wangtao
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

BBVectorize, when using VTTI, should not form types that will be split. 

This is needed so that perl's SHA can be compiled (otherwise
BBVectorize takes far too long to find its fixed point).

I'll try to come up with a reduced test case.

llvm-svn: 166738
This commit is contained in:
Hal Finkel 2012-10-25 23:47:16 +00:00
parent 9d1ee1175d
commit 20a49d6f2c
1 changed files with 19 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
llvm/lib/Transforms/Vectorize/BBVectorize.cpp
View File

@ -817,6 +817,15 @@ namespace {
IAddressSpace);
if (VCost > ICost + JCost)
return false;
// FIXME: We don't want to fuse to a type that will be split, even
// if the two input types will also be split and there is no other
// associated cost. This check depends on the fact
// that the current implementation of getMemoryOpCost returns only
// the type-splitting cost.
if (VCost > 1)
return false;
CostSavings = ICost + JCost - VCost;
}
} else {
@ -831,6 +840,16 @@ namespace {
if (VCost > ICost + JCost)
return false;
// FIXME: We don't want to fuse to a type that will be split, even
// if the two input types will also be split and there is no other
// associated cost. This check depends on the fact
// that the current implementation of getMemoryOpCost returns only
// the type-splitting cost (and does nothing else).
unsigned VTypeCost = VTTI->getMemoryOpCost(I->getOpcode(), VT1, 0, 0);
if (VTypeCost > 1)
return false;
CostSavings = ICost + JCost - VCost;
}