[clang] Use TargetInfo to determine device kernel calling convention (#144728)

We should abstract this logic away to `TargetInfo`. See
#137882 for more information.

---------

Signed-off-by: Sarnie, Nick <[email protected]>

Author: sarnex

clang/lib/CodeGen/CGCall.cpp

@@ -83,17 +83,8 @@ unsigned CodeGenTypes::ClangCallConvToLLVMCallConv(CallingConv CC) {
     return llvm::CallingConv::AArch64_SVE_VectorCall;
   case CC_SpirFunction:
     return llvm::CallingConv::SPIR_FUNC;
-  case CC_DeviceKernel: {
-    if (CGM.getLangOpts().OpenCL)
-      return CGM.getTargetCodeGenInfo().getOpenCLKernelCallingConv();
-    if (CGM.getTriple().isSPIROrSPIRV())
-      return llvm::CallingConv::SPIR_KERNEL;
-    if (CGM.getTriple().isAMDGPU())
-      return llvm::CallingConv::AMDGPU_KERNEL;
-    if (CGM.getTriple().isNVPTX())
-      return llvm::CallingConv::PTX_Kernel;
-    llvm_unreachable("Unknown kernel calling convention");
-  }
+  case CC_DeviceKernel:
+    return CGM.getTargetCodeGenInfo().getDeviceKernelCallingConv();
   case CC_PreserveMost:
     return llvm::CallingConv::PreserveMost;
   case CC_PreserveAll:

clang/lib/CodeGen/TargetInfo.cpp

@@ -103,18 +103,21 @@ TargetCodeGenInfo::getDependentLibraryOption(llvm::StringRef Lib,
   Opt += Lib;
 }
 
-unsigned TargetCodeGenInfo::getOpenCLKernelCallingConv() const {
-  // OpenCL kernels are called via an explicit runtime API with arguments
-  // set with clSetKernelArg(), not as normal sub-functions.
-  // Return SPIR_KERNEL by default as the kernel calling convention to
-  // ensure the fingerprint is fixed such way that each OpenCL argument
-  // gets one matching argument in the produced kernel function argument
-  // list to enable feasible implementation of clSetKernelArg() with
-  // aggregates etc. In case we would use the default C calling conv here,
-  // clSetKernelArg() might break depending on the target-specific
-  // conventions; different targets might split structs passed as values
-  // to multiple function arguments etc.
-  return llvm::CallingConv::SPIR_KERNEL;
+unsigned TargetCodeGenInfo::getDeviceKernelCallingConv() const {
+  if (getABIInfo().getContext().getLangOpts().OpenCL) {
+    // Device kernels are called via an explicit runtime API with arguments,
+    // such as set with clSetKernelArg() for OpenCL, not as normal
+    // sub-functions. Return SPIR_KERNEL by default as the kernel calling
+    // convention to ensure the fingerprint is fixed such way that each kernel
+    // argument gets one matching argument in the produced kernel function
+    // argument list to enable feasible implementation of clSetKernelArg() with
+    // aggregates etc. In case we would use the default C calling conv here,
+    // clSetKernelArg() might break depending on the target-specific
+    // conventions; different targets might split structs passed as values
+    // to multiple function arguments etc.
+    return llvm::CallingConv::SPIR_KERNEL;
+  }
+  llvm_unreachable("Unknown kernel calling convention");
 }
 
 void TargetCodeGenInfo::setOCLKernelStubCallingConvention(

clang/lib/CodeGen/TargetInfo.h

@@ -298,8 +298,8 @@ class TargetCodeGenInfo {
                                        llvm::StringRef Value,
                                        llvm::SmallString<32> &Opt) const {}
 
-  /// Get LLVM calling convention for OpenCL kernel.
-  virtual unsigned getOpenCLKernelCallingConv() const;
+  /// Get LLVM calling convention for device kernels.
+  virtual unsigned getDeviceKernelCallingConv() const;
 
   /// Get target specific null pointer.
   /// \param T is the LLVM type of the null pointer.

clang/lib/CodeGen/Targets/AMDGPU.cpp

@@ -304,7 +304,7 @@ class AMDGPUTargetCodeGenInfo : public TargetCodeGenInfo {
 
   void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                            CodeGen::CodeGenModule &M) const override;
-  unsigned getOpenCLKernelCallingConv() const override;
+  unsigned getDeviceKernelCallingConv() const override;
 
   llvm::Constant *getNullPointer(const CodeGen::CodeGenModule &CGM,
       llvm::PointerType *T, QualType QT) const override;
@@ -431,7 +431,7 @@ void AMDGPUTargetCodeGenInfo::setTargetAttributes(
     F->addFnAttr("amdgpu-ieee", "false");
 }
 
-unsigned AMDGPUTargetCodeGenInfo::getOpenCLKernelCallingConv() const {
+unsigned AMDGPUTargetCodeGenInfo::getDeviceKernelCallingConv() const {
   return llvm::CallingConv::AMDGPU_KERNEL;
 }
 

clang/lib/CodeGen/Targets/NVPTX.cpp

@@ -78,7 +78,7 @@ class NVPTXTargetCodeGenInfo : public TargetCodeGenInfo {
     return true;
   }
 
-  unsigned getOpenCLKernelCallingConv() const override {
+  unsigned getDeviceKernelCallingConv() const override {
     return llvm::CallingConv::PTX_Kernel;
   }
 

clang/lib/CodeGen/Targets/SPIR.cpp

@@ -51,7 +51,7 @@ class CommonSPIRTargetCodeGenInfo : public TargetCodeGenInfo {
         getABIInfo().getDataLayout().getAllocaAddrSpace());
   }
 
-  unsigned getOpenCLKernelCallingConv() const override;
+  unsigned getDeviceKernelCallingConv() const override;
   llvm::Type *getOpenCLType(CodeGenModule &CGM, const Type *T) const override;
   llvm::Type *
   getHLSLType(CodeGenModule &CGM, const Type *Ty,
@@ -219,7 +219,7 @@ void computeSPIRKernelABIInfo(CodeGenModule &CGM, CGFunctionInfo &FI) {
 }
 }
 
-unsigned CommonSPIRTargetCodeGenInfo::getOpenCLKernelCallingConv() const {
+unsigned CommonSPIRTargetCodeGenInfo::getDeviceKernelCallingConv() const {
   return llvm::CallingConv::SPIR_KERNEL;
 }