This is an archive of the discontinued LLVM Phabricator instance.

Differential D99270

[RISCV] Optimize select-like vector shuffles
ClosedPublic

Authored by frasercrmck on Mar 24 2021, 8:30 AM.

Details

Reviewers
craig.topper
evandro
rogfer01
HsiangKai
khchen
arcbbb
Commits
rG99211352c1ac: [RISCV] Optimize select-like vector shuffles
Summary

This patch adds a small optimization for vector shuffle lowering,
detecting shuffles which can be re-expressed as vector selects.

Diff Detail

Event Timeline

frasercrmck created this revision.Mar 24 2021, 8:30 AM
Herald added subscribers: vkmr, luismarques, apazos and 22 others. · View Herald TranscriptMar 24 2021, 8:30 AM
frasercrmck requested review of this revision.Mar 24 2021, 8:30 AM
Herald added a project: Restricted Project. · View Herald TranscriptMar 24 2021, 8:30 AM
Herald added subscribers: llvm-commits, MaskRay. · View Herald Transcript
frasercrmck added a comment.Mar 24 2021, 8:31 AM

This test file doesn't currently exist: I will/would pre-commit it before merging.

I do wonder if this should be offered as a target-independent optimization, but as it stands this is the best way of catching this pattern.

Harbormaster completed remote builds in B95502: Diff 333006.Mar 24 2021, 8:54 AM
craig.topper accepted this revision.Mar 24 2021, 5:12 PM

LGTM

This revision is now accepted and ready to land.Mar 24 2021, 5:12 PM
frasercrmck mentioned this in rG1e56e8717f09: [RISCV] Pre-commit shuffle test cases for D99270.Mar 25 2021, 3:50 AM
frasercrmck updated this revision to Diff 333250.Mar 25 2021, 3:52 AM
  • rebase on top of main + pre-committed test cases
frasercrmck updated this revision to Diff 333251.Mar 25 2021, 3:56 AM
  • remove unused check prefixes
Harbormaster completed remote builds in B95659: Diff 333250.Mar 25 2021, 4:23 AM
Harbormaster completed remote builds in B95660: Diff 333251.Mar 25 2021, 4:43 AM
Closed by commit rG99211352c1ac: [RISCV] Optimize select-like vector shuffles (authored by frasercrmck). · Explain WhyMar 25 2021, 4:46 AM
This revision was automatically updated to reflect the committed changes.
frasercrmck added a commit: rG99211352c1ac: [RISCV] Optimize select-like vector shuffles.

Revision Contents

PathSize
llvm/
lib/
Target/
RISCV/
29 lines
test/
CodeGen/
RISCV/
rvv/
309 lines
159 lines

Diff 333261

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 1,365 Lines • ▼ Show 20 Linesstatic SDValue lowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
} }
return SDValue(); return SDValue();
} }
static SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG, static SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) { const RISCVSubtarget &Subtarget) {
SDValue V1 = Op.getOperand(0); SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
SDLoc DL(Op); SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op.getSimpleValueType(); MVT VT = Op.getSimpleValueType();
unsigned NumElts = VT.getVectorNumElements();
ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op.getNode()); ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op.getNode());
if (SVN->isSplat()) { if (SVN->isSplat()) {
int Lane = SVN->getSplatIndex(); int Lane = SVN->getSplatIndex();
if (Lane >= 0) { if (Lane >= 0) {
MVT ContainerVT = RISCVTargetLowering::getContainerForFixedLengthVector( MVT ContainerVT = RISCVTargetLowering::getContainerForFixedLengthVector(
DAG, VT, Subtarget); DAG, VT, Subtarget);
V1 = convertToScalableVector(ContainerVT, V1, DAG, Subtarget); V1 = convertToScalableVector(ContainerVT, V1, DAG, Subtarget);
assert(Lane < (int)VT.getVectorNumElements() && "Unexpected lane!"); assert(Lane < (int)NumElts && "Unexpected lane!");
SDValue Mask, VL; SDValue Mask, VL;
std::tie(Mask, VL) = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget); std::tie(Mask, VL) = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
MVT XLenVT = Subtarget.getXLenVT();
SDValue Gather = SDValue Gather =
DAG.getNode(RISCVISD::VRGATHER_VX_VL, DL, ContainerVT, V1, DAG.getNode(RISCVISD::VRGATHER_VX_VL, DL, ContainerVT, V1,
DAG.getConstant(Lane, DL, XLenVT), Mask, VL); DAG.getConstant(Lane, DL, XLenVT), Mask, VL);
return convertFromScalableVector(VT, Gather, DAG, Subtarget); return convertFromScalableVector(VT, Gather, DAG, Subtarget);
} }
} }
// Detect shuffles which can be re-expressed as vector selects.
SmallVector<SDValue> MaskVals;
// By default we preserve the original operand order, and select LHS as true
// and RHS as false. However, since RVV vector selects may feature splats but
// only on the LHS, we may choose to invert our mask and instead select
// between RHS and LHS.
bool SwapOps = DAG.isSplatValue(V2) && !DAG.isSplatValue(V1);
bool IsSelect = all_of(enumerate(SVN->getMask()), [&](const auto &MaskIdx) {
int MaskIndex = MaskIdx.value();
bool SelectMaskVal = (MaskIndex < (int)NumElts) ^ SwapOps;
MaskVals.push_back(DAG.getConstant(SelectMaskVal, DL, XLenVT));
return MaskIndex < 0 || MaskIdx.index() == (unsigned)MaskIndex % NumElts;
});
if (IsSelect) {
assert(MaskVals.size() == NumElts && "Unexpected select-like shuffle");
MVT MaskVT = MVT::getVectorVT(MVT::i1, NumElts);
SDValue SelectMask = DAG.getBuildVector(MaskVT, DL, MaskVals);
return DAG.getNode(ISD::VSELECT, DL, VT, SelectMask, SwapOps ? V2 : V1,
SwapOps ? V1 : V2);
}
return SDValue(); return SDValue();
} }
static SDValue getRVVFPExtendOrRound(SDValue Op, MVT VT, MVT ContainerVT, static SDValue getRVVFPExtendOrRound(SDValue Op, MVT VT, MVT ContainerVT,
SDLoc DL, SelectionDAG &DAG, SDLoc DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) { const RISCVSubtarget &Subtarget) {
if (VT.isScalableVector()) if (VT.isScalableVector())
return DAG.getFPExtendOrRound(Op, DL, VT); return DAG.getFPExtendOrRound(Op, DL, VT);
▲ Show 20 Lines • Show All 6,108 Lines • Show Last 20 Lines

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-shuffles.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=riscv32 -mattr=+d,+experimental-zfh,+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32 ; RUN: llc -mtriple=riscv32 -mattr=+d,+experimental-zfh,+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK
; RUN: llc -mtriple=riscv64 -mattr=+d,+experimental-zfh,+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64 ; RUN: llc -mtriple=riscv64 -mattr=+d,+experimental-zfh,+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK
define <4 x half> @shuffle_v4f16(<4 x half> %x, <4 x half> %y) { define <4 x half> @shuffle_v4f16(<4 x half> %x, <4 x half> %y) {
; CHECK-LABEL: shuffle_v4f16: ; CHECK-LABEL: shuffle_v4f16:
; CHECK: # %bb.0: ; CHECK: # %bb.0:
; CHECK-NEXT: addi sp, sp, -16 ; CHECK-NEXT: addi a0, zero, 11
; CHECK-NEXT: .cfi_def_cfa_offset 16 ; CHECK-NEXT: vsetivli a1, 1, e8,m1,ta,mu
; CHECK-NEXT: vsetivli a0, 1, e16,m1,ta,mu ; CHECK-NEXT: vmv.s.x v0, a0
; CHECK-NEXT: vslidedown.vi v25, v8, 3
; CHECK-NEXT: vfmv.f.s ft0, v25
; CHECK-NEXT: vsetivli a0, 2, e16,m1,ta,mu
; CHECK-NEXT: vfmv.v.f v25, ft0
; CHECK-NEXT: vsetivli a0, 1, e16,m1,ta,mu
; CHECK-NEXT: vslidedown.vi v26, v9, 2
; CHECK-NEXT: vfmv.f.s ft0, v26
; CHECK-NEXT: vsetivli a0, 2, e16,m1,ta,mu
; CHECK-NEXT: vfmv.s.f v25, ft0
; CHECK-NEXT: addi a0, sp, 12
; CHECK-NEXT: vse16.v v25, (a0)
; CHECK-NEXT: vsetivli a0, 1, e16,m1,ta,mu
; CHECK-NEXT: vslidedown.vi v25, v8, 1
; CHECK-NEXT: vfmv.f.s ft0, v25
; CHECK-NEXT: vsetivli a0, 2, e16,m1,ta,mu
; CHECK-NEXT: vfmv.v.f v25, ft0
; CHECK-NEXT: vfmv.f.s ft0, v8
; CHECK-NEXT: vfmv.s.f v25, ft0
; CHECK-NEXT: addi a0, sp, 8
; CHECK-NEXT: vse16.v v25, (a0)
; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu ; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu
; CHECK-NEXT: addi a0, sp, 8 ; CHECK-NEXT: vmerge.vvm v8, v9, v8, v0
; CHECK-NEXT: vle16.v v8, (a0)
; CHECK-NEXT: addi sp, sp, 16
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%s = shufflevector <4 x half> %x, <4 x half> %y, <4 x i32> <i32 0, i32 1, i32 6, i32 3> %s = shufflevector <4 x half> %x, <4 x half> %y, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
ret <4 x half> %s ret <4 x half> %s
} }
define <8 x float> @shuffle_v8f32(<8 x float> %x, <8 x float> %y) { define <8 x float> @shuffle_v8f32(<8 x float> %x, <8 x float> %y) {
; RV32-LABEL: shuffle_v8f32: ; CHECK-LABEL: shuffle_v8f32:
; RV32: # %bb.0: ; CHECK: # %bb.0:
; RV32-NEXT: addi sp, sp, -64 ; CHECK-NEXT: addi a0, zero, 236
; RV32-NEXT: .cfi_def_cfa_offset 64 ; CHECK-NEXT: vsetivli a1, 1, e8,m1,ta,mu
; RV32-NEXT: sw ra, 60(sp) # 4-byte Folded Spill ; CHECK-NEXT: vmv.s.x v0, a0
; RV32-NEXT: sw s0, 56(sp) # 4-byte Folded Spill ; CHECK-NEXT: vsetivli a0, 8, e32,m2,ta,mu
; RV32-NEXT: .cfi_offset ra, -4 ; CHECK-NEXT: vmerge.vvm v8, v10, v8, v0
; RV32-NEXT: .cfi_offset s0, -8 ; CHECK-NEXT: ret
; RV32-NEXT: addi s0, sp, 64
; RV32-NEXT: .cfi_def_cfa s0, 0
; RV32-NEXT: andi sp, sp, -32
; RV32-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v8, 7
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV32-NEXT: vfmv.v.f v25, ft0
; RV32-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v8, 6
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV32-NEXT: vfmv.s.f v25, ft0
; RV32-NEXT: addi a0, sp, 24
; RV32-NEXT: vse32.v v25, (a0)
; RV32-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v8, 5
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV32-NEXT: vfmv.v.f v25, ft0
; RV32-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v10, 4
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV32-NEXT: vfmv.s.f v25, ft0
; RV32-NEXT: addi a0, sp, 16
; RV32-NEXT: vse32.v v25, (a0)
; RV32-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v8, 3
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV32-NEXT: vfmv.v.f v25, ft0
; RV32-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v8, 2
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV32-NEXT: vfmv.s.f v25, ft0
; RV32-NEXT: addi a0, sp, 8
; RV32-NEXT: vse32.v v25, (a0)
; RV32-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v10, 1
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV32-NEXT: vfmv.v.f v25, ft0
; RV32-NEXT: vsetvli zero, zero, e32,m2,ta,mu
; RV32-NEXT: vfmv.f.s ft0, v10
; RV32-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV32-NEXT: vfmv.s.f v25, ft0
; RV32-NEXT: vse32.v v25, (sp)
; RV32-NEXT: vsetivli a0, 8, e32,m2,ta,mu
; RV32-NEXT: vle32.v v8, (sp)
; RV32-NEXT: addi sp, s0, -64
; RV32-NEXT: lw s0, 56(sp) # 4-byte Folded Reload
; RV32-NEXT: lw ra, 60(sp) # 4-byte Folded Reload
; RV32-NEXT: addi sp, sp, 64
; RV32-NEXT: ret
;
; RV64-LABEL: shuffle_v8f32:
; RV64: # %bb.0:
; RV64-NEXT: addi sp, sp, -64
; RV64-NEXT: .cfi_def_cfa_offset 64
; RV64-NEXT: sd ra, 56(sp) # 8-byte Folded Spill
; RV64-NEXT: sd s0, 48(sp) # 8-byte Folded Spill
; RV64-NEXT: .cfi_offset ra, -8
; RV64-NEXT: .cfi_offset s0, -16
; RV64-NEXT: addi s0, sp, 64
; RV64-NEXT: .cfi_def_cfa s0, 0
; RV64-NEXT: andi sp, sp, -32
; RV64-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v8, 7
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV64-NEXT: vfmv.v.f v25, ft0
; RV64-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v8, 6
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV64-NEXT: vfmv.s.f v25, ft0
; RV64-NEXT: addi a0, sp, 24
; RV64-NEXT: vse32.v v25, (a0)
; RV64-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v8, 5
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV64-NEXT: vfmv.v.f v25, ft0
; RV64-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v10, 4
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV64-NEXT: vfmv.s.f v25, ft0
; RV64-NEXT: addi a0, sp, 16
; RV64-NEXT: vse32.v v25, (a0)
; RV64-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v8, 3
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV64-NEXT: vfmv.v.f v25, ft0
; RV64-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v8, 2
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV64-NEXT: vfmv.s.f v25, ft0
; RV64-NEXT: addi a0, sp, 8
; RV64-NEXT: vse32.v v25, (a0)
; RV64-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v10, 1
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV64-NEXT: vfmv.v.f v25, ft0
; RV64-NEXT: vsetvli zero, zero, e32,m2,ta,mu
; RV64-NEXT: vfmv.f.s ft0, v10
; RV64-NEXT: vsetivli a0, 2, e32,m1,ta,mu
; RV64-NEXT: vfmv.s.f v25, ft0
; RV64-NEXT: vse32.v v25, (sp)
; RV64-NEXT: vsetivli a0, 8, e32,m2,ta,mu
; RV64-NEXT: vle32.v v8, (sp)
; RV64-NEXT: addi sp, s0, -64
; RV64-NEXT: ld s0, 48(sp) # 8-byte Folded Reload
; RV64-NEXT: ld ra, 56(sp) # 8-byte Folded Reload
; RV64-NEXT: addi sp, sp, 64
; RV64-NEXT: ret
%s = shufflevector <8 x float> %x, <8 x float> %y, <8 x i32> <i32 8, i32 9, i32 2, i32 3, i32 12, i32 5, i32 6, i32 7> %s = shufflevector <8 x float> %x, <8 x float> %y, <8 x i32> <i32 8, i32 9, i32 2, i32 3, i32 12, i32 5, i32 6, i32 7>
ret <8 x float> %s ret <8 x float> %s
} }
define <4 x double> @shuffle_fv_v4i16(<4 x double> %x) { define <4 x double> @shuffle_fv_v4i16(<4 x double> %x) {
; RV32-LABEL: shuffle_fv_v4i16: ; CHECK-LABEL: shuffle_fv_v4i16:
; RV32: # %bb.0: ; CHECK: # %bb.0:
; RV32-NEXT: addi sp, sp, -64 ; CHECK-NEXT: addi a0, zero, 9
; RV32-NEXT: .cfi_def_cfa_offset 64 ; CHECK-NEXT: lui a1, %hi(.LCPI2_0)
; RV32-NEXT: sw ra, 60(sp) # 4-byte Folded Spill ; CHECK-NEXT: fld ft0, %lo(.LCPI2_0)(a1)
; RV32-NEXT: sw s0, 56(sp) # 4-byte Folded Spill ; CHECK-NEXT: vsetivli a1, 1, e8,m1,ta,mu
; RV32-NEXT: .cfi_offset ra, -4 ; CHECK-NEXT: vmv.s.x v0, a0
; RV32-NEXT: .cfi_offset s0, -8 ; CHECK-NEXT: vsetivli a0, 4, e64,m2,ta,mu
; RV32-NEXT: addi s0, sp, 64 ; CHECK-NEXT: vfmerge.vfm v8, v8, ft0, v0
; RV32-NEXT: .cfi_def_cfa s0, 0 ; CHECK-NEXT: ret
; RV32-NEXT: andi sp, sp, -32
; RV32-NEXT: lui a0, %hi(.LCPI2_0)
; RV32-NEXT: fld ft0, %lo(.LCPI2_0)(a0)
; RV32-NEXT: fsd ft0, 24(sp)
; RV32-NEXT: fsd ft0, 0(sp)
; RV32-NEXT: vsetivli a0, 1, e64,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v8, 2
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: fsd ft0, 16(sp)
; RV32-NEXT: vslidedown.vi v26, v8, 1
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: fsd ft0, 8(sp)
; RV32-NEXT: vsetivli a0, 4, e64,m2,ta,mu
; RV32-NEXT: vle64.v v8, (sp)
; RV32-NEXT: addi sp, s0, -64
; RV32-NEXT: lw s0, 56(sp) # 4-byte Folded Reload
; RV32-NEXT: lw ra, 60(sp) # 4-byte Folded Reload
; RV32-NEXT: addi sp, sp, 64
; RV32-NEXT: ret
;
; RV64-LABEL: shuffle_fv_v4i16:
; RV64: # %bb.0:
; RV64-NEXT: addi sp, sp, -64
; RV64-NEXT: .cfi_def_cfa_offset 64
; RV64-NEXT: sd ra, 56(sp) # 8-byte Folded Spill
; RV64-NEXT: sd s0, 48(sp) # 8-byte Folded Spill
; RV64-NEXT: .cfi_offset ra, -8
; RV64-NEXT: .cfi_offset s0, -16
; RV64-NEXT: addi s0, sp, 64
; RV64-NEXT: .cfi_def_cfa s0, 0
; RV64-NEXT: andi sp, sp, -32
; RV64-NEXT: lui a0, %hi(.LCPI2_0)
; RV64-NEXT: fld ft0, %lo(.LCPI2_0)(a0)
; RV64-NEXT: fsd ft0, 24(sp)
; RV64-NEXT: fsd ft0, 0(sp)
; RV64-NEXT: vsetivli a0, 1, e64,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v8, 2
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: fsd ft0, 16(sp)
; RV64-NEXT: vslidedown.vi v26, v8, 1
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: fsd ft0, 8(sp)
; RV64-NEXT: vsetivli a0, 4, e64,m2,ta,mu
; RV64-NEXT: vle64.v v8, (sp)
; RV64-NEXT: addi sp, s0, -64
; RV64-NEXT: ld s0, 48(sp) # 8-byte Folded Reload
; RV64-NEXT: ld ra, 56(sp) # 8-byte Folded Reload
; RV64-NEXT: addi sp, sp, 64
; RV64-NEXT: ret
%s = shufflevector <4 x double> <double 2.0, double 2.0, double 2.0, double 2.0>, <4 x double> %x, <4 x i32> <i32 0, i32 5, i32 6, i32 3> %s = shufflevector <4 x double> <double 2.0, double 2.0, double 2.0, double 2.0>, <4 x double> %x, <4 x i32> <i32 0, i32 5, i32 6, i32 3>
ret <4 x double> %s ret <4 x double> %s
} }
define <4 x double> @shuffle_vf_v4i16(<4 x double> %x) { define <4 x double> @shuffle_vf_v4i16(<4 x double> %x) {
; RV32-LABEL: shuffle_vf_v4i16: ; CHECK-LABEL: shuffle_vf_v4i16:
; RV32: # %bb.0: ; CHECK: # %bb.0:
; RV32-NEXT: addi sp, sp, -64 ; CHECK-NEXT: addi a0, zero, 6
; RV32-NEXT: .cfi_def_cfa_offset 64 ; CHECK-NEXT: lui a1, %hi(.LCPI3_0)
; RV32-NEXT: sw ra, 60(sp) # 4-byte Folded Spill ; CHECK-NEXT: fld ft0, %lo(.LCPI3_0)(a1)
; RV32-NEXT: sw s0, 56(sp) # 4-byte Folded Spill ; CHECK-NEXT: vsetivli a1, 1, e8,m1,ta,mu
; RV32-NEXT: .cfi_offset ra, -4 ; CHECK-NEXT: vmv.s.x v0, a0
; RV32-NEXT: .cfi_offset s0, -8 ; CHECK-NEXT: vsetivli a0, 4, e64,m2,ta,mu
; RV32-NEXT: addi s0, sp, 64 ; CHECK-NEXT: vfmerge.vfm v8, v8, ft0, v0
; RV32-NEXT: .cfi_def_cfa s0, 0 ; CHECK-NEXT: ret
; RV32-NEXT: andi sp, sp, -32
; RV32-NEXT: lui a0, %hi(.LCPI3_0)
; RV32-NEXT: fld ft0, %lo(.LCPI3_0)(a0)
; RV32-NEXT: fsd ft0, 16(sp)
; RV32-NEXT: fsd ft0, 8(sp)
; RV32-NEXT: vsetvli zero, zero, e64,m2,ta,mu
; RV32-NEXT: vfmv.f.s ft0, v8
; RV32-NEXT: fsd ft0, 0(sp)
; RV32-NEXT: vsetivli a0, 1, e64,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v8, 3
; RV32-NEXT: vfmv.f.s ft0, v26
; RV32-NEXT: fsd ft0, 24(sp)
; RV32-NEXT: vsetivli a0, 4, e64,m2,ta,mu
; RV32-NEXT: vle64.v v8, (sp)
; RV32-NEXT: addi sp, s0, -64
; RV32-NEXT: lw s0, 56(sp) # 4-byte Folded Reload
; RV32-NEXT: lw ra, 60(sp) # 4-byte Folded Reload
; RV32-NEXT: addi sp, sp, 64
; RV32-NEXT: ret
;
; RV64-LABEL: shuffle_vf_v4i16:
; RV64: # %bb.0:
; RV64-NEXT: addi sp, sp, -64
; RV64-NEXT: .cfi_def_cfa_offset 64
; RV64-NEXT: sd ra, 56(sp) # 8-byte Folded Spill
; RV64-NEXT: sd s0, 48(sp) # 8-byte Folded Spill
; RV64-NEXT: .cfi_offset ra, -8
; RV64-NEXT: .cfi_offset s0, -16
; RV64-NEXT: addi s0, sp, 64
; RV64-NEXT: .cfi_def_cfa s0, 0
; RV64-NEXT: andi sp, sp, -32
; RV64-NEXT: lui a0, %hi(.LCPI3_0)
; RV64-NEXT: fld ft0, %lo(.LCPI3_0)(a0)
; RV64-NEXT: fsd ft0, 16(sp)
; RV64-NEXT: fsd ft0, 8(sp)
; RV64-NEXT: vsetvli zero, zero, e64,m2,ta,mu
; RV64-NEXT: vfmv.f.s ft0, v8
; RV64-NEXT: fsd ft0, 0(sp)
; RV64-NEXT: vsetivli a0, 1, e64,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v8, 3
; RV64-NEXT: vfmv.f.s ft0, v26
; RV64-NEXT: fsd ft0, 24(sp)
; RV64-NEXT: vsetivli a0, 4, e64,m2,ta,mu
; RV64-NEXT: vle64.v v8, (sp)
; RV64-NEXT: addi sp, s0, -64
; RV64-NEXT: ld s0, 48(sp) # 8-byte Folded Reload
; RV64-NEXT: ld ra, 56(sp) # 8-byte Folded Reload
; RV64-NEXT: addi sp, sp, 64
; RV64-NEXT: ret
%s = shufflevector <4 x double> %x, <4 x double> <double 2.0, double 2.0, double 2.0, double 2.0>, <4 x i32> <i32 0, i32 5, i32 6, i32 3> %s = shufflevector <4 x double> %x, <4 x double> <double 2.0, double 2.0, double 2.0, double 2.0>, <4 x i32> <i32 0, i32 5, i32 6, i32 3>
ret <4 x double> %s ret <4 x double> %s
} }

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=riscv32 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32 ; RUN: llc -mtriple=riscv32 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefix=CHECK
; RUN: llc -mtriple=riscv64 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64 ; RUN: llc -mtriple=riscv64 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefix=CHECK
define <4 x i16> @shuffle_v4i16(<4 x i16> %x, <4 x i16> %y) { define <4 x i16> @shuffle_v4i16(<4 x i16> %x, <4 x i16> %y) {
; CHECK-LABEL: shuffle_v4i16: ; CHECK-LABEL: shuffle_v4i16:
; CHECK: # %bb.0: ; CHECK: # %bb.0:
; CHECK-NEXT: addi sp, sp, -16 ; CHECK-NEXT: addi a0, zero, 11
; CHECK-NEXT: .cfi_def_cfa_offset 16 ; CHECK-NEXT: vsetivli a1, 1, e8,m1,ta,mu
; CHECK-NEXT: vsetvli zero, zero, e16,m1,ta,mu ; CHECK-NEXT: vmv.s.x v0, a0
; CHECK-NEXT: vmv.x.s a0, v8
; CHECK-NEXT: sh a0, 8(sp)
; CHECK-NEXT: vsetivli a0, 1, e16,m1,ta,mu
; CHECK-NEXT: vslidedown.vi v25, v8, 3
; CHECK-NEXT: vmv.x.s a0, v25
; CHECK-NEXT: sh a0, 14(sp)
; CHECK-NEXT: vslidedown.vi v25, v9, 2
; CHECK-NEXT: vmv.x.s a0, v25
; CHECK-NEXT: sh a0, 12(sp)
; CHECK-NEXT: vslidedown.vi v25, v8, 1
; CHECK-NEXT: vmv.x.s a0, v25
; CHECK-NEXT: sh a0, 10(sp)
; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu ; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu
; CHECK-NEXT: addi a0, sp, 8 ; CHECK-NEXT: vmerge.vvm v8, v9, v8, v0
; CHECK-NEXT: vle16.v v8, (a0)
; CHECK-NEXT: addi sp, sp, 16
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%s = shufflevector <4 x i16> %x, <4 x i16> %y, <4 x i32> <i32 0, i32 1, i32 6, i32 3> %s = shufflevector <4 x i16> %x, <4 x i16> %y, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
ret <4 x i16> %s ret <4 x i16> %s
} }
define <8 x i32> @shuffle_v8i32(<8 x i32> %x, <8 x i32> %y) { define <8 x i32> @shuffle_v8i32(<8 x i32> %x, <8 x i32> %y) {
; RV32-LABEL: shuffle_v8i32: ; CHECK-LABEL: shuffle_v8i32:
; RV32: # %bb.0: ; CHECK: # %bb.0:
; RV32-NEXT: addi sp, sp, -64 ; CHECK-NEXT: addi a0, zero, 203
; RV32-NEXT: .cfi_def_cfa_offset 64 ; CHECK-NEXT: vsetivli a1, 1, e8,m1,ta,mu
; RV32-NEXT: sw ra, 60(sp) # 4-byte Folded Spill ; CHECK-NEXT: vmv.s.x v0, a0
; RV32-NEXT: sw s0, 56(sp) # 4-byte Folded Spill ; CHECK-NEXT: vsetivli a0, 8, e32,m2,ta,mu
; RV32-NEXT: .cfi_offset ra, -4 ; CHECK-NEXT: vmerge.vvm v8, v10, v8, v0
; RV32-NEXT: .cfi_offset s0, -8 ; CHECK-NEXT: ret
; RV32-NEXT: addi s0, sp, 64
; RV32-NEXT: .cfi_def_cfa s0, 0
; RV32-NEXT: andi sp, sp, -32
; RV32-NEXT: vsetvli zero, zero, e32,m2,ta,mu
; RV32-NEXT: vmv.x.s a0, v8
; RV32-NEXT: sw a0, 0(sp)
; RV32-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV32-NEXT: vslidedown.vi v26, v8, 7
; RV32-NEXT: vmv.x.s a0, v26
; RV32-NEXT: sw a0, 28(sp)
; RV32-NEXT: vslidedown.vi v26, v8, 6
; RV32-NEXT: vmv.x.s a0, v26
; RV32-NEXT: sw a0, 24(sp)
; RV32-NEXT: vslidedown.vi v26, v10, 5
; RV32-NEXT: vmv.x.s a0, v26
; RV32-NEXT: sw a0, 20(sp)
; RV32-NEXT: vslidedown.vi v26, v10, 4
; RV32-NEXT: vmv.x.s a0, v26
; RV32-NEXT: sw a0, 16(sp)
; RV32-NEXT: vslidedown.vi v26, v8, 3
; RV32-NEXT: vmv.x.s a0, v26
; RV32-NEXT: sw a0, 12(sp)
; RV32-NEXT: vslidedown.vi v26, v10, 2
; RV32-NEXT: vmv.x.s a0, v26
; RV32-NEXT: sw a0, 8(sp)
; RV32-NEXT: vslidedown.vi v26, v8, 1
; RV32-NEXT: vmv.x.s a0, v26
; RV32-NEXT: sw a0, 4(sp)
; RV32-NEXT: vsetivli a0, 8, e32,m2,ta,mu
; RV32-NEXT: vle32.v v8, (sp)
; RV32-NEXT: addi sp, s0, -64
; RV32-NEXT: lw s0, 56(sp) # 4-byte Folded Reload
; RV32-NEXT: lw ra, 60(sp) # 4-byte Folded Reload
; RV32-NEXT: addi sp, sp, 64
; RV32-NEXT: ret
;
; RV64-LABEL: shuffle_v8i32:
; RV64: # %bb.0:
; RV64-NEXT: addi sp, sp, -64
; RV64-NEXT: .cfi_def_cfa_offset 64
; RV64-NEXT: sd ra, 56(sp) # 8-byte Folded Spill
; RV64-NEXT: sd s0, 48(sp) # 8-byte Folded Spill
; RV64-NEXT: .cfi_offset ra, -8
; RV64-NEXT: .cfi_offset s0, -16
; RV64-NEXT: addi s0, sp, 64
; RV64-NEXT: .cfi_def_cfa s0, 0
; RV64-NEXT: andi sp, sp, -32
; RV64-NEXT: vsetvli zero, zero, e32,m2,ta,mu
; RV64-NEXT: vmv.x.s a0, v8
; RV64-NEXT: sw a0, 0(sp)
; RV64-NEXT: vsetivli a0, 1, e32,m2,ta,mu
; RV64-NEXT: vslidedown.vi v26, v8, 7
; RV64-NEXT: vmv.x.s a0, v26
; RV64-NEXT: sw a0, 28(sp)
; RV64-NEXT: vslidedown.vi v26, v8, 6
; RV64-NEXT: vmv.x.s a0, v26
; RV64-NEXT: sw a0, 24(sp)
; RV64-NEXT: vslidedown.vi v26, v10, 5
; RV64-NEXT: vmv.x.s a0, v26
; RV64-NEXT: sw a0, 20(sp)
; RV64-NEXT: vslidedown.vi v26, v10, 4
; RV64-NEXT: vmv.x.s a0, v26
; RV64-NEXT: sw a0, 16(sp)
; RV64-NEXT: vslidedown.vi v26, v8, 3
; RV64-NEXT: vmv.x.s a0, v26
; RV64-NEXT: sw a0, 12(sp)
; RV64-NEXT: vslidedown.vi v26, v10, 2
; RV64-NEXT: vmv.x.s a0, v26
; RV64-NEXT: sw a0, 8(sp)
; RV64-NEXT: vslidedown.vi v26, v8, 1
; RV64-NEXT: vmv.x.s a0, v26
; RV64-NEXT: sw a0, 4(sp)
; RV64-NEXT: vsetivli a0, 8, e32,m2,ta,mu
; RV64-NEXT: vle32.v v8, (sp)
; RV64-NEXT: addi sp, s0, -64
; RV64-NEXT: ld s0, 48(sp) # 8-byte Folded Reload
; RV64-NEXT: ld ra, 56(sp) # 8-byte Folded Reload
; RV64-NEXT: addi sp, sp, 64
; RV64-NEXT: ret
%s = shufflevector <8 x i32> %x, <8 x i32> %y, <8 x i32> <i32 0, i32 1, i32 10, i32 3, i32 12, i32 13, i32 6, i32 7> %s = shufflevector <8 x i32> %x, <8 x i32> %y, <8 x i32> <i32 0, i32 1, i32 10, i32 3, i32 12, i32 13, i32 6, i32 7>
ret <8 x i32> %s ret <8 x i32> %s
} }
define <4 x i16> @shuffle_xv_v4i16(<4 x i16> %x) { define <4 x i16> @shuffle_xv_v4i16(<4 x i16> %x) {
; CHECK-LABEL: shuffle_xv_v4i16: ; CHECK-LABEL: shuffle_xv_v4i16:
; CHECK: # %bb.0: ; CHECK: # %bb.0:
; CHECK-NEXT: addi sp, sp, -16 ; CHECK-NEXT: addi a0, zero, 9
; CHECK-NEXT: .cfi_def_cfa_offset 16 ; CHECK-NEXT: vsetivli a1, 1, e8,m1,ta,mu
; CHECK-NEXT: addi a0, zero, 5 ; CHECK-NEXT: vmv.s.x v0, a0
; CHECK-NEXT: sh a0, 14(sp)
; CHECK-NEXT: sh a0, 8(sp)
; CHECK-NEXT: vsetivli a0, 1, e16,m1,ta,mu
; CHECK-NEXT: vslidedown.vi v25, v8, 2
; CHECK-NEXT: vmv.x.s a0, v25
; CHECK-NEXT: sh a0, 12(sp)
; CHECK-NEXT: vslidedown.vi v25, v8, 1
; CHECK-NEXT: vmv.x.s a0, v25
; CHECK-NEXT: sh a0, 10(sp)
; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu ; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu
; CHECK-NEXT: addi a0, sp, 8 ; CHECK-NEXT: vmerge.vim v8, v8, 5, v0
; CHECK-NEXT: vle16.v v8, (a0)
; CHECK-NEXT: addi sp, sp, 16
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%s = shufflevector <4 x i16> <i16 5, i16 5, i16 5, i16 5>, <4 x i16> %x, <4 x i32> <i32 0, i32 5, i32 6, i32 3> %s = shufflevector <4 x i16> <i16 5, i16 5, i16 5, i16 5>, <4 x i16> %x, <4 x i32> <i32 0, i32 5, i32 6, i32 3>
ret <4 x i16> %s ret <4 x i16> %s
} }
define <4 x i16> @shuffle_vx_v4i16(<4 x i16> %x) { define <4 x i16> @shuffle_vx_v4i16(<4 x i16> %x) {
; CHECK-LABEL: shuffle_vx_v4i16: ; CHECK-LABEL: shuffle_vx_v4i16:
; CHECK: # %bb.0: ; CHECK: # %bb.0:
; CHECK-NEXT: addi sp, sp, -16 ; CHECK-NEXT: addi a0, zero, 6
; CHECK-NEXT: .cfi_def_cfa_offset 16 ; CHECK-NEXT: vsetivli a1, 1, e8,m1,ta,mu
; CHECK-NEXT: addi a0, zero, 5 ; CHECK-NEXT: vmv.s.x v0, a0
; CHECK-NEXT: sh a0, 12(sp)
; CHECK-NEXT: sh a0, 10(sp)
; CHECK-NEXT: vsetvli zero, zero, e16,m1,ta,mu
; CHECK-NEXT: vmv.x.s a0, v8
; CHECK-NEXT: sh a0, 8(sp)
; CHECK-NEXT: vsetivli a0, 1, e16,m1,ta,mu
; CHECK-NEXT: vslidedown.vi v25, v8, 3
; CHECK-NEXT: vmv.x.s a0, v25
; CHECK-NEXT: sh a0, 14(sp)
; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu ; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu
; CHECK-NEXT: addi a0, sp, 8 ; CHECK-NEXT: vmerge.vim v8, v8, 5, v0
; CHECK-NEXT: vle16.v v8, (a0)
; CHECK-NEXT: addi sp, sp, 16
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%s = shufflevector <4 x i16> %x, <4 x i16> <i16 5, i16 5, i16 5, i16 5>, <4 x i32> <i32 0, i32 5, i32 6, i32 3> %s = shufflevector <4 x i16> %x, <4 x i16> <i16 5, i16 5, i16 5, i16 5>, <4 x i32> <i32 0, i32 5, i32 6, i32 3>
ret <4 x i16> %s ret <4 x i16> %s
} }