nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 21 Packages Projects Releases Wiki Activity

metal : fix event synchronization (#22260)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-04-23 08:22:49 +03:00
committed by GitHub
parent 930e0210d1
commit 8635e221c8
1 changed files with 11 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-metal/ggml-metal-device.m
+11 -12
View File
@@ -931,13 +931,13 @@ void ggml_metal_device_rsets_keep_alive(ggml_metal_device_t dev) {
} }
struct ggml_metal_event { struct ggml_metal_event {
void * obj; // id<MTLEvent> void * obj; // id<MTLSharedEvent>
atomic_int value; atomic_int value;
}; };
void ggml_metal_event_encode_signal(ggml_metal_event_t ev, ggml_metal_cmd_buf_t cmd_buf_raw) { void ggml_metal_event_encode_signal(ggml_metal_event_t ev, ggml_metal_cmd_buf_t cmd_buf_raw) {
id<MTLEvent> event = (id<MTLEvent>)ev->obj; id<MTLSharedEvent> event = (id<MTLSharedEvent>)ev->obj;
id<MTLCommandBuffer> cmd_buf = (id<MTLCommandBuffer>) cmd_buf_raw; id<MTLCommandBuffer> cmd_buf = (id<MTLCommandBuffer>) cmd_buf_raw;
@@ -945,7 +945,7 @@ void ggml_metal_event_encode_signal(ggml_metal_event_t ev, ggml_metal_cmd_buf_t
} }
void ggml_metal_event_encode_wait(ggml_metal_event_t ev, ggml_metal_cmd_buf_t cmd_buf_raw) { void ggml_metal_event_encode_wait(ggml_metal_event_t ev, ggml_metal_cmd_buf_t cmd_buf_raw) {
id<MTLEvent> event = (id<MTLEvent>)ev->obj; id<MTLSharedEvent> event = (id<MTLSharedEvent>)ev->obj;
id<MTLCommandBuffer> cmd_buf = (id<MTLCommandBuffer>) cmd_buf_raw; id<MTLCommandBuffer> cmd_buf = (id<MTLCommandBuffer>) cmd_buf_raw;
@@ -953,7 +953,7 @@ void ggml_metal_event_encode_wait(ggml_metal_event_t ev, ggml_metal_cmd_buf_t cm
} }
ggml_metal_event_t ggml_metal_device_event_init(ggml_metal_device_t dev) { ggml_metal_event_t ggml_metal_device_event_init(ggml_metal_device_t dev) {
id<MTLEvent> event = [dev->mtl_device newEvent]; id<MTLSharedEvent> event = [dev->mtl_device newSharedEvent];
ggml_metal_event_t ev = calloc(1, sizeof(struct ggml_metal_event)); ggml_metal_event_t ev = calloc(1, sizeof(struct ggml_metal_event));
@@ -964,7 +964,7 @@ ggml_metal_event_t ggml_metal_device_event_init(ggml_metal_device_t dev) {
} }
void ggml_metal_device_event_free(ggml_metal_device_t dev, ggml_metal_event_t ev) { void ggml_metal_device_event_free(ggml_metal_device_t dev, ggml_metal_event_t ev) {
id<MTLEvent> event = ev->obj; id<MTLSharedEvent> event = ev->obj;
[event release]; [event release];
free(ev); free(ev);
@@ -973,14 +973,13 @@ void ggml_metal_device_event_free(ggml_metal_device_t dev, ggml_metal_event_t ev
} }
void ggml_metal_device_event_synchronize(ggml_metal_device_t dev, ggml_metal_event_t ev) { void ggml_metal_device_event_synchronize(ggml_metal_device_t dev, ggml_metal_event_t ev) {
@autoreleasepool { id<MTLSharedEvent> event = ev->obj;
id<MTLEvent> event = ev->obj; const bool res = [event waitUntilSignaledValue:atomic_load_explicit(&ev->value, memory_order_relaxed) timeoutMS:60000];
if (!res) {
id<MTLCommandBuffer> cmd_buf = [dev->mtl_queue commandBuffer]; GGML_ABORT("%s: failed to wait for event\n", __func__);
[cmd_buf encodeWaitForEvent:event value:atomic_load_explicit(&ev->value, memory_order_relaxed)];
[cmd_buf commit];
[cmd_buf waitUntilCompleted];
} }
GGML_UNUSED(dev);
} }
void ggml_metal_device_get_memory(ggml_metal_device_t dev, size_t * free, size_t * total) { void ggml_metal_device_get_memory(ggml_metal_device_t dev, size_t * free, size_t * total) {