In which I discover the existence of loops. Threading an operation

over a phi node by applying it to each operand may be wrong if the
operation and the phi node are mutually interdependent (the testcase
has a simple example of this).  So only do this transform if it would
be correct to perform the operation in each predecessor of the block
containing the phi, i.e. if the other operands all dominate the phi.
This should fix the FFMPEG snow.c regression reported by İsmail Dönmez.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119347 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Duncan Sands
2010-11-16 12:16:38 +00:00
parent 2c92085034
commit 1845009290
4 changed files with 133 additions and 73 deletions

View File

@ -217,7 +217,7 @@ Value *PHITransAddr::PHITranslateSubExpr(Value *V, BasicBlock *CurBB,
return GEP;
// Simplify the GEP to handle 'gep x, 0' -> x etc.
if (Value *V = SimplifyGEPInst(&GEPOps[0], GEPOps.size(), TD)) {
if (Value *V = SimplifyGEPInst(&GEPOps[0], GEPOps.size(), TD, DT)) {
for (unsigned i = 0, e = GEPOps.size(); i != e; ++i)
RemoveInstInputs(GEPOps[i], InstInputs);
@ -273,7 +273,7 @@ Value *PHITransAddr::PHITranslateSubExpr(Value *V, BasicBlock *CurBB,
}
// See if the add simplifies away.
if (Value *Res = SimplifyAddInst(LHS, RHS, isNSW, isNUW, TD)) {
if (Value *Res = SimplifyAddInst(LHS, RHS, isNSW, isNUW, TD, DT)) {
// If we simplified the operands, the LHS is no longer an input, but Res
// is.
RemoveInstInputs(LHS, InstInputs);