Package Details: stable-diffusion.cpp-cublas-git r629.caa823a-1

Git Clone URL: https://aur.archlinux.org/stable-diffusion.cpp-cublas-git.git (read-only, click to copy)
Package Base: stable-diffusion.cpp-cublas-git
Description: Stable Diffusion in pure C/C++ (with NVIDIA CUDA optimizations)
Upstream URL: https://github.com/leejet/stable-diffusion.cpp
Licenses: MIT
Conflicts: stable-diffusion.cpp, stable-diffusion.cpp-cublas
Provides: stable-diffusion.cpp, stable-diffusion.cpp-cublas
Submitter: robertfoster
Maintainer: yataro
Last Packager: yataro
Votes: 1
Popularity: 0.003033
First Submitted: 2025-04-12 14:27 (UTC)
Last Updated: 2026-05-19 09:19 (UTC)

Latest Comments

yataro commented on 2026-05-19 09:28 (UTC)

Package has been refactored, I simplified PKGBUILD and adjusted missing dependencies for frontend build. The package now correctly installs the shared stable‑diffusion library, which also led to a nearly x2 size reduction because previously sd‑cli and sd‑server were linked with a static one.

Davidyz commented on 2026-05-07 07:12 (UTC)

The build fails with the recent upgrade of gcc16. The current PKGBUILD hardcode nvcc to use gcc, which now points to gcc16. It's probably related to cuda not supporting gcc16 yet.

Suggested fix: change the gcc in makedeps to gcc15 (it's now in the official extra repo) and remove the hardcoded NVCC_CCBIN="gcc" in the build() function. This will make nvcc honour the NVCC_CCBIN="/usr/bin/g++-15" setting in /etc/profile.d/cuda.sh provided by the cuda package, and should hopefully fix the building.

mven commented on 2025-04-28 01:05 (UTC)

I'm also experiencing the same error @dcNdrew. It's currently preventing me from making this package.

DcNdrew commented on 2025-04-25 17:58 (UTC)

There is an error. error: too few arguments to function ‘ggml_tensor ggml_upscale_ext(ggml_context, ggml_tensor*, int, int, int, int, ggml_scale_mode)’