update other backends

Author: slaren

.github/workflows/bench.yml.disabled

@@ -27,10 +27,10 @@ on:
   push:
     branches:
       - master
-    paths: ['llama.cpp', 'ggml.c', 'ggml-backend.c', 'ggml-quants.c', '**/*.cu', 'examples/server/*.h*', 'examples/server/*.cpp']
+    paths: ['llama.cpp', 'ggml.c', 'ggml-backend.cpp', 'ggml-quants.c', '**/*.cu', 'examples/server/*.h*', 'examples/server/*.cpp']
   pull_request_target:
     types: [opened, synchronize, reopened]
-    paths: ['llama.cpp', 'ggml.c', 'ggml-backend.c', 'ggml-quants.c', '**/*.cu', 'examples/server/*.h*', 'examples/server/*.cpp']
+    paths: ['llama.cpp', 'ggml.c', 'ggml-backend.cpp', 'ggml-quants.c', '**/*.cu', 'examples/server/*.h*', 'examples/server/*.cpp']
   schedule:
     -  cron: '04 2 * * *'
 

Package.swift

@@ -11,7 +11,7 @@ var sources = [
     "src/unicode-data.cpp",
     "ggml/src/ggml.c",
     "ggml/src/ggml-alloc.c",
-    "ggml/src/ggml-backend.c",
+    "ggml/src/ggml-backend.cpp",
     "ggml/src/ggml-quants.c",
     "ggml/src/ggml-aarch64.c",
 ]

ggml/src/ggml-blas.cpp

@@ -322,11 +322,8 @@ static struct ggml_backend_i blas_backend_i = {
     /* .supports_op             = */ ggml_backend_blas_supports_op,
     /* .supports_buft           = */ ggml_backend_blas_supports_buft,
     /* .offload_op              = */ NULL,
-    /* .event_new               = */ NULL,
-    /* .event_free              = */ NULL,
     /* .event_record            = */ NULL,
     /* .event_wait              = */ NULL,
-    /* .event_synchronize       = */ NULL,
 };
 
 static ggml_guid_t ggml_backend_blas_guid(void) {
@@ -340,6 +337,7 @@ ggml_backend_t ggml_backend_blas_init(void) {
     ggml_backend_t backend = new ggml_backend {
         /* .guid      = */ ggml_backend_blas_guid(),
         /* .interface = */ blas_backend_i,
+        /* .device    = */ nullptr,
         /* .context   = */ ctx,
     };
 

ggml/src/ggml-cann.cpp

@@ -1326,6 +1326,7 @@ ggml_backend_buffer_type_t ggml_backend_cann_host_buffer_type() {
             /* .get_alloc_size   = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
             /* .is_host          = */ ggml_backend_cpu_buffer_type()->iface.is_host,
         },
+        /* .device   = */ nullptr,
         /* .context  = */ nullptr,
     };
 
@@ -2021,11 +2022,8 @@ static ggml_backend_i ggml_backend_cann_interface = {
     /* .supports_op             = */ ggml_backend_cann_supports_op,
     /* .supports_buft           = */ ggml_backend_cann_supports_buft,
     /* .offload_op              = */ ggml_backend_cann_offload_op,
-    /* .event_new               = */ ggml_backend_cann_event_new,
-    /* .event_free              = */ ggml_backend_cann_event_free,
     /* .event_record            = */ ggml_backend_cann_event_record,
     /* .event_wait              = */ ggml_backend_cann_event_wait,
-    /* .event_synchronize       = */ ggml_backend_cann_event_synchronize,
 };
 
 /**
@@ -2058,6 +2056,7 @@ ggml_backend_t ggml_backend_cann_init(int32_t device) {
     ggml_backend_t cann_backend =
         new ggml_backend{/* .guid      = */ ggml_backend_cann_guid(),
                          /* .interface = */ ggml_backend_cann_interface,
+                         /* .device    = */ nullptr,
                          /* .context   = */ ctx};
 
     return cann_backend;
@@ -2105,28 +2104,3 @@ static ggml_backend_t ggml_backend_reg_cann_init(const char* params,
 
     GGML_UNUSED(params);
 }
-
-extern "C" int ggml_backend_cann_reg_devices();
-
-/**
- * @brief Registers CANN (Ascend) devices as backend options.
- *
- * This function initializes ACL, retrieves the number of available CANN
- * devices, and registers each device as a backend option using
- * `ggml_backend_register`. Each device is given a unique name based on
- * `GGML_CANN_NAME` followed by its index.
- *
- * @return int The number of CANN devices registered.
- */
-int ggml_backend_cann_reg_devices() {
-    uint32_t device_count = ggml_backend_cann_get_device_count();
-    // initialization
-    for (uint32_t i = 0; i < device_count; i++) {
-        char name[128];
-        snprintf(name, sizeof(name), "CANN%d", i);
-        ggml_backend_register(name, ggml_backend_reg_cann_init,
-                              ggml_backend_cann_buffer_type(i),
-                              (void*)(intptr_t)i);
-    }
-    return device_count;
-}

ggml/src/ggml-kompute.cpp

@@ -1921,6 +1921,7 @@ ggml_backend_buffer_type_t ggml_backend_kompute_buffer_type(int device) {
         for (const auto & dev : devices) {
             vec.push_back({
                 /* .iface   = */ ggml_backend_kompute_buffer_type_interface,
+                /* .device  = */ nullptr,
                 /* .context = */ new ggml_backend_kompute_buffer_type_context(dev.index, dev.bufferAlignment, dev.maxAlloc)
             });
         }
@@ -1989,11 +1990,8 @@ static struct ggml_backend_i kompute_backend_i = {
     /* .supports_op             = */ ggml_backend_kompute_supports_op,
     /* .supports_buft           = */ ggml_backend_kompute_supports_buft,
     /* .offload_op              = */ NULL,
-    /* .event_new               = */ NULL,
-    /* .event_free              = */ NULL,
     /* .event_record            = */ NULL,
     /* .event_wait              = */ NULL,
-    /* .event_synchronize       = */ NULL,
 };
 
 static ggml_guid_t ggml_backend_kompute_guid() {
@@ -2008,6 +2006,7 @@ ggml_backend_t ggml_backend_kompute_init(int device) {
     ggml_backend_t kompute_backend = new ggml_backend {
         /* .guid      = */ ggml_backend_kompute_guid(),
         /* .interface = */ kompute_backend_i,
+        /* .device    = */ nullptr,
         /* .context   = */ s_kompute_context,
     };
 
@@ -2022,18 +2021,3 @@ static ggml_backend_t ggml_backend_reg_kompute_init(const char * params, void *
     GGML_UNUSED(params);
     return ggml_backend_kompute_init(intptr_t(user_data));
 }
-
-extern "C" int ggml_backend_kompute_reg_devices();
-
-int ggml_backend_kompute_reg_devices() {
-    auto devices = ggml_vk_available_devices_internal(0);
-    for (const auto & device : devices) {
-        ggml_backend_register(
-            ggml_kompute_format_name(device.index).c_str(),
-            ggml_backend_reg_kompute_init,
-            ggml_backend_kompute_buffer_type(device.index),
-            reinterpret_cast<void *>(intptr_t(device.index))
-        );
-    }
-    return devices.size();
-}

ggml/src/ggml-metal.m

@@ -3380,6 +3380,7 @@ ggml_backend_buffer_type_t ggml_backend_metal_buffer_type(void) {
             /* .get_alloc_size   = */ NULL, // defaults to ggml_nbytes
             /* .is_host          = */ ggml_backend_metal_buffer_type_is_host,
         },
+        /* .device  = */ NULL,
         /* .context = */ NULL,
     };
 
@@ -3539,11 +3540,8 @@ static void ggml_backend_metal_set_n_cb(ggml_backend_t backend, int n_cb) {
     /* .supports_op             = */ ggml_backend_metal_supports_op,
     /* .supports_buft           = */ ggml_backend_metal_supports_buft,
     /* .offload_op              = */ NULL,
-    /* .event_new               = */ NULL,
-    /* .event_free              = */ NULL,
     /* .event_record            = */ NULL,
     /* .event_wait              = */ NULL,
-    /* .event_synchronize       = */ NULL,
 };
 
 void ggml_backend_metal_log_set_callback(ggml_log_callback log_callback, void * user_data) {
@@ -3568,6 +3566,7 @@ ggml_backend_t ggml_backend_metal_init(void) {
     *backend = (struct ggml_backend) {
         /* .guid      = */ ggml_backend_metal_guid(),
         /* .interface = */ ggml_backend_metal_i,
+        /* .device    = */ NULL,
         /* .context   = */ ctx,
     };
 

ggml/src/ggml-rpc.cpp

@@ -662,11 +662,8 @@ static ggml_backend_i ggml_backend_rpc_interface = {
     /* .supports_op             = */ ggml_backend_rpc_supports_op,
     /* .supports_buft           = */ ggml_backend_rpc_supports_buft,
     /* .offload_op              = */ NULL,
-    /* .event_new               = */ NULL,
-    /* .event_free              = */ NULL,
     /* .event_record            = */ NULL,
     /* .event_wait              = */ NULL,
-    /* .event_synchronize       = */ NULL,
 };
 
 GGML_API ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * endpoint) {
@@ -694,6 +691,7 @@ GGML_API ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * en
 
     ggml_backend_buffer_type_t buft = new ggml_backend_buffer_type {
         /* .iface   = */ ggml_backend_rpc_buffer_type_interface,
+        /* .device  = */ nullptr,
         /* .context = */ buft_ctx
     };
     buft_map[endpoint] = buft;
@@ -709,6 +707,7 @@ ggml_backend_t ggml_backend_rpc_init(const char * endpoint) {
     ggml_backend_t backend = new ggml_backend {
         /* .guid      = */ ggml_backend_rpc_guid(),
         /* .interface = */ ggml_backend_rpc_interface,
+        /* .device    = */ nullptr,
         /* .context   = */ ctx
     };
     return backend;

ggml/src/ggml-sycl.cpp

@@ -4424,6 +4424,7 @@ ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int device) {
             queue_ptr stream = &(device_i.default_queue());
             ggml_backend_sycl_buffer_types[i] = {
                 /* .iface    = */ ggml_backend_sycl_buffer_type_interface,
+                /* .device   = */ nullptr,
                 /* .context  = */ new ggml_backend_sycl_buffer_type_context{i, GGML_SYCL_NAME + std::to_string(i), stream},
             };
         }
@@ -4449,6 +4450,7 @@ ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(ggml_backend_sycl_conte
         for (int i = 0; i < ggml_sycl_info().device_count; i++) {
             ggml_backend_sycl_buffer_types[i] = {
                 /* .iface    = */ ggml_backend_sycl_buffer_type_interface,
+                /* .device   = */ nullptr,
                 /* .context  = */ new ggml_backend_sycl_buffer_type_context{i, GGML_SYCL_NAME + std::to_string(i), ctx->stream(i, 0)},
             };
         }
@@ -4837,6 +4839,7 @@ ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * ten
 
     struct ggml_backend_buffer_type buft {
         /* .iface   = */ ggml_backend_sycl_split_buffer_type_interface,
+        /* .device  = */ nullptr,
         /* .context = */ new ggml_backend_sycl_split_buffer_type_context{tensor_split_arr},
     };
 
@@ -4890,6 +4893,7 @@ ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type() {
             /* .get_alloc_size   = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
             /* .is_host          = */ ggml_backend_cpu_buffer_type()->iface.is_host,
         },
+        /* .device   = */ nullptr,
         /* .context  = */ nullptr,
     };
 
@@ -5197,11 +5201,8 @@ static ggml_backend_i ggml_backend_sycl_interface = {
     /* .supports_op             = */ ggml_backend_sycl_supports_op,
     /* .supports_buft           = */ ggml_backend_sycl_supports_buft,
     /* .offload_op              = */ ggml_backend_sycl_offload_op,
-    /* .event_new               = */ NULL,
-    /* .event_free              = */ NULL,
     /* .event_record            = */ NULL,
     /* .event_wait              = */ NULL,
-    /* .event_synchronize       = */ NULL,
 };
 
 static ggml_guid_t ggml_backend_sycl_guid() {
@@ -5224,6 +5225,7 @@ ggml_backend_t ggml_backend_sycl_init(int device) {
     ggml_backend_t sycl_backend = new ggml_backend {
         /* .guid      = */ ggml_backend_sycl_guid(),
         /* .interface = */ ggml_backend_sycl_interface,
+        /* .device    = */ nullptr,
         /* .context   = */ ctx
     };
 
@@ -5245,15 +5247,3 @@ static ggml_backend_t ggml_backend_reg_sycl_init(const char * params, void * use
 
     UNUSED(params);
 }
-
-extern "C" int ggml_backend_sycl_reg_devices();
-
-int ggml_backend_sycl_reg_devices() {
-    assert(ggml_sycl_info().device_count>0);
-    for (int i = 0; i < ggml_sycl_info().device_count; i++) {
-        char name[128];
-        snprintf(name, sizeof(name), "%s%d", GGML_SYCL_NAME, i);
-        ggml_backend_register(name, ggml_backend_reg_sycl_init, ggml_backend_sycl_buffer_type(i), (void *) (intptr_t) i);
-    }
-    return ggml_sycl_info().device_count;
-}

ggml/src/ggml-vulkan.cpp

@@ -1938,6 +1938,7 @@ static vk_device ggml_vk_get_device(size_t idx) {
 
         device->buffer_type = {
             /* .iface    = */ ggml_backend_vk_buffer_type_interface,
+            /* .device   = */ nullptr,
             /* .context  = */ new ggml_backend_vk_buffer_type_context{ device->name, device },
         };
 
@@ -6409,6 +6410,7 @@ ggml_backend_buffer_type_t ggml_backend_vk_host_buffer_type() {
             /* .get_alloc_size   = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
             /* .is_host          = */ ggml_backend_cpu_buffer_type()->iface.is_host,
         },
+        /* .device   = */ nullptr,
         /* .context  = */ nullptr,
     };
 
@@ -6771,11 +6773,8 @@ static ggml_backend_i ggml_backend_vk_interface = {
     /* .supports_op             = */ ggml_backend_vk_supports_op,
     /* .supports_buft           = */ ggml_backend_vk_supports_buft,
     /* .offload_op              = */ ggml_backend_vk_offload_op,
-    /* .event_new               = */ NULL,
-    /* .event_free              = */ NULL,
     /* .event_record            = */ NULL,
     /* .event_wait              = */ NULL,
-    /* .event_synchronize       = */ NULL,
 };
 
 static ggml_guid_t ggml_backend_vk_guid() {
@@ -6792,6 +6791,7 @@ ggml_backend_t ggml_backend_vk_init(size_t dev_num) {
     ggml_backend_t vk_backend = new ggml_backend {
         /* .guid      = */ ggml_backend_vk_guid(),
         /* .interface = */ ggml_backend_vk_interface,
+        /* .device    = */ nullptr,
         /* .context   = */ ctx,
     };
 
@@ -6834,19 +6834,6 @@ static ggml_backend_t ggml_backend_reg_vk_init(const char * params, void * user_
     UNUSED(params);
 }
 
-extern "C" int ggml_backend_vk_reg_devices();
-
-int ggml_backend_vk_reg_devices() {
-    ggml_vk_instance_init();
-
-    for (size_t i = 0; i < vk_instance.device_indices.size(); i++) {
-        char name[128];
-        snprintf(name, sizeof(name), "%s%ld", GGML_VK_NAME, i);
-        ggml_backend_register(name, ggml_backend_reg_vk_init, ggml_backend_vk_buffer_type(i), (void *) (intptr_t) i);  // NOLINT
-    }
-    return vk_instance.device_indices.size();
-}
-
 // Extension availability
 static bool ggml_vk_instance_validation_ext_available(const std::vector<vk::ExtensionProperties>& instance_extensions) {
 #ifdef GGML_VULKAN_VALIDATE

scripts/sync-ggml-am.sh

@@ -122,7 +122,7 @@ if [ -f $SRC_LLAMA/ggml-src.patch ]; then
     # src/ggml-aarch64.h      -> ggml/src/ggml-aarch64.h
     # src/ggml-alloc.c        -> ggml/src/ggml-alloc.c
     # src/ggml-backend-impl.h -> ggml/src/ggml-backend-impl.h
-    # src/ggml-backend.c      -> ggml/src/ggml-backend.c
+    # src/ggml-backend.cpp    -> ggml/src/ggml-backend.cpp
     # src/ggml-cann/*         -> ggml/src/ggml-cann/
     # src/ggml-cann.cpp       -> ggml/src/ggml-cann.cpp
     # src/ggml-common.h       -> ggml/src/ggml-common.h
@@ -169,7 +169,7 @@ if [ -f $SRC_LLAMA/ggml-src.patch ]; then
         -e 's/([[:space:]]|[ab]\/)src\/ggml-aarch64\.h/\1ggml\/src\/ggml-aarch64.h/g' \
         -e 's/([[:space:]]|[ab]\/)src\/ggml-alloc\.c/\1ggml\/src\/ggml-alloc.c/g' \
         -e 's/([[:space:]]|[ab]\/)src\/ggml-backend-impl\.h/\1ggml\/src\/ggml-backend-impl.h/g' \
-        -e 's/([[:space:]]|[ab]\/)src\/ggml-backend\.c/\1ggml\/src\/ggml-backend.c/g' \
+        -e 's/([[:space:]]|[ab]\/)src\/ggml-backend\.cpp/\1ggml\/src\/ggml-backend.cpp/g' \
         -e 's/([[:space:]]|[ab]\/)src\/ggml-cann\//\1ggml\/src\/ggml-cann\//g' \
         -e 's/([[:space:]]|[ab]\/)src\/ggml-cann\.cpp/\1ggml\/src\/ggml-cann.cpp/g' \
         -e 's/([[:space:]]|[ab]\/)src\/ggml-common\.h/\1ggml\/src\/ggml-common.h/g' \