[clang][SPIR-V] Add support for AMDGCN flavoured SPIRV (#89796)

This change seeks to add support for vendor flavoured SPIRV - more
specifically, AMDGCN flavoured SPIRV. The aim is to generate SPIRV that
carries some extra bits of information that are only usable by AMDGCN
targets, forfeiting absolute genericity to obtain greater expressiveness
for target features:

- AMDGCN inline ASM is allowed/supported, under the assumption that the
[SPV_INTEL_inline_assembly](https://github.com/intel/llvm/blob/sycl/sycl/doc/design/spirv-extensions/SPV_INTEL_inline_assembly.asciidoc)
extension is enabled/used
- AMDGCN target specific builtins are allowed/supported, under the
assumption that e.g. the `--spirv-allow-unknown-intrinsics` option is
enabled when using the downstream translator
- the featureset matches the union of AMDGCN targets' features
- the datalayout string is overspecified to affix both the program
address space and the alloca address space, the latter under the
assumption that the
[SPV_INTEL_function_pointers](https://github.com/intel/llvm/blob/sycl/sycl/doc/design/spirv-extensions/SPV_INTEL_function_pointers.asciidoc)
extension is enabled/used, case in which the extant SPIRV datalayout
string would lead to pointers to function pointing to the private
address space, which would be wrong.

Existing AMDGCN tests are extended to cover this new target. It is
currently dormant / will require some additional changes, but I thought
I'd rather put it up for review to get feedback as early as possible. I
will note that an alternative option is to place this under AMDGPU, but
that seems slightly less natural, since this is still SPIRV, albeit
relaxed in terms of preconditions & constrained in terms of
postconditions, and only guaranteed to be usable on AMDGCN targets (it
is still possible to obtain pristine portable SPIRV through usage of the
flavoured target, though).

Author: AlexVlx

clang/lib/Basic/Targets.cpp

@@ -673,8 +673,11 @@ std::unique_ptr<TargetInfo> AllocateTarget(const llvm::Triple &Triple,
   }
   case llvm::Triple::spirv64: {
     if (os != llvm::Triple::UnknownOS ||
-        Triple.getEnvironment() != llvm::Triple::UnknownEnvironment)
+        Triple.getEnvironment() != llvm::Triple::UnknownEnvironment) {
+      if (os == llvm::Triple::OSType::AMDHSA)
+        return std::make_unique<SPIRV64AMDGCNTargetInfo>(Triple, Opts);
       return nullptr;
+    }
     return std::make_unique<SPIRV64TargetInfo>(Triple, Opts);
   }
   case llvm::Triple::wasm32:

clang/lib/Basic/Targets/SPIR.cpp

@@ -11,7 +11,9 @@
 //===----------------------------------------------------------------------===//
 
 #include "SPIR.h"
+#include "AMDGPU.h"
 #include "Targets.h"
+#include "llvm/TargetParser/TargetParser.h"
 
 using namespace clang;
 using namespace clang::targets;
@@ -54,3 +56,76 @@ void SPIRV64TargetInfo::getTargetDefines(const LangOptions &Opts,
   BaseSPIRVTargetInfo::getTargetDefines(Opts, Builder);
   DefineStd(Builder, "SPIRV64", Opts);
 }
+
+static const AMDGPUTargetInfo AMDGPUTI(llvm::Triple("amdgcn-amd-amdhsa"), {});
+
+ArrayRef<const char *> SPIRV64AMDGCNTargetInfo::getGCCRegNames() const {
+  return AMDGPUTI.getGCCRegNames();
+}
+
+bool SPIRV64AMDGCNTargetInfo::initFeatureMap(
+    llvm::StringMap<bool> &Features, DiagnosticsEngine &Diags, StringRef,
+    const std::vector<std::string> &FeatureVec) const {
+  llvm::AMDGPU::fillAMDGPUFeatureMap({}, getTriple(), Features);
+
+  return TargetInfo::initFeatureMap(Features, Diags, {}, FeatureVec);
+}
+
+bool SPIRV64AMDGCNTargetInfo::validateAsmConstraint(
+    const char *&Name, TargetInfo::ConstraintInfo &Info) const {
+  return AMDGPUTI.validateAsmConstraint(Name, Info);
+}
+
+std::string
+SPIRV64AMDGCNTargetInfo::convertConstraint(const char *&Constraint) const {
+  return AMDGPUTI.convertConstraint(Constraint);
+}
+
+ArrayRef<Builtin::Info> SPIRV64AMDGCNTargetInfo::getTargetBuiltins() const {
+  return AMDGPUTI.getTargetBuiltins();
+}
+
+void SPIRV64AMDGCNTargetInfo::getTargetDefines(const LangOptions &Opts,
+                                               MacroBuilder &Builder) const {
+  BaseSPIRVTargetInfo::getTargetDefines(Opts, Builder);
+  DefineStd(Builder, "SPIRV64", Opts);
+
+  Builder.defineMacro("__AMD__");
+  Builder.defineMacro("__AMDGPU__");
+  Builder.defineMacro("__AMDGCN__");
+}
+
+void SPIRV64AMDGCNTargetInfo::setAuxTarget(const TargetInfo *Aux) {
+  assert(Aux && "Cannot invoke setAuxTarget without a valid auxiliary target!");
+
+  // This is a 1:1 copy of AMDGPUTargetInfo::setAuxTarget()
+  assert(HalfFormat == Aux->HalfFormat);
+  assert(FloatFormat == Aux->FloatFormat);
+  assert(DoubleFormat == Aux->DoubleFormat);
+
+  // On x86_64 long double is 80-bit extended precision format, which is
+  // not supported by AMDGPU. 128-bit floating point format is also not
+  // supported by AMDGPU. Therefore keep its own format for these two types.
+  auto SaveLongDoubleFormat = LongDoubleFormat;
+  auto SaveFloat128Format = Float128Format;
+  auto SaveLongDoubleWidth = LongDoubleWidth;
+  auto SaveLongDoubleAlign = LongDoubleAlign;
+  copyAuxTarget(Aux);
+  LongDoubleFormat = SaveLongDoubleFormat;
+  Float128Format = SaveFloat128Format;
+  LongDoubleWidth = SaveLongDoubleWidth;
+  LongDoubleAlign = SaveLongDoubleAlign;
+  // For certain builtin types support on the host target, claim they are
+  // supported to pass the compilation of the host code during the device-side
+  // compilation.
+  // FIXME: As the side effect, we also accept `__float128` uses in the device
+  // code. To reject these builtin types supported in the host target but not in
+  // the device target, one approach would support `device_builtin` attribute
+  // so that we could tell the device builtin types from the host ones. This
+  // also solves the different representations of the same builtin type, such
+  // as `size_t` in the MSVC environment.
+  if (Aux->hasFloat128Type()) {
+    HasFloat128 = true;
+    Float128Format = DoubleFormat;
+  }
+}

clang/lib/Basic/Targets/SPIR.h

@@ -364,6 +364,57 @@ class LLVM_LIBRARY_VISIBILITY SPIRV64TargetInfo : public BaseSPIRVTargetInfo {
                         MacroBuilder &Builder) const override;
 };
 
+class LLVM_LIBRARY_VISIBILITY SPIRV64AMDGCNTargetInfo final
+    : public BaseSPIRVTargetInfo {
+public:
+  SPIRV64AMDGCNTargetInfo(const llvm::Triple &Triple, const TargetOptions &Opts)
+      : BaseSPIRVTargetInfo(Triple, Opts) {
+    assert(Triple.getArch() == llvm::Triple::spirv64 &&
+           "Invalid architecture for 64-bit AMDGCN SPIR-V.");
+    assert(Triple.getVendor() == llvm::Triple::VendorType::AMD &&
+           "64-bit AMDGCN SPIR-V target must use AMD vendor");
+    assert(getTriple().getOS() == llvm::Triple::OSType::AMDHSA &&
+           "64-bit AMDGCN SPIR-V target must use AMDHSA OS");
+    assert(getTriple().getEnvironment() == llvm::Triple::UnknownEnvironment &&
+           "64-bit SPIR-V target must use unknown environment type");
+    PointerWidth = PointerAlign = 64;
+    SizeType = TargetInfo::UnsignedLong;
+    PtrDiffType = IntPtrType = TargetInfo::SignedLong;
+
+    resetDataLayout("e-i64:64-v16:16-v24:32-v32:32-v48:64-"
+                    "v96:128-v192:256-v256:256-v512:512-v1024:1024-G1-P4-A0");
+
+    BFloat16Width = BFloat16Align = 16;
+    BFloat16Format = &llvm::APFloat::BFloat();
+
+    HasLegalHalfType = true;
+    HasFloat16 = true;
+    HalfArgsAndReturns = true;
+  }
+
+  bool hasBFloat16Type() const override { return true; }
+
+  ArrayRef<const char *> getGCCRegNames() const override;
+
+  bool initFeatureMap(llvm::StringMap<bool> &Features, DiagnosticsEngine &Diags,
+                      StringRef,
+                      const std::vector<std::string> &) const override;
+
+  bool validateAsmConstraint(const char *&Name,
+                             TargetInfo::ConstraintInfo &Info) const override;
+
+  std::string convertConstraint(const char *&Constraint) const override;
+
+  ArrayRef<Builtin::Info> getTargetBuiltins() const override;
+
+  void getTargetDefines(const LangOptions &Opts,
+                        MacroBuilder &Builder) const override;
+
+  void setAuxTarget(const TargetInfo *Aux) override;
+
+  bool hasInt128Type() const override { return TargetInfo::hasInt128Type(); }
+};
+
 } // namespace targets
 } // namespace clang
 #endif // LLVM_CLANG_LIB_BASIC_TARGETS_SPIR_H

clang/lib/CodeGen/CGBuiltin.cpp

@@ -6012,6 +6012,9 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
       llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch());
   if (!Prefix.empty()) {
     IntrinsicID = Intrinsic::getIntrinsicForClangBuiltin(Prefix.data(), Name);
+    if (IntrinsicID == Intrinsic::not_intrinsic && Prefix == "spv" &&
+        getTarget().getTriple().getOS() == llvm::Triple::OSType::AMDHSA)
+      IntrinsicID = Intrinsic::getIntrinsicForClangBuiltin("amdgcn", Name);
     // NOTE we don't need to perform a compatibility flag check here since the
     // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the
     // MS builtins via ALL_MS_LANGUAGES and are filtered earlier.
@@ -6182,6 +6185,10 @@ static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF,
   case llvm::Triple::riscv32:
   case llvm::Triple::riscv64:
     return CGF->EmitRISCVBuiltinExpr(BuiltinID, E, ReturnValue);
+  case llvm::Triple::spirv64:
+    if (CGF->getTarget().getTriple().getOS() != llvm::Triple::OSType::AMDHSA)
+      return nullptr;
+    return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E);
   default:
     return nullptr;
   }

clang/test/CodeGen/target-data.c

@@ -268,3 +268,7 @@
 // RUN: %clang_cc1 -triple ve -o - -emit-llvm %s | \
 // RUN: FileCheck %s -check-prefix=VE
 // VE: target datalayout = "e-m:e-i64:64-n32:64-S128-v64:64:64-v128:64:64-v256:64:64-v512:64:64-v1024:64:64-v2048:64:64-v4096:64:64-v8192:64:64-v16384:64:64"
+
+// RUN: %clang_cc1 -triple spirv64-amd -o - -emit-llvm %s | \
+// RUN: FileCheck %s -check-prefix=SPIR64
+// AMDGPUSPIRV64: target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-G1-P4-A0"