pali112
/
ollama
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vulkan: get GPU ID (ollama v0.11.5) 

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
This commit is contained in:
Xiaodong Ye 2025-08-18 17:49:32 +08:00 committed by Inforithmics
parent 1fc3239582
commit 603d3ab0ca
3 changed files with 160 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
discover/gpu_info_vulkan.c
View File

@ -203,6 +203,16 @@ void vk_check_vram(vk_handle_t rh, int i, mem_info_t *resp) {
snprintf(&resp->gpu_id[0], GPU_ID_LEN, "%d", i);
resp->gpu_name[GPU_NAME_LEN - 1] = '\0';
strncpy(&resp->gpu_name[0], properties.deviceName, GPU_NAME_LEN - 1);
resp->gpu_name[GPU_NAME_LEN - 1] = '\0';
const uint8_t *uuid = properties.pipelineCacheUUID;
snprintf(&resp->gpu_id[0], GPU_ID_LEN,
"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]
);
resp->total = (uint64_t) device_memory_total_size;
resp->free = (uint64_t) device_memory_heap_budget;
resp->major = VK_API_VERSION_MAJOR(properties.apiVersion);

115
llama/patches/0023-vulkan-get-GPU-ID-ollama-v0.11.5.patch Normal file
View File

@ -0,0 +1,115 @@
From e0ba120c913a2931010a31e0fdf160697a15b9f1 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>
---
discover/gpu_info_vulkan.c | 9 +++++
.../ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp | 35 +++++++++++++++++++
2 files changed, 44 insertions(+)
diff --git a/discover/gpu_info_vulkan.c b/discover/gpu_info_vulkan.c
index 6d67353d..afac97dd 100644
--- a/discover/gpu_info_vulkan.c
+++ b/discover/gpu_info_vulkan.c
@@ -171,6 +171,15 @@ void vk_check_vram(vk_handle_t rh, int i, mem_info_t *resp) {
snprintf(&resp->gpu_id[0], GPU_ID_LEN, "%d", i);
strncpy(&resp->gpu_name[0], properties.deviceName, GPU_NAME_LEN - 1);
resp->gpu_name[GPU_NAME_LEN - 1] = '\0';
+ const uint8_t *uuid = properties.pipelineCacheUUID;
+ snprintf(&resp->gpu_id[0], GPU_ID_LEN,
+ "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]
+ );
resp->total = (uint64_t) device_memory_total_size;
resp->free = (uint64_t) device_memory_heap_budget;
resp->major = VK_API_VERSION_MAJOR(properties.apiVersion);
diff --git a/ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp
index 4070e248..1c8c15d5 100644
--- a/ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -10194,6 +10194,27 @@ 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::PhysicalDeviceProperties props;
+ devices[device].getProperties(&props);
+
+ const auto& uuid = props.pipelineCacheUUID;
+ 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
@@ -10790,6 +10811,12 @@ 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());
@@ -10812,6 +10839,7 @@ struct ggml_backend_vk_device_context {
size_t device;
std::string name;
std::string description;
+ std::string id;
};
static const char * ggml_backend_vk_device_get_name(ggml_backend_dev_t dev) {
@@ -10824,6 +10852,11 @@ 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);
@@ -10847,6 +10880,7 @@ static enum ggml_backend_dev_type ggml_backend_vk_device_get_type(ggml_backend_d
static void ggml_backend_vk_device_get_props(ggml_backend_dev_t dev, struct ggml_backend_dev_props * props) {
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);
ggml_backend_vk_device_get_memory(dev, &props->memory_free, &props->memory_total);
props->caps = {
@@ -11265,6 +11299,7 @@ static ggml_backend_dev_t ggml_backend_vk_reg_get_device(ggml_backend_reg_t reg,
ctx->device = i;
ctx->name = GGML_VK_NAME + std::to_string(i);
ctx->description = desc;
+ ctx->id = ggml_backend_vk_get_device_id(i);
devices.push_back(new ggml_backend_device {
/* .iface = */ ggml_backend_vk_device_i,
/* .reg = */ reg,
--
2.25.1

35
ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp vendored
View File

@ -10194,6 +10194,27 @@ 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::PhysicalDeviceProperties props;
devices[device].getProperties(&props);
const auto& uuid = props.pipelineCacheUUID;
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
@ -10790,6 +10811,12 @@ 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());
@ -10812,6 +10839,7 @@ struct ggml_backend_vk_device_context {
size_t device;
std::string name;
std::string description;
std::string id;
};
static const char * ggml_backend_vk_device_get_name(ggml_backend_dev_t dev) {
@ -10824,6 +10852,11 @@ 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);
@ -10847,6 +10880,7 @@ static enum ggml_backend_dev_type ggml_backend_vk_device_get_type(ggml_backend_d
static void ggml_backend_vk_device_get_props(ggml_backend_dev_t dev, struct ggml_backend_dev_props * props) {
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);
ggml_backend_vk_device_get_memory(dev, &props->memory_free, &props->memory_total);
props->caps = {
@ -11265,6 +11299,7 @@ static ggml_backend_dev_t ggml_backend_vk_reg_get_device(ggml_backend_reg_t reg,
ctx->device = i;
ctx->name = GGML_VK_NAME + std::to_string(i);
ctx->description = desc;
ctx->id = ggml_backend_vk_get_device_id(i);
devices.push_back(new ggml_backend_device {
/* .iface = */ ggml_backend_vk_device_i,
/* .reg = */ reg,