Skip to content

Commit 8081249

Browse files
LiqinWengLiqinWeng
committed
fix the comments
1. add the mask Op in VPlanTransform
1 parent a0bfd0e commit 8081249

File tree

3 files changed

+22
-28
lines changed

3 files changed

+22
-28
lines changed

‎llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp‎

Lines changed: 3 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -960,20 +960,6 @@ void VPWidenIntrinsicRecipe::execute(VPTransformState &State) {
960960
Args.push_back(Arg);
961961
}
962962

963-
if (VPIntrinsic::isVPIntrinsic(VectorIntrinsicID)) {
964-
Value *EVL = Args.back();
965-
Args.pop_back();
966-
// Add EVL && Mask Ops for vector-predication intrinsics.
967-
if (VPIntrinsic::getMaskParamPos(VectorIntrinsicID)) {
968-
Value *Mask =
969-
State.Builder.CreateVectorSplat(State.VF, State.Builder.getTrue());
970-
Args.push_back(Mask);
971-
}
972-
if (VPIntrinsic::getVectorLengthParamPos(VectorIntrinsicID)) {
973-
Args.push_back(EVL);
974-
}
975-
}
976-
977963
// Use vector version of the intrinsic.
978964
Module *M = State.Builder.GetInsertBlock()->getModule();
979965
Function *VectorF =
@@ -1023,9 +1009,9 @@ InstructionCost VPWidenIntrinsicRecipe::computeCost(ElementCount VF,
10231009
VPIntrinsic::getFunctionalIntrinsicIDForVP(VectorIntrinsicID);
10241010
if (ID) {
10251011
FID = ID.value();
1026-
NumOperands = getNumOperands() - 1;
1027-
// Remove the EVL from arg_operands
1028-
arg_operands = make_range(op_begin(), op_begin() + getNumOperands() - 1);
1012+
NumOperands = getNumOperands() - 2;
1013+
// Remove the Mask && EVL from arg_operands
1014+
arg_operands = make_range(op_begin(), op_begin() + getNumOperands() - 2);
10291015
}
10301016
}
10311017

‎llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp‎

Lines changed: 10 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1490,9 +1490,17 @@ static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
14901490
return nullptr;
14911491

14921492
SmallVector<VPValue *> Ops(CInst->operands());
1493-
Ops.push_back(&EVL);
1493+
if (VPIntrinsic::getMaskParamPos(VPID)) {
1494+
VPValue *Mask = Plan.getOrAddLiveIn(ConstantInt::getTrue(
1495+
IntegerType::getInt1Ty(CI->getContext())));
1496+
Ops.push_back(Mask);
1497+
}
1498+
if (VPIntrinsic::getVectorLengthParamPos(VPID)) {
1499+
Ops.push_back(&EVL);
1500+
}
14941501
return new VPWidenIntrinsicRecipe(
1495-
*CI, VPID, Ops, CI->getType(), CI->getDebugLoc());
1502+
*CI, VPID, Ops, TypeInfo.inferScalarType(CInst),
1503+
CInst->getDebugLoc());
14961504
})
14971505
.Case<VPWidenSelectRecipe>([&](VPWidenSelectRecipe *Sel) {
14981506
SmallVector<VPValue *> Ops(Sel->operands());

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

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -27,7 +27,7 @@ define void @vp_smax(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) {
2727
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%c>, vp<[[ST]]>
2828
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
2929
; IF-EVL-NEXT: WIDEN ir<[[LD2:%.+]]> = vp.load vp<[[PTR2]]>, vp<[[EVL]]>
30-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[SMAX:%.+]]> = call llvm.vp.smax(ir<[[LD1]]>, ir<[[LD2]]>, vp<[[EVL]]>)
30+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[SMAX:%.+]]> = call llvm.vp.smax(ir<[[LD1]]>, ir<[[LD2]]>, ir<true>, vp<[[EVL]]>)
3131
; IF-EVL-NEXT: CLONE ir<[[GEP3:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
3232
; IF-EVL-NEXT: vp<[[PTR3:%[0-9]+]]> = vector-pointer ir<[[GEP3]]>
3333
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR3]]>, ir<[[SMAX]]>, vp<[[EVL]]>
@@ -80,7 +80,7 @@ define void @vp_smin(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) {
8080
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%c>, vp<[[ST]]>
8181
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
8282
; IF-EVL-NEXT: WIDEN ir<[[LD2:%.+]]> = vp.load vp<[[PTR2]]>, vp<[[EVL]]>
83-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[SMIN:%.+]]> = call llvm.vp.smin(ir<[[LD1]]>, ir<[[LD2]]>, vp<[[EVL]]>)
83+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[SMIN:%.+]]> = call llvm.vp.smin(ir<[[LD1]]>, ir<[[LD2]]>, ir<true>, vp<[[EVL]]>)
8484
; IF-EVL-NEXT: CLONE ir<[[GEP3:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
8585
; IF-EVL-NEXT: vp<[[PTR3:%[0-9]+]]> = vector-pointer ir<[[GEP3]]>
8686
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR3]]>, ir<[[SMIN]]>, vp<[[EVL]]>
@@ -133,7 +133,7 @@ define void @vp_umax(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) {
133133
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%c>, vp<[[ST]]>
134134
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
135135
; IF-EVL-NEXT: WIDEN ir<[[LD2:%.+]]> = vp.load vp<[[PTR2]]>, vp<[[EVL]]>
136-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[UMAX:%.+]]> = call llvm.vp.umax(ir<[[LD1]]>, ir<[[LD2]]>, vp<[[EVL]]>)
136+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[UMAX:%.+]]> = call llvm.vp.umax(ir<[[LD1]]>, ir<[[LD2]]>, ir<true>, vp<[[EVL]]>)
137137
; IF-EVL-NEXT: CLONE ir<[[GEP3:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
138138
; IF-EVL-NEXT: vp<[[PTR3:%[0-9]+]]> = vector-pointer ir<[[GEP3]]>
139139
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR3]]>, ir<[[UMAX]]>, vp<[[EVL]]>
@@ -186,7 +186,7 @@ define void @vp_umin(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) {
186186
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%c>, vp<[[ST]]>
187187
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
188188
; IF-EVL-NEXT: WIDEN ir<[[LD2:%.+]]> = vp.load vp<[[PTR2]]>, vp<[[EVL]]>
189-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[UMIN:%.+]]> = call llvm.vp.umin(ir<[[LD1]]>, ir<[[LD2]]>, vp<[[EVL]]>)
189+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[UMIN:%.+]]> = call llvm.vp.umin(ir<[[LD1]]>, ir<[[LD2]]>, ir<true>, vp<[[EVL]]>)
190190
; IF-EVL-NEXT: CLONE ir<[[GEP3:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
191191
; IF-EVL-NEXT: vp<[[PTR3:%[0-9]+]]> = vector-pointer ir<[[GEP3]]>
192192
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR3]]>, ir<[[UMIN]]>, vp<[[EVL]]>
@@ -237,7 +237,7 @@ define void @vp_ctlz(ptr noalias %a, ptr noalias %b, i64 %N) {
237237
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
238238
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
239239
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
240-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[CTLZ:%.+]]> = call llvm.vp.ctlz(ir<[[LD1]]>, ir<true>, vp<[[EVL]]>)
240+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[CTLZ:%.+]]> = call llvm.vp.ctlz(ir<[[LD1]]>, ir<true>, ir<true>, vp<[[EVL]]>)
241241
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
242242
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
243243
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[CTLZ]]>, vp<[[EVL]]>
@@ -285,7 +285,7 @@ define void @vp_cttz(ptr noalias %a, ptr noalias %b, i64 %N) {
285285
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
286286
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
287287
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
288-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[CTTZ:%.+]]> = call llvm.vp.cttz(ir<[[LD1]]>, ir<true>, vp<[[EVL]]>)
288+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[CTTZ:%.+]]> = call llvm.vp.cttz(ir<[[LD1]]>, ir<true>, ir<true>, vp<[[EVL]]>)
289289
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
290290
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
291291
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[CTTZ]]>, vp<[[EVL]]>
@@ -334,7 +334,7 @@ define void @vp_lrint(ptr noalias %a, ptr noalias %b, i64 %N) {
334334
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
335335
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
336336
; IF-EVL-NEXT: WIDEN-CAST ir<[[FPEXT:%.+]]> = fpext ir<[[LD1]]> to double
337-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[LRINT:%.+]]> = call llvm.vp.lrint(ir<[[FPEXT]]>, vp<[[EVL]]>)
337+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[LRINT:%.+]]> = call llvm.vp.lrint(ir<[[FPEXT]]>, ir<true>, vp<[[EVL]]>)
338338
; IF-EVL-NEXT: WIDEN-CAST ir<[[TRUNC:%.+]]> = trunc ir<[[LRINT]]> to i32
339339
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
340340
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
@@ -386,7 +386,7 @@ define void @vp_llrint(ptr noalias %a, ptr noalias %b, i64 %N) {
386386
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
387387
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
388388
; IF-EVL-NEXT: WIDEN-CAST ir<[[FPEXT:%.+]]> = fpext ir<[[LD1]]> to double
389-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[LLRINT:%.+]]> = call llvm.vp.llrint(ir<[[FPEXT]]>, vp<[[EVL]]>)
389+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[LLRINT:%.+]]> = call llvm.vp.llrint(ir<[[FPEXT]]>, ir<true>, vp<[[EVL]]>)
390390
; IF-EVL-NEXT: WIDEN-CAST ir<[[TRUNC:%.+]]> = trunc ir<[[LLRINT]]> to i32
391391
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
392392
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
@@ -437,7 +437,7 @@ define void @vp_abs(ptr noalias %a, ptr noalias %b, i64 %N) {
437437
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
438438
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
439439
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
440-
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[ABS:%.+]]> = call llvm.vp.abs(ir<[[LD1]]>, ir<true>, vp<[[EVL]]>)
440+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[ABS:%.+]]> = call llvm.vp.abs(ir<[[LD1]]>, ir<true>, ir<true>, vp<[[EVL]]>)
441441
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
442442
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
443443
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR2]]>, ir<[[ABS]]>, vp<[[EVL]]>

0 commit comments

Comments
 (0)