Skip to content

opencl: use max_alloc_size in backend ctx instead of querying again #12705

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

Merged
merged 1 commit into from
Apr 3, 2025
Merged

opencl: use max_alloc_size in backend ctx instead of querying again #12705

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
15 changes: 6 additions & 9 deletions ggml/src/ggml-opencl/ggml-opencl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -924,27 +924,24 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
// TODO: fixme: these sizes are hardcoded for now.
// they should be allocated based on the model's size
// and the device's max alloc size
size_t max_alloc_size;
CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(size_t), &max_alloc_size, NULL));

// Allocate intermediate buffers and images
size_t required_A_q_d_bytes = 311164928;
size_t required_A_s_d_bytes = 38895616;
size_t required_B_d_bytes = 45088768;

// Ensure buffer sizes do not exceed the maximum allocation size
size_t max_A_q_d_bytes = MIN(required_A_q_d_bytes, max_alloc_size);
size_t max_A_s_d_bytes = MIN(required_A_s_d_bytes, max_alloc_size);
size_t max_B_d_bytes = MIN(required_B_d_bytes, max_alloc_size);
if (required_A_q_d_bytes > max_alloc_size) {
size_t max_A_q_d_bytes = MIN(required_A_q_d_bytes, backend_ctx->max_alloc_size);
size_t max_A_s_d_bytes = MIN(required_A_s_d_bytes, backend_ctx->max_alloc_size);
size_t max_B_d_bytes = MIN(required_B_d_bytes, backend_ctx->max_alloc_size);
if (required_A_q_d_bytes > backend_ctx->max_alloc_size) {
GGML_LOG_WARN("ggml_opencl: A_q_d buffer size reduced from %zu to %zu due to device limitations.\n",
required_A_q_d_bytes, max_A_q_d_bytes);
}
if (required_A_s_d_bytes > max_alloc_size) {
if (required_A_s_d_bytes > backend_ctx->max_alloc_size) {
GGML_LOG_WARN("ggml_opencl: A_s_d buffer size reduced from %zu to %zu due to device limitations.\n",
required_A_s_d_bytes, max_A_s_d_bytes);
}
if (required_B_d_bytes > max_alloc_size) {
if (required_B_d_bytes > backend_ctx->max_alloc_size) {
GGML_LOG_WARN("ggml_opencl: B_d buffer size reduced from %zu to %zu due to device limitations.\n",
required_B_d_bytes, max_B_d_bytes);
}
Expand Down
Loading