Replies: 1 comment 2 replies
-
Good thinking, but I fear this is going to be a headache on Windows. I thought about defining the functions in |
Beta Was this translation helpful? Give feedback.
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Summary
Split ggml into shared libraries for each supported CPU hardware platform, that are dynamically loaded at runtime.
Background
ld.so
has a feature to load shared libraries based on the executing CPU hardware platform at runtime/load-time (not compile time).to quote
man ld.so
In addition manual loading of shared libraries could be done via
dlopen()
for platforms thatld.so
doesn't recognize, like CUDA.Pros
#ifdef
hellCons
This
ld.so
feature came to mind, given current efforts with CUDA and the creation ofggml-cuda.cu
. Why not split up the code more cleanly with a generalized hardware based structure?Beta Was this translation helpful? Give feedback.
All reactions