[PATCH] [X86][SSE] Vectorized i8 and i16 shift operators

[Simon Pilgrim]

Hi Fiona,

I did look at implementing more of the shifts as multiplies, but we end up using a lot of instructions (+ registers) for the extension/truncation stages. It is easier with SSE41 instructions, but at that point we can use PBLENDVB which gives a much tighter solution. It might still be worthwhile for constant shift amounts but that would be a later patch.


REPOSITORY
  rL LLVM

http://reviews.llvm.org/D9474

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