[PATCH] D37803: [X86] When applying the shuffle-to-zero-extend transformation on floating point, bitcast to integer first.

[Simon Pilgrim via Phabricator via llvm-commits]

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

LGTM (one minor) - I'm surprised how bad the AVX512 codegen is but that's a job for another day.....



================
Comment at: test/CodeGen/X86/vector-shuffle-combining-avx2.ll:901
+
+define internal fastcc <8 x float> @PR34577(<8 x float> %inp0, <8 x float> %inp1, <8 x float> %inp2) #0 {
+; X32-AVX2-LABEL: PR34577:
----------------
Drop the #0 ?


https://reviews.llvm.org/D37803