Skip to content

Conversation

@danielzgtg
Copy link
Contributor

This allows and defaults to setting a separate compiler for ROCm.

Before

  • AMD ROCm Clang is much slower than GCC
  • I get errors when trying to embed this in a Qt project while using ROCm Clang
  • It's extra friction to set the compiler separately

After

  • Build times are faster now that it's using GCC
  • The compiler is detected automatically after checking the LLAMA_HIPBLAS box in cmake-gui for non-Windows. You can still override it with LLAMA_ROCM_PROJECT_CMAKE_ARGS

Alternatives

  • I wanted to statically link the part split out to recombine it into one executable as before and avoid the new libggml-rocm.so. However, I was unable to get CMake to link it without errors, and I would appreciate any ideas how to.

A separate CMake project allows setting a separate compiler specifically
for ROCm. This also makes the compile config easier eliminating the need
for setting the ROCm compiler path. Most importantly, this lets me build
llama.cpp with GCC as I get errors while trying to embed this in a Qt
application otherwise.
@mofosyne mofosyne marked this pull request as draft May 25, 2024 14:12
@mofosyne mofosyne added the obsolete? Marker for potentially obsolete PR label May 25, 2024
@danielzgtg
Copy link
Contributor Author

Superseded by #10256

@danielzgtg danielzgtg closed this Jan 10, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

obsolete? Marker for potentially obsolete PR

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants