vendor/llvm-project/llvmorg-15-init-17485-ga3e38b4a206b

1 files changed, 13 insertions, 1 deletions

diff --git a/clang/lib/CodeGen/TargetInfo.cpp b/clang/lib/CodeGen/TargetInfo.cpp
index 8eaed1db8e7d..d1ee61eab9d6 100644
--- a/clang/lib/CodeGen/TargetInfo.cpp
+++ b/clang/lib/CodeGen/TargetInfo.cpp
@@ -35,7 +35,7 @@
 #include "llvm/IR/Type.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
-#include <algorithm> // std::sort
+#include <algorithm>
 
 using namespace clang;
 using namespace CodeGen;
@@ -443,6 +443,9 @@ static Address emitMergePHI(CodeGenFunction &CGF,
   return Address(PHI, Addr1.getElementType(), Align);
 }
 
+TargetCodeGenInfo::TargetCodeGenInfo(std::unique_ptr<ABIInfo> Info)
+    : Info(std::move(Info)) {}
+
 TargetCodeGenInfo::~TargetCodeGenInfo() = default;
 
 // If someone can figure out a general rule for this, that would be great.
@@ -10446,6 +10449,15 @@ ABIArgInfo SPIRVABIInfo::classifyKernelArgumentType(QualType Ty) const {
       LTy = llvm::PointerType::getWithSamePointeeType(PtrTy, GlobalAS);
       return ABIArgInfo::getDirect(LTy, 0, nullptr, false);
     }
+
+    // Force copying aggregate type in kernel arguments by value when
+    // compiling CUDA targeting SPIR-V. This is required for the object
+    // copied to be valid on the device.
+    // This behavior follows the CUDA spec
+    // https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#global-function-argument-processing,
+    // and matches the NVPTX implementation.
+    if (isAggregateTypeForABI(Ty))
+      return getNaturalAlignIndirect(Ty, /* byval */ true);
   }
   return classifyArgumentType(Ty);
 }