Are there plans to support acceleration for the stable-diffusion.cpp image generation project on OpenHarmony?
There are currently no plans for this. If you’re interested, you can download the code and the cross-compilation toolchain at https://archive.spacemit.com/toolchain/spacemit-toolchain-linux-musl-x86_64-oh-20260630.tar.xz and try compiling it yourself. We can provide support if you encounter any issues. Please note that our official acceleration code has not been submitted to this project, so you may not be able to use the accelerator cores. Support for K3 inference acceleration operators is available in ggml in the official llama.cpp source code.
The latest version of ggml already includes the corresponding acceleration code, so you can compile it yourself to verify.
OK, I’ll try it myself.