[R600] Replicate old DAGCombiner behavior in target specific DAG combine.
authorQuentin Colombet <qcolombet@apple.com>
Tue, 30 Jul 2013 00:27:16 +0000 (00:27 +0000)
committerQuentin Colombet <qcolombet@apple.com>
Tue, 30 Jul 2013 00:27:16 +0000 (00:27 +0000)
commit15d1b85094cf4c1520fdfd12db2111cd36a194db
treed311acd461d93ab0ef068737aeef6fa114044b37
parent75c9433b49b1e4e2d7e61249c3cd0e3ce910d5c8
[R600] Replicate old DAGCombiner behavior in target specific DAG combine.

build_vector is lowered to REG_SEQUENCE, which is something the register
allocator does a good job at optimizing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187397 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/R600/R600ISelLowering.cpp
test/CodeGen/R600/swizzle-export.ll