[clang] [CIR][CUDA] Support built-in CUDA surface type (PR #196079)

[Orphic Dreamer via cfe-commits]

Siya-05 wrote:

@Men-cotton 

I investigated the failures under clang/test/CIR/CodeGenCUDA/ and noticed they seem related to the NVPTX CIR target implementation rather than the CUDA surface type change itself.

Originally, the NVPTX code lived in TargetInfo.cpp. After the review suggestion to move the NVPTX-specific implementation into Targets/NVPTX.cpp, I tried that refactor, but the build failed due to references to CIR APIs that don't appear to exist in the current tree (e.g. cir::CallingConv::PTXKernel, FuncOp::setCallingConv, MissingFeatures::emitNVVMMetadata, and MissingFeatures::handleCUDALaunchBoundsAttr).

I then moved the implementation back into TargetInfo.cpp to verify whether this was a file-placement issue, but the same errors remained, Can you clarify what exactly is happening?

https://github.com/llvm/llvm-project/pull/196079