ollama/llama/patches/0027-vulkan-get-GPU-ID-olla...

95 lines
4.1 KiB
Diff

From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Xiaodong Ye <xiaodong.ye@mthreads.com>
Date: Mon, 18 Aug 2025 12:48:07 +0800
Subject: [PATCH] vulkan: get GPU ID (ollama v0.11.5)
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
---
ggml/src/ggml-vulkan/ggml-vulkan.cpp | 37 ++++++++++++++++++++++++++++
1 file changed, 37 insertions(+)
diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
index 061cd078..adea7783 100644
--- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -11588,6 +11588,29 @@ static void ggml_vk_get_device_description(int device, char * description, size_
snprintf(description, description_size, "%s", props.deviceName.data());
}
+static std::string ggml_vk_get_device_id(int device) {
+ ggml_vk_instance_init();
+
+ std::vector<vk::PhysicalDevice> devices = vk_instance.instance.enumeratePhysicalDevices();
+
+ vk::PhysicalDeviceProperties2 props;
+ vk::PhysicalDeviceIDProperties deviceIDProps;
+ props.pNext = &deviceIDProps;
+ devices[device].getProperties2(&props);
+
+ const auto& uuid = deviceIDProps.deviceUUID;
+ char id[64];
+ snprintf(id, sizeof(id),
+ "GPU-%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x",
+ uuid[0], uuid[1], uuid[2], uuid[3],
+ uuid[4], uuid[5],
+ uuid[6], uuid[7],
+ uuid[8], uuid[9],
+ uuid[10], uuid[11], uuid[12], uuid[13], uuid[14], uuid[15]
+ );
+ return std::string(id);
+}
+
// backend interface
#define UNUSED GGML_UNUSED
@@ -12394,6 +12417,12 @@ void ggml_backend_vk_get_device_description(int device, char * description, size
ggml_vk_get_device_description(dev_idx, description, description_size);
}
+std::string ggml_backend_vk_get_device_id(int device) {
+ GGML_ASSERT(device < (int) vk_instance.device_indices.size());
+ int dev_idx = vk_instance.device_indices[device];
+ return ggml_vk_get_device_id(dev_idx);
+}
+
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());
@@ -12481,6 +12510,7 @@ struct ggml_backend_vk_device_context {
std::string description;
bool is_integrated_gpu;
std::string pci_bus_id;
+ std::string id;
};
static const char * ggml_backend_vk_device_get_name(ggml_backend_dev_t dev) {
@@ -12493,6 +12523,11 @@ static const char * ggml_backend_vk_device_get_description(ggml_backend_dev_t de
return ctx->description.c_str();
}
+static const char * ggml_backend_vk_device_get_id(ggml_backend_dev_t dev) {
+ ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
+ return ctx->id.c_str();
+}
+
static void ggml_backend_vk_device_get_memory(ggml_backend_dev_t device, size_t * free, size_t * total) {
ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)device->context;
ggml_backend_vk_get_device_memory(ctx->device, free, total);
@@ -12519,6 +12554,7 @@ static void ggml_backend_vk_device_get_props(ggml_backend_dev_t dev, struct ggml
props->name = ggml_backend_vk_device_get_name(dev);
props->description = ggml_backend_vk_device_get_description(dev);
+ props->id = ggml_backend_vk_device_get_id(dev);
props->type = ggml_backend_vk_device_get_type(dev);
props->device_id = ctx->pci_bus_id.empty() ? nullptr : ctx->pci_bus_id.c_str();
ggml_backend_vk_device_get_memory(dev, &props->memory_free, &props->memory_total);
@@ -12965,6 +13001,7 @@ static ggml_backend_dev_t ggml_backend_vk_reg_get_device(ggml_backend_reg_t reg,
ctx->description = desc;
ctx->is_integrated_gpu = ggml_backend_vk_get_device_type(i) == vk::PhysicalDeviceType::eIntegratedGpu;
ctx->pci_bus_id = ggml_backend_vk_get_device_pci_id(i);
+ ctx->id = ggml_backend_vk_get_device_id(i);
devices.push_back(new ggml_backend_device {
/* .iface = */ ggml_backend_vk_device_i,
/* .reg = */ reg,
--
2.51.0