ggml-backend : add GGML_BACKEND_DEVICE_TYPE_IGPU device type (llama/15797)

* ggml-backend : add GGML_BACKEND_DEVICE_TYPE_IGPU device type

ggml-backend : add device id to device props

llama : only use iGPU devices if there are no GPU devices

llama : do not use multiple devices from different backends with the same device id

diff --git a/include/ggml-backend.h b/include/ggml-backend.h
index 4f246f6ccd62922283d0545a851d648c058dce47..ab297e0c6f69fd4fb84baa1aee421067e939c511 100644 (file)
--- a/include/ggml-backend.h
+++ b/include/ggml-backend.h
@@ -132,6 +132,8 @@ extern "C" {
         GGML_BACKEND_DEVICE_TYPE_CPU,
         // GPU device using dedicated memory
         GGML_BACKEND_DEVICE_TYPE_GPU,
+        // integrated GPU device using host memory
+        GGML_BACKEND_DEVICE_TYPE_IGPU,
         // accelerator devices intended to be used together with the CPU backend (e.g. BLAS or AMX)
         GGML_BACKEND_DEVICE_TYPE_ACCEL
     };
@@ -150,11 +152,21 @@ extern "C" {
 
     // all the device properties
     struct ggml_backend_dev_props {
+        // device name
         const char * name;
+        // device description
         const char * description;
+        // device free memory in bytes
         size_t memory_free;
+        // device total memory in bytes
         size_t memory_total;
+        // device type
         enum ggml_backend_dev_type type;
+        // device id
+        //   for PCI devices, this should be the PCI bus id formatted as "domain:bus:device.function" (e.g. "0000:01:00.0")
+        //   if the id is unknown, this should be NULL
+        const char * device_id;
+        // device capabilities
         struct ggml_backend_dev_caps caps;
     };
 

diff --git a/src/ggml-backend-impl.h b/src/ggml-backend-impl.h
index 2db5c4e0f85e5934ade50fe891a5b450084d5774..89d80db6e603d13e9c30c2f5d0554966ee325467 100644 (file)
--- a/src/ggml-backend-impl.h
+++ b/src/ggml-backend-impl.h
@@ -8,7 +8,7 @@
 extern "C" {
 #endif
 
-    #define GGML_BACKEND_API_VERSION 1
+    #define GGML_BACKEND_API_VERSION 2
 
     //
     // Backend buffer type

diff --git a/src/ggml-backend-reg.cpp b/src/ggml-backend-reg.cpp
index 5f02a710a14e003554201c8bb3e92fce0185479d..7002cb07e001526d76e9e9f1b2a191cfffd0b87b 100644 (file)
--- a/src/ggml-backend-reg.cpp
+++ b/src/ggml-backend-reg.cpp
@@ -400,9 +400,8 @@ ggml_backend_t ggml_backend_init_by_type(enum ggml_backend_dev_type type, const
 
 ggml_backend_t ggml_backend_init_best(void) {
     ggml_backend_dev_t dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_GPU);
-    if (!dev) {
-        dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
-    }
+    dev = dev ? dev : ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_IGPU);
+    dev = dev ? dev : ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
     if (!dev) {
         return nullptr;
     }

diff --git a/src/ggml-cuda/ggml-cuda.cu b/src/ggml-cuda/ggml-cuda.cu
index 0f68d685348d27b52ebddaae51d58512d5c43424..9ea8f4589d71de8c764d2aee2cec619abd2afe33 100644 (file)
--- a/src/ggml-cuda/ggml-cuda.cu
+++ b/src/ggml-cuda/ggml-cuda.cu
@@ -3210,6 +3210,7 @@ struct ggml_backend_cuda_device_context {
     int device;
     std::string name;
     std::string description;
+    std::string pci_bus_id;
 };
 
 static const char * ggml_backend_cuda_device_get_name(ggml_backend_dev_t dev) {
@@ -3234,9 +3235,12 @@ static enum ggml_backend_dev_type ggml_backend_cuda_device_get_type(ggml_backend
 }
 
 static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_backend_dev_props * props) {
+    ggml_backend_cuda_device_context * ctx = (ggml_backend_cuda_device_context *)dev->context;
+
     props->name        = ggml_backend_cuda_device_get_name(dev);
     props->description = ggml_backend_cuda_device_get_description(dev);
     props->type        = ggml_backend_cuda_device_get_type(dev);
+    props->device_id   = ctx->pci_bus_id.empty() ? nullptr : ctx->pci_bus_id.c_str();
     ggml_backend_cuda_device_get_memory(dev, &props->memory_free, &props->memory_total);
 
     bool host_buffer = getenv("GGML_CUDA_NO_PINNED") == nullptr;
@@ -3804,6 +3808,10 @@ ggml_backend_reg_t ggml_backend_cuda_reg() {
                 CUDA_CHECK(cudaGetDeviceProperties(&prop, i));
                 dev_ctx->description = prop.name;
 
+                char pci_bus_id[16] = {};
+                snprintf(pci_bus_id, sizeof(pci_bus_id), "%04x:%02x:%02x.0", prop.pciDomainID, prop.pciBusID, prop.pciDeviceID);
+                dev_ctx->pci_bus_id = pci_bus_id;
+
                 ggml_backend_dev_t dev = new ggml_backend_device {
                     /* .iface   = */ ggml_backend_cuda_device_interface,
                     /* .reg     = */ &reg,

diff --git a/src/ggml-vulkan/ggml-vulkan.cpp b/src/ggml-vulkan/ggml-vulkan.cpp
index cb379fe94ec73ef841bcf1baa451a92f98ee6089..178d8eb3de5feb08356c0a7443694d89045a7fb4 100644 (file)
--- a/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/src/ggml-vulkan/ggml-vulkan.cpp
@@ -12113,6 +12113,7 @@ static ggml_backend_buffer_type_t ggml_backend_vk_device_get_host_buffer_type(gg
 
 static enum ggml_backend_dev_type ggml_backend_vk_device_get_type(ggml_backend_dev_t dev) {
     UNUSED(dev);
+    // TODO: return GGML_BACKEND_DEVICE_TYPE_IGPU for integrated GPUs
     return GGML_BACKEND_DEVICE_TYPE_GPU;
 }
 
@@ -12120,6 +12121,7 @@ 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->type        = ggml_backend_vk_device_get_type(dev);
+    // TODO: set props->device_id to PCI bus id
     ggml_backend_vk_device_get_memory(dev, &props->memory_free, &props->memory_total);
     props->caps = {
         /* .async                 = */ false,