Version v2.0.0

Author: RTISTM

Binary/network_data.hex

@@ -1,6 +1,6 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<workspace>
-    <project>
-        <path>$WS_DIR$\Project.ewp</path>
-    </project>
-</workspace>
+<?xml version="1.0" encoding="UTF-8"?>
+<workspace>
+    <project>
+        <path>$WS_DIR$\x-cube-n6-ai-power-measurement.ewp</path>
+    </project>
+</workspace>

Binary/x-cube-n6-ai-power-measurement.hex

@@ -1,6 +1,6 @@
  /**
  ******************************************************************************
- * @file    cam_vd55g1.h
+ * @file    app_cam.h
  * @author  GPM Application Team
  *
  ******************************************************************************
@@ -15,8 +15,8 @@
  *
  ******************************************************************************
  */
-#ifndef CAM_CAMERA
-#define CAM_CAMERA
+#ifndef APP_CAM_H
+#define APP_CAM_H
 
 #define CAMERA_FPS 30
 
@@ -30,4 +30,4 @@ void CAM_IspUpdate(void);
 void CAM_Sensor_Start(void);
 void CAM_Sensor_Stop(void);
 
-#endif
+#endif /* APP_CAM_H */

EWARM/Project.ewd renamed to EWARM/x-cube-n6-ai-power-measurement.ewd

@@ -76,14 +76,14 @@
 "person",   "not_person"}
 
 /* I/O configuration */
-#define AI_OBJDETECT_YOLOV2_PP_NB_CLASSES        (1)
-#define AI_OBJDETECT_YOLOV2_PP_NB_ANCHORS        (5)
-#define AI_OBJDETECT_YOLOV2_PP_GRID_WIDTH        (7)
-#define AI_OBJDETECT_YOLOV2_PP_GRID_HEIGHT       (7)
-#define AI_OBJDETECT_YOLOV2_PP_NB_INPUT_BOXES    (AI_OBJDETECT_YOLOV2_PP_GRID_WIDTH * AI_OBJDETECT_YOLOV2_PP_GRID_HEIGHT)
+#define AI_OD_YOLOV2_PP_NB_CLASSES        (1)
+#define AI_OD_YOLOV2_PP_NB_ANCHORS        (5)
+#define AI_OD_YOLOV2_PP_GRID_WIDTH        (7)
+#define AI_OD_YOLOV2_PP_GRID_HEIGHT       (7)
+#define AI_OD_YOLOV2_PP_NB_INPUT_BOXES    (AI_OD_YOLOV2_PP_GRID_WIDTH * AI_OD_YOLOV2_PP_GRID_HEIGHT)
 
 /* Anchor boxes */
-static const float32_t AI_OBJDETECT_YOLOV2_PP_ANCHORS[2*AI_OBJDETECT_YOLOV2_PP_NB_ANCHORS] = {
+static const float32_t AI_OD_YOLOV2_PP_ANCHORS[2*AI_OD_YOLOV2_PP_NB_ANCHORS] = {
     0.9883000000f,     3.3606000000f,
     2.1194000000f,     5.3759000000f,
     3.0520000000f,     9.1336000000f,
@@ -92,8 +92,8 @@ static const float32_t AI_OBJDETECT_YOLOV2_PP_ANCHORS[2*AI_OBJDETECT_YOLOV2_PP_N
   };
 
 /* --------  Tuning below can be modified by the application --------- */
-#define AI_OBJDETECT_YOLOV2_PP_CONF_THRESHOLD    (0.6f)
-#define AI_OBJDETECT_YOLOV2_PP_IOU_THRESHOLD     (0.3f)
-#define AI_OBJDETECT_YOLOV2_PP_MAX_BOXES_LIMIT   (10)
+#define AI_OD_YOLOV2_PP_CONF_THRESHOLD    (0.6f)
+#define AI_OD_YOLOV2_PP_IOU_THRESHOLD     (0.3f)
+#define AI_OD_YOLOV2_PP_MAX_BOXES_LIMIT   (10)
 
 #endif

EWARM/Project.ewp renamed to EWARM/x-cube-n6-ai-power-measurement.ewp

@@ -0,0 +1,20 @@
+| Component                                     | Copyright                                 | License                                       |
+|:---------                                     |:----------                                |:-------                                       |
+| CMSIS                                         | Arm Limited                               | Apache-2.0                                    |
+| STM32N6xx CMSIS                               | ARM Limited, STMicroelectronics           | Apache-2.0                                    |
+| STM32N6xx HAL/LL Drivers                      | STMicroelectronics                        | BSD-3-Clause                                  |
+| STM32N6570-DK BSP Drivers                     | STMicroelectronics                        | BSD-3-Clause                                  |
+| BSP Component aps256xx                        | STMicroelectronics                        | BSD-3-Clause                                  |
+| BSP Component Common                          | STMicroelectronics                        | BSD-3-Clause                                  |
+| BSP Component mx66uw1g45g                     | STMicroelectronics                        | BSD-3-Clause                                  |
+| BSP Component rk050hr18                       | STMicroelectronics                        | BSD-3-Clause                                  |
+| Inc                                           | STMicroelectronics                        | SLA0044                                       |
+| Src                                           | STMicroelectronics                        | SLA0044                                       |
+| Lib/Camera_Middleware                         | STMicroelectronics                        | SLA0044                                       |
+| Lib/Camera_Middleware/sensors                 | STMicroelectronics                        | BSD-3-Clause                                  |
+| Lib/Camera_Middleware/ISP_Library             | LACROIX - Impulse, STMicroelectronics     | SLA0044                                       |
+| Lib/lib_vision_models_pp/lib_vision_models_pp | STMicroelectronics                        | SLA0044                                       |
+| Lib/AI_Runtime                                | STMicroelectronics                        | SLA0044                                       |
+| Binary                                        | STMicroelectronics                        | SLA0044                                       |
+| Model                                         | STMicroelectronics                        | SLA0044                                       |
+| Utilities/pwr_scripts                         | STMicroelectronics                        | SLA0044                                       |

EWARM/Project.eww renamed to EWARM/x-cube-n6-ai-power-measurement.eww

@@ -564,6 +564,9 @@ AI_INTERFACE_ENTRY void ai_not_equal_buffer_u8(ai_handle out, const ai_handle a,
 AI_INTERFACE_ENTRY void ai_squared_diff(ai_handle out, const ai_handle a, const ai_handle b);
 AI_INTERFACE_ENTRY void ai_squared_diff_buffer(ai_handle out, const ai_handle a, const ai_handle b, const ai_size loop);
 
+AI_INTERFACE_ENTRY void ai_atan2(ai_handle out, const ai_handle a, const ai_handle b);
+AI_INTERFACE_ENTRY void ai_atan2_buffer(ai_handle out, const ai_handle a, const ai_handle b, const ai_size loop);
+
 AI_API_DECLARE_END
 
 #endif /* AI_MATH_HELPERS_H */

Inc/app_cam.h

@@ -830,6 +830,16 @@ typedef struct ai_network_report_ {
   ai_signature                    signature;
 } ai_network_report;
 
+/*!
+ * @enum ai_scatter_nd_reduction
+ * @ingroup ai_platform
+ * @brief reduction operation in scatter_nd layer
+ */
+typedef enum {
+  AI_SCATTER_ND_NONE = 0x0,
+  AI_SCATTER_ND_OPERATION
+} ai_scatter_nd_reduction;
+
 /*!
  * @enum ai_upsample_mode
  * @ingroup ai_platform

Inc/app_config.h

@@ -72,6 +72,12 @@ ai_bool core_timer_start(core_timer* t);
 AI_INTERNAL_API
 core_timer_ts core_timer_stop(core_timer* t);
 
+#if defined HAS_STM32
+void core_timer_init_(void);
+core_timer_ts core_timer_ticks_start_(void);
+core_timer_ts core_timer_ticks_(void);
+#endif
+
 AI_API_DECLARE_END
 
 #endif    /*CORE_BENCHMARK_H_*/

Inc/postprocess_conf.h

@@ -33,7 +33,7 @@
 #define AI_PLATFORM_RUNTIME_MAJOR    (10)
 #endif
 #ifndef AI_PLATFORM_RUNTIME_MINOR
-#define AI_PLATFORM_RUNTIME_MINOR    (0)
+#define AI_PLATFORM_RUNTIME_MINOR    (1)
 #endif
 #ifndef AI_PLATFORM_RUNTIME_MICRO
 #define AI_PLATFORM_RUNTIME_MICRO    (0)

LICENSE.md

@@ -219,7 +219,6 @@ const char* ai_layer_type_name(const ai_layer_type type);
 AI_INTERNAL_API
 ai_bool ai_layer_type_is_valid(const ai_layer_type type);
 
-#ifdef HAS_AI_ASSERT
 
 /*!
  * @brief check scratch size computed with actual scratch buffer size
@@ -248,6 +247,7 @@ void ai_layer_check_scratch_size( ai_layer_type layer_type, ai_array_format fmt,
                           ai_bool is_sssa, ai_u32 tensor_scratch_size_bytes,
                           const char *p_function_name, const int line_nb);
 
+#ifdef HAS_AI_ASSERT
 #define CHECK_SCRATCH_BUFFER_SIZE( layer_type, fmt, \
                                    filt_width, filt_height, \
                                    n_channel_in, n_channel_out, \
@@ -261,7 +261,15 @@ void ai_layer_check_scratch_size( ai_layer_type layer_type, ai_array_format fmt,
                                  is_depthwise, is_ch1st, is_ch_wise, \
                                  is_sssa_ch, tensor_scratch_size_bytes, \
                                  __FUNCTION__, __LINE__);
-
+#else
+#define CHECK_SCRATCH_BUFFER_SIZE( layer_type, fmt, \
+                                   filt_width, filt_height, \
+                                   n_channel_in, n_channel_out, \
+                                   is_pointwise, is_rgb, \
+                                   is_depthwise, is_ch1st, is_ch_wise, \
+                                   is_sssa_ch, tensor_scratch_size_bytes) \
+	AI_WRAP_FUNC(/*NULL*/)
+#endif
 #define IS_PW 1
 #define IS_RGB 1
 #define IS_DW 1
@@ -276,7 +284,6 @@ void ai_layer_check_scratch_size( ai_layer_type layer_type, ai_array_format fmt,
 #define NOT_CH_WISE 0
 #define NOT_SSSA_CH 0
 
-#endif
 
 AI_API_DECLARE_END
 

Lib/AI_Runtime/Inc/ai_math_helpers.h

@@ -253,6 +253,14 @@ void forward_conv2d_is8os8ws8_sssa_ch(ai_layer *pLayer);
 AI_INTERNAL_API
 void forward_dw_sssa8_ch(ai_layer *pLayer);
 
+/*!
+ * @brief Computes the activations of a int8 quantized DW layer
+ *        for SSSA per channel quantized scheme Optimized for HSP
+ * @ingroup layers_conv2d
+ * @param layer the convolutional (conv) layer
+ */
+AI_INTERNAL_API
+void forward_dw_hsp_1step_sssa8_ch(ai_layer *pLayer);
 
 /*!
  * @brief Computes the activations of a int8 quantized DW layer
@@ -263,6 +271,15 @@ void forward_dw_sssa8_ch(ai_layer *pLayer);
 AI_INTERNAL_API
 void forward_dw_hsp_sssa8_ch(ai_layer *pLayer);
 
+/*!
+ * @brief Computes the activations of a int8 quantized DW layer
+ *        for SSSA per channel quantized scheme Optimized for HSP
+ * @ingroup layers_conv2d
+ * @param layer the convolutional (conv) layer
+ */
+AI_INTERNAL_API
+void forward_dw_hsp_3step_sssa8_ch(ai_layer *pLayer);
+
 /*!
  * @brief Computes the activations of a int8 quantized DW layer
  *        for SSSA per channel quantized scheme, with 3x3 kernels
@@ -328,6 +345,26 @@ void forward_pw_sssa8_ch(ai_layer *pLayer);
 AI_INTERNAL_API
 void forward_pw_hsp_sssa8_ch(ai_layer *pLayer);
 
+/*!
+ * @brief Computes the activations of a int8 quantized PW layer
+ *        for SSSA per channel quantized scheme. Optimized for HSP
+ *        1Step version (nb input channel <= 4)
+ * @ingroup layers_conv2d
+ * @param layer the convolutional (conv) layer
+ */
+AI_INTERNAL_API
+void forward_pw_hsp_1step_sssa8_ch(ai_layer *pLayer);
+
+/*!
+ * @brief Computes the activations of a int8 quantized PW layer
+ *        for SSSA per channel quantized scheme. Optimized for HSP
+ *        3 Step variant
+ * @ingroup layers_conv2d
+ * @param layer the convolutional (conv) layer
+ */
+AI_INTERNAL_API
+void forward_pw_hsp_3step_sssa8_ch(ai_layer *pLayer);
+
 /*!
  * @brief Computes the activations of a int8 quantized dilated Conv2d layer
  *        for SSSA per channel quantized scheme (valid padding)
@@ -373,9 +410,15 @@ void forward_conv2d_sssa8_ch(ai_layer *pLayer);
  * @ingroup layers_conv2d
  * @param layer the convolutional (conv) layer
  */
+AI_INTERNAL_API
+void forward_conv2d_hsp_1step_sssa8_ch(ai_layer *pLayer);
+
 AI_INTERNAL_API
 void forward_conv2d_hsp_sssa8_ch(ai_layer *pLayer);
 
+AI_INTERNAL_API
+void forward_conv2d_hsp_3step_sssa8_ch(ai_layer *pLayer);
+
 
 /*!
  * @brief Computes the activations of a int8 quantized Conv2d layer
@@ -639,6 +682,15 @@ void forward_dense_integer_SSSA(ai_layer *pLayer);
 AI_INTERNAL_API
 void forward_dense_hsp_sssa8(ai_layer *pLayer);
 
+/*!
+ * @brief Computes the activations of a integer dense (fully connected) layer
+ *        for SSSA per layer quantized scheme Optimized for HSP, 3Step loop (out_ch)
+ * @ingroup layers_dense
+ * @param layer the dense layer
+ */
+AI_INTERNAL_API
+void forward_dense_hsp_3step_sssa8(ai_layer *pLayer);
+
 /*!
  * @brief Computes the activations of a integer dense (fully connected) layer
  *        for SSSA per channel quantized scheme: HSP variant

Lib/AI_Runtime/Inc/ai_platform.h

@@ -140,7 +140,6 @@ typedef AI_ALIGNED_TYPE(struct, 4) ai_layer_tile_{
   AI_CONST ai_array* repeats;  /*!< numbers of repeated copies along each dimension */
 } ai_layer_tile;
 
-
 /*!
  * @struct ai_layer_shape
  * @ingroup layers generic
@@ -368,6 +367,22 @@ typedef AI_ALIGNED_TYPE(struct, 4) ai_layer_eltwise_integer_ {
   func_buffer_binary_integer buffer_operation; /*!< operation to apply elementwise */
 } ai_layer_eltwise_integer;
 
+/*!
+ * @struct ai_layer_scatter_nd
+ * @ingroup layers_generic
+ * @brief ScatterND layer definition
+ *
+ * This layer defines the params of a scattering layer (ND). It is intended to be used
+ * by his associated forward function @ref forward_scatter_nd
+ */
+typedef AI_ALIGNED_TYPE(struct, 4) ai_layer_scatter_nd_ {
+  AI_LAYER_COMMON_FIELDS_DECLARE
+  ai_tensor* indices;  /*!< Indices of corrisponding slices of inputs*/
+  ai_tensor* updates;  /*!< Updates of corrisponding slices of inputs*/
+  func_binary operation;    /*!< operation to apply elementwise */
+  ai_scatter_nd_reduction reduction; /*!< Reduction operation in ScatterND layer*/
+} ai_layer_scatter_nd;
+
 /*!
  * @struct ai_layer_reduce
  * @ingroup layers_generic
@@ -589,6 +604,14 @@ void forward_gather_nd(ai_layer* layer);
 AI_INTERNAL_API
 void forward_gather_nd_channel_first(ai_layer* layer);
 
+/*!
+ * @brief ScatterND an input tensor
+ * @ingroup layers_generic
+ * @param layer the scattered layer (ND)
+ */
+AI_INTERNAL_API
+void forward_scatter_nd(ai_layer* layer);
+
 /*!
  * @brief Slice an input tensors
  * @ingroup layers_generic

Lib/AI_Runtime/Inc/core_benchmark.h

@@ -142,6 +142,8 @@ LAYER_ENTRY(POOL_DQNN, LAYER_ID(49), pool_dqnn, forward_maxpool_is1os1, NULL, NU
 LAYER_ENTRY(DENSE_DQNN, LAYER_ID(50), dense_dqnn, forward_dense_is1os1ws1, NULL, NULL)
 /*!< Reverse layer */
 LAYER_ENTRY(REVERSE, LAYER_ID(51), reverse, forward_reverse, NULL, NULL)
+/*!< ScatterND layer */
+LAYER_ENTRY(SCATTER_ND, LAYER_ID(69), scatter_nd, forward_scatter_nd, NULL, NULL)
 
 /*!< TFLite wrapper */
 LAYER_ENTRY(TFLITE_WRAPPER, LAYER_ID(52), tflite_wrapper, NULL, NULL, NULL)