Skip to content

Change USE_CUDA_GRAPH #3594

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Jamaika1 wants to merge 4 commits into
base: master
Choose a base branch
from
Open

Change USE_CUDA_GRAPH #3594

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
334e68b
Change USE_CUDA_GRAPH
Jamaika1 Jan 6, 2026
5675842
Change USE_CUDA_GRAPH
Jamaika1 Jan 6, 2026
7132fdb
Changing brackets
Jamaika1 Jan 9, 2026
19c108c
Changing brackets
Jamaika1 Jan 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 21 additions & 13 deletions ggml/src/ggml-cuda/mean.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,22 +28,29 @@ void ggml_cuda_op_mean(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
#ifdef USE_CUDA_GRAPH
cudaStreamCaptureStatus iscapturing;
CUDA_CHECK(cudaStreamIsCapturing(stream, &iscapturing));
if (((ctx.cuda_graph->instance == nullptr) && (iscapturing == cudaStreamCaptureStatusNone) ||
(ctx.cuda_graph->disable_due_to_gpu_arch || ctx.cuda_graph->disable_due_to_too_many_updates ||
ctx.cuda_graph->disable_due_to_failed_graph_capture))) {
#endif // USE_CUDA_GRAPH
if ((nrows == 1) &&
if ((nrows == 1) && (ncols > 65536)) {
// CUDA_GRAPHS_DISABLED
// Single row - use device-wide reduction
size_t tmp_size = 0;
ggml_cuda_pool & pool = ctx.pool();

DeviceReduce::Sum(nullptr, tmp_size, src0_d, dst_d, ncols, stream);

ggml_cuda_pool_alloc<uint8_t> tmp_alloc(pool, tmp_size);
DeviceReduce::Sum(tmp_alloc.ptr, tmp_size, src0_d, dst_d, ncols, stream);

// Divide by ncols
divide_by_count<float>/*<<<1, 1, 0, stream>>>*/(dst_d, ncols);
return;
}
#ifdef USE_CUDA_GRAPH
// CUDA_GRAPHS_DISABLED
((ncols > 65536) &&
((ctx.cuda_graph->instance == nullptr) && (iscapturing == cudaStreamCaptureStatusNone) ||
ctx.cuda_graph->disable_due_to_gpu_arch || ctx.cuda_graph->disable_due_to_too_many_updates ||
ctx.cuda_graph->disable_due_to_failed_graph_capture)) ||
}
else if ((nrows == 1) && (ncols > 32768)) {
// CUDA_GRAPHS ENABLED
((ncols > 32768) &&
!((ctx.cuda_graph->instance == nullptr) && (iscapturing == cudaStreamCaptureStatusNone) ||
ctx.cuda_graph->disable_due_to_gpu_arch || ctx.cuda_graph->disable_due_to_too_many_updates ||
ctx.cuda_graph->disable_due_to_failed_graph_capture))) {
#else
(ncols > 65536)) {
#endif // USE_CUDA_GRAPH
// Single row - use device-wide reduction
size_t tmp_size = 0;
ggml_cuda_pool & pool = ctx.pool();
Expand All @@ -57,6 +64,7 @@ void ggml_cuda_op_mean(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
divide_by_count<float><<<1, 1, 0, stream>>>(dst_d, ncols);
return;
}
#endif // USE_CUDA_GRAPH
#endif // GGML_CUDA_USE_CUB

const dim3 block_nums(nrows, 1, 1);
Expand Down