[PATCH] D30741: [DAG] recognize div/rem by 0 as undef before trying constant folding

[Sanjay Patel via Phabricator via llvm-commits]

spatel created this revision.
Herald added a subscriber: mcrosier.

As discussed in the review thread for https://reviews.llvm.org/rL297026, this is actually 2 changes that would independently fix all of the test cases in the patch:

1. Return undef in FoldConstantArithmetic for div/rem by 0.
2. Move basic undef simplifications for div/rem (simplifyDivRem()) before foldBinopIntoSelect() as a matter of efficiency.

I can split it up, but I'm not sure yet how to expose the failures once either one of the fixes goes in.

I will handle the case of vectors with any zero element as a follow-up unless we want to do it all in one step. That change is the DAG sibling for https://reviews.llvm.org/D30665 + adding a check of vector elements to FoldConstantArithmetic().

There are likely more improvements for undef handling needed for functions like isConstantSplatVector(), so I can also try to fix those too.


https://reviews.llvm.org/D30741

Files:
  lib/CodeGen/SelectionDAG/DAGCombiner.cpp
  lib/CodeGen/SelectionDAG/SelectionDAG.cpp
  test/CodeGen/X86/div-rem-simplify.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D30741.91023.patch
Type: text/x-patch
Size: 5167 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170308/5c529464/attachment.bin>