Skip to content

Feat add nvcomp - #9

Open
EmilDohne wants to merge 33 commits into
masterfrom
feat-add_nvcomp
Open

Feat add nvcomp#9
EmilDohne wants to merge 33 commits into
masterfrom
feat-add_nvcomp

Conversation

@EmilDohne

Copy link
Copy Markdown
Owner

No description provided.

EmilDohne and others added 3 commits July 10, 2026 23:26
Detect the installed CUDA toolkit and drive both the nvcomp
redistributable variant and GPU architectures from it instead of
hardcoding, fixing illegal-memory-access crashes on newer GPUs
(e.g. Blackwell/sm_120) whose device code was never built.

- New cmake/setup_cuda.cmake: correct find_package(CUDAToolkit),
  resolve COMPRESSED_IMAGE_CUDA_MAJOR (AUTO or 11/12/13 override),
  and default CMAKE_CUDA_ARCHITECTURES to "native" (overridable).
- fetch_nvcomp.cmake: pick nvcomp version + SHA256 per CUDA major
  (11 -> 5.0.0.6, 12/13 -> 5.2.0.10, since 5.2 dropped cuda11).
- Remove the broken find_package(CUDAToolkit REQUIRED <optname>)
  and the misused option() for the CUDA version.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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