[PATCH] [DAGCombiner] Remove extra bitcasts surrounding vector shuffles
qcolombet at apple.com
Tue Apr 21 11:23:05 PDT 2015
This looks mostly good, just one question.
Are the tests you added enough to cover all the code you add?
In particular, it is not obvious to me that we cover the case where the bitcast operand of the shuffle are undef, integer constant, or fp constant.
If this is indeed not covered, please add a test case.
Comment at: test/CodeGen/X86/combine-or.ll:290
@@ +289,3 @@
+ %1 = bitcast <2 x double> %a0 to <2 x i64>
+ %2 = bitcast <2 x double> %a1 to <2 x i64>
+ %3 = and <2 x i64> %1, <i64 0, i64 -1>
Please use instnamer to have easier to update tests.
More information about the llvm-commits