merge main into amd-staging (#350)

Author: ronlieb

clang/docs/OpenMPSupport.rst

@@ -256,7 +256,7 @@ implementation.
 +------------------------------+--------------------------------------------------------------+--------------------------+-----------------------------------------------------------------------+
 | device                       | device-specific environment variables                        | :none:`unclaimed`        |                                                                       |
 +------------------------------+--------------------------------------------------------------+--------------------------+-----------------------------------------------------------------------+
-| device                       | omp_target_is_accessible routine                             | :part:`In Progress`      | https://github.com/llvm/llvm-project/pull/138294                      |
+| device                       | omp_target_is_accessible routine                             | :good:`done`             | https://github.com/llvm/llvm-project/pull/138294                      |
 +------------------------------+--------------------------------------------------------------+--------------------------+-----------------------------------------------------------------------+
 | device                       | omp_get_mapped_ptr routine                                   | :good:`done`             | D141545                                                               |
 +------------------------------+--------------------------------------------------------------+--------------------------+-----------------------------------------------------------------------+

clang/docs/ReleaseNotes.rst

@@ -695,6 +695,7 @@ Bug Fixes to Attribute Support
 - Using ``[[gnu::cleanup(some_func)]]`` where some_func is annotated with
   ``[[gnu::error("some error")]]`` now correctly triggers an error. (#GH146520)
 - Fix a crash when the function name is empty in the `swift_name` attribute. (#GH157075)
+- Fixes crashes or missing diagnostics with the `device_kernel` attribute. (#GH161905)
 
 Bug Fixes to C++ Support
 ^^^^^^^^^^^^^^^^^^^^^^^^

clang/include/clang/Basic/Attr.td

@@ -1623,7 +1623,7 @@ def SYCLKernel : InheritableAttr {
   let Documentation = [SYCLKernelDocs];
 }
 
-def DeviceKernel : DeclOrTypeAttr {
+def DeviceKernel : InheritableAttr {
   let Spellings = [Clang<"device_kernel">,
                    Clang<"nvptx_kernel">, Clang<"amdgpu_kernel">,
                    CustomKeyword<"__kernel">, CustomKeyword<"kernel">];

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -4122,6 +4122,9 @@ def warn_missing_sdksettings_for_availability_checking : Warning<
   "%0 availability is ignored without a valid 'SDKSettings.json' in the SDK">,
   InGroup<DiagGroup<"ignored-availability-without-sdk-settings">>;
 
+def err_hidden_device_kernel
+    : Error<"%0 is specified as a device kernel but it is not externally visible">;
+
 // Thread Safety Attributes
 def warn_thread_attribute_ignored : Warning<
   "ignoring %0 attribute because its argument is invalid">,

clang/lib/AST/ASTConcept.cpp

@@ -86,7 +86,7 @@ void ConstraintSatisfaction::Profile(llvm::FoldingSetNodeID &ID,
   ID.AddPointer(ConstraintOwner);
   ID.AddInteger(TemplateArgs.size());
   for (auto &Arg : TemplateArgs)
-    C.getCanonicalTemplateArgument(Arg).Profile(ID, C);
+    Arg.Profile(ID, C);
 }
 
 ConceptReference *

clang/lib/AST/TypePrinter.cpp

@@ -2147,9 +2147,6 @@ void TypePrinter::printAttributedAfter(const AttributedType *T,
   }
   case attr::AArch64VectorPcs: OS << "aarch64_vector_pcs"; break;
   case attr::AArch64SVEPcs: OS << "aarch64_sve_pcs"; break;
-  case attr::DeviceKernel:
-    OS << T->getAttr()->getSpelling();
-    break;
   case attr::IntelOclBicc:
     OS << "inteloclbicc";
     break;

clang/lib/Basic/Targets/NVPTX.h

@@ -200,7 +200,7 @@ class LLVM_LIBRARY_VISIBILITY NVPTXTargetInfo : public TargetInfo {
     // a host function.
     if (HostTarget)
       return HostTarget->checkCallingConvention(CC);
-    return CCCR_Warning;
+    return CC == CC_DeviceKernel ? CCCR_OK : CCCR_Warning;
   }
 
   bool hasBitIntType() const override { return true; }

clang/lib/CodeGen/Targets/AMDGPU.cpp

@@ -439,9 +439,11 @@ void AMDGPUTargetCodeGenInfo::setTargetAttributes(
     return;
 
   const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D);
-  if (FD)
+  if (FD) {
     setFunctionDeclAttributes(FD, F, M);
-
+    if (FD->hasAttr<DeviceKernelAttr>() && !M.getLangOpts().OpenCL)
+      F->setCallingConv(getDeviceKernelCallingConv());
+  }
   if (!getABIInfo().getCodeGenOpts().EmitIEEENaNCompliantInsts)
     F->addFnAttr("amdgpu-ieee", "false");
 }
@@ -658,7 +660,7 @@ llvm::Value *AMDGPUTargetCodeGenInfo::createEnqueuedBlockKernel(
   // kernel address (only the kernel descriptor).
   auto *F = llvm::Function::Create(FT, llvm::GlobalValue::InternalLinkage, Name,
                                    &Mod);
-  F->setCallingConv(llvm::CallingConv::AMDGPU_KERNEL);
+  F->setCallingConv(getDeviceKernelCallingConv());
 
   llvm::AttrBuilder KernelAttrs(C);
   // FIXME: The invoke isn't applying the right attributes either

clang/lib/CodeGen/Targets/NVPTX.cpp

@@ -264,7 +264,7 @@ void NVPTXTargetCodeGenInfo::setTargetAttributes(
       // And kernel functions are not subject to inlining
       F->addFnAttr(llvm::Attribute::NoInline);
       if (FD->hasAttr<CUDAGlobalAttr>()) {
-        F->setCallingConv(llvm::CallingConv::PTX_Kernel);
+        F->setCallingConv(getDeviceKernelCallingConv());
 
         for (auto IV : llvm::enumerate(FD->parameters()))
           if (IV.value()->hasAttr<CUDAGridConstantAttr>())
@@ -278,7 +278,7 @@ void NVPTXTargetCodeGenInfo::setTargetAttributes(
   }
   // Attach kernel metadata directly if compiling for NVPTX.
   if (FD->hasAttr<DeviceKernelAttr>())
-    F->setCallingConv(llvm::CallingConv::PTX_Kernel);
+    F->setCallingConv(getDeviceKernelCallingConv());
 }
 
 void NVPTXTargetCodeGenInfo::addNVVMMetadata(llvm::GlobalValue *GV,

clang/lib/CodeGen/Targets/SPIR.cpp

@@ -64,6 +64,8 @@ class CommonSPIRTargetCodeGenInfo : public TargetCodeGenInfo {
   llvm::Constant *getNullPointer(const CodeGen::CodeGenModule &CGM,
                                  llvm::PointerType *T,
                                  QualType QT) const override;
+  void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
+                           CodeGen::CodeGenModule &M) const override;
 };
 class SPIRVTargetCodeGenInfo : public CommonSPIRTargetCodeGenInfo {
 public:
@@ -268,6 +270,22 @@ CommonSPIRTargetCodeGenInfo::getNullPointer(const CodeGen::CodeGenModule &CGM,
       llvm::ConstantPointerNull::get(NPT), PT);
 }
 
+void CommonSPIRTargetCodeGenInfo::setTargetAttributes(
+    const Decl *D, llvm::GlobalValue *GV, CodeGen::CodeGenModule &M) const {
+  if (M.getLangOpts().OpenCL || GV->isDeclaration())
+    return;
+
+  const FunctionDecl *FD = dyn_cast<FunctionDecl>(D);
+  if (!FD)
+    return;
+
+  llvm::Function *F = dyn_cast<llvm::Function>(GV);
+  assert(F && "Expected GlobalValue to be a Function");
+
+  if (FD->hasAttr<DeviceKernelAttr>())
+    F->setCallingConv(getDeviceKernelCallingConv());
+}
+
 LangAS
 SPIRVTargetCodeGenInfo::getGlobalVarAddressSpace(CodeGenModule &CGM,
                                                  const VarDecl *D) const {
@@ -292,19 +310,23 @@ SPIRVTargetCodeGenInfo::getGlobalVarAddressSpace(CodeGenModule &CGM,
 
 void SPIRVTargetCodeGenInfo::setTargetAttributes(
     const Decl *D, llvm::GlobalValue *GV, CodeGen::CodeGenModule &M) const {
-  if (!M.getLangOpts().HIP ||
-      M.getTarget().getTriple().getVendor() != llvm::Triple::AMD)
-    return;
   if (GV->isDeclaration())
     return;
 
-  auto F = dyn_cast<llvm::Function>(GV);
-  if (!F)
+  const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D);
+  if (!FD)
     return;
 
-  auto FD = dyn_cast_or_null<FunctionDecl>(D);
-  if (!FD)
+  llvm::Function *F = dyn_cast<llvm::Function>(GV);
+  assert(F && "Expected GlobalValue to be a Function");
+
+  if (FD->hasAttr<DeviceKernelAttr>())
+    F->setCallingConv(getDeviceKernelCallingConv());
+
+  if (!M.getLangOpts().HIP ||
+      M.getTarget().getTriple().getVendor() != llvm::Triple::AMD)
     return;
+
   if (!FD->hasAttr<CUDAGlobalAttr>())
     return;