llama-cpp-capacitor 0.0.6 → 0.0.7

package/cpp/ggml-backend-reg.cpp

@@ -0,0 +1,600 @@
+#include "ggml-backend-impl.h"
+#include "ggml-backend.h"
+#include "ggml-impl.h"
+#include <algorithm>
+#include <cstring>
+#include <filesystem>
+#include <memory>
+#include <string>
+#include <type_traits>
+#include <vector>
+#include <cctype>
+
+#ifdef _WIN32
+#    define WIN32_LEAN_AND_MEAN
+#    ifndef NOMINMAX
+#        define NOMINMAX
+#    endif
+#    include <windows.h>
+#elif defined(__APPLE__)
+#    include <mach-o/dyld.h>
+#    include <dlfcn.h>
+#else
+#    include <dlfcn.h>
+#    include <unistd.h>
+#endif
+
+// Backend registry
+#ifdef LM_GGML_USE_CPU
+#include "ggml-cpu.h"
+#endif
+
+#ifdef LM_GGML_USE_CUDA
+#include "ggml-cuda.h"
+#endif
+
+#ifdef LM_GGML_USE_METAL
+#include "ggml-metal.h"
+#endif
+
+#ifdef LM_GGML_USE_SYCL
+#include "ggml-sycl.h"
+#endif
+
+#ifdef LM_GGML_USE_VULKAN
+#include "ggml-vulkan.h"
+#endif
+
+#ifdef LM_GGML_USE_WEBGPU
+#include "ggml-webgpu.h"
+#endif
+
+#ifdef LM_GGML_USE_ZDNN
+#include "ggml-zdnn.h"
+#endif
+
+#ifdef LM_GGML_USE_OPENCL
+#include "ggml-opencl.h"
+#endif
+
+#ifdef LM_GGML_USE_BLAS
+#include "ggml-blas.h"
+#endif
+
+#ifdef LM_GGML_USE_RPC
+#include "ggml-rpc.h"
+#endif
+
+#ifdef LM_GGML_USE_CANN
+#include "ggml-cann.h"
+#endif
+
+// disable C++17 deprecation warning for std::codecvt_utf8
+#if defined(__clang__)
+#    pragma clang diagnostic push
+#    pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#elif defined(__GNUC__)
+#    pragma GCC diagnostic push
+#    pragma GCC diagnostic ignored "-Wdeprecated-declarations"
+#endif
+
+namespace fs = std::filesystem;
+
+static std::string path_str(const fs::path & path) {
+    std::string u8path;
+    try {
+#if defined(__cpp_lib_char8_t)
+        // C++20 and later: u8string() returns std::u8string
+        std::u8string u8str = path.u8string();
+        u8path = std::string(reinterpret_cast<const char*>(u8str.c_str()));
+#else
+        // C++17: u8string() returns std::string
+        u8path = path.u8string();
+#endif
+    } catch (...) {
+    }
+    return u8path;
+}
+
+#if defined(__clang__)
+#    pragma clang diagnostic pop
+#elif defined(__GNUC__)
+#    pragma GCC diagnostic pop
+#endif
+
+#ifdef _WIN32
+
+using dl_handle = std::remove_pointer_t<HMODULE>;
+
+struct dl_handle_deleter {
+    void operator()(HMODULE handle) {
+        FreeLibrary(handle);
+    }
+};
+
+static dl_handle * dl_load_library(const fs::path & path) {
+    // suppress error dialogs for missing DLLs
+    DWORD old_mode = SetErrorMode(SEM_FAILCRITICALERRORS);
+    SetErrorMode(old_mode | SEM_FAILCRITICALERRORS);
+
+    HMODULE handle = LoadLibraryW(path.wstring().c_str());
+
+    SetErrorMode(old_mode);
+
+    return handle;
+}
+
+static void * dl_get_sym(dl_handle * handle, const char * name) {
+    DWORD old_mode = SetErrorMode(SEM_FAILCRITICALERRORS);
+    SetErrorMode(old_mode | SEM_FAILCRITICALERRORS);
+
+    void * p = (void *) GetProcAddress(handle, name);
+
+    SetErrorMode(old_mode);
+
+    return p;
+}
+
+#else
+
+using dl_handle = void;
+
+struct dl_handle_deleter {
+    void operator()(void * handle) {
+        dlclose(handle);
+    }
+};
+
+static void * dl_load_library(const fs::path & path) {
+    dl_handle * handle = dlopen(path.string().c_str(), RTLD_NOW | RTLD_LOCAL);
+
+    return handle;
+}
+
+static void * dl_get_sym(dl_handle * handle, const char * name) {
+    return dlsym(handle, name);
+}
+
+#endif
+
+using dl_handle_ptr = std::unique_ptr<dl_handle, dl_handle_deleter>;
+
+struct lm_ggml_backend_reg_entry {
+    lm_ggml_backend_reg_t reg;
+    dl_handle_ptr handle;
+};
+
+struct lm_ggml_backend_registry {
+    std::vector<lm_ggml_backend_reg_entry> backends;
+    std::vector<lm_ggml_backend_dev_t> devices;
+
+    lm_ggml_backend_registry() {
+#ifdef LM_GGML_USE_CUDA
+        register_backend(lm_ggml_backend_cuda_reg());
+#endif
+#ifdef LM_GGML_USE_METAL
+        register_backend(lm_ggml_backend_metal_reg());
+#endif
+#ifdef LM_GGML_USE_SYCL
+        register_backend(lm_ggml_backend_sycl_reg());
+#endif
+#ifdef LM_GGML_USE_VULKAN
+        register_backend(lm_ggml_backend_vk_reg());
+#endif
+#ifdef LM_GGML_USE_WEBGPU
+        register_backend(lm_ggml_backend_webgpu_reg());
+#endif
+#ifdef LM_GGML_USE_ZDNN
+        register_backend(lm_ggml_backend_zdnn_reg());
+#endif
+#ifdef LM_GGML_USE_OPENCL
+        register_backend(lm_ggml_backend_opencl_reg());
+#endif
+#ifdef LM_GGML_USE_CANN
+        register_backend(lm_ggml_backend_cann_reg());
+#endif
+#ifdef LM_GGML_USE_BLAS
+        register_backend(lm_ggml_backend_blas_reg());
+#endif
+#ifdef LM_GGML_USE_RPC
+        register_backend(lm_ggml_backend_rpc_reg());
+#endif
+#ifdef LM_GGML_USE_CPU
+        register_backend(lm_ggml_backend_cpu_reg());
+#endif
+    }
+
+    ~lm_ggml_backend_registry() {
+        // FIXME: backends cannot be safely unloaded without a function to destroy all the backend resources,
+        // since backend threads may still be running and accessing resources from the dynamic library
+        for (auto & entry : backends) {
+            if (entry.handle) {
+                entry.handle.release(); // NOLINT
+            }
+        }
+    }
+
+    void register_backend(lm_ggml_backend_reg_t reg, dl_handle_ptr handle = nullptr) {
+        if (!reg) {
+            return;
+        }
+
+#ifndef NDEBUG
+        LM_GGML_LOG_DEBUG("%s: registered backend %s (%zu devices)\n",
+            __func__, lm_ggml_backend_reg_name(reg), lm_ggml_backend_reg_dev_count(reg));
+#endif
+        backends.push_back({ reg, std::move(handle) });
+        for (size_t i = 0; i < lm_ggml_backend_reg_dev_count(reg); i++) {
+            register_device(lm_ggml_backend_reg_dev_get(reg, i));
+        }
+    }
+
+    void register_device(lm_ggml_backend_dev_t device) {
+#ifndef NDEBUG
+        LM_GGML_LOG_DEBUG("%s: registered device %s (%s)\n", __func__, lm_ggml_backend_dev_name(device), lm_ggml_backend_dev_description(device));
+#endif
+        devices.push_back(device);
+    }
+
+    lm_ggml_backend_reg_t load_backend(const fs::path & path, bool silent) {
+        dl_handle_ptr handle { dl_load_library(path) };
+        if (!handle) {
+            if (!silent) {
+                LM_GGML_LOG_ERROR("%s: failed to load %s\n", __func__, path_str(path).c_str());
+            }
+            return nullptr;
+        }
+
+        auto score_fn = (lm_ggml_backend_score_t) dl_get_sym(handle.get(), "lm_ggml_backend_score");
+        if (score_fn && score_fn() == 0) {
+            if (!silent) {
+                LM_GGML_LOG_INFO("%s: backend %s is not supported on this system\n", __func__, path_str(path).c_str());
+            }
+            return nullptr;
+        }
+
+        auto backend_init_fn = (lm_ggml_backend_init_t) dl_get_sym(handle.get(), "lm_ggml_backend_init");
+        if (!backend_init_fn) {
+            if (!silent) {
+                LM_GGML_LOG_ERROR("%s: failed to find lm_ggml_backend_init in %s\n", __func__, path_str(path).c_str());
+            }
+            return nullptr;
+        }
+
+        lm_ggml_backend_reg_t reg = backend_init_fn();
+        if (!reg || reg->api_version != LM_GGML_BACKEND_API_VERSION) {
+            if (!silent) {
+                if (!reg) {
+                    LM_GGML_LOG_ERROR("%s: failed to initialize backend from %s: lm_ggml_backend_init returned NULL\n",
+                        __func__, path_str(path).c_str());
+                } else {
+                    LM_GGML_LOG_ERROR("%s: failed to initialize backend from %s: incompatible API version (backend: %d, current: %d)\n",
+                        __func__, path_str(path).c_str(), reg->api_version, LM_GGML_BACKEND_API_VERSION);
+                }
+            }
+            return nullptr;
+        }
+
+        LM_GGML_LOG_INFO("%s: loaded %s backend from %s\n", __func__, lm_ggml_backend_reg_name(reg), path_str(path).c_str());
+
+        register_backend(reg, std::move(handle));
+
+        return reg;
+    }
+
+    void unload_backend(lm_ggml_backend_reg_t reg, bool silent) {
+        auto it = std::find_if(backends.begin(), backends.end(),
+                               [reg](const lm_ggml_backend_reg_entry & entry) { return entry.reg == reg; });
+
+        if (it == backends.end()) {
+            if (!silent) {
+                LM_GGML_LOG_ERROR("%s: backend not found\n", __func__);
+            }
+            return;
+        }
+
+        if (!silent) {
+            LM_GGML_LOG_DEBUG("%s: unloading %s backend\n", __func__, lm_ggml_backend_reg_name(reg));
+        }
+
+        // remove devices
+        devices.erase(
+            std::remove_if(devices.begin(), devices.end(),
+                            [reg](lm_ggml_backend_dev_t dev) { return lm_ggml_backend_dev_backend_reg(dev) == reg; }),
+            devices.end());
+
+        // remove backend
+        backends.erase(it);
+    }
+};
+
+static lm_ggml_backend_registry & get_reg() {
+    static lm_ggml_backend_registry reg;
+    return reg;
+}
+
+// Internal API
+void lm_ggml_backend_register(lm_ggml_backend_reg_t reg) {
+    get_reg().register_backend(reg);
+}
+
+void lm_ggml_backend_device_register(lm_ggml_backend_dev_t device) {
+    get_reg().register_device(device);
+}
+
+// Backend (reg) enumeration
+static bool striequals(const char * a, const char * b) {
+    for (; *a && *b; a++, b++) {
+        if (std::tolower(*a) != std::tolower(*b)) {
+            return false;
+        }
+    }
+    return *a == *b;
+}
+
+size_t lm_ggml_backend_reg_count() {
+    return get_reg().backends.size();
+}
+
+lm_ggml_backend_reg_t lm_ggml_backend_reg_get(size_t index) {
+    LM_GGML_ASSERT(index < lm_ggml_backend_reg_count());
+    return get_reg().backends[index].reg;
+}
+
+lm_ggml_backend_reg_t lm_ggml_backend_reg_by_name(const char * name) {
+    for (size_t i = 0; i < lm_ggml_backend_reg_count(); i++) {
+        lm_ggml_backend_reg_t reg = lm_ggml_backend_reg_get(i);
+        if (striequals(lm_ggml_backend_reg_name(reg), name)) {
+            return reg;
+        }
+    }
+    return nullptr;
+}
+
+// Device enumeration
+size_t lm_ggml_backend_dev_count() {
+    return get_reg().devices.size();
+}
+
+lm_ggml_backend_dev_t lm_ggml_backend_dev_get(size_t index) {
+    LM_GGML_ASSERT(index < lm_ggml_backend_dev_count());
+    return get_reg().devices[index];
+}
+
+lm_ggml_backend_dev_t lm_ggml_backend_dev_by_name(const char * name) {
+    for (size_t i = 0; i < lm_ggml_backend_dev_count(); i++) {
+        lm_ggml_backend_dev_t dev = lm_ggml_backend_dev_get(i);
+        if (striequals(lm_ggml_backend_dev_name(dev), name)) {
+            return dev;
+        }
+    }
+    return nullptr;
+}
+
+lm_ggml_backend_dev_t lm_ggml_backend_dev_by_type(enum lm_ggml_backend_dev_type type) {
+    for (size_t i = 0; i < lm_ggml_backend_dev_count(); i++) {
+        lm_ggml_backend_dev_t dev = lm_ggml_backend_dev_get(i);
+        if (lm_ggml_backend_dev_type(dev) == type) {
+            return dev;
+        }
+    }
+    return nullptr;
+}
+
+// Convenience functions
+lm_ggml_backend_t lm_ggml_backend_init_by_name(const char * name, const char * params) {
+    lm_ggml_backend_dev_t dev = lm_ggml_backend_dev_by_name(name);
+    if (!dev) {
+        return nullptr;
+    }
+    return lm_ggml_backend_dev_init(dev, params);
+}
+
+lm_ggml_backend_t lm_ggml_backend_init_by_type(enum lm_ggml_backend_dev_type type, const char * params) {
+    lm_ggml_backend_dev_t dev = lm_ggml_backend_dev_by_type(type);
+    if (!dev) {
+        return nullptr;
+    }
+    return lm_ggml_backend_dev_init(dev, params);
+}
+
+lm_ggml_backend_t lm_ggml_backend_init_best(void) {
+    lm_ggml_backend_dev_t dev = lm_ggml_backend_dev_by_type(LM_GGML_BACKEND_DEVICE_TYPE_GPU);
+    if (!dev) {
+        dev = lm_ggml_backend_dev_by_type(LM_GGML_BACKEND_DEVICE_TYPE_CPU);
+    }
+    if (!dev) {
+        return nullptr;
+    }
+    return lm_ggml_backend_dev_init(dev, nullptr);
+}
+
+// Dynamic loading
+lm_ggml_backend_reg_t lm_ggml_backend_load(const char * path) {
+    return get_reg().load_backend(path, false);
+}
+
+void lm_ggml_backend_unload(lm_ggml_backend_reg_t reg) {
+    get_reg().unload_backend(reg, true);
+}
+
+static fs::path get_executable_path() {
+#if defined(__APPLE__)
+    // get executable path
+    std::vector<char> path;
+    uint32_t size;
+    while (true) {
+        size = path.size();
+        if (_NSGetExecutablePath(path.data(), &size) == 0) {
+            break;
+        }
+        path.resize(size);
+    }
+    std::string base_path(path.data(), size);
+    // remove executable name
+    auto last_slash = base_path.find_last_of('/');
+    if (last_slash != std::string::npos) {
+        base_path = base_path.substr(0, last_slash);
+    }
+    return base_path + "/";
+#elif defined(__linux__) || defined(__FreeBSD__)
+    std::string base_path = ".";
+    std::vector<char> path(1024);
+    while (true) {
+        // get executable path
+#    if defined(__linux__)
+        ssize_t len = readlink("/proc/self/exe", path.data(), path.size());
+#    elif defined(__FreeBSD__)
+        ssize_t len = readlink("/proc/curproc/file", path.data(), path.size());
+#    endif
+        if (len == -1) {
+            break;
+        }
+        if (len < (ssize_t) path.size()) {
+            base_path = std::string(path.data(), len);
+            // remove executable name
+            auto last_slash = base_path.find_last_of('/');
+            if (last_slash != std::string::npos) {
+                base_path = base_path.substr(0, last_slash);
+            }
+            break;
+        }
+        path.resize(path.size() * 2);
+    }
+
+    return base_path + "/";
+#elif defined(_WIN32)
+    std::vector<wchar_t> path(MAX_PATH);
+    DWORD len = GetModuleFileNameW(NULL, path.data(), path.size());
+    if (len == 0) {
+        return {};
+    }
+    std::wstring base_path(path.data(), len);
+    // remove executable name
+    auto last_slash = base_path.find_last_of('\\');
+    if (last_slash != std::string::npos) {
+        base_path = base_path.substr(0, last_slash);
+    }
+    return base_path + L"\\";
+#else
+    return {};
+#endif
+}
+
+static fs::path backend_filename_prefix() {
+#ifdef _WIN32
+    return fs::u8path("ggml-");
+#else
+    return fs::u8path("libggml-");
+#endif
+}
+
+static fs::path backend_filename_extension() {
+#ifdef _WIN32
+    return fs::u8path(".dll");
+#else
+    return fs::u8path(".so");
+#endif
+}
+
+static lm_ggml_backend_reg_t lm_ggml_backend_load_best(const char * name, bool silent, const char * user_search_path) {
+    // enumerate all the files that match [lib]ggml-name-*.[so|dll] in the search paths
+    const fs::path name_path = fs::u8path(name);
+    const fs::path file_prefix = backend_filename_prefix().native() + name_path.native() + fs::u8path("-").native();
+    const fs::path file_extension = backend_filename_extension();
+
+    std::vector<fs::path> search_paths;
+    if (user_search_path == nullptr) {
+#ifdef LM_GGML_BACKEND_DIR
+        search_paths.push_back(fs::u8path(LM_GGML_BACKEND_DIR));
+#endif
+        // default search paths: executable directory, current directory
+        search_paths.push_back(get_executable_path());
+        search_paths.push_back(fs::current_path());
+    } else {
+        search_paths.push_back(fs::u8path(user_search_path));
+    }
+
+    int best_score = 0;
+    fs::path best_path;
+
+    for (const auto & search_path : search_paths) {
+        if (!fs::exists(search_path)) {
+            LM_GGML_LOG_DEBUG("%s: search path %s does not exist\n", __func__, path_str(search_path).c_str());
+            continue;
+        }
+        fs::directory_iterator dir_it(search_path, fs::directory_options::skip_permission_denied);
+        for (const auto & entry : dir_it) {
+            if (entry.is_regular_file()) {
+                auto filename = entry.path().filename();
+                auto ext = entry.path().extension();
+                if (filename.native().find(file_prefix) == 0 && ext == file_extension) {
+                    dl_handle_ptr handle { dl_load_library(entry) };
+                    if (!handle && !silent) {
+                        LM_GGML_LOG_ERROR("%s: failed to load %s\n", __func__, path_str(entry.path()).c_str());
+                    }
+                    if (handle) {
+                        auto score_fn = (lm_ggml_backend_score_t) dl_get_sym(handle.get(), "lm_ggml_backend_score");
+                        if (score_fn) {
+                            int s = score_fn();
+#ifndef NDEBUG
+                            LM_GGML_LOG_DEBUG("%s: %s score: %d\n", __func__, path_str(entry.path()).c_str(), s);
+#endif
+                            if (s > best_score) {
+                                best_score = s;
+                                best_path = entry.path();
+                            }
+                        } else {
+                            if (!silent) {
+                                LM_GGML_LOG_INFO("%s: failed to find lm_ggml_backend_score in %s\n", __func__, path_str(entry.path()).c_str());
+                            }
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    if (best_score == 0) {
+        // try to load the base backend
+        for (const auto & search_path : search_paths) {
+            fs::path filename = backend_filename_prefix().native() + name_path.native() + backend_filename_extension().native();
+            fs::path path = search_path / filename;
+            if (fs::exists(path)) {
+                return get_reg().load_backend(path, silent);
+            }
+        }
+        return nullptr;
+    }
+
+    return get_reg().load_backend(best_path, silent);
+}
+
+void lm_ggml_backend_load_all() {
+    lm_ggml_backend_load_all_from_path(nullptr);
+}
+
+void lm_ggml_backend_load_all_from_path(const char * dir_path) {
+#ifdef NDEBUG
+    bool silent = true;
+#else
+    bool silent = false;
+#endif
+
+    lm_ggml_backend_load_best("blas", silent, dir_path);
+    lm_ggml_backend_load_best("cann", silent, dir_path);
+    lm_ggml_backend_load_best("cuda", silent, dir_path);
+    lm_ggml_backend_load_best("hip", silent, dir_path);
+    lm_ggml_backend_load_best("metal", silent, dir_path);
+    lm_ggml_backend_load_best("rpc", silent, dir_path);
+    lm_ggml_backend_load_best("sycl", silent, dir_path);
+    lm_ggml_backend_load_best("vulkan", silent, dir_path);
+    lm_ggml_backend_load_best("opencl", silent, dir_path);
+    lm_ggml_backend_load_best("musa", silent, dir_path);
+    lm_ggml_backend_load_best("cpu", silent, dir_path);
+    // check the environment variable LM_GGML_BACKEND_PATH to load an out-of-tree backend
+    const char * backend_path = std::getenv("LM_GGML_BACKEND_PATH");
+    if (backend_path) {
+        lm_ggml_backend_load(backend_path);
+    }
+}