summaryrefslogtreecommitdiffstats
path: root/llvm/lib
diff options
context:
space:
mode:
authorSimon Pilgrim <llvm-dev@redking.me.uk>2016-07-20 10:41:28 +0000
committerSimon Pilgrim <llvm-dev@redking.me.uk>2016-07-20 10:41:28 +0000
commit1b4f511aaa0062f47fffcd7acf983cdab48b98c3 (patch)
treeddbb69bd392604efd07ac84bbc50345b14862cf9 /llvm/lib
parentdcad913ab66fde15d00c8d5116b00efef082cc41 (diff)
downloadbcm5719-llvm-1b4f511aaa0062f47fffcd7acf983cdab48b98c3.tar.gz
bcm5719-llvm-1b4f511aaa0062f47fffcd7acf983cdab48b98c3.zip
[X86][SSE] Add cost model values for CTPOP of vectors
This patch adds costs for the vectorized implementations of CTPOP, the default values were seriously underestimating the cost of these and was encouraging vectorization on targets where serialized use of POPCNT would be much better. Differential Revision: https://reviews.llvm.org/D22456 llvm-svn: 276104
Diffstat (limited to 'llvm/lib')
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp2
-rw-r--r--llvm/lib/Target/X86/X86TargetTransformInfo.cpp31
2 files changed, 29 insertions, 4 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index f247908387c..ce88155fdee 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -21052,6 +21052,8 @@ static SDValue LowerVectorCTPOPBitmath(SDValue Op, const SDLoc &DL,
DAG);
}
+// Please ensure that any codegen change from LowerVectorCTPOP is reflected in
+// updated cost models in X86TTIImpl::getIntrinsicInstrCost.
static SDValue LowerVectorCTPOP(SDValue Op, const X86Subtarget &Subtarget,
SelectionDAG &DAG) {
MVT VT = Op.getSimpleValueType();
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index f44a8c66202..1081340d446 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -945,6 +945,10 @@ int X86TTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy) {
int X86TTIImpl::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
ArrayRef<Type *> Tys, FastMathFlags FMF) {
+ // Costs should match the codegen from:
+ // BITREVERSE: llvm\test\CodeGen\X86\vector-bitreverse.ll
+ // BSWAP: llvm\test\CodeGen\X86\bswap-vector.ll
+ // CTPOP: llvm\test\CodeGen\X86\vector-popcnt-*.ll
static const CostTblEntry XOPCostTbl[] = {
{ ISD::BITREVERSE, MVT::v4i64, 4 },
{ ISD::BITREVERSE, MVT::v8i32, 4 },
@@ -966,7 +970,11 @@ int X86TTIImpl::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
{ ISD::BITREVERSE, MVT::v32i8, 5 },
{ ISD::BSWAP, MVT::v4i64, 1 },
{ ISD::BSWAP, MVT::v8i32, 1 },
- { ISD::BSWAP, MVT::v16i16, 1 }
+ { ISD::BSWAP, MVT::v16i16, 1 },
+ { ISD::CTPOP, MVT::v4i64, 7 },
+ { ISD::CTPOP, MVT::v8i32, 11 },
+ { ISD::CTPOP, MVT::v16i16, 9 },
+ { ISD::CTPOP, MVT::v32i8, 6 }
};
static const CostTblEntry AVX1CostTbl[] = {
{ ISD::BITREVERSE, MVT::v4i64, 10 },
@@ -975,7 +983,11 @@ int X86TTIImpl::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
{ ISD::BITREVERSE, MVT::v32i8, 10 },
{ ISD::BSWAP, MVT::v4i64, 4 },
{ ISD::BSWAP, MVT::v8i32, 4 },
- { ISD::BSWAP, MVT::v16i16, 4 }
+ { ISD::BSWAP, MVT::v16i16, 4 },
+ { ISD::CTPOP, MVT::v4i64, 14 },
+ { ISD::CTPOP, MVT::v8i32, 22 },
+ { ISD::CTPOP, MVT::v16i16, 18 },
+ { ISD::CTPOP, MVT::v32i8, 12 }
};
static const CostTblEntry SSSE3CostTbl[] = {
{ ISD::BITREVERSE, MVT::v2i64, 5 },
@@ -984,12 +996,20 @@ int X86TTIImpl::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
{ ISD::BITREVERSE, MVT::v16i8, 5 },
{ ISD::BSWAP, MVT::v2i64, 1 },
{ ISD::BSWAP, MVT::v4i32, 1 },
- { ISD::BSWAP, MVT::v8i16, 1 }
+ { ISD::BSWAP, MVT::v8i16, 1 },
+ { ISD::CTPOP, MVT::v2i64, 7 },
+ { ISD::CTPOP, MVT::v4i32, 11 },
+ { ISD::CTPOP, MVT::v8i16, 9 },
+ { ISD::CTPOP, MVT::v16i8, 6 }
};
static const CostTblEntry SSE2CostTbl[] = {
{ ISD::BSWAP, MVT::v2i64, 7 },
{ ISD::BSWAP, MVT::v4i32, 7 },
- { ISD::BSWAP, MVT::v8i16, 7 }
+ { ISD::BSWAP, MVT::v8i16, 7 },
+ { ISD::CTPOP, MVT::v2i64, 12 },
+ { ISD::CTPOP, MVT::v4i32, 15 },
+ { ISD::CTPOP, MVT::v8i16, 13 },
+ { ISD::CTPOP, MVT::v16i8, 10 }
};
unsigned ISD = ISD::DELETED_NODE;
@@ -1002,6 +1022,9 @@ int X86TTIImpl::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
case Intrinsic::bswap:
ISD = ISD::BSWAP;
break;
+ case Intrinsic::ctpop:
+ ISD = ISD::CTPOP;
+ break;
}
// Legalize the type.
OpenPOWER on IntegriCloud