happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

metal : fix build(#17799) 

* metal : fix build

* tests : fix context destruction
This commit is contained in:
Georgi Gerganov 2025-12-06 09:33:59 +02:00 committed by GitHub
parent 67788f6846
commit 8ce774a102
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 23 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
examples/save-load-state/save-load-state.cpp
View File

@ -241,6 +241,12 @@ int main(int argc, char ** argv) {
llama_batch_free(batch); llama_batch_free(batch);
// this one is managed by common_init_result
//llama_free(ctx);
llama_free(ctx2);
llama_free(ctx3);
if (result0 != result2) { if (result0 != result2) {
fprintf(stderr, "\n%s : error : the seq restore generation is different\n", __func__); fprintf(stderr, "\n%s : error : the seq restore generation is different\n", __func__);
return 1; return 1;

8
ggml/src/ggml-metal/ggml-metal-device.m
View File

@ -574,6 +574,8 @@ ggml_metal_rsets_t ggml_metal_rsets_init(void) {
// the requests stop after a certain amount of time (keep_alive_s) of inactivity // the requests stop after a certain amount of time (keep_alive_s) of inactivity
dispatch_queue_t d_queue = dispatch_get_global_queue(QOS_CLASS_DEFAULT, 0); dispatch_queue_t d_queue = dispatch_get_global_queue(QOS_CLASS_DEFAULT, 0);
dispatch_group_async(res->d_group, d_queue, ^{ dispatch_group_async(res->d_group, d_queue, ^{
#if defined(GGML_METAL_HAS_RESIDENCY_SETS)
if (@available(macOS 15.0, iOS 18.0, tvOS 18.0, visionOS 2.0, *)) {
while (!atomic_load_explicit(&res->d_stop, memory_order_relaxed)) { while (!atomic_load_explicit(&res->d_stop, memory_order_relaxed)) {
if (atomic_load_explicit(&res->d_loop, memory_order_relaxed) > 0) { if (atomic_load_explicit(&res->d_loop, memory_order_relaxed) > 0) {
[res->lock lock]; [res->lock lock];
@ -590,6 +592,8 @@ ggml_metal_rsets_t ggml_metal_rsets_init(void) {
// half a second // half a second
usleep(500 * 1000); usleep(500 * 1000);
} }
}
#endif
}); });
return res; return res;
@ -600,6 +604,7 @@ void ggml_metal_rsets_free(ggml_metal_rsets_t rsets) {
return; return;
} }
// note: if you hit this assert, most likely you haven't deallocated all Metal resources before exiting
GGML_ASSERT([rsets->data count] == 0); GGML_ASSERT([rsets->data count] == 0);
atomic_store_explicit(&rsets->d_stop, true, memory_order_relaxed); atomic_store_explicit(&rsets->d_stop, true, memory_order_relaxed);
@ -787,9 +792,6 @@ ggml_metal_device_t ggml_metal_device_init(void) {
dev->rsets = nil; dev->rsets = nil;
} }
// --------------------------------------------------
// print MTL GPU family: // print MTL GPU family:
GGML_LOG_INFO("%s: GPU name: %s\n", __func__, dev->props.name); GGML_LOG_INFO("%s: GPU name: %s\n", __func__, dev->props.name);