[LoopIdiom] BCmp: loop exit count must not be wider than size_t that `bcmp` takes

As reported by Joerg Sonnenberger in IRC, for 32-bit systems,
where pointer and size_t are 32-bit, if you use 64-bit-wide variable
in the loop, you could end up with loop exit count being of the type
wider than the size_t. Now, i'm not sure if we can produce `bcmp`
from that (just truncate?), but we certainly should not assert/miscompile.

llvm-svn: 374811

1 files changed, 6 insertions, 0 deletions

diff --git a/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp b/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
index 116bdd7470f..4f5ae5000a9 100644
--- a/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
@@ -2080,6 +2080,10 @@ bool LoopIdiomRecognize::recognizeBCmpLoopSCEV(uint64_t BCmpTyBytes,
 
   LLVM_DEBUG(dbgs() << "SCEV expressions for loads are acceptable.\n");
 
+  // bcmp / memcmp take length argument as size_t, so let's conservatively
+  // assume that the iteration count should be not wider than that.
+  Type *CmpFuncSizeTy = DL->getIntPtrType(SE->getContext());
+
   // For how many iterations is loop guaranteed not to exit via LoopLatch?
   // This is one less than the maximal number of comparisons,and is:  n + -1
   const SCEV *LoopExitCount =
@@ -2089,6 +2093,8 @@ bool LoopIdiomRecognize::recognizeBCmpLoopSCEV(uint64_t BCmpTyBytes,
   // Exit count, similarly, must be loop-invant that dominates the loop header.
   if (LoopExitCount == SE->getCouldNotCompute() ||
       !LoopExitCount->getType()->isIntOrPtrTy() ||
+      LoopExitCount->getType()->getScalarSizeInBits() >
+          CmpFuncSizeTy->getScalarSizeInBits() ||
       !SE->isAvailableAtLoopEntry(LoopExitCount, CurLoop)) {
     LLVM_DEBUG(dbgs() << "Unsupported SCEV expression for loop latch exit.\n");
     return false;