[Clang][SME2] Add multi-vector add/sub builtins #69725
Merged
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
kmclaughlin-arm
added
clang
kmclaughlin-arm
requested review from
dtemirbulatov,
bryanpkc,
sdesmalen-arm,
CarolineConcatto,
david-arm and
MDevereau
llvmbot
added
the
clang:frontend
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-clang-codegen Author: Kerry McLaughlin (kmclaughlin-arm) ChangesAdds the following SME2 builtins:
Other changes in this patch:
See https://github.com/ARM-software/acle/pull/217/files Patch is 405.53 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/69725.diff 14 Files Affected:
diff --git a/clang/include/clang/Basic/arm_sme.td b/clang/include/clang/Basic/arm_sme.td
index 8d85327a86b1aaf..b5655afdf419ecf 100644
--- a/clang/include/clang/Basic/arm_sme.td
+++ b/clang/include/clang/Basic/arm_sme.td
@@ -263,3 +263,38 @@ multiclass ZAFPOuterProd<string n_suffix> {
defm SVMOPA : ZAFPOuterProd<"mopa">;
defm SVMOPS : ZAFPOuterProd<"mops">;
+
+////////////////////////////////////////////////////////////////////////////////
+// SME2 - ADD, SUB
+
+multiclass ZAAddSub<string n_suffix> {
+ let TargetGuard = "sme2" in {
+ def NAME # _WRITE_SINGLE_ZA32_VG1X2_I32 : Inst<"sv" # n_suffix # "_write[_single]_za32[_{d}]_vg1x2", "vm2d", "iUi", MergeNone, "aarch64_sme_" # n_suffix # "_write_single_za_vg1x2", [IsStreaming, IsSharedZA], []>;
+ def NAME # _WRITE_SINGLE_ZA32_VG1X4_I32 : Inst<"sv" # n_suffix # "_write[_single]_za32[_{d}]_vg1x4", "vm4d", "iUi", MergeNone, "aarch64_sme_" # n_suffix # "_write_single_za_vg1x4", [IsStreaming, IsSharedZA], []>;
+
+ def NAME # _WRITE_ZA32_VG1X2_I32 : Inst<"sv" # n_suffix # "_write_za32[_{d}]_vg1x2", "vm22", "iUi", MergeNone, "aarch64_sme_" # n_suffix # "_write_za_vg1x2", [IsStreaming, IsSharedZA], []>;
+ def NAME # _WRITE_ZA32_VG1X4_I32 : Inst<"sv" # n_suffix # "_write_za32[_{d}]_vg1x4", "vm44", "iUi", MergeNone, "aarch64_sme_" # n_suffix # "_write_za_vg1x4", [IsStreaming, IsSharedZA], []>;
+
+ def NAME # _ZA32_VG1x2_I32 : Inst<"sv" # n_suffix # "_za32[_{d}]_vg1x2", "vm2", "iUif", MergeNone, "aarch64_sme_" # n_suffix # "_za32_vg1x2", [IsStreaming, IsSharedZA], []>;
+ def NAME # _ZA32_VG1X4_I32 : Inst<"sv" # n_suffix # "_za32[_{d}]_vg1x4", "vm4", "iUif", MergeNone, "aarch64_sme_" # n_suffix # "_za32_vg1x4", [IsStreaming, IsSharedZA], []>;
+
+ let TargetGuard = "sme-i16i64" in {
+ def NAME # _WRITE_SINGLE_ZA64_VG1X2_I64 : Inst<"sv" # n_suffix # "_write[_single]_za64[_{d}]_vg1x2", "vm2d", "lUl", MergeNone, "aarch64_sme_" # n_suffix # "_write_single_za_vg1x2", [IsStreaming, IsSharedZA], []>;
+ def NAME # _WRITE_SINGLE_ZA64_VG1X4_I64 : Inst<"sv" # n_suffix # "_write[_single]_za64[_{d}]_vg1x4", "vm4d", "lUl", MergeNone, "aarch64_sme_" # n_suffix # "_write_single_za_vg1x4", [IsStreaming, IsSharedZA], []>;
+
+ def NAME # _WRITE_ZA64_VG1x2_I64 : Inst<"sv" # n_suffix # "_write_za64[_{d}]_vg1x2", "vm22", "lUl", MergeNone, "aarch64_sme_" # n_suffix # "_write_za_vg1x2", [IsStreaming, IsSharedZA], []>;
+ def NAME # _WRITE_ZA64_VG1x4_I64 : Inst<"sv" # n_suffix # "_write_za64[_{d}]_vg1x4", "vm44", "lUl", MergeNone, "aarch64_sme_" # n_suffix # "_write_za_vg1x4", [IsStreaming, IsSharedZA], []>;
+
+ def NAME # _ZA64_VG1X2_I64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x2", "vm2", "lUl", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x2", [IsStreaming, IsSharedZA], []>;
+ def NAME # _ZA64_VG1X4_I64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x4", "vm4", "lUl", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x4", [IsStreaming, IsSharedZA], []>;
+ }
+
+ let TargetGuard = "sme-f64f64" in {
+ def NAME # _ZA64_VG1X2_F64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x2", "vm2", "d", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x2", [IsStreaming, IsSharedZA], []>;
+ def NAME # _ZA64_VG1X4_F64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x4", "vm4", "d", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x4", [IsStreaming, IsSharedZA], []>;
+ }
+ }
+}
+
+defm SVADD : ZAAddSub<"add">;
+defm SVSUB : ZAAddSub<"sub">;
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 8f9bdd18829ff6b..a0b0ba4f2976052 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -1893,3 +1893,12 @@ def SVPSEL_D : SInst<"svpsel_lane_b64", "PPPm", "Pl", MergeNone, "", [], []>;
def SVCNTP_COUNT : SInst<"svcntp_{d}", "n}i", "QcQsQiQl", MergeNone, "aarch64_sve_cntp_{d}", [IsOverloadNone], [ImmCheck<1, ImmCheck2_4_Mul2>]>;
}
+
+////////////////////////////////////////////////////////////////////////////////
+// SME2
+
+let TargetGuard = "sme2" in {
+// == ADD (vectors) ==
+ def SVADD_SINGLE_X2 : SInst<"svadd[_single_{d}_x2]", "22d", "cUcsUsiUilUl", MergeNone, "aarch64_sve_add_single_x2", [IsStreaming], []>;
+ def SVADD_SINGLE_X4 : SInst<"svadd[_single_{d}_x4]", "44d", "cUcsUsiUilUl", MergeNone, "aarch64_sve_add_single_x4", [IsStreaming], []>;
+}
diff --git a/clang/include/clang/Basic/arm_sve_sme_incl.td b/clang/include/clang/Basic/arm_sve_sme_incl.td
index c3a6dc4e4d44abe..a484ed19efc7e0b 100644
--- a/clang/include/clang/Basic/arm_sve_sme_incl.td
+++ b/clang/include/clang/Basic/arm_sve_sme_incl.td
@@ -256,7 +256,7 @@ class ImmCheck<int arg, ImmCheckType kind, int eltSizeArg = -1> {
}
class Inst<string n, string p, string t, MergeType mt, string i,
- list<FlagType> ft, list<ImmCheck> ch, MemEltType met> {
+ list<FlagType> ft, list<ImmCheck> ch, MemEltType met = MemEltTyDefault> {
string Name = n;
string Prototype = p;
string Types = t;
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 2b341b8090fad7d..59a31505b8993fa 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 +9888,37 @@ Value *CodeGenFunction::FormSVEBuiltinResult(Value *Call) {
return Call;
}
-Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
- const CallExpr *E) {
+void CodeGenFunction::GetAArch64SMEProcessedOperands(
+ 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);
- }
+ 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)
+ if (!IsTupleGetOrSet && (ICEArguments & (1 << i)) == 0) {
+ 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 ((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
@@ -9927,9 +9935,25 @@ 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);
+ GetAArch64SMEProcessedOperands(BuiltinID, E, Ops, TypeFlags);
+
if (TypeFlags.isLoad())
return EmitSVEMaskedLoad(E, Ty, Ops, Builtin->LLVMIntrinsic,
TypeFlags.isZExtReturn());
@@ -9943,10 +9967,10 @@ 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);
else if (TypeFlags.isTupleCreate())
@@ -10202,13 +10226,8 @@ 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:
@@ -10272,29 +10291,13 @@ Value *CodeGenFunction::EmitAArch64SMEBuiltinExpr(unsigned BuiltinID,
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);
+ GetAArch64SMEProcessedOperands(BuiltinID, E, Ops, TypeFlags);
+
if (TypeFlags.isLoad() || TypeFlags.isStore())
return EmitSMELd1St1(TypeFlags, Ops, Builtin->LLVMIntrinsic);
else if (TypeFlags.isReadZA() || TypeFlags.isWriteZA())
@@ -10307,21 +10310,24 @@ 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..ed50b051e77f96e 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -4297,6 +4297,11 @@ class CodeGenFunction : public CodeGenTypeCache {
/// the wider vector. This avoids the error when allocating space in llvm
/// for struct of scalable vectors if a function returns struct.
llvm::Value *FormSVEBuiltinResult(llvm::Value *Call);
+
+ void GetAArch64SMEProcessedOperands(unsigned BuiltinID, const CallExpr *E,
+ SmallVectorImpl<llvm::Value *> &Ops,
+ SVETypeFlags TypeFlags);
+
llvm::Value *EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
llvm::Value *EmitSMELd1St1(const SVETypeFlags &TypeFlags,
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add.c
new file mode 100644
index 000000000000000..18d7c66896d8b84
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add.c
@@ -0,0 +1,1226 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+
+// REQUIRES: aarch64-registered-target
+
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-i16i64 -target-feature +sme-f64f64 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-i16i64 -target-feature +sme-f64f64 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-i16i64 -target-feature +sme-f64f64 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-i16i64 -target-feature +sme-f64f64 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-i16i64 -target-feature +sme-f64f64 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+
+#include <arm_sme_draft_spec_subject_to_change.h>
+
+#ifdef SVE_OVERLOADED_FORMS
+// A simple used,unused... macro, long enough to represent any SVE builtin.
+#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED,A5) A1##A3##A5
+#else
+#define SVE_ACLE_FUNC(A1,A2,A3,A4,A5) A1##A2##A3##A4##A5
+#endif
+
+//
+// Single-Multi
+//
+
+// x2
+// CHECK-LABEL: @test_svadd_write_single2_s32(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[ADD:%.*]] = add i32 [[SLICE_BASE:%.*]], 7
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[ADD]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z28test_svadd_write_single2_s32j11svint32x2_tu11__SVInt32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[ADD:%.*]] = add i32 [[SLICE_BASE:%.*]], 7
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[ADD]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+void test_svadd_write_single2_s32(uint32_t slice_base, svint32x2_t zn, svint32_t zm) {
+ SVE_ACLE_FUNC(svadd_write,_single,_za32,_s32,_vg1x2)(slice_base + 7, zn, zm);
+}
+
+// CHECK-LABEL: @test_svadd_write_single2_u32(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[ADD:%.*]] = add i32 [[SLICE_BASE:%.*]], 7
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[ADD]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z28test_svadd_write_single2_u32j12svuint32x2_tu12__SVUint32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[ADD:%.*]] = add i32 [[SLICE_BASE:%.*]], 7
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[ADD]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+void test_svadd_write_single2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32_t zm) {
+ SVE_ACLE_FUNC(svadd_write,_single,_za32,_u32,_vg1x2)(slice_base + 7, zn, zm);
+}
+
+// CHECK-LABEL: @test_svadd_write_single2_s64(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[ADD:%.*]] = add i32 [[SLICE_BASE:%.*]], 7
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
+// CHECK-NEX...
[truncated]
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
david-arm
reviewed
Oct 20, 2023
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { | ||
| if ((ICEArguments & (1 << i)) == 0) | ||
| if (!IsTupleGetOrSet && (ICEArguments & (1 << i)) == 0) { |
There was a problem hiding this comment.
Perhaps you can create a temp variable and reuse it so it's a bit clearer, i.e.
bool IsICE = ICEArguments & (1 << i);
if (!IsTupleGetOrSet && !IsICE) {
...
} else if (!IsICE) {
...
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| @@ -9893,24 +9888,37 @@ Value *CodeGenFunction::FormSVEBuiltinResult(Value *Call) { | |||
| return Call; | |||
| } | |||
|
|
|||
| Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, | |||
| const CallExpr *E) { | |||
| void CodeGenFunction::GetAArch64SMEProcessedOperands( | |||
There was a problem hiding this comment.
I wonder if actually this is better named as GetAArch64SVEProcessedOperands because if we have to choose a name that's common to both the SME and SVE builtins, choosing SVE might make more sense. That's because we're specifically dealing with scalable vectors in general here and not something that's intrinsically linked to SME.
There was a problem hiding this comment.
I think this makes sense, I've changed it to GetAArch64SVEProcessedOperands
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { | ||
| if ((ICEArguments & (1 << i)) == 0) |
There was a problem hiding this comment.
Might be worth adding a comment explaining why we explicitly ignore tuple get/set functions?
There was a problem hiding this comment.
Added the following comment:
// Tuple set/get only requires one insert/extract vector, which is
// created by EmitSVETupleSetOrGet.
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| @@ -10272,29 +10291,13 @@ Value *CodeGenFunction::EmitAArch64SMEBuiltinExpr(unsigned BuiltinID, | |||
| getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
There was a problem hiding this comment.
Do we still need this code given we're now checking the ICE arguments in GetAArch64SMEProcessedOperands?
There was a problem hiding this comment.
I've removed this, it wasn't needed here now that it's checked in GetAArch64SVEProcessedOperands.
| << (SMEAttrs.empty() ? "__builtin_sve_" : "__builtin_sme_") | ||
| << FullName << ")"; | ||
| if (!SMEAttrs.empty()) | ||
| OS << SMEAttrs; |
There was a problem hiding this comment.
It looks like we're no longer printing out the attributes for the builtin - is this because the attributes are dealt with explicitly elsewhere in clang and so they are no longer needed?
There was a problem hiding this comment.
The attributes aren't currently handled elsewhere in Clang so this shouldn't have been removed. I've added most of this code back in, but kept the ACLEKind switch in below as this is needed when the builtin kind is SVE but it has SME attributes.
kmclaughlin-arm
commented
Oct 24, 2023
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| @@ -9893,24 +9888,37 @@ Value *CodeGenFunction::FormSVEBuiltinResult(Value *Call) { | |||
| return Call; | |||
| } | |||
|
|
|||
| Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, | |||
| const CallExpr *E) { | |||
| void CodeGenFunction::GetAArch64SMEProcessedOperands( | |||
There was a problem hiding this comment.
I think this makes sense, I've changed it to GetAArch64SVEProcessedOperands
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) { | ||
| if ((ICEArguments & (1 << i)) == 0) |
There was a problem hiding this comment.
Added the following comment:
// Tuple set/get only requires one insert/extract vector, which is
// created by EmitSVETupleSetOrGet.
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| @@ -10272,29 +10291,13 @@ Value *CodeGenFunction::EmitAArch64SMEBuiltinExpr(unsigned BuiltinID, | |||
| getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments); | |||
There was a problem hiding this comment.
I've removed this, it wasn't needed here now that it's checked in GetAArch64SVEProcessedOperands.
| << (SMEAttrs.empty() ? "__builtin_sve_" : "__builtin_sme_") | ||
| << FullName << ")"; | ||
| if (!SMEAttrs.empty()) | ||
| OS << SMEAttrs; |
There was a problem hiding this comment.
The attributes aren't currently handled elsewhere in Clang so this shouldn't have been removed. I've added most of this code back in, but kept the ACLEKind switch in below as this is needed when the builtin kind is SVE but it has SME attributes.
david-arm
reviewed
Oct 25, 2023
There was a problem hiding this comment.
This looks a lot better now @kmclaughlin-arm - thanks for the changes! I just have a couple of comments about the tests that I missed previously...
| // | ||
| void test_svsub_za64_vg1x4_u64(uint32_t slice_base, svuint64x4_t zn) { | ||
| SVE_ACLE_FUNC(svsub_za64,,_u64,,_vg1x4)(slice_base + 6, zn); | ||
| } |
There was a problem hiding this comment.
It looks like this file is missing many tests compared to the add file. For example, I think we're missing the signed variants when looking at the changes to arm_sme.td.
| // CPP-CHECK-NEXT: ret void | ||
| // | ||
| void test_svadd_za32_vg1x2_f32(uint32_t slice_base, svfloat32x2_t zn) { | ||
| SVE_ACLE_FUNC(svadd_za32,,_f32,,_vg1x2)(slice_base + 6, zn); |
There was a problem hiding this comment.
Perhaps it would be better to add 7 here instead, since that's the limit of the instruction? Having said that, since the LLVM intrinsics don't accept an immediate offset anyway maybe there isn't any value in adding immediate values for any of the add/sub tests?
david-arm
approved these changes
Oct 27, 2023
| //////////////////////////////////////////////////////////////////////////////// | ||
| // SME2 - ADD, SUB | ||
|
|
||
| multiclass ZAAddSub<string n_suffix> { |
There was a problem hiding this comment.
nit:
Suggested change
| multiclass ZAAddSub<string n_suffix> { | |
| multiclass AddSubZA<string n_suffix> { |
There was a problem hiding this comment.
Hi @sdesmalen-arm, I'm happy to change this, though the reason I chose this name was to match other multiclasses in the file such as ZAStore, ZAAdd, ZAFPOuterProd, etc.
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| @@ -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)) | |||
There was a problem hiding this comment.
EmitSVEStructStore does not handle any more intrinsics. Should this be part of a different patch?
There was a problem hiding this comment.
This change was intended to be part of this patch. Now that EmitSVEStructLoad is being called after GetAArch64SVEProcessedOperands, the first operand will have been split into a series of vector extracts and the number of operands expected here will change depending on the value of N.
clang/lib/CodeGen/CGBuiltin.cpp
Outdated
| @@ -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. | |||
There was a problem hiding this comment.
The changes to EmitAArch64SMEBuiltinExpr seem like a (significant) non-functional change. Can you separate it out as such so that it can be committed separately?
There was a problem hiding this comment.
I've moved these changes to #70662 and I'll update this patch shortly to remove the non-functional changes.
kmclaughlin-arm
force-pushed
the
sme2-builtins-add-sub
branch
from
e8bf03a
to
5d0b048
Compare
Adds the following SME2 builtins: - sv(add|sub) - sv(add|sub)_za32/za64, - sv(add|sub)_write_za32/za64 Other changes in this patch: - SveEmitter.cpp: Adds the createCoreHeaderIntrinsics function to remove duplicated code in createHeader & createSMEHeader. Uses a new enum (ACLEKind) to choose either "__builtin_sme_" or "__builtin_sve_" when emitting the intrinsics. - CGBuiltin.cpp: The GetAArch64SMEProcessedOperands function is created to avoid duplicating existing code from EmitAArch64SVEBuiltinExpr. - arm_sve.td: The add/sub SME2 builtins which do not operate on ZA have been added to arm_sve.td, matching the corrosponding LLVM IR intrinsic names which start with @llvm.aarch64.sve for this reason. See https://github.com/ARM-software/acle/pull/217/files
…ands - Removed code checking for ICEArguments from EmitAArch64SMEBuiltinExpr - Added comment over check for IsTupleGetOrSet - Added the code to print SME attributes back into emitIntrinsic
- Added missing tests to acle_sme2_sub.c
kmclaughlin-arm
force-pushed
the
sme2-builtins-add-sub
branch
from
9961f3d
to
4cf4256
Compare
- Ran clang-format to fix formatting issues in SveEmitter.cpp - Added a comment to arm_sve.td to explain why some SME2 intrinsics are included
| @@ -257,7 +257,7 @@ class ImmCheck<int arg, ImmCheckType kind, int eltSizeArg = -1> { | |||
| } | |||
|
|
|||
| class Inst<string n, string p, string t, MergeType mt, string i, | |||
| list<FlagType> ft, list<ImmCheck> ch, MemEltType met> { | |||
| list<FlagType> ft, list<ImmCheck> ch, MemEltType met = MemEltTyDefault> { | |||
There was a problem hiding this comment.
I am not sure about propose of this change "met = MemEltTyDefault"? The last parameter in multiclass ZAAddSub with Inst class usage is always []?
There was a problem hiding this comment.
Hi @dtemirbulatov, this change is just to set the default MemEltType to MemEltTyDefault, so that we don't have to provide this in every builtin definition.
sdesmalen-arm
approved these changes
Nov 7, 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.
Adds the following SME2 builtins:
Other changes in this patch:
to avoid duplicating existing code from EmitAArch64SVEBuiltinExpr.
been added to arm_sve.td, matching the corrosponding LLVM IR intrinsic
names which start with @llvm.aarch64.sve for this reason.
duplicated code in createHeader & createSMEHeader. Uses a new enum
(ACLEKind) to choose either "_builtin_sme" or "_builtin_sve" when
emitting the intrinsics.
See https://github.com/ARM-software/acle/pull/217/files