[PATCH] D14399: [X86][SSE] Combine UNPCKL with vector_shuffle into UNPCKH to save one instruction for sext from v16i8 to v16i16 and v8i16 to v8i32.

[Cong Hou via llvm-commits]

congh added a comment.

In http://reviews.llvm.org/D14399#283403, @RKSimon wrote:

> Please can you either add tests for 256-bit shuffles or (if you can't get this to occur) limit the combine to 128-bit vectors only?


This combine is not for AVX or couldn't happen on 256-bit: SSE4.1 provides pmovsxbw and pmovsxwd  which are better than unpack+shift. As I could not compose test cases for 256-bit vectors, I still added a check of 128-bit vector in this patch. Thank you for the advice!


http://reviews.llvm.org/D14399