RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2025-01-09 05:31:19 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

AMDGPU: Clean up VOP3NoMods pattern

Browse Source 
There is no need to copy the operands or inspect the sources.
Also remove some unnecessary clamp/omod usage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@301363 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Matt Arsenault 2017-04-25 21:17:38 +00:00
parent c2cc67a772
commit 185d3194ea
3 changed files with 22 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
View File

@ -153,7 +153,7 @@ private:
bool SelectVOP3Mods_NNaN(SDValue In, SDValue &Src, SDValue &SrcMods) const;
bool SelectVOP3Mods(SDValue In, SDValue &Src, SDValue &SrcMods) const;
bool SelectVOP3NoMods(SDValue In, SDValue &Src, SDValue &SrcMods) const;
bool SelectVOP3NoMods(SDValue In, SDValue &Src) const;
bool SelectVOP3Mods0(SDValue In, SDValue &Src, SDValue &SrcMods,
SDValue &Clamp, SDValue &Omod) const;
bool SelectVOP3NoMods0(SDValue In, SDValue &Src, SDValue &SrcMods,
@ -1667,38 +1667,28 @@ bool AMDGPUDAGToDAGISel::SelectVOP3Mods_NNaN(SDValue In, SDValue &Src,
return isNoNanSrc(Src);
}
bool AMDGPUDAGToDAGISel::SelectVOP3NoMods(SDValue In, SDValue &Src,
SDValue &SrcMods) const {
bool Res = SelectVOP3Mods(In, Src, SrcMods);
return Res && cast<ConstantSDNode>(SrcMods)->isNullValue();
bool AMDGPUDAGToDAGISel::SelectVOP3NoMods(SDValue In, SDValue &Src) const {
if (In.getOpcode() == ISD::FABS || In.getOpcode() == ISD::FNEG)
return false;
Src = In;
return true;
}
bool AMDGPUDAGToDAGISel::SelectVOP3Mods0(SDValue In, SDValue &Src,
SDValue &SrcMods, SDValue &Clamp,
SDValue &Omod) const {
SDLoc DL(In);
// FIXME: Handle Clamp and Omod
Clamp = CurDAG->getTargetConstant(0, DL, MVT::i32);
Omod = CurDAG->getTargetConstant(0, DL, MVT::i32);
Clamp = CurDAG->getTargetConstant(0, DL, MVT::i1);
Omod = CurDAG->getTargetConstant(0, DL, MVT::i1);
return SelectVOP3Mods(In, Src, SrcMods);
}
bool AMDGPUDAGToDAGISel::SelectVOP3NoMods0(SDValue In, SDValue &Src,
SDValue &SrcMods, SDValue &Clamp,
SDValue &Omod) const {
bool Res = SelectVOP3Mods0(In, Src, SrcMods, Clamp, Omod);
return Res && cast<ConstantSDNode>(SrcMods)->isNullValue() &&
cast<ConstantSDNode>(Clamp)->isNullValue() &&
cast<ConstantSDNode>(Omod)->isNullValue();
}
bool AMDGPUDAGToDAGISel::SelectVOP3Mods0Clamp(SDValue In, SDValue &Src,
SDValue &SrcMods,
SDValue &Omod) const {
// FIXME: Handle Omod
Omod = CurDAG->getTargetConstant(0, SDLoc(In), MVT::i32);
Omod = CurDAG->getTargetConstant(0, SDLoc(In), MVT::i1);
return SelectVOP3Mods(In, Src, SrcMods);
}
@ -1716,9 +1706,8 @@ bool AMDGPUDAGToDAGISel::SelectVOP3OMods(SDValue In, SDValue &Src,
Src = In;
SDLoc DL(In);
// FIXME: Handle Clamp and Omod
Clamp = CurDAG->getTargetConstant(0, DL, MVT::i32);
Omod = CurDAG->getTargetConstant(0, DL, MVT::i32);
Clamp = CurDAG->getTargetConstant(0, DL, MVT::i1);
Omod = CurDAG->getTargetConstant(0, DL, MVT::i1);
return true;
}

3
lib/Target/AMDGPU/SIInstrInfo.td
View File

@ -646,11 +646,10 @@ def DS64Bit4ByteAligned : ComplexPattern<i32, 3, "SelectDS64Bit4ByteAligned">;
def MOVRELOffset : ComplexPattern<i32, 2, "SelectMOVRELOffset">;
def VOP3Mods0 : ComplexPattern<untyped, 4, "SelectVOP3Mods0">;
def VOP3NoMods0 : ComplexPattern<untyped, 4, "SelectVOP3NoMods0">;
def VOP3Mods0Clamp : ComplexPattern<untyped, 3, "SelectVOP3Mods0Clamp">;
def VOP3Mods0Clamp0OMod : ComplexPattern<untyped, 4, "SelectVOP3Mods0Clamp0OMod">;
def VOP3Mods : ComplexPattern<untyped, 2, "SelectVOP3Mods">;
def VOP3NoMods : ComplexPattern<untyped, 2, "SelectVOP3NoMods">;
def VOP3NoMods : ComplexPattern<untyped, 1, "SelectVOP3NoMods">;
// VOP3Mods, but the input source is known to never be NaN.
def VOP3Mods_nnan : ComplexPattern<fAny, 2, "SelectVOP3Mods_NNaN">;

19
lib/Target/AMDGPU/SIInstructions.td
View File

@ -472,8 +472,8 @@ def : Pat <
// fp_to_fp16 patterns
def : Pat <
(i32 (AMDGPUfp_to_f16 (f32 (VOP3Mods0 f32:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)))),
(V_CVT_F16_F32_e64 $src0_modifiers, f32:$src0, $clamp, $omod)
(i32 (AMDGPUfp_to_f16 (f32 (VOP3Mods f32:$src0, i32:$src0_modifiers)))),
(V_CVT_F16_F32_e64 $src0_modifiers, f32:$src0, DSTCLAMP.NONE, DSTOMOD.NONE)
>;
def : Pat <
@ -502,11 +502,11 @@ def : Pat <
multiclass FMADPat <ValueType vt, Instruction inst> {
def : Pat <
(vt (fmad (VOP3NoMods0 vt:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod),
(VOP3NoMods vt:$src1, i32:$src1_modifiers),
(VOP3NoMods vt:$src2, i32:$src2_modifiers))),
(inst $src0_modifiers, $src0, $src1_modifiers, $src1,
$src2_modifiers, $src2, $clamp, $omod)
(vt (fmad (VOP3NoMods vt:$src0),
(VOP3NoMods vt:$src1),
(VOP3NoMods vt:$src2))),
(inst SRCMODS.NONE, $src0, SRCMODS.NONE, $src1,
SRCMODS.NONE, $src2, DSTCLAMP.NONE, DSTOMOD.NONE)
>;
}
@ -676,10 +676,9 @@ def : BitConvert <v16f32, v16i32, VReg_512>;
// If denormals are not enabled, it only impacts the compare of the
// inputs. The output result is not flushed.
class ClampPat<Instruction inst, ValueType vt> : Pat <
(vt (AMDGPUclamp
(VOP3Mods0Clamp vt:$src0, i32:$src0_modifiers, i32:$omod))),
(vt (AMDGPUclamp (VOP3Mods vt:$src0, i32:$src0_modifiers))),
(inst i32:$src0_modifiers, vt:$src0,
i32:$src0_modifiers, vt:$src0, DSTCLAMP.ENABLE, $omod)
i32:$src0_modifiers, vt:$src0, DSTCLAMP.ENABLE, DSTOMOD.NONE)
>;
def : ClampPat<V_MAX_F32_e64, f32>;