ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-29 19:01:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

Slightly better graph parallel for Qwen3-Next (#1307)

Browse Source 
* Make sure we pick the reduced tensor from the right GPU

* Minor
This commit is contained in:
Kawrakow
2026-02-24 15:22:30 +01:00
committed by GitHub
parent cfb6747776
commit 7065488135
4 changed files with 30 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
ggml/include/ggml-backend.h
View File

@@ -186,6 +186,7 @@ extern "C" {
GGML_API int ggml_backend_sched_get_n_backends(ggml_backend_sched_t sched);
GGML_API ggml_backend_t ggml_backend_sched_get_backend(ggml_backend_sched_t sched, int i);
GGML_API int ggml_backend_sched_get_backend_idx(ggml_backend_sched_t sched, ggml_backend_buffer_t buffer);
// Get the number of splits of the last graph
GGML_API int ggml_backend_sched_get_n_splits(ggml_backend_sched_t sched);

10
ggml/src/ggml-backend.cpp
View File

@@ -2664,6 +2664,16 @@ ggml_backend_t ggml_backend_sched_get_backend(ggml_backend_sched_t sched, int i)
return sched->backends[i];
}
int ggml_backend_sched_get_backend_idx(ggml_backend_sched_t sched, ggml_backend_buffer_t buffer) {
if (!buffer || !buffer->buft) return -1;
if (buffer && buffer->buft) {
for (int i = 0; i < sched->n_backends; ++i) {
if (ggml_backend_get_default_buffer_type(sched->backends[i]) == buffer->buft) return i;
}
}
return -1;
}
size_t ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend) {
int backend_index = ggml_backend_sched_backend_id(sched, backend);
GGML_ASSERT(backend_index >= 0 && backend_index < sched->n_backends);