zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-12-16 01:49:05 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Buffer Donation (#519)

Browse Source 
* buffer donation

* fix to move shared pointer

* format

* gpu in place for copy and binary

* revert ops test

* cpu in place

* a little cleanup

* remove useless bench
This commit is contained in:
Awni Hannun
2024-01-26 16:30:33 -08:00
committed by GitHub
parent 07f35c9d8a
commit 8993382aaa
12 changed files with 199 additions and 178 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
mlx/backend/common/unary.h
View File

@@ -64,15 +64,24 @@ struct RoundOp {
}
};
void set_unary_output_data(const array& in, array& out) {
if (in.is_donatable() && in.itemsize() == out.itemsize()) {
out.copy_shared_buffer(in);
} else {
auto size = in.data_size();
out.set_data(
allocator::malloc_or_wait(size * out.itemsize()),
size,
in.strides(),
in.flags());
}
}
template <typename T, typename Op>
void unary_op(const array& a, array& out, Op op) {
const T* a_ptr = a.data<T>();
if (a.flags().contiguous) {
out.set_data(
allocator::malloc_or_wait(a.data_size() * out.itemsize()),
a.data_size(),
a.strides(),
a.flags());
set_unary_output_data(a, out);
T* dst = out.data<T>();
for (size_t i = 0; i < a.data_size(); ++i) {
dst[i] = op(a_ptr[i]);