ggml: unify backend logging mechanism (#9709)
* Add scaffolding for ggml logging macros * Metal backend now uses GGML logging * Cuda backend now uses GGML logging * Cann backend now uses GGML logging * Add enum tag to parameters * Use C memory allocation funcs * Fix compile error * Use GGML_LOG instead of GGML_PRINT * Rename llama_state to llama_logger_state * Prevent null format string * Fix whitespace * Remove log callbacks from ggml backends * Remove cuda log statement
This commit is contained in:
bandoti
GitHub
parent
e3c355ba65
commit
d6fe7abf04
13 changed files with 197 additions and 340 deletions
|
|
@ -319,26 +319,63 @@ void ggml_abort(const char * file, int line, const char * fmt, ...) {
|
|||
// logging
|
||||
//
|
||||
|
||||
struct ggml_logger_state {
|
||||
ggml_log_callback log_callback;
|
||||
void * log_callback_user_data;
|
||||
};
|
||||
static struct ggml_logger_state g_logger_state = {ggml_log_callback_default, NULL};
|
||||
|
||||
static void ggml_log_internal_v(enum ggml_log_level level, const char * format, va_list args) {
|
||||
if (format == NULL)
|
||||
return;
|
||||
va_list args_copy;
|
||||
va_copy(args_copy, args);
|
||||
char buffer[128];
|
||||
int len = vsnprintf(buffer, 128, format, args);
|
||||
if (len < 128) {
|
||||
g_logger_state.log_callback(level, buffer, g_logger_state.log_callback_user_data);
|
||||
} else {
|
||||
char * buffer2 = (char *) calloc(len + 1, sizeof(char));
|
||||
vsnprintf(buffer2, len + 1, format, args_copy);
|
||||
buffer2[len] = 0;
|
||||
g_logger_state.log_callback(level, buffer2, g_logger_state.log_callback_user_data);
|
||||
free(buffer2);
|
||||
}
|
||||
va_end(args_copy);
|
||||
}
|
||||
|
||||
void ggml_log_internal(enum ggml_log_level level, const char * format, ...) {
|
||||
va_list args;
|
||||
va_start(args, format);
|
||||
ggml_log_internal_v(level, format, args);
|
||||
va_end(args);
|
||||
}
|
||||
|
||||
void ggml_log_callback_default(enum ggml_log_level level, const char * text, void * user_data) {
|
||||
(void) level;
|
||||
(void) user_data;
|
||||
fputs(text, stderr);
|
||||
fflush(stderr);
|
||||
}
|
||||
|
||||
#if (GGML_DEBUG >= 1)
|
||||
#define GGML_PRINT_DEBUG(...) printf(__VA_ARGS__)
|
||||
#define GGML_PRINT_DEBUG(...) GGML_LOG_DEBUG(__VA_ARGS__)
|
||||
#else
|
||||
#define GGML_PRINT_DEBUG(...)
|
||||
#endif
|
||||
|
||||
#if (GGML_DEBUG >= 5)
|
||||
#define GGML_PRINT_DEBUG_5(...) printf(__VA_ARGS__)
|
||||
#define GGML_PRINT_DEBUG_5(...) GGML_LOG_DEBUG(__VA_ARGS__)
|
||||
#else
|
||||
#define GGML_PRINT_DEBUG_5(...)
|
||||
#endif
|
||||
|
||||
#if (GGML_DEBUG >= 10)
|
||||
#define GGML_PRINT_DEBUG_10(...) printf(__VA_ARGS__)
|
||||
#define GGML_PRINT_DEBUG_10(...) GGML_LOG_DEBUG(__VA_ARGS__)
|
||||
#else
|
||||
#define GGML_PRINT_DEBUG_10(...)
|
||||
#endif
|
||||
|
||||
#define GGML_PRINT(...) printf(__VA_ARGS__)
|
||||
|
||||
//
|
||||
// end of logging block
|
||||
//
|
||||
|
|
@ -355,7 +392,7 @@ void ggml_abort(const char * file, int line, const char * fmt, ...) {
|
|||
#else
|
||||
inline static void * ggml_aligned_malloc(size_t size) {
|
||||
if (size == 0) {
|
||||
GGML_PRINT("WARNING: Behavior may be unexpected when allocating 0 bytes for ggml_aligned_malloc!\n");
|
||||
GGML_LOG_WARN("Behavior may be unexpected when allocating 0 bytes for ggml_aligned_malloc!\n");
|
||||
return NULL;
|
||||
}
|
||||
void * aligned_memory = NULL;
|
||||
|
|
@ -377,7 +414,7 @@ inline static void * ggml_aligned_malloc(size_t size) {
|
|||
error_desc = "insufficient memory";
|
||||
break;
|
||||
}
|
||||
GGML_PRINT("%s: %s (attempted to allocate %6.2f MB)\n", __func__, error_desc, size/(1024.0*1024.0));
|
||||
GGML_LOG_ERROR("%s: %s (attempted to allocate %6.2f MB)\n", __func__, error_desc, size/(1024.0*1024.0));
|
||||
GGML_ABORT("fatal error");
|
||||
return NULL;
|
||||
}
|
||||
|
|
@ -393,12 +430,12 @@ inline static void * ggml_aligned_malloc(size_t size) {
|
|||
|
||||
inline static void * ggml_malloc(size_t size) {
|
||||
if (size == 0) {
|
||||
GGML_PRINT("WARNING: Behavior may be unexpected when allocating 0 bytes for ggml_malloc!\n");
|
||||
GGML_LOG_WARN("Behavior may be unexpected when allocating 0 bytes for ggml_malloc!\n");
|
||||
return NULL;
|
||||
}
|
||||
void * result = malloc(size);
|
||||
if (result == NULL) {
|
||||
GGML_PRINT("%s: failed to allocate %6.2f MB\n", __func__, size/(1024.0*1024.0));
|
||||
GGML_LOG_ERROR("%s: failed to allocate %6.2f MB\n", __func__, size/(1024.0*1024.0));
|
||||
GGML_ABORT("fatal error");
|
||||
}
|
||||
return result;
|
||||
|
|
@ -407,12 +444,12 @@ inline static void * ggml_malloc(size_t size) {
|
|||
// calloc
|
||||
inline static void * ggml_calloc(size_t num, size_t size) {
|
||||
if (num == 0 || size == 0) {
|
||||
GGML_PRINT("WARNING: Behavior may be unexpected when allocating 0 bytes for ggml_calloc!\n");
|
||||
GGML_LOG_WARN("Behavior may be unexpected when allocating 0 bytes for ggml_calloc!\n");
|
||||
return NULL;
|
||||
}
|
||||
void * result = calloc(num, size);
|
||||
if (result == NULL) {
|
||||
GGML_PRINT("%s: failed to allocate %6.2f MB\n", __func__, size/(1024.0*1024.0));
|
||||
GGML_LOG_ERROR("%s: failed to allocate %6.2f MB\n", __func__, size/(1024.0*1024.0));
|
||||
GGML_ABORT("fatal error");
|
||||
}
|
||||
return result;
|
||||
|
|
@ -3347,7 +3384,7 @@ void ggml_numa_init(enum ggml_numa_strategy numa_flag) {
|
|||
if (fptr != NULL) {
|
||||
char buf[42];
|
||||
if (fgets(buf, sizeof(buf), fptr) && strncmp(buf, "0\n", sizeof(buf)) != 0) {
|
||||
GGML_PRINT("WARNING: /proc/sys/kernel/numa_balancing is enabled, this has been observed to impair performance\n");
|
||||
GGML_LOG_WARN("/proc/sys/kernel/numa_balancing is enabled, this has been observed to impair performance\n");
|
||||
}
|
||||
fclose(fptr);
|
||||
}
|
||||
|
|
@ -3365,21 +3402,21 @@ bool ggml_is_numa(void) {
|
|||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
void ggml_print_object(const struct ggml_object * obj) {
|
||||
GGML_PRINT(" - ggml_object: type = %d, offset = %zu, size = %zu, next = %p\n",
|
||||
GGML_LOG_INFO(" - ggml_object: type = %d, offset = %zu, size = %zu, next = %p\n",
|
||||
obj->type, obj->offs, obj->size, (const void *) obj->next);
|
||||
}
|
||||
|
||||
void ggml_print_objects(const struct ggml_context * ctx) {
|
||||
struct ggml_object * obj = ctx->objects_begin;
|
||||
|
||||
GGML_PRINT("%s: objects in context %p:\n", __func__, (const void *) ctx);
|
||||
GGML_LOG_INFO("%s: objects in context %p:\n", __func__, (const void *) ctx);
|
||||
|
||||
while (obj != NULL) {
|
||||
ggml_print_object(obj);
|
||||
obj = obj->next;
|
||||
}
|
||||
|
||||
GGML_PRINT("%s: --- end ---\n", __func__);
|
||||
GGML_LOG_INFO("%s: --- end ---\n", __func__);
|
||||
}
|
||||
|
||||
int64_t ggml_nelements(const struct ggml_tensor * tensor) {
|
||||
|
|
@ -3962,7 +3999,7 @@ static struct ggml_object * ggml_new_object(struct ggml_context * ctx, enum ggml
|
|||
struct ggml_object * const obj_new = (struct ggml_object *)(mem_buffer + cur_end);
|
||||
|
||||
if (cur_end + size_needed + GGML_OBJECT_SIZE > ctx->mem_size) {
|
||||
GGML_PRINT("%s: not enough space in the context's memory pool (needed %zu, available %zu)\n",
|
||||
GGML_LOG_WARN("%s: not enough space in the context's memory pool (needed %zu, available %zu)\n",
|
||||
__func__, cur_end + size_needed + GGML_OBJECT_SIZE, ctx->mem_size);
|
||||
assert(false);
|
||||
return NULL;
|
||||
|
|
@ -4026,7 +4063,7 @@ static struct ggml_tensor * ggml_new_tensor_impl(
|
|||
if (ctx->scratch.data != NULL) {
|
||||
// allocate tensor data in the scratch buffer
|
||||
if (ctx->scratch.offs + data_size > ctx->scratch.size) {
|
||||
GGML_PRINT("%s: not enough space in the scratch memory pool (needed %zu, available %zu)\n",
|
||||
GGML_LOG_WARN("%s: not enough space in the scratch memory pool (needed %zu, available %zu)\n",
|
||||
__func__, ctx->scratch.offs + data_size, ctx->scratch.size);
|
||||
assert(false);
|
||||
return NULL;
|
||||
|
|
@ -20013,7 +20050,7 @@ enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cpl
|
|||
}
|
||||
#else
|
||||
if (n_threads > threadpool->n_threads_max) {
|
||||
GGML_PRINT("WARNING: cplan requested more threads (%d) than available (%d)\n", n_threads, threadpool->n_threads_max);
|
||||
GGML_LOG_WARN("cplan requested more threads (%d) than available (%d)\n", n_threads, threadpool->n_threads_max);
|
||||
n_threads = threadpool->n_threads_max;
|
||||
}
|
||||
|
||||
|
|
@ -20552,30 +20589,30 @@ struct ggml_cgraph * ggml_graph_import(const char * fname, struct ggml_context *
|
|||
}
|
||||
|
||||
void ggml_graph_print(const struct ggml_cgraph * cgraph) {
|
||||
GGML_PRINT("=== GRAPH ===\n");
|
||||
GGML_LOG_INFO("=== GRAPH ===\n");
|
||||
|
||||
GGML_PRINT("n_nodes = %d\n", cgraph->n_nodes);
|
||||
GGML_LOG_INFO("n_nodes = %d\n", cgraph->n_nodes);
|
||||
for (int i = 0; i < cgraph->n_nodes; i++) {
|
||||
struct ggml_tensor * node = cgraph->nodes[i];
|
||||
|
||||
GGML_PRINT(" - %3d: [ %5" PRId64 ", %5" PRId64 ", %5" PRId64 "] %16s %s\n",
|
||||
GGML_LOG_INFO(" - %3d: [ %5" PRId64 ", %5" PRId64 ", %5" PRId64 "] %16s %s\n",
|
||||
i,
|
||||
node->ne[0], node->ne[1], node->ne[2],
|
||||
ggml_op_name(node->op), (node->flags & GGML_TENSOR_FLAG_PARAM) ? "x" : node->grad ? "g" : " ");
|
||||
}
|
||||
|
||||
GGML_PRINT("n_leafs = %d\n", cgraph->n_leafs);
|
||||
GGML_LOG_INFO("n_leafs = %d\n", cgraph->n_leafs);
|
||||
for (int i = 0; i < cgraph->n_leafs; i++) {
|
||||
struct ggml_tensor * node = cgraph->leafs[i];
|
||||
|
||||
GGML_PRINT(" - %3d: [ %5" PRId64 ", %5" PRId64 "] %8s %16s\n",
|
||||
GGML_LOG_INFO(" - %3d: [ %5" PRId64 ", %5" PRId64 "] %8s %16s\n",
|
||||
i,
|
||||
node->ne[0], node->ne[1],
|
||||
ggml_op_name(node->op),
|
||||
ggml_get_name(node));
|
||||
}
|
||||
|
||||
GGML_PRINT("========================================\n");
|
||||
GGML_LOG_INFO("========================================\n");
|
||||
}
|
||||
|
||||
// check if node is part of the graph
|
||||
|
|
@ -20746,7 +20783,7 @@ void ggml_graph_dump_dot(const struct ggml_cgraph * gb, const struct ggml_cgraph
|
|||
|
||||
fclose(fp);
|
||||
|
||||
GGML_PRINT("%s: dot -Tpng %s -o %s.png && open %s.png\n", __func__, filename, filename, filename);
|
||||
GGML_LOG_INFO("%s: dot -Tpng %s -o %s.png && open %s.png\n", __func__, filename, filename, filename);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
|
@ -23241,4 +23278,9 @@ int ggml_cpu_get_sve_cnt(void) {
|
|||
return 0;
|
||||
#endif
|
||||
}
|
||||
|
||||
void ggml_log_set(ggml_log_callback log_callback, void * user_data) {
|
||||
g_logger_state.log_callback = log_callback ? log_callback : ggml_log_callback_default;
|
||||
g_logger_state.log_callback_user_data = user_data;
|
||||
}
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue