aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Revert change to lEmitVaryingSelect().

Browse Source 
Using vector select versus a store and masked load for varying vector
selects seems to give worse code.  This may be related to
http://llvm.org/bugs/show_bug.cgi?id=16941.
This commit is contained in:
Matt Pharr
2013-08-22 07:50:25 -07:00
parent 502f8fd76b
commit 611477e214
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
expr.cpp
View File

@@ -3123,7 +3123,10 @@ static llvm::Value *
lEmitVaryingSelect(FunctionEmitContext *ctx, llvm::Value *test, lEmitVaryingSelect(FunctionEmitContext *ctx, llvm::Value *test,
llvm::Value *expr1, llvm::Value *expr2, llvm::Value *expr1, llvm::Value *expr2,
const Type *type) { const Type *type) {
#if !defined(LLVM_3_1) #if 0 // !defined(LLVM_3_1)
// Though it should be equivalent, this seems to cause non-trivial
// performance regressions versus the below. This may be related to
// http://llvm.org/bugs/show_bug.cgi?id=16941.
if (test->getType() != LLVMTypes::Int1VectorType) if (test->getType() != LLVMTypes::Int1VectorType)
test = ctx->TruncInst(test, LLVMTypes::Int1VectorType); test = ctx->TruncInst(test, LLVMTypes::Int1VectorType);
return ctx->SelectInst(test, expr1, expr2, "select"); return ctx->SelectInst(test, expr1, expr2, "select");