Wagner Bruna

I don't think OpenCL support _itself_ could cause this, because it's a build-time option, disabled by default (and it'd appear as "Using OpenCL backend" if it were enabled by mistake:...

By the way, any chance you are running sd.exe with an unmatched stable-diffusion.dll (from another release, for instance)? Even that shouldn't cause problems between these two releases, but...

Sounds like missing quant types for GET_ROWS (as in #851 ). What types do you see in src0->type and src1->type ?

> Ok with today's version it works, although the output is gibberish and actually all outputs with today's version with Ubuntu and Vulkan produce gibberish with most models :( This...

> https://github.com/CLDawes/ggml/tree/patch-qwen-image I guess this would also fix #857 ? @CLDawes , ggml changes are usually applied to llama.cpp first, then extracted to the lib, then pulled in by sd.cpp;...

Also, IIRC OpenCL benefits from both `--diffusion-conv-direct` and `--vae-conv-direct` (faster inference, reduced memory usage).

Just came up with an alternative, too: https://github.com/wbruna/stable-diffusion.cpp/commit/2050ffe5816653a381769ab6b0e5e7f52402cefd (looks like the same algorithm): | boost | 2050ffe58166 | |---|---| | | |

> For me that looks exactly different, lol, so like my first fake implementaion without boost or simple, compare it with the actual implementaion its exact what boost does but...

This also has a nice side-effect on some low-CFG distilled models: the skipped steps help avoid the "overcooked" effect when using too many steps. > I came up with the...

> I have thought about that, but I'm planning to make more changes to make it depth-aware, so it will diverge from the original EasyCache implementation. Since the latter is...