ggml-backend : add device and backend reg interfaces (llama/9707)

Also:

- metal : fix compute pass descriptor autorelease crash
- ggml-backend : add device description to CPU backend
- ggml: unify backend logging mechanism

ggml/include/ggml-backend.h

@@ -164,7 +164,7 @@ extern "C" {
     GGML_API size_t             ggml_backend_reg_dev_count(ggml_backend_reg_t reg);
     GGML_API ggml_backend_dev_t ggml_backend_reg_dev_get(ggml_backend_reg_t reg, size_t index);
     GGML_API void *             ggml_backend_reg_get_proc_address(ggml_backend_reg_t reg, const char * name);
-    GGML_API void               ggml_backend_reg_set_log_callback(ggml_backend_reg_t reg, ggml_log_callback log_callback, void * user_data);
+
 
     // Functions that may be obtained using ggml_backend_reg_get_proc_address
     typedef ggml_backend_buffer_type_t (*ggml_backend_split_buffer_type_t)(const float *);
@@ -184,9 +184,6 @@ extern "C" {
     GGML_API ggml_backend_dev_t ggml_backend_dev_by_name(const char * name);
     GGML_API ggml_backend_dev_t ggml_backend_dev_by_type(enum ggml_backend_dev_type type);
 
-    // Set the log callback for all registered backends
-    GGML_API void ggml_backend_set_log_callback(ggml_log_callback log_callback, void * user_data);
-
     // Direct backend (stream) initialization
     // = ggml_backend_dev_init(ggml_backend_dev_by_name(name), params)
     GGML_API ggml_backend_t ggml_backend_init_by_name(const char * name, const char * params);

ggml/include/ggml-cann.h

@@ -116,17 +116,6 @@ GGML_API void ggml_backend_cann_get_device_memory(int32_t device,
                                                   size_t* free,
                                                   size_t* total);
 
-/**
- * @brief Set the logging callback for GGML.
- *
- * This function sets the logging callback and user data for logging.
- *
- * @param log_callback The logging callback to set.
- * @param user_data User data to pass to the logging callback.
- */
-GGML_API void ggml_backend_cann_log_set_callback(ggml_log_callback log_callback,
-                                                 void* user_data);
-
 #ifdef __cplusplus
 }
 #endif

ggml/include/ggml-cuda.h

@@ -40,8 +40,6 @@ GGML_API void ggml_backend_cuda_get_device_memory(int device, size_t * free, siz
 GGML_API bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size);
 GGML_API void ggml_backend_cuda_unregister_host_buffer(void * buffer);
 
-GGML_API void ggml_backend_cuda_log_set_callback(ggml_log_callback log_callback, void * user_data);
-
 GGML_API ggml_backend_reg_t ggml_backend_cuda_reg(void);
 
 #ifdef  __cplusplus

ggml/include/ggml-metal.h

@@ -39,8 +39,6 @@ extern "C" {
 // user-code should use only these functions
 //
 
-GGML_API void ggml_backend_metal_log_set_callback(ggml_log_callback log_callback, void * user_data);
-
 GGML_API ggml_backend_t ggml_backend_metal_init(void);
 
 GGML_API bool ggml_backend_is_metal(ggml_backend_t backend);

ggml/include/ggml.h

@@ -2174,6 +2174,10 @@ extern "C" {
     typedef void (*ggml_opt_callback)(void * data, int accum_step, float * sched, bool * cancel);
     typedef void (*ggml_log_callback)(enum ggml_log_level level, const char * text, void * user_data);
 
+    // Set callback for all future logging events.
+    // If this is not called, or NULL is supplied, everything is output on stderr.
+    GGML_API void ggml_log_set(ggml_log_callback log_callback, void * user_data);
+
     // optimization parameters
     //
     //   see ggml.c (ggml_opt_default_params) for default values

ggml/src/ggml-backend-impl.h

@@ -215,9 +215,6 @@ extern "C" {
         // (optional) get a pointer to a function in the backend
         // backends can add custom functions that are not part of the standard ggml-backend interface
         void * (*get_proc_address)(ggml_backend_reg_t reg, const char * name);
-
-        // (optional) set the log callback for the backend
-        void (*set_log_callback)(ggml_backend_reg_t reg, ggml_log_callback log_callback, void * user_data);
     };
 
     struct ggml_backend_reg {

ggml/src/ggml-backend.cpp

@@ -1,5 +1,13 @@
 // Note: porting this file to C++ is a work in progress
 
+#ifdef _WIN32
+#define WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
+#   define NOMINMAX
+#endif
+#include <windows.h>
+#endif
+
 #include "ggml-backend-impl.h"
 #include "ggml-alloc.h"
 #include "ggml-impl.h"
@@ -10,9 +18,15 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-
+#include <string>
 #include <vector>
 
+#ifdef __APPLE__
+#include <sys/types.h>
+#include <sys/sysctl.h>
+#endif
+
+
 // backend buffer type
 
 const char * ggml_backend_buft_name(ggml_backend_buffer_type_t buft) {
@@ -505,12 +519,6 @@ void * ggml_backend_reg_get_proc_address(ggml_backend_reg_t reg, const char * na
     return reg->iface.get_proc_address(reg, name);
 }
 
-void ggml_backend_reg_set_log_callback(ggml_backend_reg_t reg, ggml_log_callback log_callback, void * user_data) {
-    if (reg->iface.set_log_callback) {
-        reg->iface.set_log_callback(reg, log_callback, user_data);
-    }
-}
-
 // Backend registry
 
 #ifdef GGML_USE_CUDA
@@ -614,13 +622,6 @@ ggml_backend_dev_t ggml_backend_dev_by_type(enum ggml_backend_dev_type type) {
     return NULL;
 }
 
-void ggml_backend_set_log_callback(ggml_log_callback log_callback, void * user_data) {
-    for (size_t i = 0; i < ggml_backend_reg_count(); i++) {
-        ggml_backend_reg_t reg = ggml_backend_reg_get(i);
-        ggml_backend_reg_set_log_callback(reg, log_callback, user_data);
-    }
-}
-
 // Convenience functions
 ggml_backend_t ggml_backend_init_by_name(const char * name, const char * params) {
     ggml_backend_dev_t dev = ggml_backend_dev_by_name(name);
@@ -1021,17 +1022,80 @@ ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size)
 
 ////////////////////////
 
+struct ggml_backend_cpu_device_context {
+    std::string description = "CPU";
+
+    ggml_backend_cpu_device_context() {
+#ifdef __APPLE__
+        size_t len = 0;
+        if (!sysctlbyname("machdep.cpu.brand_string", NULL, &len, NULL, 0)) {
+            description.resize(len);
+            sysctlbyname("machdep.cpu.brand_string", &description[0], &len, NULL, 0); // NOLINT
+        }
+#elif defined(__linux__)
+        FILE * f = fopen("/proc/cpuinfo", "r");
+        if (f) {
+            char buf[1024];
+            while (fgets(buf, sizeof(buf), f)) {
+                if (strncmp(buf, "model name", 10) == 0) {
+                    char * p = strchr(buf, ':');
+                    if (p) {
+                        p++;
+                        while (std::isspace(*p)) {
+                            p++;
+                        }
+                        while (std::isspace(p[strlen(p) - 1])) {
+                            p[strlen(p) - 1] = '\0';
+                        }
+                        description = p;
+                        break;
+                    }
+                }
+            }
+            fclose(f);
+        }
+#elif defined(_WIN32)
+        HKEY hKey;
+        if (RegOpenKeyEx(HKEY_LOCAL_MACHINE,
+                        TEXT("HARDWARE\\DESCRIPTION\\System\\CentralProcessor\\0"),
+                        0,
+                        KEY_READ,
+                        &hKey) == ERROR_SUCCESS) {
+            DWORD cpu_brand_size = 0;
+            if (RegQueryValueExA(hKey,
+                                TEXT("ProcessorNameString"),
+                                NULL,
+                                NULL,
+                                NULL,
+                                &cpu_brand_size) == ERROR_SUCCESS) {
+                description.resize(cpu_brand_size);
+                if (RegQueryValueExA(hKey,
+                                    TEXT("ProcessorNameString"),
+                                    NULL,
+                                    NULL,
+                                    (LPBYTE)&description[0], // NOLINT
+                                    &cpu_brand_size) == ERROR_SUCCESS) {
+                    if (description.find('\0') != std::string::npos) {
+                        description.resize(description.find('\0'));
+                    }
+                }
+            }
+            RegCloseKey(hKey);
+        }
+#endif
+    }
+};
+
 static const char * ggml_backend_cpu_device_get_name(ggml_backend_dev_t dev) {
     return "CPU";
 
     GGML_UNUSED(dev);
 }
 
 static const char * ggml_backend_cpu_device_get_description(ggml_backend_dev_t dev) {
-    // TODO
-    return "CPU";
+    struct ggml_backend_cpu_device_context * ctx = (struct ggml_backend_cpu_device_context *)dev->context;
 
-    GGML_UNUSED(dev);
+    return ctx->description.c_str();
 }
 
 static void ggml_backend_cpu_device_get_memory(ggml_backend_dev_t dev, size_t * free, size_t * total) {
@@ -1144,10 +1208,11 @@ static size_t ggml_backend_cpu_reg_get_device_count(ggml_backend_reg_t reg) {
 static ggml_backend_dev_t ggml_backend_cpu_reg_get_device(ggml_backend_reg_t reg, size_t index) {
     GGML_ASSERT(index == 0);
 
+    static ggml_backend_cpu_device_context ctx;
     static ggml_backend_device ggml_backend_cpu_device = {
         /* .iface   = */ ggml_backend_cpu_device_i,
         /* .reg     = */ reg,
-        /* .context = */ NULL,
+        /* .context = */ &ctx,
     };
 
     return &ggml_backend_cpu_device;
@@ -1161,7 +1226,6 @@ static const struct ggml_backend_reg_i ggml_backend_cpu_reg_i = {
     /* .get_device_count = */ ggml_backend_cpu_reg_get_device_count,
     /* .get_device       = */ ggml_backend_cpu_reg_get_device,
     /* .get_proc_address = */ NULL,
-    /* .set_log_callback = */ NULL,
 };
 
 ggml_backend_reg_t ggml_backend_cpu_reg(void) {