[AArch64][Clang] Refactor code to emit SVE & SME builtins #70662
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This patch removes duplicated code in EmitAArch64SVEBuiltinExpr and EmitAArch64SMEBuiltinExpr by creating a new function called GetAArch64SVEProcessedOperands which handles splitting up multi-vector arguments using vector extracts. These changes are non-functional.
kmclaughlin-arm
added
clang
kmclaughlin-arm
requested review from
dtemirbulatov,
bryanpkc,
sdesmalen-arm,
CarolineConcatto,
david-arm and
MDevereau
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-clang-codegen Author: Kerry McLaughlin (kmclaughlin-arm) ChangesThis patch removes duplicated code in EmitAArch64SVEBuiltinExpr and These changes are non-functional. Patch is 247.88 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/70662.diff 8 Files Affected:
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 2b341b8090fad7d..727c33e348cfa7e 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -9571,22 +9571,17 @@ Value *CodeGenFunction::EmitSVEStructStore(const SVETypeFlags &TypeFlags,
Value *BasePtr = Ops[1];
// Does the store have an offset?
- if (Ops.size() > 3)
+ if (Ops.size() > (2 + N))
BasePtr = Builder.CreateGEP(VTy, BasePtr, Ops[2]);
- Value *Val = Ops.back();
-
// The llvm.aarch64.sve.st2/3/4 intrinsics take legal part vectors, so we
// need to break up the tuple vector.
SmallVector<llvm::Value*, 5> Operands;
- unsigned MinElts = VTy->getElementCount().getKnownMinValue();
- for (unsigned I = 0; I < N; ++I) {
- Value *Idx = ConstantInt::get(CGM.Int64Ty, I * MinElts);
- Operands.push_back(Builder.CreateExtractVector(VTy, Val, Idx));
- }
+ for (unsigned I = Ops.size() - N; I < Ops.size(); ++I)
+ Operands.push_back(Ops[I]);
Operands.append({Predicate, BasePtr});
-
Function *F = CGM.getIntrinsic(IntID, { VTy });
+
return Builder.CreateCall(F, Operands);
}
@@ -9893,24 +9888,40 @@ Value *CodeGenFunction::FormSVEBuiltinResult(Value *Call) {
return Call;
}
-Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
- const CallExpr *E) {
+void CodeGenFunction::GetAArch64SVEProcessedOperands(
+ unsigned BuiltinID, const CallExpr *E, SmallVectorImpl<Value *> &Ops,
+ SVETypeFlags TypeFlags) {
// Find out if any arguments are required to be integer constant expressions.
unsigned ICEArguments = 0;
ASTContext::GetBuiltinTypeError Error;
getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
assert(Error == ASTContext::GE_None && "Should not codegen an error");
- llvm::Type *Ty = ConvertType(E->getType());
- if (BuiltinID >= SVE::BI__builtin_sve_reinterpret_s8_s8 &&
- BuiltinID <= SVE::BI__builtin_sve_reinterpret_f64_f64) {
- Value *Val = EmitScalarExpr(E->getArg(0));
- return EmitSVEReinterpret(Val, Ty);
- }
+ // Tuple set/get only requires one insert/extract vector, which is
+ // created by EmitSVETupleSetOrGet.
+ bool IsTupleGetOrSet = TypeFlags.isTupleSet() || TypeFlags.isTupleGet();
- llvm::SmallVector<Value *, 4> Ops;
for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) {
- if ((ICEArguments & (1 << i)) == 0)
+ bool IsICE = ICEArguments & (1 << i);
+ if (!IsTupleGetOrSet && !IsICE) {
+ Value *Arg = EmitScalarExpr(E->getArg(i));
+ if (auto *VTy = dyn_cast<ScalableVectorType>(Arg->getType())) {
+ unsigned MinElts = VTy->getMinNumElements();
+ bool IsPred = VTy->getElementType()->isIntegerTy(1);
+ unsigned N =
+ (MinElts * VTy->getScalarSizeInBits()) / (IsPred ? 16 : 128);
+ for (unsigned I = 0; I < N; ++I) {
+ Value *Idx = ConstantInt::get(CGM.Int64Ty, (I * MinElts) / N);
+ auto *NewVTy =
+ ScalableVectorType::get(VTy->getElementType(), MinElts / N);
+ if (N == 1 && VTy == NewVTy)
+ Ops.push_back(Arg);
+ else
+ Ops.push_back(Builder.CreateExtractVector(NewVTy, Arg, Idx));
+ }
+ } else
+ Ops.push_back(Arg);
+ } else if (!IsICE)
Ops.push_back(EmitScalarExpr(E->getArg(i)));
else {
// If this is required to be a constant, constant fold it so that we know
@@ -9927,9 +9938,25 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
}
}
+ return;
+}
+
+Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
+ const CallExpr *E) {
+ llvm::Type *Ty = ConvertType(E->getType());
+ if (BuiltinID >= SVE::BI__builtin_sve_reinterpret_s8_s8 &&
+ BuiltinID <= SVE::BI__builtin_sve_reinterpret_f64_f64) {
+ Value *Val = EmitScalarExpr(E->getArg(0));
+ return EmitSVEReinterpret(Val, Ty);
+ }
+
auto *Builtin = findARMVectorIntrinsicInMap(AArch64SVEIntrinsicMap, BuiltinID,
AArch64SVEIntrinsicsProvenSorted);
+
+ llvm::SmallVector<Value *, 4> Ops;
SVETypeFlags TypeFlags(Builtin->TypeModifier);
+ GetAArch64SVEProcessedOperands(BuiltinID, E, Ops, TypeFlags);
+
if (TypeFlags.isLoad())
return EmitSVEMaskedLoad(E, Ty, Ops, Builtin->LLVMIntrinsic,
TypeFlags.isZExtReturn());
@@ -9943,14 +9970,14 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
return EmitSVEPrefetchLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic);
else if (TypeFlags.isGatherPrefetch())
return EmitSVEGatherPrefetch(TypeFlags, Ops, Builtin->LLVMIntrinsic);
- else if (TypeFlags.isStructLoad())
- return EmitSVEStructLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic);
- else if (TypeFlags.isStructStore())
- return EmitSVEStructStore(TypeFlags, Ops, Builtin->LLVMIntrinsic);
+ else if (TypeFlags.isStructLoad())
+ return EmitSVEStructLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic);
+ else if (TypeFlags.isStructStore())
+ return EmitSVEStructStore(TypeFlags, Ops, Builtin->LLVMIntrinsic);
else if (TypeFlags.isTupleSet() || TypeFlags.isTupleGet())
- return EmitSVETupleSetOrGet(TypeFlags, Ty, Ops);
+ return EmitSVETupleSetOrGet(TypeFlags, Ty, Ops);
else if (TypeFlags.isTupleCreate())
- return EmitSVETupleCreate(TypeFlags, Ty, Ops);
+ return EmitSVETupleCreate(TypeFlags, Ty, Ops);
else if (TypeFlags.isUndef())
return UndefValue::get(Ty);
else if (Builtin->LLVMIntrinsic != 0) {
@@ -10202,13 +10229,8 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
case SVE::BI__builtin_sve_svtbl2_f64: {
SVETypeFlags TF(Builtin->TypeModifier);
auto VTy = cast<llvm::ScalableVectorType>(getSVEType(TF));
- Value *V0 = Builder.CreateExtractVector(VTy, Ops[0],
- ConstantInt::get(CGM.Int64Ty, 0));
- unsigned MinElts = VTy->getMinNumElements();
- Value *V1 = Builder.CreateExtractVector(
- VTy, Ops[0], ConstantInt::get(CGM.Int64Ty, MinElts));
Function *F = CGM.getIntrinsic(Intrinsic::aarch64_sve_tbl2, VTy);
- return Builder.CreateCall(F, {V0, V1, Ops[1]});
+ return Builder.CreateCall(F, Ops);
}
case SVE::BI__builtin_sve_svset_neonq_s8:
@@ -10266,35 +10288,13 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
Value *CodeGenFunction::EmitAArch64SMEBuiltinExpr(unsigned BuiltinID,
const CallExpr *E) {
- // Find out if any arguments are required to be integer constant expressions.
- unsigned ICEArguments = 0;
- ASTContext::GetBuiltinTypeError Error;
- getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
- assert(Error == ASTContext::GE_None && "Should not codegen an error");
-
- llvm::Type *Ty = ConvertType(E->getType());
- llvm::SmallVector<Value *, 4> Ops;
- for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) {
- if ((ICEArguments & (1 << i)) == 0)
- Ops.push_back(EmitScalarExpr(E->getArg(i)));
- else {
- // If this is required to be a constant, constant fold it so that we know
- // that the generated intrinsic gets a ConstantInt.
- std::optional<llvm::APSInt> Result =
- E->getArg(i)->getIntegerConstantExpr(getContext());
- assert(Result && "Expected argument to be a constant");
-
- // Immediates for SVE llvm intrinsics are always 32bit. We can safely
- // truncate because the immediate has been range checked and no valid
- // immediate requires more than a handful of bits.
- *Result = Result->extOrTrunc(32);
- Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), *Result));
- }
- }
-
auto *Builtin = findARMVectorIntrinsicInMap(AArch64SMEIntrinsicMap, BuiltinID,
AArch64SMEIntrinsicsProvenSorted);
+
+ llvm::SmallVector<Value *, 4> Ops;
SVETypeFlags TypeFlags(Builtin->TypeModifier);
+ GetAArch64SVEProcessedOperands(BuiltinID, E, Ops, TypeFlags);
+
if (TypeFlags.isLoad() || TypeFlags.isStore())
return EmitSMELd1St1(TypeFlags, Ops, Builtin->LLVMIntrinsic);
else if (TypeFlags.isReadZA() || TypeFlags.isWriteZA())
@@ -10307,21 +10307,24 @@ Value *CodeGenFunction::EmitAArch64SMEBuiltinExpr(unsigned BuiltinID,
BuiltinID == SME::BI__builtin_sme_svldr_za ||
BuiltinID == SME::BI__builtin_sme_svstr_za)
return EmitSMELdrStr(TypeFlags, Ops, Builtin->LLVMIntrinsic);
- else if (Builtin->LLVMIntrinsic != 0) {
- // Predicates must match the main datatype.
- for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (auto PredTy = dyn_cast<llvm::VectorType>(Ops[i]->getType()))
- if (PredTy->getElementType()->isIntegerTy(1))
- Ops[i] = EmitSVEPredicateCast(Ops[i], getSVEType(TypeFlags));
- Function *F = CGM.getIntrinsic(Builtin->LLVMIntrinsic,
- getSVEOverloadTypes(TypeFlags, Ty, Ops));
- Value *Call = Builder.CreateCall(F, Ops);
- return Call;
- }
+ // Should not happen!
+ if (Builtin->LLVMIntrinsic == 0)
+ return nullptr;
- /// Should not happen
- return nullptr;
+ // Predicates must match the main datatype.
+ for (unsigned i = 0, e = Ops.size(); i != e; ++i)
+ if (auto PredTy = dyn_cast<llvm::VectorType>(Ops[i]->getType()))
+ if (PredTy->getElementType()->isIntegerTy(1))
+ Ops[i] = EmitSVEPredicateCast(Ops[i], getSVEType(TypeFlags));
+
+ Function *F =
+ TypeFlags.isOverloadNone()
+ ? CGM.getIntrinsic(Builtin->LLVMIntrinsic)
+ : CGM.getIntrinsic(Builtin->LLVMIntrinsic, {getSVEType(TypeFlags)});
+ Value *Call = Builder.CreateCall(F, Ops);
+
+ return FormSVEBuiltinResult(Call);
}
Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
index e82115e2d706cf1..ee2c4b1e10afba5 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -4311,6 +4311,11 @@ class CodeGenFunction : public CodeGenTypeCache {
llvm::Value *EmitSMELdrStr(const SVETypeFlags &TypeFlags,
llvm::SmallVectorImpl<llvm::Value *> &Ops,
unsigned IntID);
+
+ void GetAArch64SVEProcessedOperands(unsigned BuiltinID, const CallExpr *E,
+ SmallVectorImpl<llvm::Value *> &Ops,
+ SVETypeFlags TypeFlags);
+
llvm::Value *EmitAArch64SMEBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
llvm::Value *EmitAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
diff --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2-bfloat.c
index abe59567b038719..89a7d011293682a 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2-bfloat.c
@@ -16,18 +16,18 @@
#endif
// CHECK-LABEL: @test_svst2_bf16(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA:%.*]], i64 0)
-// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA]], i64 8)
-// CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8bf16(<vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x i1> [[TMP2]], ptr [[BASE:%.*]])
// CHECK-NEXT: ret void
//
// CPP-CHECK-LABEL: @_Z15test_svst2_bf16u10__SVBool_tPu6__bf1614svbfloat16x2_t(
// CPP-CHECK-NEXT: entry:
-// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA:%.*]], i64 0)
-// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA]], i64 8)
-// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8bf16(<vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x i1> [[TMP2]], ptr [[BASE:%.*]])
// CPP-CHECK-NEXT: ret void
//
void test_svst2_bf16(svbool_t pg, bfloat16_t *base, svbfloat16x2_t data)
@@ -37,20 +37,20 @@ void test_svst2_bf16(svbool_t pg, bfloat16_t *base, svbfloat16x2_t data)
// CHECK-LABEL: @test_svst2_vnum_bf16(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
-// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA:%.*]], i64 0)
-// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA]], i64 8)
-// CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8bf16(<vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT: [[TMP3:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
+// CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x i1> [[TMP2]], ptr [[TMP3]])
// CHECK-NEXT: ret void
//
// CPP-CHECK-LABEL: @_Z20test_svst2_vnum_bf16u10__SVBool_tPu6__bf16l14svbfloat16x2_t(
// CPP-CHECK-NEXT: entry:
-// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
-// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA:%.*]], i64 0)
-// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA]], i64 8)
-// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8bf16(<vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[DATA]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x i1> [[TMP2]], ptr [[TMP3]])
// CPP-CHECK-NEXT: ret void
//
void test_svst2_vnum_bf16(svbool_t pg, bfloat16_t *base, int64_t vnum, svbfloat16x2_t data)
diff --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2.c
index 6a6632d51b1c048..7848cbc5d9abcc2 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2.c
@@ -35,18 +35,18 @@ void test_svst2_s8(svbool_t pg, int8_t *base, svint8x2_t data)
// CHECK-LABEL: @test_svst2_s16(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[DATA:%.*]], i64 0)
-// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[DATA]], i64 8)
-// CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8i16(<vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[DATA:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[DATA]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i1> [[TMP2]], ptr [[BASE:%.*]])
// CHECK-NEXT: ret void
//
// CPP-CHECK-LABEL: @_Z14test_svst2_s16u10__SVBool_tPs11svint16x2_t(
// CPP-CHECK-NEXT: entry:
-// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
-// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[DATA:%.*]], i64 0)
-// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[DATA]], i64 8)
-// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8i16(<vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[DATA:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[DATA]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sve.st2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i1> [[TMP2]], ptr [[BASE:%.*]])
// CPP-CHECK-NEXT: ret void
//
void test_svst2_s16(svbool_t pg, int16_t *base, svint16x2_t data)
@@ -56,18 +56,18 @@ void test_svst2_s16(svbool_t pg, int16_t *base, svint16x2_t data)
// CHECK-LABEL: @test_svst2_s32(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> ...
[truncated]
|
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { | ||
| if ((ICEArguments & (1 << i)) == 0) | ||
| bool IsICE = ICEArguments & (1 << i); | ||
| if (!IsTupleGetOrSet && !IsICE) { |
There was a problem hiding this comment.
nit: While we're refactoring this, can this loop be rewritten a bit, e.g.
for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) {
bool IsICE = ICEArguments & (1 << i);
Value *Arg = EmitScalarExpr(E->getArg(i));
if (!isa<ScalableVectorType>(Arg->getType())) {
if (IsICE || !IsTupleGetOrSet)
Ops.push_back(Arg);
continue;
}
auto *VTy = cast<ScalableVectorType>(Arg->getType())
/* rest of the code here */
}
There was a problem hiding this comment.
Hi @sdesmalen-arm, I've tried to rewrite this in such a way that we only have to check flags such as IsICE & IsTupleGetOrSet once
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| Value *Idx = ConstantInt::get(CGM.Int64Ty, (I * MinElts) / N); | ||
| auto *NewVTy = | ||
| ScalableVectorType::get(VTy->getElementType(), MinElts / N); | ||
| if (N == 1 && VTy == NewVTy) |
There was a problem hiding this comment.
Is VTy == NewVTy implied if N == 1? If so, can we hoist the N == 1 check out of the loop?
There was a problem hiding this comment.
I don't think both checks are needed, so this can be hoisted out
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| } | ||
|
|
||
| if (IsTupleGetOrSet) { | ||
| Ops.push_back(EmitScalarExpr(E->getArg(i))); |
There was a problem hiding this comment.
This is the same as Ops.push_back(Arg); ? And if so, you can combine it with the case below.
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| } | ||
| } | ||
|
|
||
| return; |
There was a problem hiding this comment.
return is unnecessary for a function returning void.
- Removed return from GetAArch64SVEProcessedOperands
sdesmalen-arm
approved these changes
Nov 1, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This patch removes duplicated code in EmitAArch64SVEBuiltinExpr and
EmitAArch64SMEBuiltinExpr by creating a new function called
GetAArch64SVEProcessedOperands which handles splitting up multi-vector
arguments using vector extracts.
These changes are non-functional.