *StoreEv, const SCEVAddRecExpr *LoadEv,
-    const SCEV *BECount) {
+    Value *DestPtr, Value *SourcePtr, const SCEV *StoreSizeSCEV,
+    MaybeAlign StoreAlign, MaybeAlign LoadAlign, Instruction *TheStore,
+    Instruction *TheLoad, const SCEVAddRecExpr *StoreEv,
+    const SCEVAddRecExpr *LoadEv, const SCEV *BECount) {
 
   // FIXME: until llvm.memcpy.inline supports dynamic sizes, we need to
   // conservatively bail here, since otherwise we may have to transform
@@ -1274,9 +1277,14 @@ bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(
   Type *IntIdxTy = Builder.getIntNTy(DL->getIndexSizeInBits(StrAS));
 
   APInt Stride = getStoreStride(StoreEv);
+  const SCEVConstant *ConstStoreSize = dyn_cast<SCEVConstant>(StoreSizeSCEV);
+
+  // TODO: Deal with non-constant size; Currently expect constant store size
+  assert(ConstStoreSize && "store size is expected to be a constant");
+
+  int64_t StoreSize = ConstStoreSize->getValue()->getZExtValue();
   bool IsNegStride = StoreSize == -Stride;
 
-  const SCEV *StoreSizeSCEV = SE->getConstant(BECount->getType(), StoreSize);
   // Handle negative strided loops.
   if (IsNegStride)
     StrStart =
@@ -1376,8 +1384,8 @@ bool LoopIdiomRecognize::processLoopStoreOfLoopLoad(
 
   // Okay, everything is safe, we can transform this!
 
-  const SCEV *NumBytesS = getNumBytes(
-      BECount, IntIdxTy, SE->getConstant(IntIdxTy, StoreSize), CurLoop, DL, SE);
+  const SCEV *NumBytesS =
+      getNumBytes(BECount, IntIdxTy, StoreSizeSCEV, CurLoop, DL, SE);
 
   Value *NumBytes =
       Expander.expandCodeFor(NumBytesS, IntIdxTy, Preheader->getTerminator());