Skip to content

vulkan: use memory budget extension to read memory usage #15545

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 6 commits into from
Sep 1, 2025
Merged

vulkan: use memory budget extension to read memory usage #15545

Changes from 4 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
3a0dcc3
vulkan: use memory budget extension to read memory usage
giladgd Aug 24, 2025
4ccfa38
fix: formatting and names
giladgd Aug 24, 2025
ea5d796
formatting
giladgd Aug 24, 2025
6784772
fix: detect and cache memory budget extension availability on init
giladgd Aug 30, 2025
0934c3b
fix: read `budgetprops.heapBudget` instead of `heap.size` when memory…
giladgd Aug 31, 2025
a39edf4
style: lints
giladgd Aug 31, 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
43 changes: 36 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 @@ -1355,6 +1355,7 @@ struct vk_instance_t {
PFN_vkCmdInsertDebugUtilsLabelEXT pfn_vkCmdInsertDebugUtilsLabelEXT = {};

std::vector<size_t> device_indices;
std::vector<bool> device_supports_membudget;
vk_device devices[GGML_VK_MAX_DEVICES];
};

Expand Down Expand Up @@ -4202,15 +4203,16 @@ static void ggml_vk_instance_init() {
vk_instance.pfn_vkCmdBeginDebugUtilsLabelEXT = (PFN_vkCmdBeginDebugUtilsLabelEXT) vkGetInstanceProcAddr(vk_instance.instance, "vkCmdBeginDebugUtilsLabelEXT");
vk_instance.pfn_vkCmdEndDebugUtilsLabelEXT = (PFN_vkCmdEndDebugUtilsLabelEXT) vkGetInstanceProcAddr(vk_instance.instance, "vkCmdEndDebugUtilsLabelEXT");
vk_instance.pfn_vkCmdInsertDebugUtilsLabelEXT = (PFN_vkCmdInsertDebugUtilsLabelEXT) vkGetInstanceProcAddr(vk_instance.instance, "vkCmdInsertDebugUtilsLabelEXT");

}

vk_perf_logger_enabled = getenv("GGML_VK_PERF_LOGGER") != nullptr;

std::vector<vk::PhysicalDevice> devices = vk_instance.instance.enumeratePhysicalDevices();

// Emulate behavior of CUDA_VISIBLE_DEVICES for Vulkan
char * devices_env = getenv("GGML_VK_VISIBLE_DEVICES");
if (devices_env != nullptr) {
size_t num_available_devices = vk_instance.instance.enumeratePhysicalDevices().size();
size_t num_available_devices = devices.size();

std::string devices(devices_env);
std::replace(devices.begin(), devices.end(), ',', ' ');
Expand All @@ -4225,8 +4227,6 @@ static void ggml_vk_instance_init() {
vk_instance.device_indices.push_back(tmp);
}
} else {
std::vector<vk::PhysicalDevice> devices = vk_instance.instance.enumeratePhysicalDevices();

// If no vulkan devices are found, return early
if (devices.empty()) {
GGML_LOG_INFO("ggml_vulkan: No devices found.\n");
Expand Down Expand Up @@ -4331,6 +4331,19 @@ static void ggml_vk_instance_init() {
GGML_LOG_DEBUG("ggml_vulkan: Found %zu Vulkan devices:\n", vk_instance.device_indices.size());

for (size_t i = 0; i < vk_instance.device_indices.size(); i++) {
vk::PhysicalDevice vkdev = devices[vk_instance.device_indices[i]];
std::vector<vk::ExtensionProperties> extensionprops = vkdev.enumerateDeviceExtensionProperties();

bool membudget_supported = false;
for (const auto & ext : extensionprops) {
if (std::string(ext.extensionName.data()) == VK_EXT_MEMORY_BUDGET_EXTENSION_NAME) {
Copy link
Collaborator

@jeffbolznv jeffbolznv Aug 31, 2025

Choose a reason for hiding this comment

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

I'd prefer strcmp, but it's not a huge deal.

Copy link
Contributor Author

@giladgd giladgd Aug 31, 2025

Choose a reason for hiding this comment

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

Done

membudget_supported = true;
break;
}
}

vk_instance.device_supports_membudget.push_back(membudget_supported);

ggml_vk_print_gpu_info(i);
}
}
Expand Down Expand Up @@ -11441,15 +11454,31 @@ void ggml_backend_vk_get_device_description(int device, char * description, size

void ggml_backend_vk_get_device_memory(int device, size_t * free, size_t * total) {
GGML_ASSERT(device < (int) vk_instance.device_indices.size());
GGML_ASSERT(device < (int) vk_instance.device_supports_membudget.size());

vk::PhysicalDevice vkdev = vk_instance.instance.enumeratePhysicalDevices()[vk_instance.device_indices[device]];

vk::PhysicalDeviceMemoryProperties memprops = vkdev.getMemoryProperties();
bool membudget_supported = vk_instance.device_supports_membudget[device];

vk::PhysicalDeviceMemoryBudgetPropertiesEXT budgetprops;
vk::PhysicalDeviceMemoryProperties2 memprops2 = {};

if (membudget_supported) {
memprops2.pNext = &budgetprops;
vkdev.getMemoryProperties2(&memprops2);
}

for (uint32_t i = 0; i < memprops.memoryHeapCount; ++i) {
const vk::MemoryHeap & heap = memprops.memoryHeaps[i];

for (const vk::MemoryHeap& heap : memprops.memoryHeaps) {
if (heap.flags & vk::MemoryHeapFlagBits::eDeviceLocal) {
*total = heap.size;
*free = heap.size;

if (membudget_supported && i < budgetprops.heapUsage.size()) {
*free = *total - budgetprops.heapUsage[i];
} else {
*free = heap.size;
}
break;
}
}
Expand Down
Loading