[PATCH] Generate better code for shuffles

[Sanjay Patel]

================
Comment at: test/CodeGen/X86/vector-shuffle-combining.ll:1566
@@ +1565,3 @@
+; AVX:       # BB#0:
+; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX-NEXT:    vpunpcklqdq  {{.*#+}} xmm2 = xmm0[0],xmm1[0]
----------------
This isn't passing 'make check' for me. On an AVX2 machine, we generate 'vextracti128' (the integer flavor of the extract instruction).

http://reviews.llvm.org/D6678

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/