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

[Orphic Dreamer via cfe-commits]

Siya-05 wrote:

This was moved to clang/lib/CIR/CodeGen/Targets/NVPTX.cpp per the earlier review feedback. The conflict resolution preserves that refactoring.

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