From 5e9efa10fc7eea582bb1ad216364fb2c467e4477 Mon Sep 17 00:00:00 2001 From: Nadav Rotem Date: Tue, 11 Dec 2012 18:58:10 +0000 Subject: Loop Vectorize: optimize the vectorization of trunc(induction_var). The truncation is now done on scalars. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169904 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Transforms/Vectorize/LoopVectorize.cpp | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-) (limited to 'lib/Transforms/Vectorize/LoopVectorize.cpp') diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp index a69124032b..9550e8a04c 100644 --- a/lib/Transforms/Vectorize/LoopVectorize.cpp +++ b/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -1204,8 +1204,20 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal, case Instruction::Trunc: case Instruction::FPTrunc: case Instruction::BitCast: { - /// Vectorize bitcasts. CastInst *CI = dyn_cast(it); + /// Optimize the special case where the source is the induction + /// variable. Notice that we can only optimize the 'trunc' case + /// because: a. FP conversions lose precision, b. sext/zext may wrap, + /// c. other casts depend on pointer size. + if (CI->getOperand(0) == OldInduction && + it->getOpcode() == Instruction::Trunc) { + Value *ScalarCast = Builder.CreateCast(CI->getOpcode(), Induction, + CI->getType()); + Value *Broadcasted = getBroadcastInstrs(ScalarCast); + WidenMap[it] = getConsecutiveVector(Broadcasted); + break; + } + /// Vectorize casts. Value *A = getVectorValue(it->getOperand(0)); Type *DestTy = VectorType::get(CI->getType()->getScalarType(), VF); WidenMap[it] = Builder.CreateCast(CI->getOpcode(), A, DestTy); -- cgit v1.2.3-18-g5258