ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 17:20:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

Undo sync reduction (#1063)

Browse Source 
I'm finding issues for Qwen3-MoE

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-12-13 16:58:32 +01:00
committed by GitHub
parent 093cc7c380
commit 2e04b7cbef
1 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
ggml/src/ggml-backend.cpp
View File

@@ -1898,7 +1898,7 @@ static void ggml_backend_sched_copy_inputs(ggml_backend_sched_t sched, ggml_back
} else { } else {
ggml_backend_synchronize(split_backend); ggml_backend_synchronize(split_backend);
} }
needs_sync[split_backend_id] = false; //needs_sync[split_backend_id] = false;
} }
ggml_backend_tensor_copy(input, input_cpy); ggml_backend_tensor_copy(input, input_cpy);
} else { } else {
@@ -1909,7 +1909,7 @@ static void ggml_backend_sched_copy_inputs(ggml_backend_sched_t sched, ggml_back
} else { } else {
ggml_backend_synchronize(split_backend); ggml_backend_synchronize(split_backend);
} }
needs_sync[split_backend_id] = false; //needs_sync[split_backend_id] = false;
} }
ggml_tensor * node = split->graph.nodes[0]; ggml_tensor * node = split->graph.nodes[0];
@@ -1945,7 +1945,7 @@ static void ggml_backend_sched_copy_inputs(ggml_backend_sched_t sched, ggml_back
ggml_backend_tensor_get_async(ids_backend, ids_tensor, ids.data(), 0, ggml_nbytes(ids_tensor)); ggml_backend_tensor_get_async(ids_backend, ids_tensor, ids.data(), 0, ggml_nbytes(ids_tensor));
ggml_backend_synchronize(ids_backend); ggml_backend_synchronize(ids_backend);
needs_sync[tensor_backend_id(ids_tensor)] = false; //needs_sync[tensor_backend_id(ids_tensor)] = false;
unique_ids.resize((n_expert + 31)/32); unique_ids.resize((n_expert + 31)/32);
std::memset(unique_ids.data(), 0, unique_ids.size()*sizeof(uint32_t)); std::memset(unique_ids.data(), 0, unique_ids.size()*sizeof(uint32_t));
@@ -2005,7 +2005,7 @@ static void ggml_backend_sched_copy_inputs(ggml_backend_sched_t sched, ggml_back
int input_backend_id = tensor_backend_id(input); int input_backend_id = tensor_backend_id(input);
if (needs_sync[input_backend_id]) { if (needs_sync[input_backend_id]) {
ggml_backend_synchronize(input_backend); ggml_backend_synchronize(input_backend);
needs_sync[input_backend_id] = false; //needs_sync[input_backend_id] = false;
} }
if (needs_sync[split_backend_id]) { if (needs_sync[split_backend_id]) {
if (sched->events[split_backend_id][sched->cur_copy] != NULL) { if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
@@ -2013,7 +2013,7 @@ static void ggml_backend_sched_copy_inputs(ggml_backend_sched_t sched, ggml_back
} else { } else {
ggml_backend_synchronize(split_backend); ggml_backend_synchronize(split_backend);
} }
needs_sync[split_backend_id] = false; //needs_sync[split_backend_id] = false;
} }
ggml_backend_tensor_copy(input, input_cpy); ggml_backend_tensor_copy(input, input_cpy);
} }
@@ -2092,7 +2092,7 @@ static ggml_status ggml_backend_sched_compute_splits_sm_graph(ggml_backend_sched
static enum ggml_status ggml_backend_sched_compute_splits(ggml_backend_sched_t sched) { static enum ggml_status ggml_backend_sched_compute_splits(ggml_backend_sched_t sched) {
if (sched->split_mode_graph) { if (false && sched->split_mode_graph) {
return ggml_backend_sched_compute_splits_sm_graph(sched); return ggml_backend_sched_compute_splits_sm_graph(sched);
} }