[PATCH] D11477: fix invalid load folding with SSE/AVX FP logical instructions (PR22371)

[Chandler Carruth]

chandlerc accepted this revision.
chandlerc added a comment.
This revision is now accepted and ready to land.

LGTM, and looks good for the branch as well.


================
Comment at: test/CodeGen/X86/pr2656.ll:9-10
@@ -11,1 +8,4 @@
 
+; We can not fold either stack load into an 'xor' intruction because that
+; would change what should be a 4-byte load into a 16-byte load.
+
----------------
I'd specifically call out that we *can* do a 16-byte constant pool load for the xorps mask used to negate these values, it just isn't folded because it is used twice. Otherwise it's a bit confusing to read the comment followed by this particular example.


http://reviews.llvm.org/D11477