RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00
Code Issues Projects Releases Wiki Activity

[ARM] Use [SU]ABSDIFF nodes instead of intrinsics for VABD/VABA

Browse Source 
No functional change, but it preps codegen for the future when SABSDIFF
will start getting generated in anger.

llvm-svn: 242546
This commit is contained in:
James Molloy 2015-07-17 17:10:55 +00:00
parent c5d3ffedd0
commit 4e1c143536
2 changed files with 22 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
lib/Target/ARM/ARMISelLowering.cpp
View File

@ -142,6 +142,11 @@ void ARMTargetLowering::addTypeForNEON(MVT VT, MVT PromotedLdStVT,
setOperationAction(ISD::SREM, VT, Expand);
setOperationAction(ISD::UREM, VT, Expand);
setOperationAction(ISD::FREM, VT, Expand);
if (VT.isInteger()) {
setOperationAction(ISD::SABSDIFF, VT, Legal);
setOperationAction(ISD::UABSDIFF, VT, Legal);
}
}
void ARMTargetLowering::addDRTypeForNEON(MVT VT) {
@ -9717,6 +9722,15 @@ static SDValue PerformIntrinsicCombine(SDNode *N, SelectionDAG &DAG) {
// Don't do anything for most intrinsics.
break;
case Intrinsic::arm_neon_vabds:
if (!N->getValueType(0).isInteger())
return SDValue();
return DAG.getNode(ISD::SABSDIFF, SDLoc(N), N->getValueType(0),
N->getOperand(1), N->getOperand(2));
case Intrinsic::arm_neon_vabdu:
return DAG.getNode(ISD::UABSDIFF, SDLoc(N), N->getValueType(0),
N->getOperand(1), N->getOperand(2));
// Vector shifts: check for immediate versions and lower them.
// Note: This is done during DAG combining instead of DAG legalizing because
// the build_vectors for 64-bit vector element shift counts are generally

16
lib/Target/ARM/ARMInstrNEON.td
View File

@ -4999,10 +4999,10 @@ def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
// VABD : Vector Absolute Difference
defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
"vabd", "s", int_arm_neon_vabds, 1>;
"vabd", "s", sabsdiff, 1>;
defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
"vabd", "u", int_arm_neon_vabdu, 1>;
"vabd", "u", uabsdiff, 1>;
def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
"vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
@ -5010,21 +5010,21 @@ def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
// VABDL : Vector Absolute Difference Long (Q = | D - D |)
defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
"vabdl", "s", int_arm_neon_vabds, zext, 1>;
"vabdl", "s", sabsdiff, zext, 1>;
defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
"vabdl", "u", int_arm_neon_vabdu, zext, 1>;
"vabdl", "u", uabsdiff, zext, 1>;
// VABA : Vector Absolute Difference and Accumulate
defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
"vaba", "s", int_arm_neon_vabds, add>;
"vaba", "s", sabsdiff, add>;
defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
"vaba", "u", int_arm_neon_vabdu, add>;
"vaba", "u", uabsdiff, add>;
// VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
"vabal", "s", int_arm_neon_vabds, zext, add>;
"vabal", "s", sabsdiff, zext, add>;
defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
"vabal", "u", int_arm_neon_vabdu, zext, add>;
"vabal", "u", uabsdiff, zext, add>;
// Vector Maximum and Minimum.