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

[SystemZTTIImpl] Give correct cost values for vector bswap intrinsics. 

Implement getIntrinsicInstrCost() and return costs reflecting that bswap can
be done with a vperm per vector register.

Review: Ulrich Weigand
https://reviews.llvm.org/D54789

llvm-svn: 347445
This commit is contained in:
Jonas Paulsson 2018-11-22 07:17:29 +00:00
parent da91431842
commit 96782c2c0b
3 changed files with 90 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
llvm/lib/Target/SystemZ/SystemZTargetTransformInfo.cpp
View File

@ -1047,3 +1047,29 @@ int SystemZTTIImpl::getInterleavedMemoryOpCost(unsigned Opcode, Type *VecTy,
// Cost of load/store operations and the permutations needed.
return NumVectorMemOps + NumPermutes;
}
static int getVectorIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy) {
if (RetTy->isVectorTy() && ID == Intrinsic::bswap)
return getNumVectorRegs(RetTy); // VPERM
return -1;
}
int SystemZTTIImpl::getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Value *> Args,
FastMathFlags FMF, unsigned VF) {
int Cost = getVectorIntrinsicInstrCost(ID, RetTy);
if (Cost != -1)
return Cost;
return BaseT::getIntrinsicInstrCost(ID, RetTy, Args, FMF, VF);
}
int SystemZTTIImpl::getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Type *> Tys,
FastMathFlags FMF,
unsigned ScalarizationCostPassed) {
int Cost = getVectorIntrinsicInstrCost(ID, RetTy);
if (Cost != -1)
return Cost;
return BaseT::getIntrinsicInstrCost(ID, RetTy, Tys,
FMF, ScalarizationCostPassed);
}

7
llvm/lib/Target/SystemZ/SystemZTargetTransformInfo.h
View File

@ -98,6 +98,13 @@ public:
unsigned AddressSpace,
bool UseMaskForCond = false,
bool UseMaskForGaps = false);
int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Value *> Args, FastMathFlags FMF,
unsigned VF = 1);
int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Type *> Tys, FastMathFlags FMF,
unsigned ScalarizationCostPassed = UINT_MAX);
/// @}
};

57
llvm/test/Analysis/CostModel/SystemZ/intrinsics.ll Normal file
View File

@ -0,0 +1,57 @@
; RUN: opt < %s -cost-model -analyze -mtriple=systemz-unknown -mcpu=z13 | FileCheck %s
define void @bswap_i64(i64 %arg, <2 x i64> %arg2) {
; CHECK: Printing analysis 'Cost Model Analysis' for function 'bswap_i64':
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp1 = tail call i64
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp2 = tail call <2 x i64>
; CHECK: Cost Model: Found an estimated cost of 2 for instruction: %swp4 = tail call <4 x i64>
%swp1 = tail call i64 @llvm.bswap.i64(i64 %arg)
%swp2 = tail call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %arg2)
%swp4 = tail call <4 x i64> @llvm.bswap.v4i64(<4 x i64> undef)
ret void
}
define void @bswap_i32(i32 %arg, <2 x i32> %arg2, <4 x i32> %arg4) {
; CHECK: Printing analysis 'Cost Model Analysis' for function 'bswap_i32':
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp1 = tail call i32
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp2 = tail call <2 x i32>
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp4 = tail call <4 x i32>
; CHECK: Cost Model: Found an estimated cost of 2 for instruction: %swp8 = tail call <8 x i32>
%swp1 = tail call i32 @llvm.bswap.i32(i32 %arg)
%swp2 = tail call <2 x i32> @llvm.bswap.v2i32(<2 x i32> %arg2)
%swp4 = tail call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %arg4)
%swp8 = tail call <8 x i32> @llvm.bswap.v8i32(<8 x i32> undef)
ret void
}
define void @bswap_i16(i16 %arg, <2 x i16> %arg2, <4 x i16> %arg4,
<8 x i16> %arg8) {
; CHECK: Printing analysis 'Cost Model Analysis' for function 'bswap_i16':
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp1 = tail call i16 @llvm.bswap.i16(i16 %arg)
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp2 = tail call <2 x i16> @llvm.bswap.v2i16(<2 x i16> %arg2)
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp4 = tail call <4 x i16> @llvm.bswap.v4i16(<4 x i16> %arg4)
; CHECK: Cost Model: Found an estimated cost of 1 for instruction: %swp8 = tail call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %arg8)
; CHECK: Cost Model: Found an estimated cost of 2 for instruction: %swp16 = tail call <16 x i16> @llvm.bswap.v16i16(<16 x i16> undef)
%swp1 = tail call i16 @llvm.bswap.i16(i16 %arg)
%swp2 = tail call <2 x i16> @llvm.bswap.v2i16(<2 x i16> %arg2)
%swp4 = tail call <4 x i16> @llvm.bswap.v4i16(<4 x i16> %arg4)
%swp8 = tail call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %arg8)
%swp16 = tail call <16 x i16> @llvm.bswap.v16i16(<16 x i16> undef)
ret void
}
declare i64 @llvm.bswap.i64(i64)
declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>)
declare i32 @llvm.bswap.i32(i32)
declare <2 x i32> @llvm.bswap.v2i32(<2 x i32>)
declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
declare <8 x i32> @llvm.bswap.v8i32(<8 x i32>)
declare i16 @llvm.bswap.i16(i16)
declare <2 x i16> @llvm.bswap.v2i16(<2 x i16>)
declare <4 x i16> @llvm.bswap.v4i16(<4 x i16>)
declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>)
declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)