Server stop working after recent refactor on Windows

[sorasoras]

latest branch
server stop working after recently refactor.
I couldn't run server.exe on my Windows build Rocm

PS W:\git\llama.cpp\AVX2\bin> .\server.exe -h
PS W:\git\llama.cpp\AVX2\bin> .\server.exe
PS W:\git\llama.cpp\AVX2\bin>

compile with

cmake .. -G "Ninja" -DCMAKE_BUILD_TYPE=Release -DLLAMA_HIPBLAS=ON -DCMAKE_C_COMPILER="C:/Program Files/AMD/ROCm/5.7/bin/clang.exe" -DCMAKE_CXX_COMPILER="C:/Program Files/AMD/ROCm/5.7/bin/clang++.exe" -DAMDGPU_TARGETS="gfx1100;gfx1030;gfx1031"

[characharm]

The issue is not only with ROCm but also with Vulkan and even AVX2.

[FredericoPerimLopes]

I'm also unable to load server on windows build with CUDA.

[ggerganov]

Seems like a Windows-related problem, so unfortunately I won't be able to help with that.

Could you try to pin-point the exact commit after which it stops working?

[sorasoras]

Seems like a Windows-related problem, so unfortunately I won't be able to help with that.

Could you try to pin-point the exact commit after which it stops working?

#5882
this one but this is a huge pr

[slaren]

This happens because ggml_time_us is being called (in server_metrics) before ggml_time_init, which results in a division by zero as timer_freq is zero.

[ggerganov]

Should be fixed now

[phymbert]

Thanks @ggerganov is it worth the effort to add a ci workflow for server tests on Windows ? if yes, I will give it a try next week

[ggerganov]

Yes, would be useful to have a Windows run