replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[CostModel][X86] Reordered AVX1 arithmetic cost LUT into descending target order. NFCI. 

llvm-svn: 291352
This commit is contained in:
Simon Pilgrim 2017-01-07 17:03:51 +00:00
parent 32f8d560dd
commit 100eae1ee0
1 changed files with 27 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
View File

@ -440,6 +440,33 @@ int X86TTIImpl::getArithmeticInstrCost(
ISD = ISD::MUL;
}
static const CostTblEntry AVX1CostTable[] = {
// We don't have to scalarize unsupported ops. We can issue two half-sized
// operations and we only need to extract the upper YMM half.
// Two ops + 1 extract + 1 insert = 4.
{ ISD::MUL, MVT::v16i16, 4 },
{ ISD::MUL, MVT::v8i32, 4 },
{ ISD::SUB, MVT::v32i8, 4 },
{ ISD::ADD, MVT::v32i8, 4 },
{ ISD::SUB, MVT::v16i16, 4 },
{ ISD::ADD, MVT::v16i16, 4 },
{ ISD::SUB, MVT::v8i32, 4 },
{ ISD::ADD, MVT::v8i32, 4 },
{ ISD::SUB, MVT::v4i64, 4 },
{ ISD::ADD, MVT::v4i64, 4 },
// A v4i64 multiply is custom lowered as two split v2i64 vectors that then
// are lowered as a series of long multiplies(3), shifts(3) and adds(2)
// Because we believe v4i64 to be a legal type, we must also include the
// extract+insert in the cost table. Therefore, the cost here is 18
// instead of 8.
{ ISD::MUL, MVT::v4i64, 18 },
};
if (ST->hasAVX() && !ST->hasAVX2())
if (const auto *Entry = CostTableLookup(AVX1CostTable, ISD, LT.second))
return LT.first * Entry->Cost;
static const CostTblEntry SSE42CostTable[] = {
{ ISD::FDIV, MVT::f32, 14 }, // Nehalem from http://www.agner.org/
{ ISD::FDIV, MVT::v4f32, 14 }, // Nehalem from http://www.agner.org/
@ -529,33 +556,6 @@ int X86TTIImpl::getArithmeticInstrCost(
if (const auto *Entry = CostTableLookup(SSE2CostTable, ISD, LT.second))
return LT.first * Entry->Cost;
static const CostTblEntry AVX1CostTable[] = {
// We don't have to scalarize unsupported ops. We can issue two half-sized
// operations and we only need to extract the upper YMM half.
// Two ops + 1 extract + 1 insert = 4.
{ ISD::MUL, MVT::v16i16, 4 },
{ ISD::MUL, MVT::v8i32, 4 },
{ ISD::SUB, MVT::v32i8, 4 },
{ ISD::ADD, MVT::v32i8, 4 },
{ ISD::SUB, MVT::v16i16, 4 },
{ ISD::ADD, MVT::v16i16, 4 },
{ ISD::SUB, MVT::v8i32, 4 },
{ ISD::ADD, MVT::v8i32, 4 },
{ ISD::SUB, MVT::v4i64, 4 },
{ ISD::ADD, MVT::v4i64, 4 },
// A v4i64 multiply is custom lowered as two split v2i64 vectors that then
// are lowered as a series of long multiplies(3), shifts(3) and adds(2)
// Because we believe v4i64 to be a legal type, we must also include the
// extract+insert in the cost table. Therefore, the cost here is 18
// instead of 8.
{ ISD::MUL, MVT::v4i64, 18 },
};
// Look for AVX1 lowering tricks.
if (ST->hasAVX() && !ST->hasAVX2())
if (const auto *Entry = CostTableLookup(AVX1CostTable, ISD, LT.second))
return LT.first * Entry->Cost;
static const CostTblEntry SSE1CostTable[] = {
{ ISD::FDIV, MVT::f32, 17 }, // Pentium III from http://www.agner.org/
{ ISD::FDIV, MVT::v4f32, 34 }, // Pentium III from http://www.agner.org/