[PATCH] [X86][SSE] Improved (v)insertps shuffle matching

Simon Pilgrim llvm-dev at redking.me.uk
Sat Jan 10 13:37:34 PST 2015


Thanks for the feedback guys - apologies for the code style problems, they should be fixed now.

Regarding performance - its tricky to give specific numbers as insertps gets matched against a wide variety of masks, but if we assume that an insertps instruction replaces a xorps (zero) and 2 dependant shufps, on a brief test I'm seeing a 35% boost on Core2Duo.


REPOSITORY
  rL LLVM

http://reviews.llvm.org/D6879

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






More information about the llvm-commits mailing list