@jeffbolznv @0cc4m Looks like Vulkan could do with support too (but at least it didn't crash like CUDA did):
https://github.com/ggml-org/llama.cpp/actions/runs/19436977342/job/55610305113?pr=17326#step:3:7370

I thought Vulkan already supported this, but I'll check what's going on.

@jeffbolznv @0cc4m Looks like Vulkan could do with support too (but at least it didn't crash like CUDA did): https://github.com/ggml-org/llama.cpp/actions/runs/19436977342/job/55610305113?pr=17326#step:3:7370

Fixed in #17328

@slaren gentle ping

While you're at it, can you rename the function to ggml_cpy_scalar and either remove the indentation or make it consistent?

There's a system to the indentation, they are indented to similar calls, but it still looks messy because everything is interleaved, I guess we might as well remove it.

Edit: Looks nicer and much more consistent on their own line.