mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 29 Packages Projects Releases Wiki Activity

CUDA: Improve performance via less synchronizations between token (#17795)

Browse Source 
* Adds CPU-to-CUDA copy capability to
ggml_backend_cuda_cpy_tensor_async()

* Adds function to relax sync requirements between input copies on
supported backends (CUDA for now)

* Exchanges synchronous copy with async copy function.

* Adds macro guards to allow compilation in non-CUDA builds

* Reworked backend detection in ggml-backend.cpp to avoid linking
conflicts

* Relax requirement of checks in async CUDA copies from backend and buffer type to just buffer type, to avoid linking issues

* Minor cleanup

* Makes opt-in to relax use of explicit syncs more general. Backends like
vulkan which require a synchronization between HtoD copies and graph
execution could also adopt this change now.

* Reintroduces stricter check for CPU->CUDA backend async copy via
GGML_DEVICE_TYPE_CPU.

* Corrects initialization of ggml_backend_sync_mode in
ggml_backend_sched_split initialization

* Simplifies synchronizations to adhere to `saaasg` pattern.

* Apply suggestion from @ggerganov (src->buffer to buf_src)

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* Apply suggestion from @ggerganov (src->buffer to buf_src) v2

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Andreas Kieslinger
2026-03-05 12:53:21 +01:00
committed by GitHub
parent 872646b30c
commit 2cd20b72ed
2 changed files with 19 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-backend.cpp
+9 -5
View File
@@ -1455,6 +1455,10 @@ static enum ggml_status ggml_backend_sched_compute_splits(ggml_backend_sched_t s
int split_backend_id = split->backend_id;
ggml_backend_t split_backend = sched->backends[split_backend_id];
if (sched->events[split_backend_id][sched->cur_copy] == NULL) {
ggml_backend_synchronize(split_backend);
}
// copy the input tensors to the split backend
for (int input_id = 0; input_id < split->n_inputs; input_id++) {
ggml_backend_t input_backend = ggml_backend_sched_get_tensor_backend(sched, split->inputs[input_id]);
@@ -1465,16 +1469,12 @@ static enum ggml_status ggml_backend_sched_compute_splits(ggml_backend_sched_t s
// inputs from the user must be copied immediately to prevent the user overwriting the data before the copy is done
if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
ggml_backend_event_synchronize(sched->events[split_backend_id][sched->cur_copy]);
} else {
ggml_backend_synchronize(split_backend);
}
ggml_backend_tensor_copy(input, input_cpy);
ggml_backend_tensor_copy_async(input_backend, split_backend, input, input_cpy);
} else {
// wait for the split backend to finish using the input before overwriting it
if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
ggml_backend_event_wait(split_backend, sched->events[split_backend_id][sched->cur_copy]);
} else {
ggml_backend_synchronize(split_backend);
}
// when offloading MoE weights, we can reduce the amount of data copied by copying only the experts that are used
@@ -1578,6 +1578,10 @@ static enum ggml_status ggml_backend_sched_compute_splits(ggml_backend_sched_t s
}
}
if (sched->events[split_backend_id][sched->cur_copy] == NULL) {
ggml_backend_synchronize(split_backend);
}
if (!sched->callback_eval) {
enum ggml_status ec = ggml_backend_graph_compute_async(split_backend, &split->graph);
if (ec != GGML_STATUS_SUCCESS) {