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

Combine NEON VABD (absolute difference) intrinsics with ADDs to make VABA 

(absolute difference with accumulate) intrinsics.  Radar 8228576.

llvm-svn: 110170
This commit is contained in:
Bob Wilson 2010-08-04 00:12:08 +00:00
parent 794172ca9c
commit 79daf7e0ae
2 changed files with 40 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
llvm/lib/Target/ARM/ARMISelLowering.cpp
View File

@ -4248,12 +4248,28 @@ SDValue combineSelectAndUse(SDNode *N, SDValue Slct, SDValue OtherOp,
/// operands.
static SDValue PerformADDCombineWithOperands(SDNode *N, SDValue N0, SDValue N1,
TargetLowering::DAGCombinerInfo &DCI) {
SelectionDAG &DAG = DCI.DAG;
// fold (add (select cc, 0, c), x) -> (select cc, x, (add, x, c))
if (N0.getOpcode() == ISD::SELECT && N0.getNode()->hasOneUse()) {
SDValue Result = combineSelectAndUse(N, N0, N1, DCI);
if (Result.getNode()) return Result;
}
// fold (add (arm_neon_vabd a, b) c) -> (arm_neon_vaba c, a, b)
EVT VT = N->getValueType(0);
if (N0.getOpcode() == ISD::INTRINSIC_WO_CHAIN && VT.isInteger()) {
unsigned IntNo = cast<ConstantSDNode>(N0.getOperand(0))->getZExtValue();
if (IntNo == Intrinsic::arm_neon_vabds)
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, N->getDebugLoc(), VT,
DAG.getConstant(Intrinsic::arm_neon_vabas, MVT::i32),
N1, N0.getOperand(1), N0.getOperand(2));
if (IntNo == Intrinsic::arm_neon_vabdu)
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, N->getDebugLoc(), VT,
DAG.getConstant(Intrinsic::arm_neon_vabau, MVT::i32),
N1, N0.getOperand(1), N0.getOperand(2));
}
return SDValue();
}

24
llvm/test/CodeGen/ARM/vaba.ll
View File

@ -203,3 +203,27 @@ declare <2 x i64> @llvm.arm.neon.vabals.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) n
declare <8 x i16> @llvm.arm.neon.vabalu.v8i16(<8 x i16>, <8 x i8>, <8 x i8>) nounwind readnone
declare <4 x i32> @llvm.arm.neon.vabalu.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
declare <2 x i64> @llvm.arm.neon.vabalu.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
define <8 x i8> @vabd_combine_s8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
;CHECK: vabd_combine_s8:
;CHECK: vaba.s8
%tmp1 = load <8 x i8>* %A
%tmp2 = load <8 x i8>* %B
%tmp3 = call <8 x i8> @llvm.arm.neon.vabds.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
%tmp4 = add <8 x i8> %tmp2, %tmp3
ret <8 x i8> %tmp4
}
define <4 x i16> @vabd_combine_u16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
;CHECK: vabd_combine_u16:
;CHECK: vaba.u16
%tmp1 = load <4 x i16>* %A
%tmp2 = load <4 x i16>* %B
%tmp3 = call <4 x i16> @llvm.arm.neon.vabdu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
%tmp4 = add <4 x i16> %tmp3, %tmp1
ret <4 x i16> %tmp4
}
declare <8 x i8> @llvm.arm.neon.vabds.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
declare <4 x i16> @llvm.arm.neon.vabdu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone