* @date 15 Jan 2025
* @brief HAL (Hardware Abstract Layer) API for ML
* @author Yongjoo Ahn <yongjoo1.ahn@samsung.com>
- * @see https://github.com/nnstreamer/nnstremaer
+ * @see https://github.com/nnstreamer/nnstreamer
* @bug No known bugs except for NYI items
*
* @details
- * This HAL ML provides an interface for Tizen ML API (NNStreamer's tensor_filter subplugin)
- * to use hardware acceleration devices (NPU, ...).
+ * This HAL ML provides an interface for Tizen ML API (NNStreamer's
+ * tensor_filter subplugin) to use hardware acceleration devices (NPU, ...).
*/
#ifndef __HAL_ML__
* @brief Creates hal-ml-param instance
* @since HAL_MODULE_ML 1.0
* @remarks The @a param should be released using hal_ml_param_destroy().
- * @param[out] option Newly created option handle is returned.
+ * @param[out] param Newly created param handle is returned.
* @return @c 0 on success. Otherwise a negative error value.
* @retval #HAL_ML_ERROR_NONE Successful.
* @retval #HAL_ML_ERROR_INVALID_PARAMETER Fail. The parameter is invalid.
* @retval #HAL_ML_ERROR_OUT_OF_MEMORY Failed to allocate required memory.
*/
-int hal_ml_param_create(hal_ml_param_h *param);
+int hal_ml_param_create (hal_ml_param_h *param);
/**
* @brief Destroys hal-ml-param instance
* @retval #HAL_ML_ERROR_NONE Successful.
* @retval #HAL_ML_ERROR_INVALID_PARAMETER Fail. The parameter is invalid.
*/
-int hal_ml_param_destroy(hal_ml_param_h param);
+int hal_ml_param_destroy (hal_ml_param_h param);
/**
* @brief Sets a parameter for hal-ml-param instance
* @retval #HAL_ML_ERROR_NONE Successful.
* @retval #HAL_ML_ERROR_INVALID_PARAMETER Fail. The parameter is invalid.
*/
-int hal_ml_param_set(hal_ml_param_h param, const char *key, void *value);
+int hal_ml_param_set (hal_ml_param_h param, const char *key, void *value);
/**
* @brief Creates hal-ml instance
* @retval #HAL_ML_ERROR_INVALID_PARAMETER Fail. The parameter is invalid.
* @retval #HAL_ML_ERROR_OUT_OF_MEMORY Failed to allocate required memory.
*/
-int hal_ml_create(const char *backend_name, hal_ml_h *handle);
+int hal_ml_create (const char *backend_name, hal_ml_h *handle);
/**
* @brief Destroys hal-ml instance
* @retval #HAL_ML_ERROR_NONE Successful.
* @retval #HAL_ML_ERROR_INVALID_PARAMETER Fail. The parameter is invalid.
*/
-int hal_ml_destroy(hal_ml_h handle);
+int hal_ml_destroy (hal_ml_h handle);
/**
* @brief Sends a request to hal-ml instance
* @retval #HAL_ML_ERROR_NONE Successful.
* @retval #HAL_ML_ERROR_INVALID_PARAMETER Fail. The parameter is invalid.
*/
-int hal_ml_request(hal_ml_h handle, const char *request_name, hal_ml_param_h param);
+int hal_ml_request (hal_ml_h handle, const char *request_name, hal_ml_param_h param);
/**
* @brief Invokes the hal-ml instance with the given data.
* @retval #HAL_ML_ERROR_NONE Successful.
* @retval #HAL_ML_ERROR_INVALID_PARAMETER Fail. The parameter is invalid.
*/
-int hal_ml_request_invoke(hal_ml_h handle, const void *input, void *output);
+int hal_ml_request_invoke (hal_ml_h handle, const void *input, void *output);
/**
- * @brief Invoke dynamic the hal-ml instance with the given data.
+ * @brief Invokes the hal-ml instance dynamically with the given data.
* @since HAL_MODULE_ML 1.0
* @param[in] handle The handle of the instance.
* @param[in, out] prop The properties for the invoke dynamic.
* @retval #HAL_ML_ERROR_NONE Successful.
* @retval #HAL_ML_ERROR_INVALID_PARAMETER Fail. The parameter is invalid.
*/
-int hal_ml_request_invoke_dynamic(hal_ml_h handle, void *prop, const void *input, void *output);
+int hal_ml_request_invoke_dynamic (hal_ml_h handle, void *prop, const void *input, void *output);
/**
* @}
* @date 15 Jan 2025
* @brief HAL (Hardware Abstract Layer) API for ML
* @author Yongjoo Ahn <yongjoo1.ahn@samsung.com>
- * @see https://github.com/nnstreamer/nnstremaer
+ * @see https://github.com/nnstreamer/nnstreamer
* @bug No known bugs except for NYI items
*
* @details
- * This HAL ML provides an interface for Tizen ML API (NNStreamer's tensor_filter subplugin)
- * to use hardware acceleration devices (NPU, ...).
+ * This HAL ML provides an interface for Tizen ML API (NNStreamer's
+ * tensor_filter subplugin) to use hardware acceleration devices (NPU, ...).
*/
+#include <dlog.h>
#include <glib.h>
-
#include <hal/hal-common.h>
-
#include "hal-ml-interface.h"
#include "hal-ml.h"
-#include <dlog.h>
#define LOG_TAG "HAL_ML"
-#define _D(fmt, args...) SLOGD(fmt, ##args)
-#define _I(fmt, args...) SLOGI(fmt, ##args)
-#define _W(fmt, args...) SLOGW(fmt, ##args)
-#define _E(fmt, args...) SLOGE(fmt, ##args)
-
+#define _D(fmt, args...) SLOGD (fmt, ##args)
+#define _I(fmt, args...) SLOGI (fmt, ##args)
+#define _W(fmt, args...) SLOGW (fmt, ##args)
+#define _E(fmt, args...) SLOGE (fmt, ##args)
typedef struct _hal_ml_s {
void *backend_private;
void *value;
} hal_ml_param_value_s;
-static int hal_ml_create_backend(void **data, void *user_data)
+static int
+hal_ml_create_backend (void **data, void *user_data)
{
*data = g_new0 (hal_backend_ml_funcs, 1);
return HAL_ML_ERROR_NONE;
}
-static int hal_ml_exit_backend(void *data, void *user_data)
+static int
+hal_ml_exit_backend (void *data, void *user_data)
{
- hal_backend_ml_funcs *ml = (hal_backend_ml_funcs *) data;
g_free (data);
return HAL_ML_ERROR_NONE;
}
#define MAX_LIB_NAME_LENGTH 256
-static int hal_ml_scan_backends(void)
+static int
+hal_ml_scan_backends (void)
{
int i;
- _D("Scan available HAL ML backends...");
-
- hal_ml_backend_count = hal_common_get_backend_count(HAL_MODULE_ML);
- _D("hal_ml_backend_count: %d", hal_ml_backend_count);
+ _D ("Scanning available HAL ML backends...");
- hal_ml_backend_names = (char **) malloc(sizeof (char *) * hal_ml_backend_count);
+ hal_ml_backend_count = hal_common_get_backend_count (HAL_MODULE_ML);
+ _D ("hal_ml_backend_count: %d", hal_ml_backend_count);
+ hal_ml_backend_names = (char **) malloc (sizeof (char *) * hal_ml_backend_count);
for (i = 0; i < hal_ml_backend_count; i++) {
- hal_ml_backend_names[i] = (char *) malloc(sizeof (char) * MAX_LIB_NAME_LENGTH);
+ hal_ml_backend_names[i] = (char *) malloc (sizeof (char) * MAX_LIB_NAME_LENGTH);
}
- hal_common_get_backend_library_names(HAL_MODULE_ML, hal_ml_backend_names, hal_ml_backend_count, MAX_LIB_NAME_LENGTH);
+ hal_common_get_backend_library_names (HAL_MODULE_ML, hal_ml_backend_names,
+ hal_ml_backend_count, MAX_LIB_NAME_LENGTH);
for (i = 0; i < hal_ml_backend_count; i++) {
- _D("hal_ml_backend_names[%d]: %s", i, hal_ml_backend_names[i]);
+ _D ("hal_ml_backend_names[%d]: %s", i, hal_ml_backend_names[i]);
}
return 0;
}
-int hal_ml_param_create(hal_ml_param_h *param)
+int
+hal_ml_param_create (hal_ml_param_h *param)
{
if (!param) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
- hal_ml_param_s *param_s = g_new0(hal_ml_param_s, 1);
+ hal_ml_param_s *param_s = g_new0 (hal_ml_param_s, 1);
if (!param_s) {
return HAL_ML_ERROR_OUT_OF_MEMORY;
}
- param_s->table = g_hash_table_new_full(g_str_hash, g_str_equal, g_free, g_free);
+ param_s->table = g_hash_table_new_full (g_str_hash, g_str_equal, g_free, g_free);
if (!param_s->table) {
g_free (param_s);
return HAL_ML_ERROR_OUT_OF_MEMORY;
}
*param = (hal_ml_param_h) param_s;
-
return HAL_ML_ERROR_NONE;
}
-int hal_ml_param_destroy(hal_ml_param_h param)
+int
+hal_ml_param_destroy (hal_ml_param_h param)
{
- if (param == NULL) {
+ if (!param) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
hal_ml_param_s *param_s = (hal_ml_param_s *) param;
if (param_s->table) {
- g_hash_table_destroy(param_s->table);
+ g_hash_table_destroy (param_s->table);
}
g_free (param_s);
return HAL_ML_ERROR_NONE;
}
-int hal_ml_param_set(hal_ml_param_h param, const char *key, void *value)
+int
+hal_ml_param_set (hal_ml_param_h param, const char *key, void *value)
{
- if (param == NULL || key == NULL || value == NULL) {
-
+ if (!param || !key || !value) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
hal_ml_param_s *param_s = (hal_ml_param_s *) param;
- hal_ml_param_value_s *param_value = g_new0(hal_ml_param_value_s, 1);
+ hal_ml_param_value_s *param_value = g_new0 (hal_ml_param_value_s, 1);
if (!param_value) {
return HAL_ML_ERROR_OUT_OF_MEMORY;
}
param_value->value = value;
- g_hash_table_insert(param_s->table, g_strdup(key), (gpointer) param_value);
+ g_hash_table_insert (param_s->table, g_strdup (key), (gpointer) param_value);
return HAL_ML_ERROR_NONE;
}
-static int hal_ml_param_get(hal_ml_param_h param, const char *key, void **value)
+static int
+hal_ml_param_get (hal_ml_param_h param, const char *key, void **value)
{
- if (param == NULL || key == NULL || value == NULL) {
+ if (!param || !key || !value) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
hal_ml_param_s *param_s = (hal_ml_param_s *) param;
- hal_ml_param_value_s *param_value = (hal_ml_param_value_s *) g_hash_table_lookup(param_s->table, key);
+ hal_ml_param_value_s *param_value
+ = (hal_ml_param_value_s *) g_hash_table_lookup (param_s->table, key);
if (!param_value) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
*value = param_value->value;
-
return HAL_ML_ERROR_NONE;
}
-int hal_ml_create (const char *backend_name, hal_ml_h *handle)
+int
+hal_ml_create (const char *backend_name, hal_ml_h *handle)
{
int i;
- /* scan backend only once */
+ /* Scan backend only once */
static int scanned = 1;
- if (scanned == 1)
- scanned = hal_ml_scan_backends();
+ if (scanned == 1) {
+ scanned = hal_ml_scan_backends ();
+ }
- _I("init backend %s", backend_name);
+ _I ("Initializing backend %s", backend_name);
- /* find matched backend */
+ /* Find matched backend */
for (i = 0; i < hal_ml_backend_count; i++) {
if (g_strrstr (hal_ml_backend_names[i], backend_name) != NULL) {
- hal_ml_s *new_handle = NULL;
- new_handle = g_new0 (hal_ml_s, 1);
+ hal_ml_s *new_handle = g_new0 (hal_ml_s, 1);
if (!new_handle) {
return HAL_ML_ERROR_OUT_OF_MEMORY;
}
new_handle->backend_library_name = hal_ml_backend_names[i];
- /* fill functions pointers from backend */
- int ret = hal_common_get_backend_with_library_name_v2(
- HAL_MODULE_ML,
- (void **) &new_handle->funcs,
- NULL,
- hal_ml_create_backend,
+ /* Fill function pointers from backend */
+ int ret = hal_common_get_backend_with_library_name_v2 (HAL_MODULE_ML,
+ (void **) &new_handle->funcs, NULL, hal_ml_create_backend,
new_handle->backend_library_name);
if (ret != 0 || !new_handle->funcs || !new_handle->funcs->init) {
- _E("Failed to get backend");
- g_free(new_handle);
+ _E ("Failed to get backend");
+ g_free (new_handle);
return HAL_ML_ERROR_RUNTIME_ERROR;
}
- /* init backend */
+ /* Initialize backend */
ret = new_handle->funcs->init (&new_handle->backend_private);
if (ret != 0) {
- _E("Failed to initialize backend.");
- hal_common_put_backend_with_library_name_v2(
- HAL_MODULE_ML,
- (void *) new_handle->funcs,
- NULL,
- hal_ml_exit_backend,
- new_handle->backend_library_name);
+ _E ("Failed to initialize backend.");
+ hal_common_put_backend_with_library_name_v2 (HAL_MODULE_ML,
+ (void *) new_handle->funcs, NULL, hal_ml_exit_backend,
+ new_handle->backend_library_name);
return HAL_ML_ERROR_RUNTIME_ERROR;
}
+ _I ("Backend initialized successfully with %s", new_handle->backend_library_name);
*handle = (hal_ml_h) new_handle;
return ret;
}
}
- _E("No backend matched with %s", backend_name);
+ _E ("No backend matched with %s", backend_name);
return HAL_ML_ERROR_INVALID_PARAMETER;
}
-int hal_ml_destroy (hal_ml_h handle)
+int
+hal_ml_destroy (hal_ml_h handle)
{
hal_ml_s *ml = (hal_ml_s *) handle;
+
+ _I ("Deinitializing backend %s", ml->backend_library_name);
+
int ret = ml->funcs->deinit (ml->backend_private);
- ret = hal_common_put_backend_with_library_name_v2(
- HAL_MODULE_ML,
- (void *) ml->funcs,
- NULL,
- hal_ml_exit_backend,
- ml->backend_library_name);
+ ret = hal_common_put_backend_with_library_name_v2 (HAL_MODULE_ML,
+ (void *) ml->funcs, NULL, hal_ml_exit_backend, ml->backend_library_name);
g_free (ml);
- if (ret != 0)
+ if (ret != 0) {
return HAL_ML_ERROR_RUNTIME_ERROR;
+ }
return HAL_ML_ERROR_NONE;
}
-static int _hal_ml_configure_instance(hal_ml_h handle, hal_ml_param_h param) {
+static int
+_hal_ml_configure_instance (hal_ml_h handle, hal_ml_param_h param)
+{
if (!handle || !param) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
hal_ml_s *ml = (hal_ml_s *) handle;
const void *prop;
- hal_ml_param_get(param, "properties", (void **)&prop);
- return ml->funcs->configure_instance(ml->backend_private, prop);
+ hal_ml_param_get (param, "properties", (void **) &prop);
+ return ml->funcs->configure_instance (ml->backend_private, prop);
}
-static int _hal_ml_invoke(hal_ml_h handle, hal_ml_param_h param) {
+static int
+_hal_ml_invoke (hal_ml_h handle, hal_ml_param_h param)
+{
if (!handle || !param) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
hal_ml_s *ml = (hal_ml_s *) handle;
const void *input;
void *output;
- hal_ml_param_get(param, "input", (void **)&input);
- hal_ml_param_get(param, "output", (void **)&output);
- return ml->funcs->invoke(ml->backend_private, input, output);
+ hal_ml_param_get (param, "input", (void **) &input);
+ hal_ml_param_get (param, "output", (void **) &output);
+ return ml->funcs->invoke (ml->backend_private, input, output);
}
-static int _hal_ml_invoke_dynamic(hal_ml_h handle, hal_ml_param_h param) {
+static int
+_hal_ml_invoke_dynamic (hal_ml_h handle, hal_ml_param_h param)
+{
if (!handle || !param) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
void *prop;
const void *input;
void *output;
- hal_ml_param_get(param, "properties", (void **)&prop);
- hal_ml_param_get(param, "input", (void **)&input);
- hal_ml_param_get(param, "output", (void **)&output);
- return ml->funcs->invoke_dynamic(ml->backend_private, prop, input, output);
+ hal_ml_param_get (param, "properties", (void **) &prop);
+ hal_ml_param_get (param, "input", (void **) &input);
+ hal_ml_param_get (param, "output", (void **) &output);
+ return ml->funcs->invoke_dynamic (ml->backend_private, prop, input, output);
}
-static int _hal_ml_get_framework_info(hal_ml_h handle, hal_ml_param_h param) {
+static int
+_hal_ml_get_framework_info (hal_ml_h handle, hal_ml_param_h param)
+{
if (!handle || !param) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
hal_ml_s *ml = (hal_ml_s *) handle;
void *framework_info;
- hal_ml_param_get(param, "framework_info", (void **)&framework_info);
- return ml->funcs->get_framework_info(ml->backend_private, framework_info);
+ hal_ml_param_get (param, "framework_info", (void **) &framework_info);
+ return ml->funcs->get_framework_info (ml->backend_private, framework_info);
}
-static int _hal_ml_get_model_info(hal_ml_h handle, hal_ml_param_h param) {
+static int
+_hal_ml_get_model_info (hal_ml_h handle, hal_ml_param_h param)
+{
if (!handle || !param) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
int *model_info_ops;
void *in_info;
void *out_info;
- hal_ml_param_get(param, "ops", (void **)&model_info_ops);
- hal_ml_param_get(param, "in_info", (void **)&in_info);
- hal_ml_param_get(param, "out_info", (void **)&out_info);
- return ml->funcs->get_model_info(ml->backend_private, *model_info_ops, in_info, out_info);
+ hal_ml_param_get (param, "ops", (void **) &model_info_ops);
+ hal_ml_param_get (param, "in_info", (void **) &in_info);
+ hal_ml_param_get (param, "out_info", (void **) &out_info);
+ return ml->funcs->get_model_info (ml->backend_private, *model_info_ops, in_info, out_info);
}
-static int _hal_ml_event_handler(hal_ml_h handle, hal_ml_param_h param) {
+static int
+_hal_ml_event_handler (hal_ml_h handle, hal_ml_param_h param)
+{
if (!handle || !param) {
return HAL_ML_ERROR_INVALID_PARAMETER;
}
hal_ml_s *ml = (hal_ml_s *) handle;
int *event_ops;
void *data;
- hal_ml_param_get(param, "ops", (void **)&event_ops);
- hal_ml_param_get(param, "data", (void **)&data);
- return ml->funcs->event_handler(ml->backend_private, *event_ops, data);
+ hal_ml_param_get (param, "ops", (void **) &event_ops);
+ hal_ml_param_get (param, "data", (void **) &data);
+ return ml->funcs->event_handler (ml->backend_private, *event_ops, data);
}
-int hal_ml_request(hal_ml_h handle, const char *request_name, hal_ml_param_h param)
+int
+hal_ml_request (hal_ml_h handle, const char *request_name, hal_ml_param_h param)
{
- hal_ml_s *ml = (hal_ml_s *) handle;
-
- if (g_ascii_strcasecmp(request_name, "configure_instance") == 0)
- return _hal_ml_configure_instance(handle, param);
+ if (g_ascii_strcasecmp (request_name, "configure_instance") == 0)
+ return _hal_ml_configure_instance (handle, param);
- if (g_ascii_strcasecmp(request_name, "invoke") == 0)
- return _hal_ml_invoke(handle, param);
+ if (g_ascii_strcasecmp (request_name, "invoke") == 0)
+ return _hal_ml_invoke (handle, param);
- if (g_ascii_strcasecmp(request_name, "invoke_dynamic") == 0)
- return _hal_ml_invoke_dynamic(handle, param);
+ if (g_ascii_strcasecmp (request_name, "invoke_dynamic") == 0)
+ return _hal_ml_invoke_dynamic (handle, param);
- if (g_ascii_strcasecmp(request_name, "get_framework_info") == 0)
- return _hal_ml_get_framework_info(handle, param);
+ if (g_ascii_strcasecmp (request_name, "get_framework_info") == 0)
+ return _hal_ml_get_framework_info (handle, param);
- if (g_ascii_strcasecmp(request_name, "get_model_info") == 0)
- return _hal_ml_get_model_info(handle, param);
+ if (g_ascii_strcasecmp (request_name, "get_model_info") == 0)
+ return _hal_ml_get_model_info (handle, param);
- if (g_ascii_strcasecmp(request_name, "eventHandler") == 0)
- return _hal_ml_event_handler(handle, param);
+ if (g_ascii_strcasecmp (request_name, "eventHandler") == 0)
+ return _hal_ml_event_handler (handle, param);
- _E("Invalid request name %s", request_name);
+ _E ("Invalid request name %s", request_name);
return HAL_ML_ERROR_INVALID_PARAMETER;
}
-int hal_ml_request_invoke(hal_ml_h handle, const void *input, void *output)
+int
+hal_ml_request_invoke (hal_ml_h handle, const void *input, void *output)
{
hal_ml_s *ml = (hal_ml_s *) handle;
-
- return ml->funcs->invoke(ml->backend_private, input, output);
+ return ml->funcs->invoke (ml->backend_private, input, output);
}
projects / platform / hal / api / ml.git / commitdiff