[PATCH] D25722: Improved cost model for FDIV and FSQRT

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Tue Oct 25 04:28:08 PDT 2016


RKSimon added inline comments.


================
Comment at: lib/Target/X86/X86TargetTransformInfo.cpp:1088
   };
   static const CostTblEntry SSSE3CostTbl[] = {
     { ISD::BITREVERSE, MVT::v2i64,   5 },
----------------
RKSimon wrote:
> Worth adding a SSE41CostTbl for Core2 era costs?
Please add Nehalem costs (from Agner) - they're notably better than the P4 default:

FSQRT f32/4f32 : 18 f64/2f64 : 32


https://reviews.llvm.org/D25722





More information about the llvm-commits mailing list