Move to nanobind #455
Move to nanobind #455
Conversation
flying-sheep
left a comment
flying-sheep
left a comment
There was a problem hiding this comment.
Oh wow, that looks super cool! Addresses all my previous concerns about manual specification of magic numbers for chunking, repetitive kernel wrapper code, and so on!
| from rapids_singlecell._cuda import _qc_cuda as _qc | ||
|
|
||
| if sparse.isspmatrix_csr(X): | ||
| _qc.sparse_qc_csr( |
flying-sheep
force-pushed
the
move-to-nanobind
branch
from
2105d37 to
dddd9e8
Compare
flying-sheep
left a comment
flying-sheep
left a comment
There was a problem hiding this comment.
looks awesome, the module exports are much nicer now!
This is missing stub generation, otherwise really nice!
| static inline void launch_auc(std::uintptr_t ranks, int R, int C, std::uintptr_t cnct, | ||
| std::uintptr_t starts, std::uintptr_t lens, int n_sets, int n_up, | ||
| std::uintptr_t max_aucs, std::uintptr_t es, cudaStream_t stream) { | ||
| static inline void launch_auc(const int* ranks, int R, int C, const int* cnct, const int* starts, | ||
| const int* lens, int n_sets, int n_up, const float* max_aucs, | ||
| float* es, cudaStream_t stream) { |
There was a problem hiding this comment.
why do these stop being a pointer sized non-nullable ints and become a 🤷-sized signed ints? (same in many other places)
There was a problem hiding this comment.
because they are now typed by feeding them through nanobind
There was a problem hiding this comment.
so nanobind creates all this? then this shouldn’t be checked into Git and just be part of the build system.
There was a problem hiding this comment.
no this has to be specifically written for each function. I dont know if there is an abstraction that auto creates this in the build system
This for now is a test with precompiled cuda extensions.