Skip to content

[RISCV][CG]Use processShuffleMasks for per-register shuffles #120803

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Dec 23, 2024
Merged

[RISCV][CG]Use processShuffleMasks for per-register shuffles #120803

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
93 changes: 52 additions & 41 deletions llvm/lib/Target/RISCV/RISCVISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5103,7 +5103,6 @@ static SDValue lowerShuffleViaVRegSplitting(ShuffleVectorSDNode *SVN,
SDValue V1 = SVN->getOperand(0);
SDValue V2 = SVN->getOperand(1);
ArrayRef<int> Mask = SVN->getMask();
unsigned NumElts = VT.getVectorNumElements();

// If we don't know exact data layout, not much we can do. If this
// is already m1 or smaller, no point in splitting further.
Expand All @@ -5120,58 +5119,70 @@ static SDValue lowerShuffleViaVRegSplitting(ShuffleVectorSDNode *SVN,

MVT ElemVT = VT.getVectorElementType();
unsigned ElemsPerVReg = *VLen / ElemVT.getFixedSizeInBits();
unsigned VRegsPerSrc = NumElts / ElemsPerVReg;

SmallVector<std::pair<int, SmallVector<int>>>
OutMasks(VRegsPerSrc, {-1, {}});

// Check if our mask can be done as a 1-to-1 mapping from source
// to destination registers in the group without needing to
// write each destination more than once.
for (unsigned DstIdx = 0; DstIdx < Mask.size(); DstIdx++) {
int DstVecIdx = DstIdx / ElemsPerVReg;
int DstSubIdx = DstIdx % ElemsPerVReg;
int SrcIdx = Mask[DstIdx];
if (SrcIdx < 0 || (unsigned)SrcIdx >= 2 * NumElts)
continue;
int SrcVecIdx = SrcIdx / ElemsPerVReg;
int SrcSubIdx = SrcIdx % ElemsPerVReg;
if (OutMasks[DstVecIdx].first == -1)
OutMasks[DstVecIdx].first = SrcVecIdx;
if (OutMasks[DstVecIdx].first != SrcVecIdx)
// Note: This case could easily be handled by keeping track of a chain
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In addition to the functional issue which causes me to revert this change, your comments and tests don't address the profitability issue called out here. Please revise to include.

Copy link
Member Author

@alexey-bataev alexey-bataev Jan 2, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you expect that the multi-shuffles may cause perf degradation? Do you suggest to add some extra analysis and if it requires more than 1 2-vector shuffles just early exit?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Possibly, yes. In particular, using a quadratic number of shuffles (with distinct indices and masks) vs a single one is a lot of code size increase.

I suspect this needs a bit of thought and investigation. I am not proposing any particular heuristic, and am open to being convinced that the right heuristic is to just blindly expand. I'd just like to see it explored and justified.

Though 1-2 shuffles is way to low a threshold. You definitely want something which allows at least the linear expansion of the code you replaced.

// of source values and generating two element shuffles below. This is
// less an implementation question, and more a profitability one.
return SDValue();

OutMasks[DstVecIdx].second.resize(ElemsPerVReg, -1);
OutMasks[DstVecIdx].second[DstSubIdx] = SrcSubIdx;
}

EVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
MVT OneRegVT = MVT::getVectorVT(ElemVT, ElemsPerVReg);
MVT M1VT = getContainerForFixedLengthVector(DAG, OneRegVT, Subtarget);
assert(M1VT == getLMUL1VT(M1VT));
unsigned NumOpElts = M1VT.getVectorMinNumElements();
SDValue Vec = DAG.getUNDEF(ContainerVT);
unsigned NormalizedVF = ContainerVT.getVectorMinNumElements();
unsigned NumOfSrcRegs = NormalizedVF / NumOpElts;
unsigned NumOfDestRegs = NormalizedVF / NumOpElts;
// The following semantically builds up a fixed length concat_vector
// of the component shuffle_vectors. We eagerly lower to scalable here
// to avoid DAG combining it back to a large shuffle_vector again.
V1 = convertToScalableVector(ContainerVT, V1, DAG, Subtarget);
V2 = convertToScalableVector(ContainerVT, V2, DAG, Subtarget);
for (unsigned DstVecIdx = 0 ; DstVecIdx < OutMasks.size(); DstVecIdx++) {
auto &[SrcVecIdx, SrcSubMask] = OutMasks[DstVecIdx];
if (SrcVecIdx == -1)
SmallVector<SDValue> SubRegs(NumOfDestRegs);
unsigned RegCnt = 0;
unsigned PrevCnt = 0;
processShuffleMasks(
Mask, NumOfSrcRegs, NumOfDestRegs, NumOfDestRegs,
[&]() {
PrevCnt = RegCnt;
++RegCnt;
},
[&, &DAG = DAG](ArrayRef<int> SrcSubMask, unsigned SrcVecIdx,
unsigned DstVecIdx) {
SDValue SrcVec = SrcVecIdx >= NumOfSrcRegs ? V2 : V1;
unsigned ExtractIdx = (SrcVecIdx % NumOfSrcRegs) * NumOpElts;
SDValue SubVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, M1VT, SrcVec,
DAG.getVectorIdxConstant(ExtractIdx, DL));
SubVec = convertFromScalableVector(OneRegVT, SubVec, DAG, Subtarget);
SubVec = DAG.getVectorShuffle(OneRegVT, DL, SubVec, SubVec, SrcSubMask);
SubRegs[RegCnt] = convertToScalableVector(M1VT, SubVec, DAG, Subtarget);
PrevCnt = RegCnt;
++RegCnt;
},
[&, &DAG = DAG](ArrayRef<int> SrcSubMask, unsigned Idx1, unsigned Idx2) {
if (PrevCnt + 1 == RegCnt)
++RegCnt;
SDValue SubVec1 = SubRegs[PrevCnt + 1];
if (!SubVec1) {
SDValue SrcVec = Idx1 >= NumOfSrcRegs ? V2 : V1;
unsigned ExtractIdx = (Idx1 % NumOfSrcRegs) * NumOpElts;
SubVec1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, M1VT, SrcVec,
DAG.getVectorIdxConstant(ExtractIdx, DL));
}
SubVec1 = convertFromScalableVector(OneRegVT, SubVec1, DAG, Subtarget);
SDValue SrcVec = Idx2 >= NumOfSrcRegs ? V2 : V1;
unsigned ExtractIdx = (Idx2 % NumOfSrcRegs) * NumOpElts;
SDValue SubVec2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, M1VT, SrcVec,
DAG.getVectorIdxConstant(ExtractIdx, DL));
SubVec2 = convertFromScalableVector(OneRegVT, SubVec2, DAG, Subtarget);
SubVec1 =
DAG.getVectorShuffle(OneRegVT, DL, SubVec1, SubVec2, SrcSubMask);
SubVec1 = convertToScalableVector(M1VT, SubVec1, DAG, Subtarget);
SubRegs[PrevCnt + 1] = SubVec1;
});
assert(RegCnt == NumOfDestRegs && "Whole vector must be processed");
SDValue Vec = DAG.getUNDEF(ContainerVT);
for (auto [I, V] : enumerate(SubRegs)) {
if (!V)
continue;
unsigned ExtractIdx = (SrcVecIdx % VRegsPerSrc) * NumOpElts;
SDValue SrcVec = (unsigned)SrcVecIdx >= VRegsPerSrc ? V2 : V1;
SDValue SubVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, M1VT, SrcVec,
DAG.getVectorIdxConstant(ExtractIdx, DL));
SubVec = convertFromScalableVector(OneRegVT, SubVec, DAG, Subtarget);
SubVec = DAG.getVectorShuffle(OneRegVT, DL, SubVec, SubVec, SrcSubMask);
SubVec = convertToScalableVector(M1VT, SubVec, DAG, Subtarget);
unsigned InsertIdx = DstVecIdx * NumOpElts;
Vec = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, ContainerVT, Vec, SubVec,
unsigned InsertIdx = I * NumOpElts;

Vec = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, ContainerVT, Vec, V,
DAG.getVectorIdxConstant(InsertIdx, DL));
}
return convertFromScalableVector(VT, Vec, DAG, Subtarget);
Expand Down
89 changes: 41 additions & 48 deletions llvm/test/CodeGen/RISCV/rvv/fixed-vectors-shuffle-exact-vlen.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -168,12 +168,11 @@ define <4 x i64> @m2_splat_into_slide_two_source_v2_lo(<4 x i64> %v1, <4 x i64>
define <4 x i64> @m2_splat_into_slide_two_source(<4 x i64> %v1, <4 x i64> %v2) vscale_range(2,2) {
; CHECK-LABEL: m2_splat_into_slide_two_source:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 1, e8, mf8, ta, ma
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Did I miss a separate commit which added additional testing for this change? It clearly isn't NFC, and the test changes here seem fairly minimal for something this involved. If not, please commit additional coverage (including the cause of the functional bug) before reapplying.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There are at least 3 tests, which check the functionality, just the changes are in this one.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You had added a couple tests in 78ab771; I had missed finding that on my first quick search.

Please add the following:

  • A fully quadratic shuffle, possibly at multiple lmuls
  • A mostly linear case - i.e. something which isn't linear, but you think is still a good example of the generality.
  • Whatever the functional bug was.

; CHECK-NEXT: vmv.v.i v0, 12
; CHECK-NEXT: vsetivli zero, 4, e64, m2, ta, mu
; CHECK-NEXT: vsetivli zero, 2, e64, m1, ta, ma
; CHECK-NEXT: vslidedown.vi v13, v10, 1
; CHECK-NEXT: vslideup.vi v13, v11, 1
; CHECK-NEXT: vrgather.vi v12, v8, 0
; CHECK-NEXT: vslideup.vi v12, v10, 1, v0.t
; CHECK-NEXT: vmv.v.v v8, v12
; CHECK-NEXT: vmv2r.v v8, v12
; CHECK-NEXT: ret
%res = shufflevector <4 x i64> %v1, <4 x i64> %v2, <4 x i32> <i32 0, i32 0, i32 5, i32 6>
ret <4 x i64> %res
Expand All @@ -183,18 +182,17 @@ define void @shuffle1(ptr %explicit_0, ptr %explicit_1) vscale_range(2,2) {
; CHECK-LABEL: shuffle1:
; CHECK: # %bb.0:
; CHECK-NEXT: addi a0, a0, 252
; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, ma
; CHECK-NEXT: vmv.v.i v8, 0
; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
; CHECK-NEXT: vid.v v8
; CHECK-NEXT: vid.v v10
; CHECK-NEXT: vsetivli zero, 3, e32, m1, ta, ma
; CHECK-NEXT: vle32.v v9, (a0)
; CHECK-NEXT: li a0, 175
; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
; CHECK-NEXT: vsrl.vi v8, v8, 1
; CHECK-NEXT: vmv.s.x v0, a0
; CHECK-NEXT: vadd.vi v8, v8, 1
; CHECK-NEXT: vrgather.vv v11, v9, v8
; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, ma
; CHECK-NEXT: vmerge.vim v8, v10, 0, v0
; CHECK-NEXT: vle32.v v11, (a0)
; CHECK-NEXT: vmv.v.i v0, 5
; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
; CHECK-NEXT: vsrl.vi v10, v10, 1
; CHECK-NEXT: vadd.vi v10, v10, 1
; CHECK-NEXT: vrgather.vv v9, v11, v10, v0.t
; CHECK-NEXT: addi a0, a1, 672
; CHECK-NEXT: vs2r.v v8, (a0)
; CHECK-NEXT: ret
Expand All @@ -211,15 +209,15 @@ define void @shuffle1(ptr %explicit_0, ptr %explicit_1) vscale_range(2,2) {
define <16 x float> @shuffle2(<4 x float> %a) vscale_range(2,2) {
; CHECK-LABEL: shuffle2:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
; CHECK-NEXT: vid.v v9
; CHECK-NEXT: li a0, -97
; CHECK-NEXT: vadd.vv v9, v9, v9
; CHECK-NEXT: vrsub.vi v9, v9, 4
; CHECK-NEXT: vmv.s.x v0, a0
; CHECK-NEXT: vrgather.vv v13, v8, v9
; CHECK-NEXT: vsetivli zero, 16, e32, m4, ta, ma
; CHECK-NEXT: vmerge.vim v8, v12, 0, v0
; CHECK-NEXT: vmv1r.v v12, v8
; CHECK-NEXT: vmv.v.i v8, 0
; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
; CHECK-NEXT: vid.v v13
; CHECK-NEXT: vadd.vv v13, v13, v13
; CHECK-NEXT: vmv.v.i v0, 6
; CHECK-NEXT: vrsub.vi v13, v13, 4
; CHECK-NEXT: vrgather.vv v9, v12, v13, v0.t
; CHECK-NEXT: ret
%b = extractelement <4 x float> %a, i32 2
%c = insertelement <16 x float> <float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float undef, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00>, float %b, i32 5
Expand All @@ -231,16 +229,15 @@ define <16 x float> @shuffle2(<4 x float> %a) vscale_range(2,2) {
define i64 @extract_any_extend_vector_inreg_v16i64(<16 x i64> %a0, i32 %a1) vscale_range(2,2) {
; RV32-LABEL: extract_any_extend_vector_inreg_v16i64:
; RV32: # %bb.0:
; RV32-NEXT: li a1, 16
; RV32-NEXT: vsetivli zero, 16, e64, m8, ta, mu
; RV32-NEXT: vsetivli zero, 16, e64, m8, ta, ma
; RV32-NEXT: vmv.v.i v16, 0
; RV32-NEXT: vmv.s.x v0, a1
; RV32-NEXT: vsetivli zero, 2, e64, m1, ta, mu
; RV32-NEXT: vmv.v.i v0, 1
; RV32-NEXT: li a1, 32
; RV32-NEXT: vrgather.vi v16, v8, 15, v0.t
; RV32-NEXT: vsetvli zero, zero, e64, m8, ta, ma
; RV32-NEXT: vrgather.vi v18, v15, 1, v0.t
; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
; RV32-NEXT: vslidedown.vx v8, v16, a0
; RV32-NEXT: vmv.x.s a0, v8
; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
; RV32-NEXT: vsrl.vx v8, v8, a1
; RV32-NEXT: vmv.x.s a1, v8
; RV32-NEXT: ret
Expand All @@ -258,13 +255,14 @@ define i64 @extract_any_extend_vector_inreg_v16i64(<16 x i64> %a0, i32 %a1) vsca
; RV64-NEXT: addi s0, sp, 256
; RV64-NEXT: .cfi_def_cfa s0, 0
; RV64-NEXT: andi sp, sp, -128
; RV64-NEXT: li a1, -17
; RV64-NEXT: vsetivli zero, 1, e8, mf8, ta, ma
; RV64-NEXT: vmv.v.i v0, 1
; RV64-NEXT: vsetivli zero, 16, e64, m8, ta, ma
; RV64-NEXT: vmv.s.x v0, a1
; RV64-NEXT: vrgather.vi v16, v8, 15
; RV64-NEXT: vmerge.vim v8, v16, 0, v0
; RV64-NEXT: vmv.v.i v16, 0
; RV64-NEXT: vsetivli zero, 2, e64, m1, ta, mu
; RV64-NEXT: vrgather.vi v18, v15, 1, v0.t
; RV64-NEXT: mv s2, sp
; RV64-NEXT: vs8r.v v8, (s2)
; RV64-NEXT: vs8r.v v16, (s2)
; RV64-NEXT: andi a0, a0, 15
; RV64-NEXT: li a1, 8
; RV64-NEXT: call __muldi3
Expand All @@ -290,21 +288,16 @@ define i64 @extract_any_extend_vector_inreg_v16i64(<16 x i64> %a0, i32 %a1) vsca
define <4 x double> @shuffles_add(<4 x double> %0, <4 x double> %1) vscale_range(2,2) {
; CHECK-LABEL: shuffles_add:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 2, e64, m1, ta, mu
; CHECK-NEXT: vmv1r.v v13, v10
; CHECK-NEXT: vslideup.vi v13, v11, 1
; CHECK-NEXT: vmv1r.v v8, v9
; CHECK-NEXT: vmv.v.i v0, 1
; CHECK-NEXT: vrgather.vi v12, v9, 0
; CHECK-NEXT: vmv1r.v v9, v11
; CHECK-NEXT: vrgather.vi v9, v10, 1, v0.t
; CHECK-NEXT: vsetivli zero, 4, e64, m2, ta, ma
; CHECK-NEXT: vrgather.vi v12, v8, 2
; CHECK-NEXT: vsetvli zero, zero, e16, mf2, ta, ma
; CHECK-NEXT: vid.v v14
; CHECK-NEXT: vmv.v.i v0, 12
; CHECK-NEXT: vsetvli zero, zero, e64, m2, ta, ma
; CHECK-NEXT: vrgather.vi v16, v8, 3
; CHECK-NEXT: vsetvli zero, zero, e16, mf2, ta, ma
; CHECK-NEXT: vadd.vv v8, v14, v14
; CHECK-NEXT: vadd.vi v9, v8, -4
; CHECK-NEXT: vadd.vi v8, v8, -3
; CHECK-NEXT: vsetvli zero, zero, e64, m2, ta, mu
; CHECK-NEXT: vrgatherei16.vv v12, v10, v9, v0.t
; CHECK-NEXT: vrgatherei16.vv v16, v10, v8, v0.t
; CHECK-NEXT: vfadd.vv v8, v12, v16
; CHECK-NEXT: vfadd.vv v8, v12, v8
; CHECK-NEXT: ret
%3 = shufflevector <4 x double> %0, <4 x double> %1, <4 x i32> <i32 undef, i32 2, i32 4, i32 6>
%4 = shufflevector <4 x double> %0, <4 x double> %1, <4 x i32> <i32 undef, i32 3, i32 5, i32 7>
Expand Down
Loading