Skip to content

Vulkan: Add device-specific blacklist for coopmat for the AMD proprietary driver #11074

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 2 commits into from
Jan 4, 2025
Merged

Vulkan: Add device-specific blacklist for coopmat for the AMD proprietary driver #11074

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
54b11a3
Vulkan: Add device-specific blacklist for coopmat for the AMD proprie…
0cc4m Dec 29, 2024
40e3363
Add (TM) to AMD name check
0cc4m Jan 4, 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
30 changes: 23 additions & 7 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 @@ -2040,6 +2040,8 @@ static void ggml_vk_load_shaders(vk_device& device) {
std::cerr << "Done!" << std::endl;
}

static bool ggml_vk_khr_cooperative_matrix_support(const vk::PhysicalDeviceProperties& props, const vk::PhysicalDeviceDriverProperties& driver_props);

static vk_device ggml_vk_get_device(size_t idx) {
VK_LOG_DEBUG("ggml_vk_get_device(" << idx << ")");

Expand Down Expand Up @@ -2175,9 +2177,7 @@ static vk_device ggml_vk_get_device(size_t idx) {

device->fp16 = !force_disable_f16 && fp16_storage && fp16_compute;

if (device->vendor_id == VK_VENDOR_ID_INTEL || (device->vendor_id == VK_VENDOR_ID_AMD && (driver_props.driverID == vk::DriverId::eAmdProprietary || driver_props.driverID == vk::DriverId::eAmdOpenSource))) {
// Intel drivers don't support coopmat properly yet
// Only RADV supports coopmat properly on AMD
if (!ggml_vk_khr_cooperative_matrix_support(device->properties, driver_props)) {
device->coopmat_support = false;
}

Expand Down Expand Up @@ -2515,7 +2515,6 @@ static vk_device ggml_vk_get_device(size_t idx) {
return vk_instance.devices[idx];
}


static void ggml_vk_print_gpu_info(size_t idx) {
GGML_ASSERT(idx < vk_instance.device_indices.size());
size_t dev_num = vk_instance.device_indices[idx];
Expand Down Expand Up @@ -2565,9 +2564,7 @@ static void ggml_vk_print_gpu_info(size_t idx) {
}
}

if (props2.properties.vendorID == VK_VENDOR_ID_INTEL || (props2.properties.vendorID == VK_VENDOR_ID_AMD && (driver_props.driverID == vk::DriverId::eAmdProprietary || driver_props.driverID == vk::DriverId::eAmdOpenSource))) {
// Intel drivers don't support coopmat properly yet
// Only RADV supports coopmat properly on AMD
if (!ggml_vk_khr_cooperative_matrix_support(props2.properties, driver_props)) {
coopmat_support = false;
}

Expand Down Expand Up @@ -8088,6 +8085,25 @@ static bool ggml_vk_instance_portability_enumeration_ext_available(const std::ve
UNUSED(instance_extensions);
}

static bool ggml_vk_khr_cooperative_matrix_support(const vk::PhysicalDeviceProperties& props, const vk::PhysicalDeviceDriverProperties& driver_props) {
switch (props.vendorID) {
case VK_VENDOR_ID_INTEL:
// Intel drivers don't support coopmat properly yet
return false;
case VK_VENDOR_ID_AMD:
if (driver_props.driverID == vk::DriverId::eAmdProprietary || driver_props.driverID == vk::DriverId::eAmdOpenSource) {
// Workaround for AMD proprietary driver reporting support on all GPUs
const std::string name = props.deviceName;
return name.rfind("AMD Radeon RX 7", 0) == 0 || // RDNA 3 consumer GPUs
Copy link
Collaborator

@jeffbolznv jeffbolznv Jan 4, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a way to write this so newer GPUs will be enabled by default?

Copy link
Collaborator Author

@0cc4m 0cc4m Jan 4, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That would be good, but I couldn't think of an easy way. I'm hoping that AMD resolves this in their drivers, then we can just remove the check.

I'll try to figure out the proper way to report this to their Windows and amdvlk driver team.

name.rfind("AMD Radeon PRO W7", 0) == 0 || // RDNA 3 workstation GPUs
name.rfind("AMD Radeon 7", 0) == 0; // RDNA 3 APUs
}
return true;
default:
return true;
}
}

// checks

#ifdef GGML_VULKAN_CHECK_RESULTS
Expand Down
Loading