-
Notifications
You must be signed in to change notification settings - Fork 8.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Wrong detection count threads in NUMA configuration in Windows #5524
Labels
Comments
@ggerganov, can anyone confirm bug? Here is another good example of bug fixing: |
This issue was closed because it has been inactive for 14 days since being marked as stale. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
@ggerganov,
I solved similar issue in different projects on github, solution is simple to make wrapper function for geting threads, in Windows it is necessary to take into accumulate all logical processors in all NUMA nodes.
Problem lines in common, tests and examples llama.cpp:
llama.cpp/common/common.cpp
Lines 85 to 86 in 5f5808c
llama.cpp/common/common.cpp
Lines 160 to 187 in 5f5808c
llama.cpp/common/common.cpp
Lines 187 to 196 in 5f5808c
llama.cpp/common/common.cpp
Lines 1089 to 1099 in 5f5808c
Solutions:
[❌] C variant detection is not done here: git-for-windows/git#4766
[✔️] C++11 Windows XP minimal (rewriten modern variant by @mrexodia): x64dbg/x64dbg@d2f6ba7
[✔️] Modern C++17 (variant by @GermanAizek and #llvm-project maintainers): GermanAizek/llvm-project@d1fa25f
more optimized variant detection NUMA and return count threads on host, must added before
GetLogicalProcessorInformationEx
:The text was updated successfully, but these errors were encountered: