CODEWITHSUNDEEP
compiler-constructionllvmclanggpgpu
dalibocai
dalibocai

Reputation: 2347

Compiling CUDA or OpenCL to llvm IR

Clang can translate CUDA(not complete) or OpenCL to llvm. How can it preserve the "shared" keyword, such that it knows the difference between shared and global memory?

Upvotes: 2

Views: 1076

Answers (1)

Volodymyr Rudyi
Volodymyr Rudyi

Reputation: 648

I`m not familiar with CUDA or OpenCL, but can advice you to look at resulting LLVM IR that is generated by frontend and see what code is being generated. Also you can look at corresponding LLVM backend to see how it utilizes shared/global memory.

Upvotes: 0

Related Questions