[llvm] d867634 - [CostModel][X86] Update comment describing source of costs - we now use llvm-mca more than IACA
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Fri Jul 2 06:52:24 PDT 2021
Author: Simon Pilgrim
Date: 2021-07-02T14:29:32+01:00
New Revision: d867634fbdb1f1f521981c244c410a9b839c8325
URL: https://github.com/llvm/llvm-project/commit/d867634fbdb1f1f521981c244c410a9b839c8325
DIFF: https://github.com/llvm/llvm-project/commit/d867634fbdb1f1f521981c244c410a9b839c8325.diff
LOG: [CostModel][X86] Update comment describing source of costs - we now use llvm-mca more than IACA
Added:
Modified:
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index 9eb5abe4dd9b..9064cca9da52 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -2074,9 +2074,9 @@ InstructionCost X86TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
};
static const TypeConversionCostTblEntry SSE2ConversionTbl[] = {
- // These are somewhat magic numbers justified by looking at the output of
- // Intel's IACA, running some kernels and making sure when we take
- // legalization into account the throughput will be overestimated.
+ // These are somewhat magic numbers justified by comparing the
+ // output of llvm-mca for our various supported scheduler models
+ // and basing it off the worst case scenario.
{ ISD::SINT_TO_FP, MVT::v4f32, MVT::v16i8, 3 },
{ ISD::SINT_TO_FP, MVT::v2f64, MVT::v16i8, 4 },
{ ISD::SINT_TO_FP, MVT::v4f32, MVT::v8i16, 3 },
More information about the llvm-commits
mailing list