Skip to content

replace gpu_device index with opaque device handles - #129

Merged
cjpais merged 4 commits into
mainfrom
device-enumeration-api-change
Aug 17, 2026
Merged

replace gpu_device index with opaque device handles#129
cjpais merged 4 commits into
mainfrom
device-enumeration-api-change

Conversation

@cjpais

@cjpais cjpais commented Aug 15, 2026

Copy link
Copy Markdown
Contributor

aimed at fixing cjpais/Handy#1894

Basically, our device index stuff kind of got a mess when I implemented auto. So we're just going to do a proper implementation, largely following the steps of llama.cpp in terms of how we enumerate and select devices

@cjpais
cjpais merged commit 9315160 into main Aug 17, 2026
25 checks passed
cunba-ai pushed a commit to cunba-ai/transcribe.cpp that referenced this pull request Aug 20, 2026
handy-computer#129)

The merge of handy-computer handy-computer#129 renamed the device C API but the
dynload posture's symbol table/trampolines (transcribe_dyn.rs) are
fork-owned and were not covered upstream: the stale table still listed
the four removed symbols (a DLL built from the merged tree would fail
the eager dynload resolve, disabling the backend) and lacked the five
new ones.

- xtask: teach not_loaded_default the transcribe_device_t handle type
  (null is both the not-loaded default and the API's own sentinel).
- transcribe_sys.rs: declare transcribe_build_id() (present in the
  public header, missing from the committed binding).
- cargo xtask bindgen: refresh both files against abihash 7df72bf9;
  enum underlying types stay c_uint (this machine's libclang resolves
  them as c_int — kept out of the commit deliberately).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant