Skip to content

Commit 68b2d92

Browse files
LiqinWengLiqinWeng
committed
[VPlan] Use VPWidenIntrinsicRecipe to vp.cast
1 parent 55fcbcd commit 68b2d92

File tree

8 files changed

+83
-44
lines changed

8 files changed

+83
-44
lines changed

‎llvm/include/llvm/IR/VectorBuilder.h

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -99,11 +99,11 @@ class VectorBuilder {
9999
const Twine &Name = Twine());
100100

101101
/// Emit a VP reduction intrinsic call for recurrence kind.
102-
/// \param RdxID The intrinsic ID of llvm.vector.reduce.*
102+
/// \param ID The intrinsic ID of call Intrinsic
103103
/// \param ValTy The type of operand which the reduction operation is
104104
/// performed.
105105
/// \param VecOpArray The operand list.
106-
Value *createSimpleReduction(Intrinsic::ID RdxID, Type *ValTy,
106+
Value *createSimpleIntrinsic(Intrinsic::ID ID, Type *ValTy,
107107
ArrayRef<Value *> VecOpArray,
108108
const Twine &Name = Twine());
109109
};

‎llvm/lib/IR/VectorBuilder.cpp

Lines changed: 4 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -60,13 +60,12 @@ Value *VectorBuilder::createVectorInstruction(unsigned Opcode, Type *ReturnTy,
6060
return createVectorInstructionImpl(VPID, ReturnTy, InstOpArray, Name);
6161
}
6262

63-
Value *VectorBuilder::createSimpleReduction(Intrinsic::ID RdxID,
64-
Type *ValTy,
63+
Value *VectorBuilder::createSimpleIntrinsic(Intrinsic::ID ID, Type *ValTy,
6564
ArrayRef<Value *> InstOpArray,
6665
const Twine &Name) {
67-
auto VPID = VPIntrinsic::getForIntrinsic(RdxID);
68-
assert(VPReductionIntrinsic::isVPReduction(VPID) &&
69-
"No VPIntrinsic for this reduction");
66+
auto VPID = VPIntrinsic::getForIntrinsic(ID);
67+
assert(VPIntrinsic::isVPIntrinsic(VPID) &&
68+
"No VPIntrinsic for this Intrinsic");
7069
return createVectorInstructionImpl(VPID, ValTy, InstOpArray, Name);
7170
}
7271

‎llvm/lib/Transforms/Utils/LoopUtils.cpp

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1300,7 +1300,7 @@ Value *llvm::createSimpleReduction(VectorBuilder &VBuilder, Value *Src,
13001300
Type *SrcEltTy = SrcTy->getElementType();
13011301
Value *Iden = getRecurrenceIdentity(Kind, SrcEltTy, Desc.getFastMathFlags());
13021302
Value *Ops[] = {Iden, Src};
1303-
return VBuilder.createSimpleReduction(Id, SrcTy, Ops);
1303+
return VBuilder.createSimpleIntrinsic(Id, SrcTy, Ops);
13041304
}
13051305

13061306
Value *llvm::createReduction(IRBuilderBase &B,
@@ -1343,7 +1343,7 @@ Value *llvm::createOrderedReduction(VectorBuilder &VBuilder,
13431343
Intrinsic::ID Id = getReductionIntrinsicID(RecurKind::FAdd);
13441344
auto *SrcTy = cast<VectorType>(Src->getType());
13451345
Value *Ops[] = {Start, Src};
1346-
return VBuilder.createSimpleReduction(Id, SrcTy, Ops);
1346+
return VBuilder.createSimpleIntrinsic(Id, SrcTy, Ops);
13471347
}
13481348

13491349
void llvm::propagateIRFlags(Value *I, ArrayRef<Value *> VL, Value *OpValue,

‎llvm/lib/Transforms/Vectorize/VPlan.h

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1677,6 +1677,15 @@ class VPWidenIntrinsicRecipe : public VPRecipeWithIRFlags {
16771677
MayWriteToMemory(CI.mayWriteToMemory()),
16781678
MayHaveSideEffects(CI.mayHaveSideEffects()) {}
16791679

1680+
// VPWidenIntrinsicRecipe(CastInst &CI, Intrinsic::ID VectorIntrinsicID,
1681+
// ArrayRef<VPValue *> CallArguments, Type *Ty,
1682+
// DebugLoc DL = {})
1683+
// : VPRecipeWithIRFlags(VPDef::VPWidenIntrinsicSC, CallArguments, CI),
1684+
// VectorIntrinsicID(VectorIntrinsicID), ResultTy(Ty),
1685+
// MayReadFromMemory(CI.mayReadFromMemory()),
1686+
// MayWriteToMemory(CI.mayWriteToMemory()),
1687+
// MayHaveSideEffects(CI.mayHaveSideEffects()) {}
1688+
16801689
VPWidenIntrinsicRecipe(Intrinsic::ID VectorIntrinsicID,
16811690
ArrayRef<VPValue *> CallArguments, Type *Ty,
16821691
bool MayReadFromMemory, bool MayWriteToMemory,

‎llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

Lines changed: 31 additions & 18 deletions
Original file line numberDiff line numberDiff line change
@@ -987,24 +987,37 @@ void VPWidenIntrinsicRecipe::execute(VPTransformState &State) {
987987
Args.push_back(Arg);
988988
}
989989

990-
// Use vector version of the intrinsic.
991-
Module *M = State.Builder.GetInsertBlock()->getModule();
992-
Function *VectorF =
993-
Intrinsic::getOrInsertDeclaration(M, VectorIntrinsicID, TysForDecl);
994-
assert(VectorF && "Can't retrieve vector intrinsic.");
995-
996-
auto *CI = cast_or_null<CallInst>(getUnderlyingValue());
997-
SmallVector<OperandBundleDef, 1> OpBundles;
998-
if (CI)
999-
CI->getOperandBundlesAsDefs(OpBundles);
1000-
1001-
CallInst *V = State.Builder.CreateCall(VectorF, Args, OpBundles);
1002-
1003-
setFlags(V);
1004-
1005-
if (!V->getType()->isVoidTy())
1006-
State.set(this, V);
1007-
State.addMetadata(V, CI);
990+
if (VPIntrinsic::isVPIntrinsic(VectorIntrinsicID)) {
991+
// Use vector version of the vector predicate Intrinsic
992+
IRBuilderBase &BuilderIR = State.Builder;
993+
VectorBuilder VBuilder(BuilderIR);
994+
Value *Mask = BuilderIR.CreateVectorSplat(State.VF, BuilderIR.getTrue());
995+
VBuilder.setMask(Mask).setEVL(Args.back());
996+
Args.pop_back();
997+
Value *VPInst = VBuilder.createSimpleIntrinsic(
998+
VectorIntrinsicID, TysForDecl[0], Args, "vp.call");
999+
if (!VPInst->getType()->isVoidTy())
1000+
State.set(this, VPInst);
1001+
State.addMetadata(VPInst,
1002+
dyn_cast_or_null<Instruction>(getUnderlyingValue()));
1003+
} else {
1004+
// Use vector version of the intrinsic.
1005+
Module *M = State.Builder.GetInsertBlock()->getModule();
1006+
Function *VectorF =
1007+
Intrinsic::getOrInsertDeclaration(M, VectorIntrinsicID, TysForDecl);
1008+
assert(VectorF && "Can't retrieve vector intrinsic.");
1009+
1010+
auto *CI = cast_or_null<CallInst>(getUnderlyingValue());
1011+
SmallVector<OperandBundleDef, 1> OpBundles;
1012+
if (CI)
1013+
CI->getOperandBundlesAsDefs(OpBundles);
1014+
1015+
CallInst *V = State.Builder.CreateCall(VectorF, Args, OpBundles);
1016+
setFlags(V);
1017+
if (!V->getType()->isVoidTy())
1018+
State.set(this, V);
1019+
State.addMetadata(V, CI);
1020+
}
10081021
}
10091022

10101023
InstructionCost VPWidenIntrinsicRecipe::computeCost(ElementCount VF,

‎llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1385,6 +1385,24 @@ static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
13851385
VPValue *NewMask = GetNewMask(Red->getCondOp());
13861386
return new VPReductionEVLRecipe(*Red, EVL, NewMask);
13871387
})
1388+
.Case<VPWidenCastRecipe>(
1389+
[&](VPWidenCastRecipe *CInst) -> VPRecipeBase * {
1390+
auto *CI = cast<CastInst>(CInst->getUnderlyingInstr());
1391+
SmallVector<VPValue *> Ops(CInst->operands());
1392+
Ops.push_back(&EVL);
1393+
Intrinsic::ID VPID =
1394+
VPIntrinsic::getForOpcode(CI->getOpcode());
1395+
if (VPID == Intrinsic::not_intrinsic)
1396+
return nullptr;
1397+
// FIXME: In fact, can we really not pass the
1398+
// underlyingInstr? In this case, how to set the Flag and
1399+
// add metadata in VPWidenIntrinsicRecipe::execute?
1400+
// return new VPWidenIntrinsicRecipe(
1401+
// *CI, VPID, Ops, CI->getType(), CI->getDebugLoc());
1402+
return new VPWidenIntrinsicRecipe(
1403+
VPID, Ops, TypeInfo.inferScalarType(CInst), false,
1404+
false, false);
1405+
})
13881406
.Case<VPWidenSelectRecipe>([&](VPWidenSelectRecipe *Sel) {
13891407
SmallVector<VPValue *> Ops(Sel->operands());
13901408
Ops.push_back(&EVL);

‎llvm/test/Transforms/LoopVectorize/RISCV/inloop-reduction.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -166,7 +166,7 @@ define i32 @add_i16_i32(ptr nocapture readonly %x, i32 %n) {
166166
; IF-EVL-INLOOP-NEXT: [[TMP8:%.*]] = getelementptr inbounds i16, ptr [[X:%.*]], i32 [[TMP7]]
167167
; IF-EVL-INLOOP-NEXT: [[TMP9:%.*]] = getelementptr inbounds i16, ptr [[TMP8]], i32 0
168168
; IF-EVL-INLOOP-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 8 x i16> @llvm.vp.load.nxv8i16.p0(ptr align 2 [[TMP9]], <vscale x 8 x i1> shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i64 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer), i32 [[TMP6]])
169-
; IF-EVL-INLOOP-NEXT: [[VP_CAST:%.*]] = sext <vscale x 8 x i16> [[VP_OP_LOAD]] to <vscale x 8 x i32>
169+
; IF-EVL-INLOOP-NEXT: [[VP_CAST:%.*]] = call <vscale x 8 x i32> @llvm.vp.sext.nxv8i32.nxv8i16(<vscale x 8 x i16> [[VP_OP_LOAD]], <vscale x 8 x i1> shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i64 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer), i32 [[TMP6]])
170170
; IF-EVL-INLOOP-NEXT: [[TMP10:%.*]] = call i32 @llvm.vp.reduce.add.nxv8i32(i32 0, <vscale x 8 x i32> [[VP_CAST]], <vscale x 8 x i1> shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i64 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer), i32 [[TMP6]])
171171
; IF-EVL-INLOOP-NEXT: [[TMP11]] = add i32 [[TMP10]], [[VEC_PHI]]
172172
; IF-EVL-INLOOP-NEXT: [[INDEX_EVL_NEXT]] = add i32 [[TMP6]], [[EVL_BASED_IV]]

‎llvm/test/Transforms/LoopVectorize/RISCV/vplan-vp-cast-intrinsics.ll

Lines changed: 16 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -23,10 +23,10 @@ define void @vp_sext(ptr noalias %a, ptr noalias %b, i64 %N) {
2323
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
2424
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
2525
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
26-
; IF-EVL-NEXT: WIDEN-CAST ir<[[SEXT:%.+]]> = sext ir<[[LD1]]> to i64
26+
; IF-EVL-NEXT: WIDEN-INTRINSIC vp<[[SEXT:%.+]]> = call llvm.vp.sext(ir<[[LD1]]>, vp<[[EVL]]>)
2727
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
2828
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
29-
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[SEXT]]>, vp<[[EVL]]>
29+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, vp<[[SEXT]]>, vp<[[EVL]]>
3030
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
3131
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
3232
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%.+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
@@ -73,10 +73,10 @@ define void @vp_zext(ptr noalias %a, ptr noalias %b, i64 %N) {
7373
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
7474
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
7575
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
76-
; IF-EVL-NEXT: WIDEN-CAST ir<[[ZEXT:%.+]]> = zext ir<[[LD1]]> to i64
76+
; IF-EVL-NEXT: WIDEN-INTRINSIC vp<[[ZEXT:%.+]]> = call llvm.vp.zext(ir<[[LD1]]>, vp<[[EVL]]>)
7777
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
7878
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
79-
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[ZEXT]]>, vp<[[EVL]]>
79+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, vp<[[ZEXT]]>, vp<[[EVL]]>
8080
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
8181
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
8282
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%.+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
@@ -121,10 +121,10 @@ define void @vp_truncate(ptr noalias %a, ptr noalias %b, i64 %N) {
121121
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
122122
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
123123
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
124-
; IF-EVL-NEXT: WIDEN-CAST ir<[[TRUNC:%.+]]> = trunc ir<[[LD1]]> to i16
124+
; IF-EVL-NEXT: WIDEN-INTRINSIC vp<[[TRUNC:%.+]]> = call llvm.vp.trunc(ir<[[LD1]]>, vp<[[EVL]]>)
125125
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
126126
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
127-
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[TRUNC]]>, vp<[[EVL]]>
127+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, vp<[[TRUNC]]>, vp<[[EVL]]>
128128
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
129129
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
130130
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%.+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
@@ -169,10 +169,10 @@ define void @vp_fpext(ptr noalias %a, ptr noalias %b, i64 %N) {
169169
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
170170
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
171171
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
172-
; IF-EVL-NEXT: WIDEN-CAST ir<[[FPEXT:%.+]]> = fpext ir<[[LD1]]> to
172+
; IF-EVL-NEXT: WIDEN-INTRINSIC vp<[[FPEXT:%.+]]> = call llvm.vp.fpext(ir<[[LD1]]>, vp<[[EVL]]>)
173173
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
174174
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
175-
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[FPEXT]]>, vp<[[EVL]]>
175+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, vp<[[FPEXT]]>, vp<[[EVL]]>
176176
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
177177
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
178178
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%.+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
@@ -217,10 +217,10 @@ define void @vp_fptrunct(ptr noalias %a, ptr noalias %b, i64 %N) {
217217
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
218218
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
219219
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
220-
; IF-EVL-NEXT: WIDEN-CAST ir<[[FPTRUNC:%.+]]> = fptrunc ir<[[LD1]]> to float
220+
; IF-EVL-NEXT: WIDEN-INTRINSIC vp<[[FPTRUNC:%.+]]> = call llvm.vp.fptrunc(ir<[[LD1]]>, vp<[[EVL]]>)
221221
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
222222
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
223-
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[FPTRUNC]]>, vp<[[EVL]]>
223+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, vp<[[FPTRUNC]]>, vp<[[EVL]]>
224224
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
225225
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
226226
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%.+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
@@ -265,10 +265,10 @@ define void @vp_fptosi(ptr noalias %a, ptr noalias %b, i64 %N) {
265265
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
266266
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
267267
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
268-
; IF-EVL-NEXT: WIDEN-CAST ir<[[FPTOSI:%.+]]> = fptoui ir<[[LD1]]> to i32
268+
; IF-EVL-NEXT: WIDEN-INTRINSIC vp<[[FPTOSI:%.+]]> = call llvm.vp.fptoui(ir<[[LD1]]>, vp<[[EVL]]>)
269269
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
270270
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
271-
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[FPTOSI]]>, vp<[[EVL]]>
271+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, vp<[[FPTOSI]]>, vp<[[EVL]]>
272272
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
273273
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
274274
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%.+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
@@ -313,10 +313,10 @@ define void @vp_inttofp(ptr noalias %a, ptr noalias %b, i64 %N) {
313313
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
314314
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
315315
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
316-
; IF-EVL-NEXT: WIDEN-CAST ir<[[SITOFP:%.+]]> = sitofp ir<[[LD1]]> to float
316+
; IF-EVL-NEXT: WIDEN-INTRINSIC vp<[[SITOFP:%.+]]> = call llvm.vp.sitofp(ir<[[LD1]]>, vp<[[EVL]]>)
317317
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
318318
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
319-
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[SITOFP]]>, vp<[[EVL]]>
319+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, vp<[[SITOFP]]>, vp<[[EVL]]>
320320
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
321321
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
322322
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%.+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
@@ -361,10 +361,10 @@ define void @vp_uinttofp(ptr noalias %a, ptr noalias %b, i64 %N) {
361361
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
362362
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
363363
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
364-
; IF-EVL-NEXT: WIDEN-CAST ir<[[UITOFP:%.+]]> = uitofp ir<[[LD1]]> to float
364+
; IF-EVL-NEXT: WIDEN-INTRINSIC vp<[[UITOFP:%.+]]> = call llvm.vp.uitofp(ir<[[LD1]]>, vp<[[EVL]]>)
365365
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
366366
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
367-
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[UITOFP]]>, vp<[[EVL]]>
367+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, vp<[[UITOFP]]>, vp<[[EVL]]>
368368
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
369369
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
370370
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%.+]]> = add vp<[[IV]]>, vp<[[VFUF]]>

0 commit comments

Comments
 (0)