[AArch64] Add soft-float ABI

clang/include/clang/Basic/DiagnosticCommonKinds.td

@@ -356,6 +356,8 @@ def warn_target_unrecognized_env : Warning<
 def warn_knl_knm_isa_support_removed : Warning<
   "KNL, KNM related Intel Xeon Phi CPU's specific ISA's supports will be removed in LLVM 19.">,
   InGroup<DiagGroup<"knl-knm-isa-support-removed">>;
+def err_target_unsupported_abi_with_fpu : Error<
+  "'%0' ABI is not supported with FPU">;
 
 // Source manager
 def err_cannot_open_file : Error<"cannot open file '%0': %1">, DefaultFatal;

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -11314,6 +11314,8 @@ def err_omp_wrong_dependency_iterator_type : Error<
 def err_target_unsupported_type
     : Error<"%0 requires %select{|%2 bit size}1 %3 %select{|return }4type support,"
             " but target '%5' does not support it">;
+def err_target_unsupported_type_for_abi
+    : Error<"%0 requires %1 type support, but ABI '%2' does not support it">;
 def err_omp_lambda_capture_in_declare_target_not_to : Error<
   "variable captured in declare target region must appear in a to clause">;
 def err_omp_device_type_mismatch : Error<

clang/include/clang/Basic/TargetInfo.h

@@ -231,6 +231,7 @@ class TargetInfo : public TransferrableTargetInfo,
   bool HasIbm128;
   bool HasLongDouble;
   bool HasFPReturn;
+  bool HasFPTypes;
   bool HasStrictFP;
 
   unsigned char MaxAtomicPromoteWidth, MaxAtomicInlineWidth;
@@ -689,6 +690,9 @@ class TargetInfo : public TransferrableTargetInfo,
   /// on this target.
   virtual bool hasFPReturn() const { return HasFPReturn; }
 
+  /// Determine whether floating point types are supported for this target.
+  virtual bool hasFPTypes() const { return HasFPTypes; }
+
   /// Determine whether constrained floating point is supported on this target.
   virtual bool hasStrictFP() const { return HasStrictFP; }
 
@@ -1331,6 +1335,10 @@ class TargetInfo : public TransferrableTargetInfo,
     return false;
   }
 
+  /// Make changes to the supported types which depend on both the target
+  /// features and ABI.
+  virtual void setSupportedArgTypes() {}
+
   /// Use the specified unit for FP math.
   ///
   /// \return False on error (invalid unit name).

clang/lib/Basic/TargetInfo.cpp

@@ -67,6 +67,7 @@ TargetInfo::TargetInfo(const llvm::Triple &T) : Triple(T) {
   HasFullBFloat16 = false;
   HasLongDouble = true;
   HasFPReturn = true;
+  HasFPTypes = true;
   HasStrictFP = false;
   PointerWidth = PointerAlign = 32;
   BoolWidth = BoolAlign = 8;

clang/lib/Basic/Targets.cpp

@@ -830,6 +830,7 @@ TargetInfo::CreateTargetInfo(DiagnosticsEngine &Diags,
   Target->setSupportedOpenCLOpts();
   Target->setCommandLineOpenCLOpts();
   Target->setMaxAtomicWidth();
+  Target->setSupportedArgTypes();
 
   if (!Opts->DarwinTargetVariantTriple.empty())
     Target->DarwinTargetVariantTriple =

clang/lib/Basic/Targets/AArch64.cpp

@@ -11,6 +11,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "AArch64.h"
+#include "clang/Basic/Diagnostic.h"
 #include "clang/Basic/LangOptions.h"
 #include "clang/Basic/TargetBuiltins.h"
 #include "clang/Basic/TargetInfo.h"
@@ -199,13 +200,32 @@ AArch64TargetInfo::AArch64TargetInfo(const llvm::Triple &Triple,
 StringRef AArch64TargetInfo::getABI() const { return ABI; }
 
 bool AArch64TargetInfo::setABI(const std::string &Name) {
-  if (Name != "aapcs" && Name != "darwinpcs")
+  if (Name != "aapcs" && Name != "aapcs-soft" && Name != "darwinpcs")
     return false;
 
   ABI = Name;
   return true;
 }
 
+void AArch64TargetInfo::setSupportedArgTypes() {
+  if (!(FPU & FPUMode) && ABI != "aapcs-soft") {
+    // When a hard-float ABI is used on a target without an FPU, all
+    // floating-point argument and return types are rejected because they must
+    // be passed in FP registers.
+    HasFPTypes = false;
+  }
+}
+
+bool AArch64TargetInfo::validateTarget(DiagnosticsEngine &Diags) const {
+  if (hasFeature("fp") && ABI == "aapcs-soft") {
+    // aapcs-soft is not allowed for targets with an FPU, to avoid there being
+    // two incomatible ABIs.
+    Diags.Report(diag::err_target_unsupported_abi_with_fpu) << ABI;
+    return false;
+  }
+  return true;
+}
+
 bool AArch64TargetInfo::validateBranchProtection(StringRef Spec, StringRef,
                                                  BranchProtectionInfo &BPI,
                                                  StringRef &Err) const {
@@ -686,7 +706,8 @@ bool AArch64TargetInfo::hasFeature(StringRef Feature) const {
   return llvm::StringSwitch<bool>(Feature)
       .Cases("aarch64", "arm64", "arm", true)
       .Case("fmv", HasFMV)
-      .Cases("neon", "fp", "simd", FPU & NeonMode)
+      .Case("fp", FPU & FPUMode)
+      .Cases("neon", "simd", FPU & NeonMode)
       .Case("jscvt", HasJSCVT)
       .Case("fcma", HasFCMA)
       .Case("rng", HasRandGen)

clang/lib/Basic/Targets/AArch64.h

@@ -96,6 +96,7 @@ class LLVM_LIBRARY_VISIBILITY AArch64TargetInfo : public TargetInfo {
 
   StringRef getABI() const override;
   bool setABI(const std::string &Name) override;
+  void setSupportedArgTypes() override;
 
   bool validateBranchProtection(StringRef Spec, StringRef Arch,
                                 BranchProtectionInfo &BPI,
@@ -199,6 +200,8 @@ class LLVM_LIBRARY_VISIBILITY AArch64TargetInfo : public TargetInfo {
   bool hasInt128Type() const override;
 
   bool hasBitIntType() const override { return true; }
+
+  bool validateTarget(DiagnosticsEngine &Diags) const override;
 };
 
 class LLVM_LIBRARY_VISIBILITY AArch64leTargetInfo : public AArch64TargetInfo {

clang/lib/CodeGen/CodeGenModule.cpp

@@ -145,6 +145,8 @@ createTargetCodeGenInfo(CodeGenModule &CGM) {
       Kind = AArch64ABIKind::DarwinPCS;
     else if (Triple.isOSWindows())
       return createWindowsAArch64TargetCodeGenInfo(CGM, AArch64ABIKind::Win64);
+    else if (Target.getABI() == "aapcs-soft")
+      Kind = AArch64ABIKind::AAPCSSoft;
 
     return createAArch64TargetCodeGenInfo(CGM, Kind);
   }

clang/lib/CodeGen/TargetInfo.h

@@ -416,6 +416,7 @@ enum class AArch64ABIKind {
   AAPCS = 0,
   DarwinPCS,
   Win64,
+  AAPCSSoft,
 };
 
 std::unique_ptr<TargetCodeGenInfo>

clang/lib/CodeGen/Targets/AArch64.cpp

@@ -53,8 +53,8 @@ class AArch64ABIInfo : public ABIInfo {
   Address EmitDarwinVAArg(Address VAListAddr, QualType Ty,
                           CodeGenFunction &CGF) const;
 
-  Address EmitAAPCSVAArg(Address VAListAddr, QualType Ty,
-                         CodeGenFunction &CGF) const;
+  Address EmitAAPCSVAArg(Address VAListAddr, QualType Ty, CodeGenFunction &CGF,
+                         AArch64ABIKind Kind) const;
 
   Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                     QualType Ty) const override {
@@ -65,7 +65,7 @@ class AArch64ABIInfo : public ABIInfo {
 
     return Kind == AArch64ABIKind::Win64 ? EmitMSVAArg(CGF, VAListAddr, Ty)
            : isDarwinPCS()               ? EmitDarwinVAArg(VAListAddr, Ty, CGF)
-                                         : EmitAAPCSVAArg(VAListAddr, Ty, CGF);
+                           : EmitAAPCSVAArg(VAListAddr, Ty, CGF, Kind);
   }
 
   Address EmitMSVAArg(CodeGenFunction &CGF, Address VAListAddr,
@@ -482,6 +482,11 @@ bool AArch64SwiftABIInfo::isLegalVectorType(CharUnits VectorSize,
 }
 
 bool AArch64ABIInfo::isHomogeneousAggregateBaseType(QualType Ty) const {
+  // For the soft-float ABI variant, no types are considered to be homogeneous
+  // aggregates.
+  if (Kind == AArch64ABIKind::AAPCSSoft)
+    return false;
+
   // Homogeneous aggregates for AAPCS64 must have base types of a floating
   // point type or a short-vector type. This is the same as the 32-bit ABI,
   // but with the difference that any floating-point type is allowed,
@@ -513,7 +518,8 @@ bool AArch64ABIInfo::isZeroLengthBitfieldPermittedInHomogeneousAggregate()
 }
 
 Address AArch64ABIInfo::EmitAAPCSVAArg(Address VAListAddr, QualType Ty,
-                                       CodeGenFunction &CGF) const {
+                                       CodeGenFunction &CGF,
+                                       AArch64ABIKind Kind) const {
   ABIArgInfo AI = classifyArgumentType(Ty, /*IsVariadic=*/true,
                                        CGF.CurFnInfo->getCallingConvention());
   // Empty records are ignored for parameter passing purposes.
@@ -538,7 +544,8 @@ Address AArch64ABIInfo::EmitAAPCSVAArg(Address VAListAddr, QualType Ty,
     BaseTy = ArrTy->getElementType();
     NumRegs = ArrTy->getNumElements();
   }
-  bool IsFPR = BaseTy->isFloatingPointTy() || BaseTy->isVectorTy();
+  bool IsFPR = Kind != AArch64ABIKind::AAPCSSoft &&
+               (BaseTy->isFloatingPointTy() || BaseTy->isVectorTy());
 
   // The AArch64 va_list type and handling is specified in the Procedure Call
   // Standard, section B.4:

clang/lib/Sema/Sema.cpp

@@ -1944,6 +1944,21 @@ Sema::SemaDiagnosticBuilder Sema::Diag(SourceLocation Loc, unsigned DiagID,
   return DB;
 }
 
+static bool typeIsOrContainsFloat(const Type &Ty) {
+  if (Ty.isFloatingType())
+    return true;
+
+  if (const RecordDecl *Decl = Ty.getAsRecordDecl()) {
+    for (const FieldDecl *FD : Decl->fields()) {
+      const Type &FieldType = *FD->getType();
+      if (typeIsOrContainsFloat(FieldType))
+        return true;
+    }
+  }
+
+  return false;
+}
+
 void Sema::checkTypeSupport(QualType Ty, SourceLocation Loc, ValueDecl *D) {
   if (isUnevaluatedContext() || Ty.isNull())
     return;
@@ -2088,6 +2103,25 @@ void Sema::checkTypeSupport(QualType Ty, SourceLocation Loc, ValueDecl *D) {
           !Builtin::evaluateRequiredTargetFeatures("sme", CallerFeatureMap))
         Diag(D->getLocation(), diag::err_sve_vector_in_non_sve_target) << Ty;
     }
+
+    // Don't allow any floating-point types (including structs containing
+    // floats) for ABIs which do not support them.
+    if (!TI.hasFPTypes() && typeIsOrContainsFloat(*UnqualTy)) {
+      PartialDiagnostic PD = PDiag(diag::err_target_unsupported_type_for_abi);
+
+      if (D)
+        PD << D;
+      else
+        PD << "expression";
+
+      if (Diag(Loc, PD, FD) << Ty << TI.getABI()) {
+        if (D)
+          D->setInvalidDecl();
+      }
+
+      if (D)
+        targetDiag(D->getLocation(), diag::note_defined_here, FD) << D;
+    }
   };
 
   CheckType(Ty);

clang/test/CodeGen/aarch64-soft-float-abi.c

@@ -0,0 +1,56 @@
+// RUN: %clang_cc1 -triple aarch64 -target-feature +fp-armv8 -target-abi aapcs -emit-llvm -o - %s | FileCheck %s --check-prefixes=CHECK,HARD
+// RUN: %clang_cc1 -triple aarch64 -target-feature -fp-armv8 -target-abi aapcs-soft -emit-llvm -o - %s | FileCheck %s --check-prefixes=CHECK,SOFT
+
+// See also llvm/test/CodeGen/AArch64/soft-float-abi.ll, which checks the LLVM
+// backend parts of the soft-float ABI.
+
+// The va_list type does not change between the ABIs
+// CHECK: %struct.__va_list = type { ptr, ptr, ptr, i32, i32 }
+
+// Floats are passed in integer registers, this will be handled by the backend.
+// CHECK: define dso_local half @test0(half noundef %a)
+// CHECK: define dso_local bfloat @test1(bfloat noundef %a)
+// CHECK: define dso_local float @test2(float noundef %a)
+// CHECK: define dso_local double @test3(double noundef %a)
+// CHECK: define dso_local fp128 @test4(fp128 noundef %a)
+__fp16 test0(__fp16 a) { return a; }
+__bf16 test1(__bf16 a) { return a; }
+float test2(float a) { return a; }
+double test3(double a) { return a; }
+long double test4(long double a) { return a; }
+
+// No types are considered to be HFAs or HVAs by the soft-float PCS, so these
+// are converted to integer types.
+struct A {
+  float x;
+};
+// SOFT: define dso_local i32 @test10(i64 %a.coerce)
+// HARD: define dso_local %struct.A @test10([1 x float] alignstack(8) %a.coerce)
+struct A test10(struct A a) { return a; }
+
+struct B {
+  double x;
+  double y;
+};
+// SOFT: define dso_local [2 x i64] @test11([2 x i64] %a.coerce)
+// HARD: define dso_local %struct.B @test11([2 x double] alignstack(8) %a.coerce)
+struct B test11(struct B a) { return a; }
+
+#include <stdarg.h>
+
+// For variadic arguments, va_arg will always retreive 
+// CHECK-LABEL: define dso_local double @test20(i32 noundef %a, ...)
+// CHECK: %vl = alloca %struct.__va_list, align 8
+// SOFT: %gr_offs_p = getelementptr inbounds %struct.__va_list, ptr %vl, i32 0, i32 3
+// SOFT: %reg_top_p = getelementptr inbounds %struct.__va_list, ptr %vl, i32 0, i32 1
+// HARD: %vr_offs_p = getelementptr inbounds %struct.__va_list, ptr %vl, i32 0, i32 4
+// HARD: %reg_top_p = getelementptr inbounds %struct.__va_list, ptr %vl, i32 0, i32 2
+double test20(int a, ...) {
+  va_list vl;
+  va_start(vl, a);
+  return va_arg(vl, double);
+}
+
+// Vector types are only available for targets with the correct hardware, and
+// their calling-convention is left undefined by the soft-float ABI, so they
+// aren't tested here.

clang/test/CodeGen/attr-target-clones-aarch64.c

@@ -1,6 +1,6 @@
 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-attributes --check-globals --include-generated-funcs
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -S -emit-llvm -o - %s | FileCheck %s
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature -fmv -S -emit-llvm -o - %s | FileCheck %s -check-prefix=CHECK-NOFMV
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature -fp-armv8 -S -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature -fp-armv8 -target-feature -fmv -S -emit-llvm -o - %s | FileCheck %s -check-prefix=CHECK-NOFMV
 
 int __attribute__((target_clones("lse+aes", "sve2"))) ftc(void) { return 0; }
 int __attribute__((target_clones("sha2", "sha2+memtag2", " default "))) ftc_def(void) { return 1; }
@@ -414,23 +414,23 @@ inline int __attribute__((target_clones("fp16", "sve2-bitperm+fcma", "default"))
 // CHECK-NOFMV-NEXT:    ret i32 [[ADD5]]
 //
 //.
-// CHECK: attributes #[[ATTR0:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+lse,+neon" }
-// CHECK: attributes #[[ATTR1:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve,+sve2" }
-// CHECK: attributes #[[ATTR2:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-// CHECK: attributes #[[ATTR3:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+sha2" }
-// CHECK: attributes #[[ATTR4:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+mte,+neon,+sha2" }
-// CHECK: attributes #[[ATTR5:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon" }
-// CHECK: attributes #[[ATTR6:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+crc,+dotprod,+fp-armv8,+neon" }
-// CHECK: attributes #[[ATTR7:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+neon,+rand" }
-// CHECK: attributes #[[ATTR8:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+predres,+rcpc" }
-// CHECK: attributes #[[ATTR9:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sve,+sve2,+sve2-aes,+wfxt" }
-// CHECK: attributes #[[ATTR10:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon" }
-// CHECK: attributes #[[ATTR11:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+complxnum,+fp-armv8,+fullfp16,+neon,+sve,+sve2,+sve2-bitperm" }
-// CHECK: attributes #[[ATTR12:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bti" }
-// CHECK: attributes #[[ATTR13:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fp-armv8,+fullfp16,+neon,+sb,+sve" }
+// CHECK: attributes #[[ATTR0:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+lse,+neon,-fp-armv8" }
+// CHECK: attributes #[[ATTR1:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+neon,+sve,+sve2,-fp-armv8" }
+// CHECK: attributes #[[ATTR2:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fp-armv8" }
+// CHECK: attributes #[[ATTR3:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon,+sha2,-fp-armv8" }
+// CHECK: attributes #[[ATTR4:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+mte,+neon,+sha2,-fp-armv8" }
+// CHECK: attributes #[[ATTR5:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon,-fp-armv8" }
+// CHECK: attributes #[[ATTR6:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+crc,+dotprod,+neon,-fp-armv8" }
+// CHECK: attributes #[[ATTR7:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+neon,+rand,-fp-armv8" }
+// CHECK: attributes #[[ATTR8:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+predres,+rcpc,-fp-armv8" }
+// CHECK: attributes #[[ATTR9:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+neon,+sve,+sve2,+sve2-aes,+wfxt,-fp-armv8" }
+// CHECK: attributes #[[ATTR10:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+neon,-fp-armv8" }
+// CHECK: attributes #[[ATTR11:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+complxnum,+fullfp16,+neon,+sve,+sve2,+sve2-bitperm,-fp-armv8" }
+// CHECK: attributes #[[ATTR12:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+bti,-fp-armv8" }
+// CHECK: attributes #[[ATTR13:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+fullfp16,+neon,+sb,+sve,-fp-armv8" }
 //.
-// CHECK-NOFMV: attributes #[[ATTR0:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fmv" }
-// CHECK-NOFMV: attributes #[[ATTR1:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fmv" }
+// CHECK-NOFMV: attributes #[[ATTR0:[0-9]+]] = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fmv,-fp-armv8" }
+// CHECK-NOFMV: attributes #[[ATTR1:[0-9]+]] = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="-fmv,-fp-armv8" }
 //.
 // CHECK: [[META0:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
 // CHECK: [[META1:![0-9]+]] = !{!"{{.*}}clang version {{.*}}"}