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

move memory APIs into top level mlx.core (#1982)

Browse Source
This commit is contained in:
Awni Hannun
2025-03-21 07:25:12 -07:00
committed by GitHub
parent 65a38c452b
commit 4e1994e9d7
25 changed files with 418 additions and 323 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
mlx/transforms.cpp
View File

@@ -12,6 +12,7 @@
#include "mlx/backend/cpu/eval.h"
#include "mlx/backend/metal/metal_impl.h"
#include "mlx/fence.h"
#include "mlx/memory.h"
#include "mlx/ops.h"
#include "mlx/primitives.h"
#include "mlx/scheduler.h"
@@ -219,7 +220,7 @@ array eval_impl(std::vector<array> outputs, bool async) {
}
if (scheduler::n_active_tasks() > MAX_ACTIVE_TASKS ||
(metal::get_active_memory() > metal::get_memory_limit() &&
(get_active_memory() > get_memory_limit() &&
scheduler::n_active_tasks() > 0)) {
// Commit any open streams
for (auto& [_, e] : events) {
@@ -228,8 +229,7 @@ array eval_impl(std::vector<array> outputs, bool async) {
}
}
scheduler::wait_for_one();
// TODO memory api should be moved out of metal
while (metal::get_active_memory() > metal::get_memory_limit() &&
while (get_active_memory() > get_memory_limit() &&
scheduler::n_active_tasks() > 0) {
scheduler::wait_for_one();
}