Skip to content

vulkan: matmul gcn tuning #13016

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 4 commits into from
Apr 24, 2025
Merged

vulkan: matmul gcn tuning #13016

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
940817b
tune matmul for gcn
netrunnereve Apr 18, 2025
54b0053
this one is more power efficient
netrunnereve Apr 19, 2025
cc091f8
Update ggml/src/ggml-vulkan/ggml-vulkan.cpp
netrunnereve Apr 20, 2025
1a12157
disable this tune for the proprietary driver
netrunnereve Apr 24, 2025
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
7 changes: 7 additions & 0 deletions ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -246,6 +246,7 @@ struct vk_device_struct {
bool pipeline_robustness;
vk::Device device;
uint32_t vendor_id;
vk::DriverId driver_id;
vk_device_architecture architecture;
vk_queue compute_queue;
vk_queue transfer_queue;
Expand Down Expand Up @@ -1740,6 +1741,11 @@ static void ggml_vk_load_shaders(vk_device& device) {
m_warptile_mmq_int = { 128, 64, 64, 32, subgroup_size_8, 32, 2, 2, 2, 1, subgroup_size_8 };
s_warptile_mmq_int = { subgroup_size_32, 32, 32, 32, 32, 32, 2, 2, 1, 1, subgroup_size_8 };

// chip specific tuning
if ((device->architecture == AMD_GCN) && (device->driver_id != vk::DriverId::eAmdProprietary)) {
m_warptile_mmq = m_warptile_mmq_int = { 256, 64, 64, 32, 16, 16, 2, 2, 2, 1, 16 };
}

l_mmq_wg_denoms = l_wg_denoms = {128, 128, 1 };
m_mmq_wg_denoms = m_wg_denoms = { 64, 64, 1 };
s_mmq_wg_denoms = s_wg_denoms = { 32, 32, 1 };
Expand Down Expand Up @@ -2658,6 +2664,7 @@ static vk_device ggml_vk_get_device(size_t idx) {
device->physical_device.getProperties2(&props2);
device->properties = props2.properties;
device->vendor_id = device->properties.vendorID;
device->driver_id = driver_props.driverID;

const char* GGML_VK_FORCE_MAX_ALLOCATION_SIZE = getenv("GGML_VK_FORCE_MAX_ALLOCATION_SIZE");

Expand Down
Loading