[llvm] [IA][RISCV] Support VP loads/stores in InterleavedAccessPass (PR #120490)
Min-Yih Hsu via llvm-commits
llvm-commits at lists.llvm.org
Mon Jan 27 13:45:59 PST 2025
https://github.com/mshockwave updated https://github.com/llvm/llvm-project/pull/120490
>From 156661a47bfaeb5dcd0588683c34ec91eeb8402a Mon Sep 17 00:00:00 2001
From: Min-Yih Hsu <min.hsu at sifive.com>
Date: Tue, 21 Jan 2025 19:43:15 -0800
Subject: [PATCH 1/3] [IA][RISCV] Support VP intrinsics in
InterleavedAccessPass
Teach InterleavedAccessPass to recognize the following patterns:
- vp.store an interleaved scalable vector
- Deinterleaving a scalable vector loaded from vp.load
Upon recognizing these patterns, IA will collect the interleaved /
deinterleaved operands and delegate them over to their respective
newly-added TLI hooks.
For RISC-V, these patterns are lowered into segmented loads/stores
(except when we're interleaving constant splats, in which case a
unit-strde store will be generated)
Right now we only recognized power-of-two (de)interleave cases, in which
(de)interleave4/8 are synthesized from a tree of (de)interleave2.
Co-authored-by: Nikolay Panchenko <nicholas.panchenko at gmail.com>
---
llvm/include/llvm/CodeGen/TargetLowering.h | 29 +
llvm/lib/CodeGen/InterleavedAccessPass.cpp | 110 +++-
llvm/lib/Target/RISCV/RISCVISelLowering.cpp | 257 ++++++++
llvm/lib/Target/RISCV/RISCVISelLowering.h | 9 +
.../scalable-vectors-interleaved-access.ll | 619 ++++++++++++++++++
5 files changed, 1005 insertions(+), 19 deletions(-)
create mode 100644 llvm/test/CodeGen/RISCV/rvv/scalable-vectors-interleaved-access.ll
diff --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index 59743dbe4d2ea4..50fa3a39c0a0c4 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -94,6 +94,7 @@ class TargetRegisterClass;
class TargetRegisterInfo;
class TargetTransformInfo;
class Value;
+class VPIntrinsic;
namespace Sched {
@@ -3152,6 +3153,34 @@ class TargetLoweringBase {
return false;
}
+ /// Lower an interleaved load to target specific intrinsics. Return
+ /// true on success.
+ ///
+ /// \p Load is a vp.load instruction.
+ /// \p Mask is a mask value
+ /// \p DeinterleaveIntrin is vector.deinterleave intrinsic
+ /// \p DeinterleaveRes is a list of deinterleaved results.
+ virtual bool
+ lowerInterleavedScalableLoad(VPIntrinsic *Load, Value *Mask,
+ IntrinsicInst *DeinterleaveIntrin,
+ ArrayRef<Value *> DeinterleaveRes) const {
+ return false;
+ }
+
+ /// Lower an interleaved store to target specific intrinsics. Return
+ /// true on success.
+ ///
+ /// \p Store is the vp.store instruction.
+ /// \p Mask is a mask value
+ /// \p InterleaveIntrin is vector.interleave intrinsic
+ /// \p InterleaveOps is a list of values being interleaved.
+ virtual bool
+ lowerInterleavedScalableStore(VPIntrinsic *Store, Value *Mask,
+ IntrinsicInst *InterleaveIntrin,
+ ArrayRef<Value *> InterleaveOps) const {
+ return false;
+ }
+
/// Lower a deinterleave intrinsic to a target specific load intrinsic.
/// Return true on success. Currently only supports
/// llvm.vector.deinterleave2
diff --git a/llvm/lib/CodeGen/InterleavedAccessPass.cpp b/llvm/lib/CodeGen/InterleavedAccessPass.cpp
index 3f6a69ecb7d729..9b15d0351ebe18 100644
--- a/llvm/lib/CodeGen/InterleavedAccessPass.cpp
+++ b/llvm/lib/CodeGen/InterleavedAccessPass.cpp
@@ -630,11 +630,34 @@ getVectorDeinterleaveFactor(IntrinsicInst *II,
return true;
}
+/// Check the interleaved mask
+///
+/// - if a value within the optional is non-nullptr, the value corresponds to
+/// deinterleaved mask
+/// - if a value within the option is nullptr, the value corresponds to all-true
+/// mask
+/// - return nullopt if mask cannot be deinterleaved
+static std::optional<Value *> getMask(Value *WideMask, unsigned Factor) {
+ using namespace llvm::PatternMatch;
+ if (auto *IMI = dyn_cast<IntrinsicInst>(WideMask)) {
+ SmallVector<Value *, 8> Operands;
+ SmallVector<Instruction *, 8> DeadInsts;
+ if (getVectorInterleaveFactor(IMI, Operands, DeadInsts)) {
+ assert(!Operands.empty());
+ if (Operands.size() == Factor &&
+ std::equal(Operands.begin(), Operands.end(), Operands.begin()))
+ return Operands.front();
+ }
+ }
+ if (match(WideMask, m_AllOnes()))
+ return nullptr;
+ return std::nullopt;
+}
+
bool InterleavedAccessImpl::lowerDeinterleaveIntrinsic(
IntrinsicInst *DI, SmallSetVector<Instruction *, 32> &DeadInsts) {
- LoadInst *LI = dyn_cast<LoadInst>(DI->getOperand(0));
-
- if (!LI || !LI->hasOneUse() || !LI->isSimple())
+ Value *LoadedVal = DI->getOperand(0);
+ if (!LoadedVal->hasOneUse() || !isa<LoadInst, VPIntrinsic>(LoadedVal))
return false;
SmallVector<Value *, 8> DeinterleaveValues;
@@ -643,16 +666,42 @@ bool InterleavedAccessImpl::lowerDeinterleaveIntrinsic(
DeinterleaveDeadInsts))
return false;
- LLVM_DEBUG(dbgs() << "IA: Found a deinterleave intrinsic: " << *DI
- << " with factor = " << DeinterleaveValues.size() << "\n");
+ const unsigned Factor = DeinterleaveValues.size();
- // Try and match this with target specific intrinsics.
- if (!TLI->lowerDeinterleaveIntrinsicToLoad(LI, DeinterleaveValues))
- return false;
+ if (auto *VPLoad = dyn_cast<VPIntrinsic>(LoadedVal)) {
+ if (VPLoad->getIntrinsicID() != Intrinsic::vp_load)
+ return false;
+ // Check mask operand. Handle both all-true and interleaved mask.
+ Value *WideMask = VPLoad->getOperand(1);
+ std::optional<Value *> Mask = getMask(WideMask, Factor);
+ if (!Mask)
+ return false;
+
+ LLVM_DEBUG(dbgs() << "IA: Found a vp.load with deinterleave intrinsic "
+ << *DI << " and factor = " << Factor << "\n");
+
+ // Since lowerInterleaveLoad expects Shuffles and LoadInst, use special
+ // TLI function to emit target-specific interleaved instruction.
+ if (!TLI->lowerInterleavedScalableLoad(VPLoad, *Mask, DI,
+ DeinterleaveValues))
+ return false;
+
+ } else {
+ auto *LI = cast<LoadInst>(LoadedVal);
+ if (!LI->isSimple())
+ return false;
+
+ LLVM_DEBUG(dbgs() << "IA: Found a load with deinterleave intrinsic " << *DI
+ << " and factor = " << Factor << "\n");
+
+ // Try and match this with target specific intrinsics.
+ if (!TLI->lowerDeinterleaveIntrinsicToLoad(LI, DeinterleaveValues))
+ return false;
+ }
DeadInsts.insert(DeinterleaveDeadInsts.begin(), DeinterleaveDeadInsts.end());
// We now have a target-specific load, so delete the old one.
- DeadInsts.insert(LI);
+ DeadInsts.insert(cast<Instruction>(LoadedVal));
return true;
}
@@ -660,10 +709,8 @@ bool InterleavedAccessImpl::lowerInterleaveIntrinsic(
IntrinsicInst *II, SmallSetVector<Instruction *, 32> &DeadInsts) {
if (!II->hasOneUse())
return false;
-
- StoreInst *SI = dyn_cast<StoreInst>(*(II->users().begin()));
-
- if (!SI || !SI->isSimple())
+ Value *StoredBy = II->user_back();
+ if (!isa<StoreInst, VPIntrinsic>(StoredBy))
return false;
SmallVector<Value *, 8> InterleaveValues;
@@ -671,15 +718,40 @@ bool InterleavedAccessImpl::lowerInterleaveIntrinsic(
if (!getVectorInterleaveFactor(II, InterleaveValues, InterleaveDeadInsts))
return false;
- LLVM_DEBUG(dbgs() << "IA: Found an interleave intrinsic: " << *II
- << " with factor = " << InterleaveValues.size() << "\n");
+ const unsigned Factor = InterleaveValues.size();
- // Try and match this with target specific intrinsics.
- if (!TLI->lowerInterleaveIntrinsicToStore(SI, InterleaveValues))
- return false;
+ if (auto *VPStore = dyn_cast<VPIntrinsic>(StoredBy)) {
+ if (VPStore->getIntrinsicID() != Intrinsic::vp_store)
+ return false;
+
+ Value *WideMask = VPStore->getOperand(2);
+ std::optional<Value *> Mask = getMask(WideMask, Factor);
+ if (!Mask)
+ return false;
+
+ LLVM_DEBUG(dbgs() << "IA: Found a vp.store with interleave intrinsic "
+ << *II << " and factor = " << Factor << "\n");
+
+ // Since lowerInterleavedStore expects Shuffle and StoreInst, use special
+ // TLI function to emit target-specific interleaved instruction.
+ if (!TLI->lowerInterleavedScalableStore(VPStore, *Mask, II,
+ InterleaveValues))
+ return false;
+ } else {
+ auto *SI = cast<StoreInst>(StoredBy);
+ if (!SI->isSimple())
+ return false;
+
+ LLVM_DEBUG(dbgs() << "IA: Found a store with interleave intrinsic " << *II
+ << " and factor = " << Factor << "\n");
+
+ // Try and match this with target specific intrinsics.
+ if (!TLI->lowerInterleaveIntrinsicToStore(SI, InterleaveValues))
+ return false;
+ }
// We now have a target-specific store, so delete the old one.
- DeadInsts.insert(SI);
+ DeadInsts.insert(cast<Instruction>(StoredBy));
DeadInsts.insert(InterleaveDeadInsts.begin(), InterleaveDeadInsts.end());
return true;
}
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 295fd315c56daf..b5155106a506bc 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -22529,6 +22529,263 @@ bool RISCVTargetLowering::lowerInterleaveIntrinsicToStore(
return true;
}
+/// Lower an interleaved vp.load into a vlsegN intrinsic.
+///
+/// E.g. Lower an interleaved vp.load (Factor = 2):
+/// %l = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr %ptr,
+/// %mask,
+/// i32 %wide.rvl)
+/// %dl = tail call { <vscale x 32 x i8>, <vscale x 32 x i8> }
+/// @llvm.vector.deinterleave2.nxv64i8(
+/// <vscale x 64 x i8> %l)
+/// %r0 = extractvalue { <vscale x 32 x i8>, <vscale x 32 x i8> } %dl, 0
+/// %r1 = extractvalue { <vscale x 32 x i8>, <vscale x 32 x i8> } %dl, 1
+///
+/// Into:
+/// %rvl = udiv %wide.rvl, 2
+/// %sl = call { <vscale x 32 x i8>, <vscale x 32 x i8> }
+/// @llvm.riscv.vlseg2.mask.nxv32i8.i64(<vscale x 32 x i8> undef,
+/// <vscale x 32 x i8> undef,
+/// ptr %ptr,
+/// %mask,
+/// i64 %rvl,
+/// i64 1)
+/// %r0 = extractvalue { <vscale x 32 x i8>, <vscale x 32 x i8> } %sl, 0
+/// %r1 = extractvalue { <vscale x 32 x i8>, <vscale x 32 x i8> } %sl, 1
+///
+/// NOTE: the deinterleave2 intrinsic won't be touched and is expected to be
+/// removed by the caller
+bool RISCVTargetLowering::lowerInterleavedScalableLoad(
+ VPIntrinsic *Load, Value *Mask, IntrinsicInst *DeinterleaveIntrin,
+ ArrayRef<Value *> DeInterleaveResults) const {
+ assert(Load->getIntrinsicID() == Intrinsic::vp_load &&
+ "Unexpected intrinsic");
+
+ const unsigned Factor = DeInterleaveResults.size();
+
+ auto *WideVTy = cast<VectorType>(Load->getType());
+ unsigned WideNumElements = WideVTy->getElementCount().getKnownMinValue();
+ assert(WideNumElements % Factor == 0 &&
+ "ElementCount of a wide load must be divisible by interleave factor");
+ auto *VTy =
+ VectorType::get(WideVTy->getScalarType(), WideNumElements / Factor,
+ WideVTy->isScalableTy());
+ // FIXME: Should pass alignment attribute from pointer, but vectorizer needs
+ // to emit it first.
+ auto &DL = Load->getModule()->getDataLayout();
+ Align Alignment = Align(DL.getTypeStoreSize(WideVTy->getScalarType()));
+ if (!isLegalInterleavedAccessType(
+ VTy, Factor, Alignment,
+ Load->getArgOperand(0)->getType()->getPointerAddressSpace(), DL))
+ return false;
+
+ IRBuilder<> Builder(Load);
+ Value *WideEVL = Load->getArgOperand(2);
+ auto *XLenTy = Type::getIntNTy(Load->getContext(), Subtarget.getXLen());
+ Value *EVL = Builder.CreateZExtOrTrunc(
+ Builder.CreateUDiv(WideEVL, ConstantInt::get(WideEVL->getType(), Factor)),
+ XLenTy);
+
+ static const Intrinsic::ID IntrMaskIds[] = {
+ Intrinsic::riscv_vlseg2_mask, Intrinsic::riscv_vlseg3_mask,
+ Intrinsic::riscv_vlseg4_mask, Intrinsic::riscv_vlseg5_mask,
+ Intrinsic::riscv_vlseg6_mask, Intrinsic::riscv_vlseg7_mask,
+ Intrinsic::riscv_vlseg8_mask,
+ };
+
+ unsigned SEW = DL.getTypeSizeInBits(VTy->getElementType());
+ unsigned NumElts = VTy->getElementCount().getKnownMinValue();
+ Type *VecTupTy = TargetExtType::get(
+ Load->getContext(), "riscv.vector.tuple",
+ ScalableVectorType::get(Type::getInt8Ty(Load->getContext()),
+ NumElts * SEW / 8),
+ Factor);
+
+ Value *PoisonVal = PoisonValue::get(VecTupTy);
+ SmallVector<Value *> Operands;
+ Operands.append({PoisonVal, Load->getArgOperand(0)});
+
+ if (!Mask)
+ Mask = ConstantVector::getSplat(VTy->getElementCount(),
+ ConstantInt::getTrue(Load->getContext()));
+
+ Function *VlsegNFunc = Intrinsic::getOrInsertDeclaration(
+ Load->getModule(), IntrMaskIds[Factor - 2],
+ {VecTupTy, Mask->getType(), EVL->getType()});
+
+ Operands.push_back(Mask);
+
+ Operands.push_back(EVL);
+
+ // Tail-policy
+ Operands.push_back(ConstantInt::get(XLenTy, RISCVII::TAIL_AGNOSTIC));
+
+ Operands.push_back(ConstantInt::get(XLenTy, Log2_64(SEW)));
+
+ CallInst *VlsegN = Builder.CreateCall(VlsegNFunc, Operands);
+
+ SmallVector<Type *, 8> AggrTypes{Factor, VTy};
+ Value *Return =
+ PoisonValue::get(StructType::get(Load->getContext(), AggrTypes));
+ Function *VecExtractFunc = Intrinsic::getOrInsertDeclaration(
+ Load->getModule(), Intrinsic::riscv_tuple_extract, {VTy, VecTupTy});
+ for (unsigned i = 0; i < Factor; ++i) {
+ Value *VecExtract =
+ Builder.CreateCall(VecExtractFunc, {VlsegN, Builder.getInt32(i)});
+ Return = Builder.CreateInsertValue(Return, VecExtract, i);
+ }
+
+ for (auto [Idx, DIO] : enumerate(DeInterleaveResults)) {
+ // We have to create a brand new ExtractValue to replace each
+ // of these old ExtractValue instructions.
+ Value *NewEV =
+ Builder.CreateExtractValue(Return, {static_cast<unsigned>(Idx)});
+ DIO->replaceAllUsesWith(NewEV);
+ }
+
+ return true;
+}
+
+/// If we're interleaving 2 constant splats, for instance `<vscale x 8 x i32>
+/// <splat of 666>` and `<vscale x 8 x i32> <splat of 777>`, we can create a
+/// larger splat
+/// `<vscale x 4 x i64> <splat of ((777 << 32) | 666)>` first before casting it
+/// into
+/// `<vscale x 8 x i32>`. This will resuling a simple unit stride store rather
+/// than a segment store, which is more expensive in this case.
+static Value *foldInterleaved2OfConstSplats(IntrinsicInst *InterleaveIntrin,
+ VectorType *VTy,
+ const TargetLowering *TLI,
+ Instruction *VPStore) {
+ // We only handle Factor = 2 for now.
+ assert(InterleaveIntrin->arg_size() == 2);
+ auto *SplatVal0 = dyn_cast_or_null<ConstantInt>(
+ getSplatValue(InterleaveIntrin->getArgOperand(0)));
+ auto *SplatVal1 = dyn_cast_or_null<ConstantInt>(
+ getSplatValue(InterleaveIntrin->getArgOperand(1)));
+ if (!SplatVal0 || !SplatVal1)
+ return nullptr;
+
+ auto &Ctx = VPStore->getContext();
+ auto &DL = VPStore->getModule()->getDataLayout();
+
+ auto *NewVTy = VectorType::getExtendedElementVectorType(VTy);
+ if (!TLI->isTypeLegal(TLI->getValueType(DL, NewVTy)))
+ return nullptr;
+
+ // InterleavedAccessPass will remove VPStore after this but we still want to
+ // preserve it, hence clone another one here.
+ auto *ClonedVPStore = VPStore->clone();
+ ClonedVPStore->insertBefore(VPStore);
+ IRBuilder<> Builder(ClonedVPStore);
+
+ Type *ETy = VTy->getElementType();
+ unsigned Width = ETy->getIntegerBitWidth();
+
+ APInt NewSplatVal(Width * 2, SplatVal1->getZExtValue());
+ NewSplatVal <<= Width;
+ NewSplatVal |= SplatVal0->getZExtValue();
+ auto *NewSplat = ConstantVector::getSplat(NewVTy->getElementCount(),
+ ConstantInt::get(Ctx, NewSplatVal));
+ return Builder.CreateBitCast(NewSplat,
+ VectorType::getDoubleElementsVectorType(VTy));
+}
+
+/// Lower an interleaved vp.store into a vssegN intrinsic.
+///
+/// E.g. Lower an interleaved vp.store (Factor = 2):
+///
+/// %is = tail call <vscale x 64 x i8>
+/// @llvm.vector.interleave2.nxv64i8(
+/// <vscale x 32 x i8> %load0,
+/// <vscale x 32 x i8> %load1
+/// %wide.rvl = shl nuw nsw i32 %rvl, 1
+/// tail call void @llvm.vp.store.nxv64i8.p0(
+/// <vscale x 64 x i8> %is, ptr %ptr,
+/// %mask,
+/// i32 %wide.rvl)
+///
+/// Into:
+/// call void @llvm.riscv.vsseg2.mask.nxv32i8.i64(
+/// <vscale x 32 x i8> %load1,
+/// <vscale x 32 x i8> %load2, ptr %ptr,
+/// %mask,
+/// i64 %rvl)
+bool RISCVTargetLowering::lowerInterleavedScalableStore(
+ VPIntrinsic *Store, Value *Mask, IntrinsicInst *InterleaveIntrin,
+ ArrayRef<Value *> InterleaveOperands) const {
+ assert(Store->getIntrinsicID() == Intrinsic::vp_store &&
+ "Unexpected intrinsic");
+
+ const unsigned Factor = InterleaveOperands.size();
+
+ VectorType *VTy = cast<VectorType>(InterleaveOperands[0]->getType());
+
+ // FIXME: Should pass alignment attribute from pointer, but vectorizer needs
+ // to emit it first.
+ const DataLayout &DL = Store->getDataLayout();
+ Align Alignment = Align(DL.getTypeStoreSize(VTy->getScalarType()));
+ if (!isLegalInterleavedAccessType(
+ VTy, Factor, Alignment,
+ Store->getArgOperand(1)->getType()->getPointerAddressSpace(), DL))
+ return false;
+
+ if (Factor == 2)
+ if (Value *BC =
+ foldInterleaved2OfConstSplats(InterleaveIntrin, VTy, this, Store)) {
+ InterleaveIntrin->replaceAllUsesWith(BC);
+ return true;
+ }
+
+ IRBuilder<> Builder(Store);
+ Value *WideEVL = Store->getArgOperand(3);
+ auto *XLenTy = Type::getIntNTy(Store->getContext(), Subtarget.getXLen());
+ Value *EVL = Builder.CreateZExtOrTrunc(
+ Builder.CreateUDiv(WideEVL, ConstantInt::get(WideEVL->getType(), Factor)),
+ XLenTy);
+
+ static const Intrinsic::ID IntrMaskIds[] = {
+ Intrinsic::riscv_vsseg2_mask, Intrinsic::riscv_vsseg3_mask,
+ Intrinsic::riscv_vsseg4_mask, Intrinsic::riscv_vsseg5_mask,
+ Intrinsic::riscv_vsseg6_mask, Intrinsic::riscv_vsseg7_mask,
+ Intrinsic::riscv_vsseg8_mask,
+ };
+
+ unsigned SEW = DL.getTypeSizeInBits(VTy->getElementType());
+ unsigned NumElts = VTy->getElementCount().getKnownMinValue();
+ Type *VecTupTy = TargetExtType::get(
+ Store->getContext(), "riscv.vector.tuple",
+ ScalableVectorType::get(Type::getInt8Ty(Store->getContext()),
+ NumElts * SEW / 8),
+ Factor);
+
+ Function *VecInsertFunc = Intrinsic::getOrInsertDeclaration(
+ Store->getModule(), Intrinsic::riscv_tuple_insert, {VecTupTy, VTy});
+ Value *StoredVal = PoisonValue::get(VecTupTy);
+ for (unsigned i = 0; i < Factor; ++i)
+ StoredVal = Builder.CreateCall(
+ VecInsertFunc, {StoredVal, InterleaveOperands[i], Builder.getInt32(i)});
+
+ SmallVector<Value *, 5> Operands;
+ Operands.push_back(StoredVal);
+ Operands.push_back(Store->getArgOperand(1));
+
+ if (!Mask)
+ Mask = ConstantVector::getSplat(VTy->getElementCount(),
+ ConstantInt::getTrue(Store->getContext()));
+
+ Function *VssegNFunc = Intrinsic::getOrInsertDeclaration(
+ Store->getModule(), IntrMaskIds[Factor - 2],
+ {VecTupTy, Mask->getType(), EVL->getType()});
+
+ Operands.push_back(Mask);
+ Operands.push_back(EVL);
+ Operands.push_back(ConstantInt::get(XLenTy, Log2_64(SEW)));
+
+ Builder.CreateCall(VssegNFunc, Operands);
+ return true;
+}
+
MachineInstr *
RISCVTargetLowering::EmitKCFICheck(MachineBasicBlock &MBB,
MachineBasicBlock::instr_iterator &MBBI,
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.h b/llvm/lib/Target/RISCV/RISCVISelLowering.h
index 77605a3076a80a..f779f4fdc26c1b 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.h
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.h
@@ -910,6 +910,15 @@ class RISCVTargetLowering : public TargetLowering {
bool lowerInterleaveIntrinsicToStore(
StoreInst *SI, ArrayRef<Value *> InterleaveValues) const override;
+ bool lowerInterleavedScalableLoad(
+ VPIntrinsic *Load, Value *Mask, IntrinsicInst *DeinterleaveIntrin,
+ ArrayRef<Value *> DeinterleaveRes) const override;
+
+ bool
+ lowerInterleavedScalableStore(VPIntrinsic *Store, Value *Mask,
+ IntrinsicInst *InterleaveIntrin,
+ ArrayRef<Value *> InterleaveOps) const override;
+
bool supportKCFIBundles() const override { return true; }
SDValue expandIndirectJTBranch(const SDLoc &dl, SDValue Value, SDValue Addr,
diff --git a/llvm/test/CodeGen/RISCV/rvv/scalable-vectors-interleaved-access.ll b/llvm/test/CodeGen/RISCV/rvv/scalable-vectors-interleaved-access.ll
new file mode 100644
index 00000000000000..ac254792e167a8
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/scalable-vectors-interleaved-access.ll
@@ -0,0 +1,619 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc < %s -mtriple=riscv32 -mattr=+v,m -O2 | FileCheck -check-prefixes=CHECK,RV32 %s
+; RUN: llc < %s -mtriple=riscv64 -mattr=+v,m -O2 | FileCheck -check-prefixes=CHECK,RV64 %s
+
+define {<vscale x 2 x i32>, <vscale x 2 x i32>} @load_factor2_v2(ptr %ptr, i32 %rvl) {
+; RV32-LABEL: load_factor2_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 1
+; RV32-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV32-NEXT: vlseg2e32.v v8, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: load_factor2_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 1
+; RV64-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV64-NEXT: vlseg2e32.v v8, (a0)
+; RV64-NEXT: ret
+ %wide.masked.load = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr %ptr, <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i32 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 %rvl)
+ %deinterleaved.results = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %wide.masked.load)
+ %t0 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %deinterleaved.results, 0
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %deinterleaved.results, 1
+ %res0 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } undef, <vscale x 2 x i32> %t0, 0
+ %res1 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %res0, <vscale x 2 x i32> %t1, 1
+ ret { <vscale x 2 x i32>, <vscale x 2 x i32> } %res1
+}
+
+define {<vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>} @load_factor4_v2(ptr %ptr, i32 %rvl) {
+; RV32-LABEL: load_factor4_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 2
+; RV32-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV32-NEXT: vlseg4e32.v v8, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: load_factor4_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 2
+; RV64-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV64-NEXT: vlseg4e32.v v8, (a0)
+; RV64-NEXT: ret
+ %wide.masked.load = call <vscale x 8 x i32> @llvm.vp.load.nxv8i32.p0(ptr %ptr, <vscale x 8 x i1> shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i32 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer), i32 %rvl)
+ %d0 = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %wide.masked.load)
+ %d0.0 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d0, 0
+ %d0.1 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d0, 1
+ %d1 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d0.0)
+ %t0 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d1, 0
+ %t2 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d1, 1
+ %d2 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d0.1)
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d2, 0
+ %t3 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d2, 1
+
+ %res0 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } undef, <vscale x 2 x i32> %t0, 0
+ %res1 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res0, <vscale x 2 x i32> %t1, 1
+ %res2 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res1, <vscale x 2 x i32> %t2, 2
+ %res3 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res2, <vscale x 2 x i32> %t3, 3
+ ret { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res3
+}
+
+define {<vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>} @load_factor8_v2(ptr %ptr, i32 %rvl) {
+; RV32-LABEL: load_factor8_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 3
+; RV32-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV32-NEXT: vlseg8e32.v v8, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: load_factor8_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 3
+; RV64-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV64-NEXT: vlseg8e32.v v8, (a0)
+; RV64-NEXT: ret
+ %wide.masked.load = call <vscale x 16 x i32> @llvm.vp.load.nxv16i32.p0(ptr %ptr, <vscale x 16 x i1> shufflevector (<vscale x 16 x i1> insertelement (<vscale x 16 x i1> poison, i1 true, i32 0), <vscale x 16 x i1> poison, <vscale x 16 x i32> zeroinitializer), i32 %rvl)
+ %d0 = call { <vscale x 8 x i32>, <vscale x 8 x i32> } @llvm.vector.deinterleave2.nxv16i32(<vscale x 16 x i32> %wide.masked.load)
+ %d0.0 = extractvalue { <vscale x 8 x i32>, <vscale x 8 x i32> } %d0, 0
+ %d0.1 = extractvalue { <vscale x 8 x i32>, <vscale x 8 x i32> } %d0, 1
+ %d1 = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %d0.0)
+ %d1.0 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d1, 0
+ %d1.1 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d1, 1
+ %d2 = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %d0.1)
+ %d2.0 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d2, 0
+ %d2.1 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d2, 1
+
+ %d3 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d1.0)
+ %t0 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d3, 0
+ %t4 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d3, 1
+ %d4 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d1.1)
+ %t2 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d4, 0
+ %t6 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d4, 1
+ %d5 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d2.0)
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d5, 0
+ %t5 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d5, 1
+ %d6 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d2.1)
+ %t3 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d6, 0
+ %t7 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d6, 1
+
+ %res0 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } undef, <vscale x 2 x i32> %t0, 0
+ %res1 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res0, <vscale x 2 x i32> %t1, 1
+ %res2 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res1, <vscale x 2 x i32> %t2, 2
+ %res3 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res2, <vscale x 2 x i32> %t3, 3
+ %res4 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res3, <vscale x 2 x i32> %t4, 4
+ %res5 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res4, <vscale x 2 x i32> %t5, 5
+ %res6 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res5, <vscale x 2 x i32> %t6, 6
+ %res7 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res6, <vscale x 2 x i32> %t7, 7
+ ret { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res7
+}
+
+define void @store_factor2_v2(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v1, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: store_factor2_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 1
+; RV32-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV32-NEXT: vsseg2e32.v v8, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: store_factor2_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 1
+; RV64-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV64-NEXT: vsseg2e32.v v8, (a0)
+; RV64-NEXT: ret
+ %interleaved.vec = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v1)
+ call void @llvm.vp.store.nxv2i32.p0(<vscale x 2 x i32> %interleaved.vec, ptr %ptr, <vscale x 2 x i1> shufflevector (<vscale x 2 x i1> insertelement (<vscale x 2 x i1> poison, i1 true, i32 0), <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer), i32 %rvl)
+ ret void
+}
+
+; Expecting unit-stride store here rather than segmented store.
+define void @store_factor2_const_splat(ptr %dst) {
+; RV32-LABEL: store_factor2_const_splat:
+; RV32: # %bb.0:
+; RV32-NEXT: addi sp, sp, -16
+; RV32-NEXT: .cfi_def_cfa_offset 16
+; RV32-NEXT: li a1, 777
+; RV32-NEXT: li a2, 666
+; RV32-NEXT: sw a2, 8(sp)
+; RV32-NEXT: sw a1, 12(sp)
+; RV32-NEXT: addi a1, sp, 8
+; RV32-NEXT: vsetvli a2, zero, e64, m8, ta, ma
+; RV32-NEXT: vlse64.v v8, (a1), zero
+; RV32-NEXT: li a1, 87
+; RV32-NEXT: vsetvli zero, a1, e32, m8, ta, ma
+; RV32-NEXT: vse32.v v8, (a0)
+; RV32-NEXT: addi sp, sp, 16
+; RV32-NEXT: .cfi_def_cfa_offset 0
+; RV32-NEXT: ret
+;
+; RV64-LABEL: store_factor2_const_splat:
+; RV64: # %bb.0:
+; RV64-NEXT: li a1, 777
+; RV64-NEXT: slli a1, a1, 32
+; RV64-NEXT: addi a1, a1, 666
+; RV64-NEXT: vsetvli a2, zero, e64, m8, ta, ma
+; RV64-NEXT: vmv.v.x v8, a1
+; RV64-NEXT: li a1, 87
+; RV64-NEXT: vsetvli zero, a1, e32, m8, ta, ma
+; RV64-NEXT: vse32.v v8, (a0)
+; RV64-NEXT: ret
+ %interleave2 = call <vscale x 16 x i32> @llvm.vector.interleave2.nxv16i32(
+ <vscale x 8 x i32> shufflevector (<vscale x 8 x i32> insertelement (<vscale x 8 x i32> poison, i32 666, i64 0), <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer),
+ <vscale x 8 x i32> shufflevector (<vscale x 8 x i32> insertelement (<vscale x 8 x i32> poison, i32 777, i64 0), <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer)
+ )
+ call void @llvm.vp.store.nxv16i32.p0(<vscale x 16 x i32> %interleave2, ptr %dst,
+ <vscale x 16 x i1> shufflevector (<vscale x 16 x i1> insertelement (<vscale x 16 x i1> poison, i1 1, i64 0), <vscale x 16 x i1> poison, <vscale x 16 x i32> zeroinitializer),
+ i32 87)
+ ret void
+}
+
+define void @store_factor4_v2(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v1, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: store_factor4_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 2
+; RV32-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV32-NEXT: vmv1r.v v10, v8
+; RV32-NEXT: vmv1r.v v11, v9
+; RV32-NEXT: vsseg4e32.v v8, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: store_factor4_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 2
+; RV64-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV64-NEXT: vmv1r.v v10, v8
+; RV64-NEXT: vmv1r.v v11, v9
+; RV64-NEXT: vsseg4e32.v v8, (a0)
+; RV64-NEXT: ret
+ %interleaved.vec0 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v0)
+ %interleaved.vec1 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v1, <vscale x 1 x i32> %v1)
+ %interleaved.vec2 = call <vscale x 4 x i32> @llvm.vector.interleave2.nxv4i32(<vscale x 2 x i32> %interleaved.vec0, <vscale x 2 x i32> %interleaved.vec1)
+ call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> %interleaved.vec2, ptr %ptr, <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i32 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 %rvl)
+ ret void
+}
+
+define void @store_factor8_v2(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v1, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: store_factor8_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 3
+; RV32-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV32-NEXT: vmv1r.v v10, v8
+; RV32-NEXT: vmv1r.v v11, v9
+; RV32-NEXT: vmv1r.v v12, v8
+; RV32-NEXT: vmv1r.v v13, v9
+; RV32-NEXT: vmv1r.v v14, v8
+; RV32-NEXT: vmv1r.v v15, v9
+; RV32-NEXT: vsseg8e32.v v8, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: store_factor8_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 3
+; RV64-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV64-NEXT: vmv1r.v v10, v8
+; RV64-NEXT: vmv1r.v v11, v9
+; RV64-NEXT: vmv1r.v v12, v8
+; RV64-NEXT: vmv1r.v v13, v9
+; RV64-NEXT: vmv1r.v v14, v8
+; RV64-NEXT: vmv1r.v v15, v9
+; RV64-NEXT: vsseg8e32.v v8, (a0)
+; RV64-NEXT: ret
+ %interleaved.vec0 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v0)
+ %interleaved.vec1 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v0)
+ %interleaved.vec2 = call <vscale x 4 x i32> @llvm.vector.interleave2.nxv4i32(<vscale x 2 x i32> %interleaved.vec0, <vscale x 2 x i32> %interleaved.vec1)
+ %interleaved.vec3 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v1, <vscale x 1 x i32> %v1)
+ %interleaved.vec4 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v1, <vscale x 1 x i32> %v1)
+ %interleaved.vec5 = call <vscale x 4 x i32> @llvm.vector.interleave2.nxv4i32(<vscale x 2 x i32> %interleaved.vec3, <vscale x 2 x i32> %interleaved.vec4)
+ %interleaved.vec6 = call <vscale x 8 x i32> @llvm.vector.interleave2.nxv8i32(<vscale x 4 x i32> %interleaved.vec2, <vscale x 4 x i32> %interleaved.vec5)
+ call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %interleaved.vec6, ptr %ptr, <vscale x 8 x i1> shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i32 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer), i32 %rvl)
+ ret void
+}
+
+define {<vscale x 2 x i32>, <vscale x 2 x i32>} @masked_load_factor2_v2(<vscale x 2 x i1> %mask, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: masked_load_factor2_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 1
+; RV32-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV32-NEXT: vlseg2e32.v v8, (a0), v0.t
+; RV32-NEXT: ret
+;
+; RV64-LABEL: masked_load_factor2_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 1
+; RV64-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV64-NEXT: vlseg2e32.v v8, (a0), v0.t
+; RV64-NEXT: ret
+ %interleaved.mask = tail call <vscale x 4 x i1> @llvm.vector.interleave2.nxv4i1(<vscale x 2 x i1> %mask, <vscale x 2 x i1> %mask)
+ %wide.masked.load = tail call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr %ptr, <vscale x 4 x i1> %interleaved.mask, i32 %rvl)
+ %deinterleaved.results = tail call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv16i32(<vscale x 4 x i32> %wide.masked.load)
+ %t0 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %deinterleaved.results, 0
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %deinterleaved.results, 1
+ %res0 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } undef, <vscale x 2 x i32> %t0, 0
+ %res1 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %res0, <vscale x 2 x i32> %t1, 1
+ ret { <vscale x 2 x i32>, <vscale x 2 x i32> } %res1
+}
+
+define {<vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>} @masked_load_factor4_v2(<vscale x 2 x i1> %mask, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: masked_load_factor4_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 2
+; RV32-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV32-NEXT: vlseg4e32.v v8, (a0), v0.t
+; RV32-NEXT: ret
+;
+; RV64-LABEL: masked_load_factor4_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 2
+; RV64-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV64-NEXT: vlseg4e32.v v8, (a0), v0.t
+; RV64-NEXT: ret
+ %interleaved.mask0 = call <vscale x 4 x i1> @llvm.vector.interleave2.nxv4i1(<vscale x 2 x i1> %mask, <vscale x 2 x i1> %mask)
+ %interleaved.mask1 = call <vscale x 4 x i1> @llvm.vector.interleave2.nxv4i1(<vscale x 2 x i1> %mask, <vscale x 2 x i1> %mask)
+ %interleaved.mask2 = call <vscale x 8 x i1> @llvm.vector.interleave2.nxv8i1(<vscale x 4 x i1> %interleaved.mask0, <vscale x 4 x i1> %interleaved.mask1)
+ %wide.masked.load = call <vscale x 8 x i32> @llvm.vp.load.nxv8i32.p0(ptr %ptr, <vscale x 8 x i1> %interleaved.mask2, i32 %rvl)
+ %d0 = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %wide.masked.load)
+ %d0.0 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d0, 0
+ %d0.1 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d0, 1
+ %d1 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d0.0)
+ %t0 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d1, 0
+ %t2 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d1, 1
+ %d2 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d0.1)
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d2, 0
+ %t3 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d2, 1
+
+ %res0 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } undef, <vscale x 2 x i32> %t0, 0
+ %res1 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res0, <vscale x 2 x i32> %t1, 1
+ %res2 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res1, <vscale x 2 x i32> %t2, 2
+ %res3 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res2, <vscale x 2 x i32> %t3, 3
+ ret { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res3
+}
+
+define void @masked_store_factor2_v2(<vscale x 1 x i1> %mask, <vscale x 1 x i32> %v0, <vscale x 1 x i32> %v1, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: masked_store_factor2_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 1
+; RV32-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV32-NEXT: vmv1r.v v9, v8
+; RV32-NEXT: vsseg2e32.v v8, (a0), v0.t
+; RV32-NEXT: ret
+;
+; RV64-LABEL: masked_store_factor2_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 1
+; RV64-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV64-NEXT: vmv1r.v v9, v8
+; RV64-NEXT: vsseg2e32.v v8, (a0), v0.t
+; RV64-NEXT: ret
+ %interleaved.mask = tail call <vscale x 2 x i1> @llvm.vector.interleave2.nxv2i1(<vscale x 1 x i1> %mask, <vscale x 1 x i1> %mask)
+ %interleaved.vec = tail call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v0)
+ tail call void @llvm.vp.store.nxv2i32.p0(<vscale x 2 x i32> %interleaved.vec, ptr %ptr, <vscale x 2 x i1> %interleaved.mask, i32 %rvl)
+ ret void
+}
+
+define void @masked_load_store_factor2_v2_shared_mask(<vscale x 2 x i1> %mask, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: masked_load_store_factor2_v2_shared_mask:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 1
+; RV32-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV32-NEXT: vlseg2e32.v v8, (a0), v0.t
+; RV32-NEXT: vsseg2e32.v v8, (a0), v0.t
+; RV32-NEXT: ret
+;
+; RV64-LABEL: masked_load_store_factor2_v2_shared_mask:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 1
+; RV64-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV64-NEXT: vlseg2e32.v v8, (a0), v0.t
+; RV64-NEXT: vsseg2e32.v v8, (a0), v0.t
+; RV64-NEXT: ret
+ %interleaved.mask = tail call <vscale x 4 x i1> @llvm.vector.interleave2.nxv4i1(<vscale x 2 x i1> %mask, <vscale x 2 x i1> %mask)
+ %wide.masked.load = tail call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr %ptr, <vscale x 4 x i1> %interleaved.mask, i32 %rvl)
+ %deinterleaved.results = tail call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv16i32(<vscale x 4 x i32> %wide.masked.load)
+ %t0 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %deinterleaved.results, 0
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %deinterleaved.results, 1
+ %interleaved.vec = tail call <vscale x 4 x i32> @llvm.vector.interleave2.nxv4i32(<vscale x 2 x i32> %t0, <vscale x 2 x i32> %t1)
+ tail call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> %interleaved.vec, ptr %ptr, <vscale x 4 x i1> %interleaved.mask, i32 %rvl)
+ ret void
+}
+
+define i32 @masked_load_store_factor2_v2_shared_mask_extract(<vscale x 2 x i1> %mask, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: masked_load_store_factor2_v2_shared_mask_extract:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetvli a2, zero, e8, mf4, ta, ma
+; RV32-NEXT: vmv1r.v v8, v0
+; RV32-NEXT: vmv.v.i v9, 0
+; RV32-NEXT: li a2, -1
+; RV32-NEXT: vsetvli a3, zero, e8, mf2, ta, ma
+; RV32-NEXT: vmv.v.i v10, 0
+; RV32-NEXT: csrr a3, vlenb
+; RV32-NEXT: vsetvli a4, zero, e8, mf4, ta, ma
+; RV32-NEXT: vmerge.vim v11, v9, 1, v0
+; RV32-NEXT: srli a3, a3, 2
+; RV32-NEXT: vwaddu.vv v12, v11, v11
+; RV32-NEXT: vwmaccu.vx v12, a2, v11
+; RV32-NEXT: vmsne.vi v0, v12, 0
+; RV32-NEXT: vsetvli a2, zero, e8, mf2, ta, ma
+; RV32-NEXT: vslidedown.vx v11, v12, a3
+; RV32-NEXT: vmerge.vim v10, v10, 1, v0
+; RV32-NEXT: vsetvli a2, zero, e8, mf4, ta, ma
+; RV32-NEXT: vmsne.vi v0, v11, 0
+; RV32-NEXT: add a2, a3, a3
+; RV32-NEXT: vmerge.vim v9, v9, 1, v0
+; RV32-NEXT: vsetvli zero, a2, e8, mf2, ta, ma
+; RV32-NEXT: vslideup.vx v10, v9, a3
+; RV32-NEXT: vsetvli a2, zero, e8, mf2, ta, ma
+; RV32-NEXT: vmsne.vi v0, v10, 0
+; RV32-NEXT: vsetvli zero, a1, e32, m2, ta, ma
+; RV32-NEXT: vle32.v v10, (a0), v0.t
+; RV32-NEXT: li a2, 32
+; RV32-NEXT: vsetvli a3, zero, e32, m1, ta, ma
+; RV32-NEXT: vnsrl.wx v13, v10, a2
+; RV32-NEXT: vmv.x.s a2, v10
+; RV32-NEXT: vnsrl.wi v12, v10, 0
+; RV32-NEXT: srli a1, a1, 1
+; RV32-NEXT: vmv1r.v v0, v8
+; RV32-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV32-NEXT: vsseg2e32.v v12, (a0), v0.t
+; RV32-NEXT: mv a0, a2
+; RV32-NEXT: ret
+;
+; RV64-LABEL: masked_load_store_factor2_v2_shared_mask_extract:
+; RV64: # %bb.0:
+; RV64-NEXT: vsetvli a2, zero, e8, mf4, ta, ma
+; RV64-NEXT: vmv1r.v v8, v0
+; RV64-NEXT: vmv.v.i v9, 0
+; RV64-NEXT: li a2, -1
+; RV64-NEXT: vsetvli a3, zero, e8, mf2, ta, ma
+; RV64-NEXT: vmv.v.i v10, 0
+; RV64-NEXT: csrr a3, vlenb
+; RV64-NEXT: vsetvli a4, zero, e8, mf4, ta, ma
+; RV64-NEXT: vmerge.vim v11, v9, 1, v0
+; RV64-NEXT: srli a3, a3, 2
+; RV64-NEXT: vwaddu.vv v12, v11, v11
+; RV64-NEXT: vwmaccu.vx v12, a2, v11
+; RV64-NEXT: vmsne.vi v0, v12, 0
+; RV64-NEXT: vsetvli a2, zero, e8, mf2, ta, ma
+; RV64-NEXT: vslidedown.vx v11, v12, a3
+; RV64-NEXT: vmerge.vim v10, v10, 1, v0
+; RV64-NEXT: vsetvli a2, zero, e8, mf4, ta, ma
+; RV64-NEXT: vmsne.vi v0, v11, 0
+; RV64-NEXT: add a2, a3, a3
+; RV64-NEXT: vmerge.vim v9, v9, 1, v0
+; RV64-NEXT: vsetvli zero, a2, e8, mf2, ta, ma
+; RV64-NEXT: vslideup.vx v10, v9, a3
+; RV64-NEXT: slli a2, a1, 32
+; RV64-NEXT: vsetvli a3, zero, e8, mf2, ta, ma
+; RV64-NEXT: vmsne.vi v0, v10, 0
+; RV64-NEXT: srli a2, a2, 32
+; RV64-NEXT: vsetvli zero, a2, e32, m2, ta, ma
+; RV64-NEXT: vle32.v v10, (a0), v0.t
+; RV64-NEXT: li a2, 32
+; RV64-NEXT: vsetvli a3, zero, e32, m1, ta, ma
+; RV64-NEXT: vnsrl.wx v13, v10, a2
+; RV64-NEXT: vmv.x.s a2, v10
+; RV64-NEXT: vnsrl.wi v12, v10, 0
+; RV64-NEXT: srliw a1, a1, 1
+; RV64-NEXT: vmv1r.v v0, v8
+; RV64-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; RV64-NEXT: vsseg2e32.v v12, (a0), v0.t
+; RV64-NEXT: mv a0, a2
+; RV64-NEXT: ret
+ %interleaved.mask = tail call <vscale x 4 x i1> @llvm.vector.interleave2.nxv4i1(<vscale x 2 x i1> %mask, <vscale x 2 x i1> %mask)
+ %wide.masked.load = tail call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr %ptr, <vscale x 4 x i1> %interleaved.mask, i32 %rvl)
+ %deinterleaved.results = tail call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv16i32(<vscale x 4 x i32> %wide.masked.load)
+ %t0 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %deinterleaved.results, 0
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %deinterleaved.results, 1
+ %r0 = extractelement <vscale x 4 x i32> %wide.masked.load, i32 0
+ %interleaved.vec = tail call <vscale x 4 x i32> @llvm.vector.interleave2.nxv4i32(<vscale x 2 x i32> %t0, <vscale x 2 x i32> %t1)
+ tail call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> %interleaved.vec, ptr %ptr, <vscale x 4 x i1> %interleaved.mask, i32 %rvl)
+ ret i32 %r0
+}
+
+define void @masked_store_factor4_v2(<vscale x 1 x i1> %mask, <vscale x 1 x i32> %v0, <vscale x 1 x i32> %v1, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: masked_store_factor4_v2:
+; RV32: # %bb.0:
+; RV32-NEXT: srli a1, a1, 2
+; RV32-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV32-NEXT: vmv1r.v v10, v8
+; RV32-NEXT: vmv1r.v v11, v9
+; RV32-NEXT: vsseg4e32.v v8, (a0), v0.t
+; RV32-NEXT: ret
+;
+; RV64-LABEL: masked_store_factor4_v2:
+; RV64: # %bb.0:
+; RV64-NEXT: srliw a1, a1, 2
+; RV64-NEXT: vsetvli zero, a1, e32, mf2, ta, ma
+; RV64-NEXT: vmv1r.v v10, v8
+; RV64-NEXT: vmv1r.v v11, v9
+; RV64-NEXT: vsseg4e32.v v8, (a0), v0.t
+; RV64-NEXT: ret
+ %interleaved.mask0 = call <vscale x 2 x i1> @llvm.vector.interleave2.nxv2i1(<vscale x 1 x i1> %mask, <vscale x 1 x i1> %mask)
+ %interleaved.mask1 = call <vscale x 2 x i1> @llvm.vector.interleave2.nxv2i1(<vscale x 1 x i1> %mask, <vscale x 1 x i1> %mask)
+ %interleaved.mask2 = call <vscale x 4 x i1> @llvm.vector.interleave2.nxv4i1(<vscale x 2 x i1> %interleaved.mask0, <vscale x 2 x i1> %interleaved.mask1)
+ %interleaved.vec0 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v0)
+ %interleaved.vec1 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v1, <vscale x 1 x i32> %v1)
+ %interleaved.vec2 = call <vscale x 4 x i32> @llvm.vector.interleave2.nxv4i32(<vscale x 2 x i32> %interleaved.vec0, <vscale x 2 x i32> %interleaved.vec1)
+ call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> %interleaved.vec2, ptr %ptr, <vscale x 4 x i1> %interleaved.mask2, i32 %rvl)
+ ret void
+}
+
+; Negative tests
+
+; We should not transform this function because the deinterleave tree is not in a desired form.
+define {<vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>} @incorrect_extract_value_index(ptr %ptr, i32 %rvl) {
+; RV32-LABEL: incorrect_extract_value_index:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetvli zero, a1, e32, m4, ta, ma
+; RV32-NEXT: vle32.v v8, (a0)
+; RV32-NEXT: li a0, 32
+; RV32-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; RV32-NEXT: vnsrl.wi v12, v8, 0
+; RV32-NEXT: vsetvli a1, zero, e32, m1, ta, ma
+; RV32-NEXT: vnsrl.wx v9, v12, a0
+; RV32-NEXT: vnsrl.wi v8, v12, 0
+; RV32-NEXT: vmv.v.v v10, v9
+; RV32-NEXT: vmv.v.v v11, v9
+; RV32-NEXT: ret
+;
+; RV64-LABEL: incorrect_extract_value_index:
+; RV64: # %bb.0:
+; RV64-NEXT: slli a1, a1, 32
+; RV64-NEXT: srli a1, a1, 32
+; RV64-NEXT: vsetvli zero, a1, e32, m4, ta, ma
+; RV64-NEXT: vle32.v v8, (a0)
+; RV64-NEXT: li a0, 32
+; RV64-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; RV64-NEXT: vnsrl.wi v12, v8, 0
+; RV64-NEXT: vsetvli a1, zero, e32, m1, ta, ma
+; RV64-NEXT: vnsrl.wx v9, v12, a0
+; RV64-NEXT: vnsrl.wi v8, v12, 0
+; RV64-NEXT: vmv.v.v v10, v9
+; RV64-NEXT: vmv.v.v v11, v9
+; RV64-NEXT: ret
+ %wide.masked.load = call <vscale x 8 x i32> @llvm.vp.load.nxv8i32.p0(ptr %ptr, <vscale x 8 x i1> shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i32 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer), i32 %rvl)
+ %d0 = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %wide.masked.load)
+ %d0.0 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d0, 0
+ %d0.1 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d0, 0
+ %d1 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d0.0)
+ %t0 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d1, 0
+ %t2 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d1, 1
+ %d2 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d0.1)
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d2, 1
+ %t3 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d2, 1
+
+ %res0 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } undef, <vscale x 2 x i32> %t0, 0
+ %res1 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res0, <vscale x 2 x i32> %t1, 1
+ %res2 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res1, <vscale x 2 x i32> %t2, 2
+ %res3 = insertvalue { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res2, <vscale x 2 x i32> %t3, 3
+ ret { <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32>, <vscale x 2 x i32> } %res3
+}
+
+; We should not transform this function because the expression is not a balanced tree.
+define {<vscale x 4 x i32>, <vscale x 2 x i32>, <vscale x 1 x i32>, <vscale x 1 x i32>} @not_balanced_load_tree(ptr %ptr, i32 %rvl) {
+; RV32-LABEL: not_balanced_load_tree:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetvli zero, a1, e32, m4, ta, ma
+; RV32-NEXT: vle32.v v12, (a0)
+; RV32-NEXT: li a0, 32
+; RV32-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; RV32-NEXT: vnsrl.wx v8, v12, a0
+; RV32-NEXT: vnsrl.wi v16, v12, 0
+; RV32-NEXT: vsetvli a1, zero, e32, m1, ta, ma
+; RV32-NEXT: vnsrl.wi v10, v16, 0
+; RV32-NEXT: vnsrl.wx v11, v16, a0
+; RV32-NEXT: vsetvli a1, zero, e32, mf2, ta, ma
+; RV32-NEXT: vnsrl.wx v12, v11, a0
+; RV32-NEXT: vnsrl.wi v11, v11, 0
+; RV32-NEXT: ret
+;
+; RV64-LABEL: not_balanced_load_tree:
+; RV64: # %bb.0:
+; RV64-NEXT: slli a1, a1, 32
+; RV64-NEXT: srli a1, a1, 32
+; RV64-NEXT: vsetvli zero, a1, e32, m4, ta, ma
+; RV64-NEXT: vle32.v v12, (a0)
+; RV64-NEXT: li a0, 32
+; RV64-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; RV64-NEXT: vnsrl.wx v8, v12, a0
+; RV64-NEXT: vnsrl.wi v16, v12, 0
+; RV64-NEXT: vsetvli a1, zero, e32, m1, ta, ma
+; RV64-NEXT: vnsrl.wi v10, v16, 0
+; RV64-NEXT: vnsrl.wx v11, v16, a0
+; RV64-NEXT: vsetvli a1, zero, e32, mf2, ta, ma
+; RV64-NEXT: vnsrl.wx v12, v11, a0
+; RV64-NEXT: vnsrl.wi v11, v11, 0
+; RV64-NEXT: ret
+ %wide.masked.load = call <vscale x 8 x i32> @llvm.vp.load.nxv8i32.p0(ptr %ptr, <vscale x 8 x i1> shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i32 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer), i32 %rvl)
+ %d0 = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %wide.masked.load)
+ %d0.0 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d0, 0
+ %t0 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %d0, 1
+ %d1 = call { <vscale x 2 x i32>, <vscale x 2 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %d0.0)
+ %t1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d1, 0
+ %d1.1 = extractvalue { <vscale x 2 x i32>, <vscale x 2 x i32> } %d1, 1
+ %d2 = call { <vscale x 1 x i32>, <vscale x 1 x i32> } @llvm.vector.deinterleave2.nxv4i32(<vscale x 2 x i32> %d1.1)
+ %t2 = extractvalue { <vscale x 1 x i32>, <vscale x 1 x i32> } %d2, 0
+ %t3 = extractvalue { <vscale x 1 x i32>, <vscale x 1 x i32> } %d2, 1
+
+ %res0 = insertvalue { <vscale x 4 x i32>, <vscale x 2 x i32>, <vscale x 1 x i32>, <vscale x 1 x i32> } undef, <vscale x 4 x i32> %t0, 0
+ %res1 = insertvalue { <vscale x 4 x i32>, <vscale x 2 x i32>, <vscale x 1 x i32>, <vscale x 1 x i32> } %res0, <vscale x 2 x i32> %t1, 1
+ %res2 = insertvalue { <vscale x 4 x i32>, <vscale x 2 x i32>, <vscale x 1 x i32>, <vscale x 1 x i32> } %res1, <vscale x 1 x i32> %t2, 2
+ %res3 = insertvalue { <vscale x 4 x i32>, <vscale x 2 x i32>, <vscale x 1 x i32>, <vscale x 1 x i32> } %res2, <vscale x 1 x i32> %t3, 3
+ ret { <vscale x 4 x i32>, <vscale x 2 x i32>, <vscale x 1 x i32>, <vscale x 1 x i32> } %res3
+}
+
+define void @not_balanced_store_tree(<vscale x 1 x i32> %v0, <vscale x 2 x i32> %v1, <vscale x 4 x i32> %v2, ptr %ptr, i32 %rvl) {
+; RV32-LABEL: not_balanced_store_tree:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetvli a2, zero, e32, mf2, ta, ma
+; RV32-NEXT: vwaddu.vv v12, v8, v8
+; RV32-NEXT: li a2, -1
+; RV32-NEXT: csrr a3, vlenb
+; RV32-NEXT: vwmaccu.vx v12, a2, v8
+; RV32-NEXT: srli a3, a3, 3
+; RV32-NEXT: vsetvli a4, zero, e32, m1, ta, ma
+; RV32-NEXT: vslidedown.vx v8, v12, a3
+; RV32-NEXT: add a4, a3, a3
+; RV32-NEXT: vsetvli zero, a4, e32, m1, ta, ma
+; RV32-NEXT: vslideup.vx v12, v8, a3
+; RV32-NEXT: vsetvli a3, zero, e32, m1, ta, ma
+; RV32-NEXT: vwaddu.vv v14, v12, v9
+; RV32-NEXT: vwmaccu.vx v14, a2, v9
+; RV32-NEXT: vsetvli a3, zero, e32, m2, ta, ma
+; RV32-NEXT: vwaddu.vv v16, v14, v10
+; RV32-NEXT: vwmaccu.vx v16, a2, v10
+; RV32-NEXT: vsetvli zero, a1, e32, m4, ta, ma
+; RV32-NEXT: vse32.v v16, (a0)
+; RV32-NEXT: ret
+;
+; RV64-LABEL: not_balanced_store_tree:
+; RV64: # %bb.0:
+; RV64-NEXT: vsetvli a2, zero, e32, mf2, ta, ma
+; RV64-NEXT: vwaddu.vv v12, v8, v8
+; RV64-NEXT: li a2, -1
+; RV64-NEXT: csrr a3, vlenb
+; RV64-NEXT: slli a1, a1, 32
+; RV64-NEXT: vwmaccu.vx v12, a2, v8
+; RV64-NEXT: srli a3, a3, 3
+; RV64-NEXT: vsetvli a4, zero, e32, m1, ta, ma
+; RV64-NEXT: vslidedown.vx v8, v12, a3
+; RV64-NEXT: add a4, a3, a3
+; RV64-NEXT: vsetvli zero, a4, e32, m1, ta, ma
+; RV64-NEXT: vslideup.vx v12, v8, a3
+; RV64-NEXT: vsetvli a3, zero, e32, m1, ta, ma
+; RV64-NEXT: vwaddu.vv v14, v12, v9
+; RV64-NEXT: vwmaccu.vx v14, a2, v9
+; RV64-NEXT: vsetvli a3, zero, e32, m2, ta, ma
+; RV64-NEXT: vwaddu.vv v16, v14, v10
+; RV64-NEXT: vwmaccu.vx v16, a2, v10
+; RV64-NEXT: srli a1, a1, 32
+; RV64-NEXT: vsetvli zero, a1, e32, m4, ta, ma
+; RV64-NEXT: vse32.v v16, (a0)
+; RV64-NEXT: ret
+ %interleaved.vec0 = call <vscale x 2 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 1 x i32> %v0, <vscale x 1 x i32> %v0)
+ %interleaved.vec1 = call <vscale x 4 x i32> @llvm.vector.interleave2.nxv2i32(<vscale x 2 x i32> %interleaved.vec0, <vscale x 2 x i32> %v1)
+ %interleaved.vec2 = call <vscale x 8 x i32> @llvm.vector.interleave2.nxv4i32(<vscale x 4 x i32> %interleaved.vec1, <vscale x 4 x i32> %v2)
+ call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %interleaved.vec2, ptr %ptr, <vscale x 8 x i1> shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i32 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer), i32 %rvl)
+ ret void
+}
+
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK: {{.*}}
>From c869bec122d63858c7196040d6e29e679761d459 Mon Sep 17 00:00:00 2001
From: Min-Yih Hsu <min.hsu at sifive.com>
Date: Mon, 27 Jan 2025 13:34:29 -0800
Subject: [PATCH 2/3] fixup! Address review comments
- Change what getMask returns
- Change the name of TLI hooks
- Limit the TLI hooks to scalable vectors only for now
---
llvm/include/llvm/CodeGen/TargetLowering.h | 12 ++----
llvm/lib/CodeGen/InterleavedAccessPass.cpp | 19 +++++----
llvm/lib/Target/RISCV/RISCVISelLowering.cpp | 46 ++++++++++-----------
llvm/lib/Target/RISCV/RISCVISelLowering.h | 11 +++--
4 files changed, 41 insertions(+), 47 deletions(-)
diff --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index 50fa3a39c0a0c4..9411dc66b2931f 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -3158,12 +3158,10 @@ class TargetLoweringBase {
///
/// \p Load is a vp.load instruction.
/// \p Mask is a mask value
- /// \p DeinterleaveIntrin is vector.deinterleave intrinsic
/// \p DeinterleaveRes is a list of deinterleaved results.
virtual bool
- lowerInterleavedScalableLoad(VPIntrinsic *Load, Value *Mask,
- IntrinsicInst *DeinterleaveIntrin,
- ArrayRef<Value *> DeinterleaveRes) const {
+ lowerDeinterleavedIntrinsicToVPLoad(VPIntrinsic *Load, Value *Mask,
+ ArrayRef<Value *> DeinterleaveRes) const {
return false;
}
@@ -3172,12 +3170,10 @@ class TargetLoweringBase {
///
/// \p Store is the vp.store instruction.
/// \p Mask is a mask value
- /// \p InterleaveIntrin is vector.interleave intrinsic
/// \p InterleaveOps is a list of values being interleaved.
virtual bool
- lowerInterleavedScalableStore(VPIntrinsic *Store, Value *Mask,
- IntrinsicInst *InterleaveIntrin,
- ArrayRef<Value *> InterleaveOps) const {
+ lowerInterleavedIntrinsicToVPStore(VPIntrinsic *Store, Value *Mask,
+ ArrayRef<Value *> InterleaveOps) const {
return false;
}
diff --git a/llvm/lib/CodeGen/InterleavedAccessPass.cpp b/llvm/lib/CodeGen/InterleavedAccessPass.cpp
index 9b15d0351ebe18..32ca90938d7ab8 100644
--- a/llvm/lib/CodeGen/InterleavedAccessPass.cpp
+++ b/llvm/lib/CodeGen/InterleavedAccessPass.cpp
@@ -637,7 +637,7 @@ getVectorDeinterleaveFactor(IntrinsicInst *II,
/// - if a value within the option is nullptr, the value corresponds to all-true
/// mask
/// - return nullopt if mask cannot be deinterleaved
-static std::optional<Value *> getMask(Value *WideMask, unsigned Factor) {
+static Value *getMask(Value *WideMask, unsigned Factor) {
using namespace llvm::PatternMatch;
if (auto *IMI = dyn_cast<IntrinsicInst>(WideMask)) {
SmallVector<Value *, 8> Operands;
@@ -650,8 +650,9 @@ static std::optional<Value *> getMask(Value *WideMask, unsigned Factor) {
}
}
if (match(WideMask, m_AllOnes()))
- return nullptr;
- return std::nullopt;
+ return WideMask;
+
+ return nullptr;
}
bool InterleavedAccessImpl::lowerDeinterleaveIntrinsic(
@@ -673,7 +674,7 @@ bool InterleavedAccessImpl::lowerDeinterleaveIntrinsic(
return false;
// Check mask operand. Handle both all-true and interleaved mask.
Value *WideMask = VPLoad->getOperand(1);
- std::optional<Value *> Mask = getMask(WideMask, Factor);
+ Value *Mask = getMask(WideMask, Factor);
if (!Mask)
return false;
@@ -682,8 +683,8 @@ bool InterleavedAccessImpl::lowerDeinterleaveIntrinsic(
// Since lowerInterleaveLoad expects Shuffles and LoadInst, use special
// TLI function to emit target-specific interleaved instruction.
- if (!TLI->lowerInterleavedScalableLoad(VPLoad, *Mask, DI,
- DeinterleaveValues))
+ if (!TLI->lowerDeinterleavedIntrinsicToVPLoad(VPLoad, Mask,
+ DeinterleaveValues))
return false;
} else {
@@ -725,7 +726,7 @@ bool InterleavedAccessImpl::lowerInterleaveIntrinsic(
return false;
Value *WideMask = VPStore->getOperand(2);
- std::optional<Value *> Mask = getMask(WideMask, Factor);
+ Value *Mask = getMask(WideMask, Factor);
if (!Mask)
return false;
@@ -734,8 +735,8 @@ bool InterleavedAccessImpl::lowerInterleaveIntrinsic(
// Since lowerInterleavedStore expects Shuffle and StoreInst, use special
// TLI function to emit target-specific interleaved instruction.
- if (!TLI->lowerInterleavedScalableStore(VPStore, *Mask, II,
- InterleaveValues))
+ if (!TLI->lowerInterleavedIntrinsicToVPStore(VPStore, Mask,
+ InterleaveValues))
return false;
} else {
auto *SI = cast<StoreInst>(StoredBy);
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index b5155106a506bc..a93f05a85d3aa7 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -22555,15 +22555,20 @@ bool RISCVTargetLowering::lowerInterleaveIntrinsicToStore(
///
/// NOTE: the deinterleave2 intrinsic won't be touched and is expected to be
/// removed by the caller
-bool RISCVTargetLowering::lowerInterleavedScalableLoad(
- VPIntrinsic *Load, Value *Mask, IntrinsicInst *DeinterleaveIntrin,
+bool RISCVTargetLowering::lowerDeinterleavedIntrinsicToVPLoad(
+ VPIntrinsic *Load, Value *Mask,
ArrayRef<Value *> DeInterleaveResults) const {
+ assert(Mask && "Expect a valid mask");
assert(Load->getIntrinsicID() == Intrinsic::vp_load &&
"Unexpected intrinsic");
const unsigned Factor = DeInterleaveResults.size();
- auto *WideVTy = cast<VectorType>(Load->getType());
+ auto *WideVTy = dyn_cast<ScalableVectorType>(Load->getType());
+ // TODO: Support fixed vectors.
+ if (!WideVTy)
+ return false;
+
unsigned WideNumElements = WideVTy->getElementCount().getKnownMinValue();
assert(WideNumElements % Factor == 0 &&
"ElementCount of a wide load must be divisible by interleave factor");
@@ -22605,10 +22610,6 @@ bool RISCVTargetLowering::lowerInterleavedScalableLoad(
SmallVector<Value *> Operands;
Operands.append({PoisonVal, Load->getArgOperand(0)});
- if (!Mask)
- Mask = ConstantVector::getSplat(VTy->getElementCount(),
- ConstantInt::getTrue(Load->getContext()));
-
Function *VlsegNFunc = Intrinsic::getOrInsertDeclaration(
Load->getModule(), IntrMaskIds[Factor - 2],
{VecTupTy, Mask->getType(), EVL->getType()});
@@ -22653,16 +22654,12 @@ bool RISCVTargetLowering::lowerInterleavedScalableLoad(
/// into
/// `<vscale x 8 x i32>`. This will resuling a simple unit stride store rather
/// than a segment store, which is more expensive in this case.
-static Value *foldInterleaved2OfConstSplats(IntrinsicInst *InterleaveIntrin,
+static Value *foldInterleaved2OfConstSplats(Value *Op0, Value *Op1,
VectorType *VTy,
const TargetLowering *TLI,
Instruction *VPStore) {
- // We only handle Factor = 2 for now.
- assert(InterleaveIntrin->arg_size() == 2);
- auto *SplatVal0 = dyn_cast_or_null<ConstantInt>(
- getSplatValue(InterleaveIntrin->getArgOperand(0)));
- auto *SplatVal1 = dyn_cast_or_null<ConstantInt>(
- getSplatValue(InterleaveIntrin->getArgOperand(1)));
+ auto *SplatVal0 = dyn_cast_or_null<ConstantInt>(getSplatValue(Op0));
+ auto *SplatVal1 = dyn_cast_or_null<ConstantInt>(getSplatValue(Op1));
if (!SplatVal0 || !SplatVal1)
return nullptr;
@@ -22711,15 +22708,19 @@ static Value *foldInterleaved2OfConstSplats(IntrinsicInst *InterleaveIntrin,
/// <vscale x 32 x i8> %load2, ptr %ptr,
/// %mask,
/// i64 %rvl)
-bool RISCVTargetLowering::lowerInterleavedScalableStore(
- VPIntrinsic *Store, Value *Mask, IntrinsicInst *InterleaveIntrin,
+bool RISCVTargetLowering::lowerInterleavedIntrinsicToVPStore(
+ VPIntrinsic *Store, Value *Mask,
ArrayRef<Value *> InterleaveOperands) const {
+ assert(Mask && "Expect a valid mask");
assert(Store->getIntrinsicID() == Intrinsic::vp_store &&
"Unexpected intrinsic");
const unsigned Factor = InterleaveOperands.size();
- VectorType *VTy = cast<VectorType>(InterleaveOperands[0]->getType());
+ auto *VTy = dyn_cast<ScalableVectorType>(InterleaveOperands[0]->getType());
+ // TODO: Support fixed vectors.
+ if (!VTy)
+ return false;
// FIXME: Should pass alignment attribute from pointer, but vectorizer needs
// to emit it first.
@@ -22731,9 +22732,10 @@ bool RISCVTargetLowering::lowerInterleavedScalableStore(
return false;
if (Factor == 2)
- if (Value *BC =
- foldInterleaved2OfConstSplats(InterleaveIntrin, VTy, this, Store)) {
- InterleaveIntrin->replaceAllUsesWith(BC);
+ if (Value *BC = foldInterleaved2OfConstSplats(
+ InterleaveOperands[0], InterleaveOperands[1], VTy, this, Store)) {
+ // Store is guranteed to be the only user of the interleaved intrinsic.
+ Store->getOperand(0)->replaceAllUsesWith(BC);
return true;
}
@@ -22770,10 +22772,6 @@ bool RISCVTargetLowering::lowerInterleavedScalableStore(
Operands.push_back(StoredVal);
Operands.push_back(Store->getArgOperand(1));
- if (!Mask)
- Mask = ConstantVector::getSplat(VTy->getElementCount(),
- ConstantInt::getTrue(Store->getContext()));
-
Function *VssegNFunc = Intrinsic::getOrInsertDeclaration(
Store->getModule(), IntrMaskIds[Factor - 2],
{VecTupTy, Mask->getType(), EVL->getType()});
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.h b/llvm/lib/Target/RISCV/RISCVISelLowering.h
index f779f4fdc26c1b..e9dd8ff96fa37b 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.h
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.h
@@ -910,14 +910,13 @@ class RISCVTargetLowering : public TargetLowering {
bool lowerInterleaveIntrinsicToStore(
StoreInst *SI, ArrayRef<Value *> InterleaveValues) const override;
- bool lowerInterleavedScalableLoad(
- VPIntrinsic *Load, Value *Mask, IntrinsicInst *DeinterleaveIntrin,
+ bool lowerDeinterleavedIntrinsicToVPLoad(
+ VPIntrinsic *Load, Value *Mask,
ArrayRef<Value *> DeinterleaveRes) const override;
- bool
- lowerInterleavedScalableStore(VPIntrinsic *Store, Value *Mask,
- IntrinsicInst *InterleaveIntrin,
- ArrayRef<Value *> InterleaveOps) const override;
+ bool lowerInterleavedIntrinsicToVPStore(
+ VPIntrinsic *Store, Value *Mask,
+ ArrayRef<Value *> InterleaveOps) const override;
bool supportKCFIBundles() const override { return true; }
>From 3688672841431cb526b2a2e7f347c3d07437ce15 Mon Sep 17 00:00:00 2001
From: Min-Yih Hsu <min.hsu at sifive.com>
Date: Mon, 27 Jan 2025 13:45:25 -0800
Subject: [PATCH 3/3] fixup! Rename test
And add negative test case for fixed vectors.
---
...ess.ll => vp-vector-interleaved-access.ll} | 51 +++++++++++++++++++
1 file changed, 51 insertions(+)
rename llvm/test/CodeGen/RISCV/rvv/{scalable-vectors-interleaved-access.ll => vp-vector-interleaved-access.ll} (93%)
diff --git a/llvm/test/CodeGen/RISCV/rvv/scalable-vectors-interleaved-access.ll b/llvm/test/CodeGen/RISCV/rvv/vp-vector-interleaved-access.ll
similarity index 93%
rename from llvm/test/CodeGen/RISCV/rvv/scalable-vectors-interleaved-access.ll
rename to llvm/test/CodeGen/RISCV/rvv/vp-vector-interleaved-access.ll
index ac254792e167a8..648f6e324c59d8 100644
--- a/llvm/test/CodeGen/RISCV/rvv/scalable-vectors-interleaved-access.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vp-vector-interleaved-access.ll
@@ -615,5 +615,56 @@ define void @not_balanced_store_tree(<vscale x 1 x i32> %v0, <vscale x 2 x i32>
ret void
}
+; We only support scalable vectors for now.
+define {<2 x i32>, <2 x i32>, <2 x i32>, <2 x i32>} @not_scalable_vectors(ptr %ptr, i32 %rvl) {
+; RV32-LABEL: not_scalable_vectors:
+; RV32: # %bb.0:
+; RV32-NEXT: vsetvli zero, a1, e32, m2, ta, ma
+; RV32-NEXT: vle32.v v8, (a0)
+; RV32-NEXT: li a0, 32
+; RV32-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV32-NEXT: vnsrl.wx v12, v8, a0
+; RV32-NEXT: vnsrl.wi v11, v8, 0
+; RV32-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV32-NEXT: vnsrl.wx v10, v11, a0
+; RV32-NEXT: vnsrl.wi v8, v11, 0
+; RV32-NEXT: vnsrl.wx v11, v12, a0
+; RV32-NEXT: vnsrl.wi v9, v12, 0
+; RV32-NEXT: ret
+;
+; RV64-LABEL: not_scalable_vectors:
+; RV64: # %bb.0:
+; RV64-NEXT: slli a1, a1, 32
+; RV64-NEXT: srli a1, a1, 32
+; RV64-NEXT: vsetvli zero, a1, e32, m2, ta, ma
+; RV64-NEXT: vle32.v v8, (a0)
+; RV64-NEXT: li a0, 32
+; RV64-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; RV64-NEXT: vnsrl.wx v12, v8, a0
+; RV64-NEXT: vnsrl.wi v11, v8, 0
+; RV64-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; RV64-NEXT: vnsrl.wx v10, v11, a0
+; RV64-NEXT: vnsrl.wi v8, v11, 0
+; RV64-NEXT: vnsrl.wx v11, v12, a0
+; RV64-NEXT: vnsrl.wi v9, v12, 0
+; RV64-NEXT: ret
+ %wide.masked.load = call <8 x i32> @llvm.vp.load.v8i32.p0(ptr %ptr, <8 x i1> splat (i1 true), i32 %rvl)
+ %d0 = call { <4 x i32>, <4 x i32> } @llvm.vector.deinterleave2.v8i32(<8 x i32> %wide.masked.load)
+ %d0.0 = extractvalue { <4 x i32>, <4 x i32> } %d0, 0
+ %d0.1 = extractvalue { <4 x i32>, <4 x i32> } %d0, 1
+ %d1 = call { <2 x i32>, <2 x i32> } @llvm.vector.deinterleave2.v4i32(<4 x i32> %d0.0)
+ %t0 = extractvalue { <2 x i32>, <2 x i32> } %d1, 0
+ %t2 = extractvalue { <2 x i32>, <2 x i32> } %d1, 1
+ %d2 = call { <2 x i32>, <2 x i32> } @llvm.vector.deinterleave2.v4i32(<4 x i32> %d0.1)
+ %t1 = extractvalue { <2 x i32>, <2 x i32> } %d2, 0
+ %t3 = extractvalue { <2 x i32>, <2 x i32> } %d2, 1
+
+ %res0 = insertvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } undef, <2 x i32> %t0, 0
+ %res1 = insertvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %res0, <2 x i32> %t1, 1
+ %res2 = insertvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %res1, <2 x i32> %t2, 2
+ %res3 = insertvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %res2, <2 x i32> %t3, 3
+ ret { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %res3
+}
+
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; CHECK: {{.*}}
More information about the llvm-commits
mailing list