[PATCH] [X86][SSE] Added support for SSE3 lane duplication shuffle instructions
chandlerc at gmail.com
Sun Jan 18 16:46:39 PST 2015
This looks really good with only a minor ordering issue between movddup and broadcast which I commented on below.
Comment at: test/CodeGen/X86/vector-shuffle-512-v8.ll:633
@@ -632,3 +632,3 @@
; ALL: # BB#0:
-; ALL-NEXT: vbroadcastsd %xmm0, %ymm1
; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
I would expect vbroadcast to be generally a better choice than vmovddup here..
More information about the llvm-commits