SelectionDAG: Teach FoldConstantArithmetic how to deal with vectors.
authorBenjamin Kramer <benny.kra@googlemail.com>
Mon, 4 Feb 2013 15:19:18 +0000 (15:19 +0000)
committerBenjamin Kramer <benny.kra@googlemail.com>
Mon, 4 Feb 2013 15:19:18 +0000 (15:19 +0000)
commit4969310052f45b1e2e5d21735e38641a20be0e21
tree04e90cfba49332cd63f2bb6fbf495761d54f5a5c
parent17174e59004d4df12f9037f0dc99ae530d71eab9
SelectionDAG: Teach FoldConstantArithmetic how to deal with vectors.

This required disabling a PowerPC optimization that did the following:
input:
x = BUILD_VECTOR <i32 16, i32 16, i32 16, i32 16>
lowered to:
tmp = BUILD_VECTOR <i32 8, i32 8, i32 8, i32 8>
x = ADD tmp, tmp

The add now gets folded immediately and we're back at the BUILD_VECTOR we
started from. I don't see a way to fix this currently so I left it disabled
for now.

Fix some trivially foldable X86 tests too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@174325 91177308-0d34-0410-b5e6-96231b3b80d8
include/llvm/CodeGen/SelectionDAG.h
lib/CodeGen/SelectionDAG/SelectionDAG.cpp
lib/Target/PowerPC/PPCISelLowering.cpp
test/CodeGen/PowerPC/2008-07-10-SplatMiscompile.ll
test/CodeGen/X86/2011-11-30-or.ll
test/CodeGen/X86/blend-msb.ll
test/CodeGen/X86/vector-gep.ll