Skip to content

Commit ac8f0bb

Browse files
topperctopperc
committed
[RISCV] Reduce ManualCodeGen for segment load/store intrinsics. NFC
Operate directly on the existing Ops vector instead of copying to a new vector. This is similar to what the autogenerated codegen does for other intrinsics. This reduced the clang binary size by ~96kb on my local Release+Asserts build.
1 parent 46343ca commit ac8f0bb

File tree

1 file changed

+72
-115
lines changed

1 file changed

+72
-115
lines changed

‎clang/include/clang/Basic/riscv_vector.td

Lines changed: 72 additions & 115 deletions
Original file line numberDiff line numberDiff line change
@@ -721,8 +721,6 @@ multiclass RVVUnitStridedSegLoadTuple<string op> {
721721
NF = nf,
722722
ManualCodegen = [{
723723
{
724-
SmallVector<llvm::Value*, 6> Operands;
725-
726724
bool NoPassthru =
727725
(IsMasked && (PolicyAttrs & RVV_VTA) && (PolicyAttrs & RVV_VMA)) |
728726
(!IsMasked && (PolicyAttrs & RVV_VTA));
@@ -733,24 +731,18 @@ multiclass RVVUnitStridedSegLoadTuple<string op> {
733731
else
734732
IntrinsicTypes = {ResultType, Ops[Offset]->getType(), Ops.back()->getType()};
735733

736-
if (NoPassthru) { // Push poison into passthru
737-
Operands.push_back(llvm::PoisonValue::get(ResultType));
738-
} else { // Push intrinsics operands into passthru
739-
llvm::Value *PassthruOperand = IsMasked ? Ops[1] : Ops[0];
740-
Operands.push_back(PassthruOperand);
741-
}
742-
743-
Operands.push_back(Ops[Offset]); // Ptr
744734
if (IsMasked)
745-
Operands.push_back(Ops[0]);
746-
Operands.push_back(Ops[Offset + 1]); // VL
735+
std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end() - 1);
736+
if (NoPassthru)
737+
Ops.insert(Ops.begin(), llvm::PoisonValue::get(ResultType));
738+
747739
if (IsMasked)
748-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), PolicyAttrs));
749-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
740+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), PolicyAttrs));
741+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
750742

751743
llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes);
752744

753-
llvm::Value *LoadValue = Builder.CreateCall(F, Operands, "");
745+
llvm::Value *LoadValue = Builder.CreateCall(F, Ops, "");
754746
if (ReturnValue.isNull())
755747
return LoadValue;
756748
else
@@ -787,26 +779,24 @@ multiclass RVVUnitStridedSegStoreTuple<string op> {
787779
{
788780
// Masked
789781
// Builtin: (mask, ptr, v_tuple, vl)
790-
// Intrinsic: (tuple, ptr, mask, vl)
782+
// Intrinsic: (tuple, ptr, mask, vl, SegInstSEW)
791783
// Unmasked
792784
// Builtin: (ptr, v_tuple, vl)
793-
// Intrinsic: (tuple, ptr, vl)
794-
unsigned Offset = IsMasked ? 1 : 0;
785+
// Intrinsic: (tuple, ptr, vl, SegInstSEW)
795786

796-
SmallVector<llvm::Value*, 5> Operands;
797-
Operands.push_back(Ops[Offset + 1]); // tuple
798-
Operands.push_back(Ops[Offset]); // Ptr
799787
if (IsMasked)
800-
Operands.push_back(Ops[0]);
801-
Operands.push_back(Ops[Offset + 2]); // VL
802-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
788+
std::swap(Ops[0], Ops[2]);
789+
else
790+
std::swap(Ops[0], Ops[1]);
791+
792+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
803793

804794
if (IsMasked)
805-
IntrinsicTypes = {Operands[0]->getType(), Ops[Offset]->getType(), Ops[0]->getType(), Operands.back()->getType()};
795+
IntrinsicTypes = {Ops[0]->getType(), Ops[1]->getType(), Ops[2]->getType(), Ops[3]->getType()};
806796
else
807-
IntrinsicTypes = {Operands[0]->getType(), Ops[Offset]->getType(), Operands.back()->getType()};
808-
llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes);
809-
return Builder.CreateCall(F, Operands, "");
797+
IntrinsicTypes = {Ops[0]->getType(), Ops[1]->getType(), Ops[2]->getType()};
798+
799+
break;
810800
}
811801
}] in {
812802
defvar T = "(Tuple:" # nf # ")";
@@ -836,8 +826,6 @@ multiclass RVVUnitStridedSegLoadFFTuple<string op> {
836826
NF = nf,
837827
ManualCodegen = [{
838828
{
839-
SmallVector<llvm::Value*, 6> Operands;
840-
841829
bool NoPassthru =
842830
(IsMasked && (PolicyAttrs & RVV_VTA) && (PolicyAttrs & RVV_VMA)) |
843831
(!IsMasked && (PolicyAttrs & RVV_VTA));
@@ -848,24 +836,21 @@ multiclass RVVUnitStridedSegLoadFFTuple<string op> {
848836
else
849837
IntrinsicTypes = {ResultType, Ops.back()->getType(), Ops[Offset]->getType()};
850838

851-
if (NoPassthru) { // Push poison into passthru
852-
Operands.push_back(llvm::PoisonValue::get(ResultType));
853-
} else { // Push intrinsics operands into passthru
854-
llvm::Value *PassthruOperand = IsMasked ? Ops[1] : Ops[0];
855-
Operands.push_back(PassthruOperand);
856-
}
857-
858-
Operands.push_back(Ops[Offset]); // Ptr
859839
if (IsMasked)
860-
Operands.push_back(Ops[0]);
861-
Operands.push_back(Ops[Offset + 2]); // vl
840+
std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end() - 1);
841+
if (NoPassthru)
842+
Ops.insert(Ops.begin(), llvm::PoisonValue::get(ResultType));
843+
862844
if (IsMasked)
863-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), PolicyAttrs));
864-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
845+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), PolicyAttrs));
846+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
847+
848+
Value *NewVL = Ops[2];
849+
Ops.erase(Ops.begin() + 2);
865850

866851
llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes);
867852

868-
llvm::Value *LoadValue = Builder.CreateCall(F, Operands, "");
853+
llvm::Value *LoadValue = Builder.CreateCall(F, Ops, "");
869854
// Get alignment from the new vl operand
870855
clang::CharUnits Align =
871856
CGM.getNaturalPointeeTypeAlignment(E->getArg(Offset + 1)->getType());
@@ -874,7 +859,7 @@ multiclass RVVUnitStridedSegLoadFFTuple<string op> {
874859

875860
// Store new_vl
876861
llvm::Value *V = Builder.CreateExtractValue(LoadValue, 1);
877-
Builder.CreateStore(V, Address(Ops[Offset + 1], V->getType(), Align));
862+
Builder.CreateStore(V, Address(NewVL, V->getType(), Align));
878863

879864
if (ReturnValue.isNull())
880865
return ReturnTuple;
@@ -909,8 +894,6 @@ multiclass RVVStridedSegLoadTuple<string op> {
909894
NF = nf,
910895
ManualCodegen = [{
911896
{
912-
SmallVector<llvm::Value*, 7> Operands;
913-
914897
bool NoPassthru =
915898
(IsMasked && (PolicyAttrs & RVV_VTA) && (PolicyAttrs & RVV_VMA)) |
916899
(!IsMasked && (PolicyAttrs & RVV_VTA));
@@ -921,24 +904,17 @@ multiclass RVVStridedSegLoadTuple<string op> {
921904
else
922905
IntrinsicTypes = {ResultType, Ops[Offset]->getType(), Ops.back()->getType()};
923906

924-
if (NoPassthru) { // Push poison into passthru
925-
Operands.push_back(llvm::PoisonValue::get(ResultType));
926-
} else { // Push intrinsics operands into passthru
927-
llvm::Value *PassthruOperand = IsMasked ? Ops[1] : Ops[0];
928-
Operands.push_back(PassthruOperand);
929-
}
930-
931-
Operands.push_back(Ops[Offset]); // Ptr
932-
Operands.push_back(Ops[Offset + 1]); // Stride
933907
if (IsMasked)
934-
Operands.push_back(Ops[0]);
935-
Operands.push_back(Ops[Offset + 2]); // VL
908+
std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end() - 1);
909+
if (NoPassthru)
910+
Ops.insert(Ops.begin(), llvm::PoisonValue::get(ResultType));
911+
936912
if (IsMasked)
937-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), PolicyAttrs));
938-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
913+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), PolicyAttrs));
914+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
939915

940916
llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes);
941-
llvm::Value *LoadValue = Builder.CreateCall(F, Operands, "");
917+
llvm::Value *LoadValue = Builder.CreateCall(F, Ops, "");
942918

943919
if (ReturnValue.isNull())
944920
return LoadValue;
@@ -977,27 +953,23 @@ multiclass RVVStridedSegStoreTuple<string op> {
977953
{
978954
// Masked
979955
// Builtin: (mask, ptr, stride, v_tuple, vl)
980-
// Intrinsic: (tuple, ptr, stride, mask, vl)
956+
// Intrinsic: (tuple, ptr, stride, mask, vl, SegInstSEW)
981957
// Unmasked
982958
// Builtin: (ptr, stride, v_tuple, vl)
983-
// Intrinsic: (tuple, ptr, stride, vl)
984-
unsigned Offset = IsMasked ? 1 : 0;
959+
// Intrinsic: (tuple, ptr, stride, vl, SegInstSEW)
985960

986-
SmallVector<llvm::Value*, 6> Operands;
987-
Operands.push_back(Ops[Offset + 2]); // tuple
988-
Operands.push_back(Ops[Offset]); // Ptr
989-
Operands.push_back(Ops[Offset + 1]); // Stride
990961
if (IsMasked)
991-
Operands.push_back(Ops[0]);
992-
Operands.push_back(Ops[Offset + 3]); // VL
993-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
962+
std::swap(Ops[0], Ops[3]);
963+
else
964+
std::rotate(Ops.begin(), Ops.begin() + 2, Ops.begin() + 3);
965+
966+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
994967

995968
if (IsMasked)
996-
IntrinsicTypes = {Operands[0]->getType(), Operands[1]->getType(), Operands.back()->getType(), Ops[0]->getType()};
969+
IntrinsicTypes = {Ops[0]->getType(), Ops[1]->getType(), Ops[4]->getType(), Ops[3]->getType()};
997970
else
998-
IntrinsicTypes = {Operands[0]->getType(), Operands[1]->getType(), Operands.back()->getType()};
999-
llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes);
1000-
return Builder.CreateCall(F, Operands, "");
971+
IntrinsicTypes = {Ops[0]->getType(), Ops[1]->getType(), Ops[3]->getType()};
972+
break;
1001973
}
1002974
}] in {
1003975
defvar T = "(Tuple:" # nf # ")";
@@ -1022,40 +994,30 @@ multiclass RVVIndexedSegLoadTuple<string op> {
1022994
NF = nf,
1023995
ManualCodegen = [{
1024996
{
1025-
SmallVector<llvm::Value*, 7> Operands;
1026-
1027997
bool NoPassthru =
1028998
(IsMasked && (PolicyAttrs & RVV_VTA) && (PolicyAttrs & RVV_VMA)) |
1029999
(!IsMasked && (PolicyAttrs & RVV_VTA));
1030-
unsigned Offset = IsMasked ? NoPassthru ? 1 : 2 : NoPassthru ? 0 : 1;
10311000

1032-
if (NoPassthru) { // Push poison into passthru
1033-
Operands.push_back(llvm::PoisonValue::get(ResultType));
1034-
} else { // Push intrinsics operands into passthru
1035-
llvm::Value *PassthruOperand = IsMasked ? Ops[1] : Ops[0];
1036-
Operands.push_back(PassthruOperand);
1037-
}
1038-
1039-
Operands.push_back(Ops[Offset]); // Ptr
1040-
Operands.push_back(Ops[Offset + 1]); // Idx
10411001
if (IsMasked)
1042-
Operands.push_back(Ops[0]);
1043-
Operands.push_back(Ops[Offset + 2]); // VL
1002+
std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end() - 1);
1003+
if (NoPassthru)
1004+
Ops.insert(Ops.begin(), llvm::PoisonValue::get(ResultType));
1005+
10441006
if (IsMasked)
1045-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), PolicyAttrs));
1046-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
1007+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), PolicyAttrs));
1008+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
10471009

10481010
if (IsMasked)
1049-
IntrinsicTypes = {ResultType, Ops[Offset]->getType(),
1050-
Ops[Offset + 1]->getType(),
1051-
Ops[0]->getType(),
1052-
Ops.back()->getType()};
1011+
IntrinsicTypes = {ResultType, Ops[1]->getType(),
1012+
Ops[2]->getType(),
1013+
Ops[3]->getType(),
1014+
Ops[4]->getType()};
10531015
else
1054-
IntrinsicTypes = {ResultType, Ops[Offset]->getType(),
1055-
Ops[Offset + 1]->getType(),
1056-
Ops.back()->getType()};
1016+
IntrinsicTypes = {ResultType, Ops[1]->getType(),
1017+
Ops[2]->getType(),
1018+
Ops[3]->getType()};
10571019
llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes);
1058-
llvm::Value *LoadValue = Builder.CreateCall(F, Operands, "");
1020+
llvm::Value *LoadValue = Builder.CreateCall(F, Ops, "");
10591021

10601022
if (ReturnValue.isNull())
10611023
return LoadValue;
@@ -1090,30 +1052,25 @@ multiclass RVVIndexedSegStoreTuple<string op> {
10901052
{
10911053
// Masked
10921054
// Builtin: (mask, ptr, index, v_tuple, vl)
1093-
// Intrinsic: (tuple, ptr, index, mask, vl)
1055+
// Intrinsic: (tuple, ptr, index, mask, vl, SegInstSEW)
10941056
// Unmasked
10951057
// Builtin: (ptr, index, v_tuple, vl)
1096-
// Intrinsic: (tuple, ptr, index, vl)
1097-
unsigned Offset = IsMasked ? 1 : 0;
1058+
// Intrinsic: (tuple, ptr, index, vl, SegInstSEW)
10981059

1099-
SmallVector<llvm::Value*, 6> Operands;
1100-
Operands.push_back(Ops[Offset + 2]); // tuple
1101-
Operands.push_back(Ops[Offset]); // Ptr
1102-
Operands.push_back(Ops[Offset + 1]); // Idx
11031060
if (IsMasked)
1104-
Operands.push_back(Ops[0]);
1105-
Operands.push_back(Ops[Offset + 3]); // VL
1106-
Operands.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
1061+
std::swap(Ops[0], Ops[3]);
1062+
else
1063+
std::rotate(Ops.begin(), Ops.begin() + 2, Ops.begin() + 3);
1064+
1065+
Ops.push_back(ConstantInt::get(Ops.back()->getType(), SegInstSEW));
11071066

11081067
if (IsMasked)
1109-
IntrinsicTypes = {Operands[0]->getType(), Ops[Offset]->getType(), Ops[Offset + 1]->getType(),
1110-
Ops[0]->getType(),
1111-
Operands.back()->getType()};
1068+
IntrinsicTypes = {Ops[0]->getType(), Ops[1]->getType(), Ops[2]->getType(),
1069+
Ops[3]->getType(), Ops[4]->getType()};
11121070
else
1113-
IntrinsicTypes = {Operands[0]->getType(), Ops[Offset]->getType(), Ops[Offset + 1]->getType(),
1114-
Operands.back()->getType()};
1115-
llvm::Function *F = CGM.getIntrinsic(ID, IntrinsicTypes);
1116-
return Builder.CreateCall(F, Operands, "");
1071+
IntrinsicTypes = {Ops[0]->getType(), Ops[1]->getType(), Ops[2]->getType(),
1072+
Ops[3]->getType()};
1073+
break;
11171074
}
11181075
}] in {
11191076
defvar T = "(Tuple:" # nf # ")";

0 commit comments

Comments
 (0)