[Single] Add input and output ranks

This patch adds input and output ranks to ml_single.
Input and output ranks are used when user explicitly sets rank property.

Signed-off-by: Yelin Jeong <yelini.jeong@samsung.com>

diff --git a/c/src/ml-api-common.c b/c/src/ml-api-common.c
index 68d79d7..c972039 100644 (file)
--- a/c/src/ml-api-common.c
+++ b/c/src/ml-api-common.c
@@ -96,6 +96,25 @@ _ml_tensors_info_initialize (ml_tensors_info_s * info)
 }
 
 /**
+ * @brief Initializes the tensors information with default value.
+ */
+int
+_ml_tensors_rank_initialize (guint * rank)
+{
+  guint i;
+
+  if (!rank)
+    _ml_error_report_return (ML_ERROR_INVALID_PARAMETER,
+        "The parameter, rank, is NULL. Provide a valid pointer.");
+
+  for (i = 0; i < ML_TENSOR_SIZE_LIMIT; i++) {
+    rank[i] = 0;
+  }
+
+  return ML_ERROR_NONE;
+}
+
+/**
  * @brief Compares the given tensor info.
  */
 static gboolean

diff --git a/c/src/ml-api-inference-single.c b/c/src/ml-api-inference-single.c
index c12db10..804bd31 100644 (file)
--- a/c/src/ml-api-inference-single.c
+++ b/c/src/ml-api-inference-single.c
@@ -134,6 +134,10 @@ typedef struct
   ml_tensors_data_s in_tensors;    /**< input tensor wrapper for processing */
   ml_tensors_data_s out_tensors;   /**< output tensor wrapper for processing */
 
+  /** @todo Use only ml_tensor_info_s dimension instead of saving ranks value */
+  guint input_ranks[ML_TENSOR_SIZE_LIMIT];   /**< the rank list of input tensors, it is calculated based on the dimension string. */
+  guint output_ranks[ML_TENSOR_SIZE_LIMIT];  /**< the rank list of output tensors, it is calculated based on the dimension string. */
+
   GList *destroy_data_list;         /**< data to be freed by filter */
 } ml_single;
 
@@ -818,6 +822,8 @@ ml_single_create_handle (ml_nnfw_type_e nnfw)
 
   _ml_tensors_info_initialize (&single_h->in_info);
   _ml_tensors_info_initialize (&single_h->out_info);
+  _ml_tensors_rank_initialize (single_h->input_ranks);
+  _ml_tensors_rank_initialize (single_h->output_ranks);
   g_mutex_init (&single_h->mutex);
   g_cond_init (&single_h->cond);
 
@@ -1745,8 +1751,33 @@ ml_single_set_property (ml_single_h single, const char *name, const char *value)
       num = gst_tensors_info_parse_types_string (&gst_info, value);
     else if (g_str_has_suffix (name, "name"))
       num = gst_tensors_info_parse_names_string (&gst_info, value);
-    else
-      num = gst_tensors_info_parse_dimensions_string (&gst_info, value);
+    else {
+      guint *rank;
+      gchar **str_dims;
+      guint i;
+
+      if (is_input) {
+        rank = single_h->input_ranks;
+      } else {
+        rank = single_h->output_ranks;
+      }
+
+      str_dims = g_strsplit_set (value, ",.", -1);
+      num = g_strv_length (str_dims);
+
+      if (num > ML_TENSOR_SIZE_LIMIT) {
+        _ml_error_report ("Invalid param, dimensions (%d) max (%d)\n",
+            num, ML_TENSOR_SIZE_LIMIT);
+
+        num = ML_TENSOR_SIZE_LIMIT;
+      }
+
+      for (i = 0; i < num; ++i) {
+        rank[i] = gst_tensor_parse_dimension (str_dims[i],
+            gst_info.info[i].dimension);
+      }
+      g_strfreev (str_dims);
+    }
 
     if (num == gst_info.num_tensors) {
       ml_tensors_info_h ml_info;
@@ -1808,9 +1839,8 @@ ml_single_get_property (ml_single_h single, const char *name, char **value)
 
   ML_SINGLE_GET_VALID_HANDLE_LOCKED (single_h, single, 0);
 
-  if (g_str_equal (name, "input") || g_str_equal (name, "inputtype") ||
-      g_str_equal (name, "inputname") || g_str_equal (name, "inputlayout") ||
-      g_str_equal (name, "output") || g_str_equal (name, "outputtype") ||
+  if (g_str_equal (name, "inputtype") || g_str_equal (name, "inputname") ||
+      g_str_equal (name, "inputlayout") || g_str_equal (name, "outputtype") ||
       g_str_equal (name, "outputname") || g_str_equal (name, "outputlayout") ||
       g_str_equal (name, "accelerator") || g_str_equal (name, "custom")) {
     /* string */
@@ -1821,6 +1851,40 @@ ml_single_get_property (ml_single_h single, const char *name, char **value)
     /* boolean */
     g_object_get (G_OBJECT (single_h->filter), name, &bool_value, NULL);
     *value = (bool_value) ? g_strdup ("true") : g_strdup ("false");
+  } else if (g_str_equal (name, "input") || g_str_equal (name, "output")) {
+    gchar *dim_str = NULL;
+    const guint *rank;
+    gboolean is_input = g_str_has_prefix (name, "input");
+    GstTensorsInfo gst_info;
+
+    if (is_input) {
+      rank = single_h->input_ranks;
+    } else {
+      rank = single_h->output_ranks;
+    }
+
+    ml_single_get_gst_info (single_h, is_input, &gst_info);
+
+    if (gst_info.num_tensors > 0) {
+      guint i;
+      GString *dimensions = g_string_new (NULL);
+
+      for (i = 0; i < gst_info.num_tensors; ++i) {
+        dim_str =
+            gst_tensor_get_rank_dimension_string (gst_info.info[i].dimension,
+            *(rank + i));
+        g_string_append (dimensions, dim_str);
+
+        if (i < gst_info.num_tensors - 1) {
+          g_string_append (dimensions, ",");
+        }
+        g_free (dim_str);
+      }
+      dim_str = g_string_free (dimensions, FALSE);
+    } else {
+      dim_str = g_strdup ("");
+    }
+    *value = dim_str;
   } else {
     _ml_error_report
         ("The property key, '%s', is not available for get_property and not recognized by the API. It should be one of {input, inputtype, inputname, inputlayout, output, outputtype, outputname, outputlayout, accelerator, custom, is-updatable}.",

diff --git a/c/src/ml-api-internal.h b/c/src/ml-api-internal.h
index e9a6627..2c012e9 100644 (file)
--- a/c/src/ml-api-internal.h
+++ b/c/src/ml-api-internal.h
@@ -268,6 +268,16 @@ size_t _ml_tensor_info_get_size (const ml_tensor_info_s *info);
 int _ml_tensors_info_initialize (ml_tensors_info_s *info);
 
 /**
+ * @brief Initializes the rank information with default value.
+ * @since_tizen 7.5
+ * @param[in] info The rank array pointer to be initialized.
+ * @return @c 0 on success. Otherwise a negative error value.
+ * @retval #ML_ERROR_NONE Successful
+ * @retval #ML_ERROR_INVALID_PARAMETER Given parameter is invalid.
+ */
+int _ml_tensors_rank_initialize (guint *rank);
+
+/**
  * @brief Frees and initialize the data in tensors info.
  * @since_tizen 5.5
  * @param[in] info The tensors info pointer to be freed.

diff --git a/tests/capi/unittest_capi_inference_single.cc b/tests/capi/unittest_capi_inference_single.cc
index ad0fd5e..a52a84f 100644 (file)
--- a/tests/capi/unittest_capi_inference_single.cc
+++ b/tests/capi/unittest_capi_inference_single.cc
@@ -2044,6 +2044,113 @@ skip_test:
   g_free (test_model);
 }
 
+/**
+ * @brief Test for input_ranks and output_ranks property of the ml_single
+ * @details Given dimension string, check its value.
+ */
+TEST (nnstreamer_capi_singleshot, property_05_p)
+{
+  ml_single_h single;
+  char *prop_value;
+  int status;
+
+  const gchar *root_path = g_getenv ("MLAPI_SOURCE_ROOT_PATH");
+  gchar *test_model;
+
+  /* supposed to run test in build directory */
+  if (root_path == NULL)
+    root_path = "..";
+
+  /** add.tflite adds value 2 to all the values in the input */
+  test_model = g_build_filename (
+      root_path, "tests", "test_models", "models", "add.tflite", NULL);
+  ASSERT_TRUE (g_file_test (test_model, G_FILE_TEST_EXISTS));
+
+  status = ml_single_open (&single, test_model, NULL, NULL,
+      ML_NNFW_TYPE_TENSORFLOW_LITE, ML_NNFW_HW_ANY);
+  if (is_enabled_tensorflow_lite) {
+    EXPECT_EQ (status, ML_ERROR_NONE);
+  } else {
+    EXPECT_NE (status, ML_ERROR_NONE);
+    goto skip_test;
+  }
+
+  status = ml_single_set_property (single, "input", "5:1:1:1");
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  status = ml_single_get_property (single, "input", &prop_value);
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  EXPECT_STREQ (prop_value, "5:1:1:1");
+  g_free (prop_value);
+
+  status = ml_single_set_property (single, "input", "5:1:1");
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  status = ml_single_get_property (single, "input", &prop_value);
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  EXPECT_STREQ (prop_value, "5:1:1");
+  g_free (prop_value);
+
+  status = ml_single_set_property (single, "input", "5:1");
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  status = ml_single_get_property (single, "input", &prop_value);
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  EXPECT_STREQ (prop_value, "5:1");
+  g_free (prop_value);
+
+  status = ml_single_set_property (single, "input", "5");
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  status = ml_single_get_property (single, "input", &prop_value);
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  EXPECT_STREQ (prop_value, "5");
+  g_free (prop_value);
+
+  status = ml_single_set_property (single, "output", "5:1:1:1");
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  status = ml_single_get_property (single, "output", &prop_value);
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  EXPECT_STREQ (prop_value, "5:1:1:1");
+  g_free (prop_value);
+
+  status = ml_single_set_property (single, "output", "5:1:1");
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  status = ml_single_get_property (single, "output", &prop_value);
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  EXPECT_STREQ (prop_value, "5:1:1");
+  g_free (prop_value);
+
+  status = ml_single_set_property (single, "output", "5:1");
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  status = ml_single_get_property (single, "output", &prop_value);
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  EXPECT_STREQ (prop_value, "5:1");
+  g_free (prop_value);
+
+  status = ml_single_set_property (single, "output", "5");
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  status = ml_single_get_property (single, "output", &prop_value);
+  EXPECT_EQ (status, ML_ERROR_NONE);
+
+  EXPECT_STREQ (prop_value, "5");
+  g_free (prop_value);
+
+skip_test:
+  g_free (test_model);
+}
+
 #ifdef ENABLE_NNFW_RUNTIME
 /**
  * @brief Test NNStreamer single shot (nnfw backend)