Update internal for 22Q2 release (#432)

* Update internal for 22Q2 release

update to internal commit-id: e96103281b08404cabb9b65306587627cfa3cb93

Signed-off-by: yuenan.li <[email protected]>

* Update prebuilt for 22Q2 release

Signed-off-by: yuenan.li <[email protected]>

Co-authored-by: yuenan.li <[email protected]>

.gitignore

@@ -336,3 +336,4 @@ ASALocalRun/
 # IDE
 .settings/
 build/
+*_build/

prebuilt-sdk/x86_64_linux/VERSION

@@ -1 +1 @@
-REL/6.4.10.2
+6.4.11

prebuilt-sdk/x86_64_linux/include/VX/vx_kernels.h

@@ -499,6 +499,8 @@ enum vx_kernel_e {
 
     VX_KERNEL_NN_DECONV_3D_LAYER = VX_KERNEL_BASE(VX_ID_VIVANTE, VX_LIBRARY_KHR_BASE) + 0x31,
 
+    VX_KERNEL_STREAM_PROCESSOR = VX_KERNEL_BASE(VX_ID_VIVANTE, VX_LIBRARY_KHR_BASE) + 0x32,
+
     VX_KERNEL_MAX_1_2, /*!< \internal Used for VX1.2 bounds checking in the conformance test. */
 };
 

prebuilt-sdk/x86_64_linux/include/VX/vx_khr_compatible.h

@@ -196,4 +196,45 @@ VX_DECONV_3D_API_SUPPORT is used to declare that vsi openvx driver can support d
 #define VX_TENSOR_STRIDE_X_BITS_SUPPORT 1
 #endif
 
+/*
+VX_REMOVE_RESHAPE_SUPPORT is used to declare if graph opt support to remove reshape op, if support, it's not need to remove reshape in ovxlib.
+ 0: not support
+ 1: support
+*/
+/*
+#ifndef VX_REMOVE_RESHAPE_SUPPORT
+#define VX_REMOVE_RESHAPE_SUPPORT 0
+#endif
+*/
+
+/*
+VX_STREAM_PROCESSOR_SUPPORT is used to declare that vsi openvx driver can support vxStreamProcessorNode API
+ [value]
+ 0: not support
+ 1: support
+*/
+#ifndef VX_STREAM_PROCESSOR_SUPPORT
+#define VX_STREAM_PROCESSOR_SUPPORT 0
+#endif
+
+/*
+ VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL is used to declare that this tensor connect to fixed DMA channel.
+ [value]
+ 0: not support
+ 1: support
+*/
+#ifndef VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL
+#define VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL 1
+#endif
+
+/*
+ VX_SCALE_EXTRA_PARAMETER_SUPPORT is used to declare that RESIZE can support align_cornor and half_pixel_center parameter
+ [value]
+ 0: not support
+ 1: support
+*/
+#ifndef VX_SCALE_EXTRA_PARAMETER_SUPPORT
+#define VX_SCALE_EXTRA_PARAMETER_SUPPORT 1
+#endif
+
 #endif /* __VX_KHR_COMPATIBLE_H__ */

prebuilt-sdk/x86_64_linux/include/VX/vx_khr_nn.h

@@ -57,6 +57,12 @@ enum vx_graph_attribute_internal_type_e
     VX_GRAPH_AXI_SRAM_PRE_LOAD                    =  VX_ATTRIBUTE_BASE(VX_ID_VIVANTE, VX_TYPE_GRAPH) + 0x2,
     /*! \brief Queries a graph for its running priority (read-write. Use a <tt>\ref vx_uint32</tt> parameter. */
     VX_GRAPH_PRIORITY_VALUE_VIV                   =  VX_ATTRIBUTE_BASE(VX_ID_VIVANTE, VX_TYPE_GRAPH) + 0x3,
+    VX_GRAPH_PSI_EXTRATOR_PARAMETER               = VX_ATTRIBUTE_BASE(VX_ID_VIVANTE, VX_TYPE_GRAPH) + 0x4,
+    VX_GRAPH_PSI_FILLER_PARAMETER                 = VX_ATTRIBUTE_BASE(VX_ID_VIVANTE, VX_TYPE_GRAPH) + 0x5,
+    VX_GRAPH_DENOISE_POSTPROCESS_PARAMETER        = VX_ATTRIBUTE_BASE(VX_ID_VIVANTE, VX_TYPE_GRAPH) + 0x6,
+    VX_GRAPH_DATA_COMPRESSION_RATIO               = VX_ATTRIBUTE_BASE(VX_ID_VIVANTE, VX_TYPE_GRAPH) + 0x7,
+    VX_GRAPH_ISP_EMULATION_PARAMETER              = VX_ATTRIBUTE_BASE(VX_ID_VIVANTE, VX_TYPE_GRAPH) + 0x8,
+    VX_GRAPH_PROCESS_FPS                          = VX_ATTRIBUTE_BASE(VX_ID_VIVANTE, VX_TYPE_GRAPH) + 0x9,
 };
 
 /*! \brief Size Alignment of User Memory
@@ -209,7 +215,8 @@ enum vx_nn_activation_function_e
     VX_NN_ACTIVATION_LEAKYRELU_MAX_POOLING = VX_ENUM_BASE(VX_ID_VIVANTE, VX_ENUM_NN_ACTIVATION_FUNCTION_TYPE) + 0x4,
     VX_NN_ACTIVATION_SWISH = VX_ENUM_BASE(VX_ID_VIVANTE, VX_ENUM_NN_ACTIVATION_FUNCTION_TYPE) + 0x5,
     VX_NN_ACTIVATION_HSWISH = VX_ENUM_BASE(VX_ID_VIVANTE, VX_ENUM_NN_ACTIVATION_FUNCTION_TYPE) + 0x6,
-    VX_NN_ACTIVATION_NONE = VX_ENUM_BASE(VX_ID_VIVANTE, VX_ENUM_NN_ACTIVATION_FUNCTION_TYPE) + 0x7,
+    VX_NN_ACTIVATION_CUSTOM = VX_ENUM_BASE(VX_ID_VIVANTE, VX_ENUM_NN_ACTIVATION_FUNCTION_TYPE) + 0x7,
+    VX_NN_ACTIVATION_NONE = VX_ENUM_BASE(VX_ID_VIVANTE, VX_ENUM_NN_ACTIVATION_FUNCTION_TYPE) + 0x8,
 };
 
 /*! \brief  The Convolutional network type
@@ -285,6 +292,59 @@ enum vx_tensor_rank_type_e
     VX_TENSOR_RANK_SN,
 };
 
+/*! \brief The attribute of tensor.
+ * \ingroup group_tensor
+ * \version 0.4
+ */
+enum vx_tensor_priority_e
+{
+    /*! \brief no special requirement */
+    VX_TENSOR_DEFAULT = 0,
+
+    /*! \brief  2nd input(reference) */
+    /*VX_TENSOR_2ND_INPUT_FOR       = 1,*/
+    VX_TENSOR_FOR_GRAPH_REFERENCE = 1,
+};
+
+
+/*! \brief The attribute of tensor memory.
+ * \ingroup group_tensor
+ * \version 0.4
+ */
+enum vx_tensor_memory_attribute_e
+{
+    /*! \brief no special requirement */
+    VX_TENSOR_MEMORY_DEFAULT = 0,
+
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_0 = (0x1 << 0),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_1 = (0x1 << 1),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_2 = (0x1 << 2),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_3 = (0x1 << 3),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_4 = (0x1 << 4),
+    /*
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_5 = (0x1 << VX_DMA5_IN_ISP_OCM_PSI),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_6 = (0x1 << VX_DMA6_DDR_DECOMPRESS),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_7 = (0x1 << VX_DMA7_POSTOUT_OCM_ISP),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_8 = (0x1 << VX_DMA8_COMPRESS_DDR),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_9 = (0x1 << VX_DMA9_ISP_PATTERN_GENERATOR),
+    VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL_10 = (0x1 << VX_DMA10_ISP_CHECKSUM_GENERATOR),
+    */
+    /*! \brief DMA transfer data to VIP and enable circular buffer */
+#if !VX_TENSOR_MEMORY_CONNECT_DMA_CHANNEL
+    VX_TENSOR_MEMORY_ENABLE_CIRCULAR_BY_DMA = 0xFFFFFFFF,
+#endif
+};
+
+enum vx_dma_extrator_pad_mode_e
+{
+    /*! \brief no special requirement */
+    VX_DMA_EXTRATOR_PAD_CONST = 0,
+
+    /*! \brief DMA extrator pad with nearest edge */
+    VX_DMA_EXTRATOR_PAD_WITH_NEAREAST_EDGE = 1,
+};
+
+
 /*! \brief The precision of tensor.
  * \ingroup group_tensor
  * \version 0.4
@@ -601,6 +661,19 @@ VX_API_ENTRY vx_tensor VX_API_CALL vxReshapeTensor(vx_tensor tensor, vx_int32* n
  */
 VX_API_ENTRY vx_status VX_API_CALL vxSetTensorAttribute(vx_tensor tensor, vx_enum attribute, const void *ptr, vx_size size);
 
+/*! \brief Creates an opaque reference to a tensor data buffer.
+ * \details The tensor is a dummy tensor which will not allocate any memory. And it cannot reshape or view.
+ * Not guaranteed to exist until the <tt>vx_graph</tt> containing it has been verified.
+ * \param [in] context The reference to the implementation context.
+ * \param [in] number_of_dims The number of dimensions.
+ * \param [in] dims Dimensions sizes in elements.
+ * \param [in] data_format The <tt>\ref vx_type_e</tt> that represents the data format of the tensor data elements.
+ * \return A tensor data reference or zero when an error is encountered.
+ * \ingroup group_tensor
+ * \version 0.3
+ */
+VX_API_ENTRY vx_tensor VX_API_CALL vxCreateDummyTensor(vx_context context, vx_size number_of_dims, const vx_size *dims, vx_enum data_format);
+
 
 /*! \brief The type enumeration lists all NN extension types.
  * \ingroup group_cnn
@@ -1317,6 +1390,13 @@ typedef struct _vx_nn_scale_params_t
     vx_enum type;             /*!< \brief  The interpolation type, only support VX_INTERPOLATION_BILINEAR.  */
 } vx_nn_scale_params_t, * vx_nn_scale_params;
 
+typedef struct _vx_nn_scale_params_ext_t
+{
+    vx_nn_scale_params_t base;
+    vx_bool align_corners;
+    vx_bool half_pixel_centers;
+} vx_nn_scale_params_ext_t, * vx_nn_scale_params_ext;
+
 /*! \brief [Graph] Creates a scale Layer Node.
  * \param [in] graph The reference to the parent graph.
  * \param [in] input The input tensor data to scale.
@@ -2054,8 +2134,15 @@ typedef struct _vx_hardware_caps_params_ext_t
     vx_hardware_caps_params_t base;
     vx_uint32 subGroupSize;        /*!< \brief  shader sub-group size.*/
     vx_bool   supportVA40;         /*!< \brief  support 40bit virtual address.*/
+    vx_uint32 supportStreamProcessor; /*!< \brief  support stream processor.*/
 } vx_hardware_caps_params_ext_t;
 
+typedef struct _vx_hardware_caps_params_ext2_t
+{
+    vx_hardware_caps_params_ext_t base;
+    vx_uint32 streamProcessorExecCount;     /*!< \brief  streamprocess execution count.  */
+} vx_hardware_caps_params_ext2_t;
+
 /*! \brief Queries hardware caps information.
  * \param [in] context The reference to the context.
  * \param [in] hardware_caps_params <tt>\ref vx_hardware_caps_params_t </tt>.

prebuilt-sdk/x86_64_linux/include/VX/vx_khr_nn_internal.h

@@ -219,6 +219,15 @@ typedef struct _vx_nn_convolution_relu_pooling_params_ext4_t
     vx_bool         enable_nn_tensor_add_relu;  /*!< \brief  Enable Relu function after tensor add. */
 } vx_nn_convolution_relu_pooling_params_ext4_t, * vx_nn_convolution_relu_pooling_params_ext4;
 
+typedef struct _vx_nn_convolution_relu_pooling_params_ext5_t
+{
+    vx_nn_convolution_relu_pooling_params_ext4_t ext4;  /*!< \brief convolution relu pooling params <tt>\ref vx_nn_convolution_relu_pooling_params_ext_t</tt> */
+
+    vx_object_array inputs_list;
+    vx_object_array outputs_list;
+    vx_spinst       spinst_obj;
+} vx_nn_convolution_relu_pooling_params_ext5_t, * vx_nn_convolution_relu_pooling_params_ext5;
+
 /*! \brief [Graph] Creates a Convolutional Network Convolution and Activation(Relu) and Pooling Layer Node, this fucntion match kronos NN Extension 1.2 verion.
  * \details This function implement Convolutional Network Convolution and Activation(Relu) and Pooling layer.
  *  For fixed-point data types, a fixed point calculation is performed with round and saturate according to the number of accumulator bits. The number of the accumulator bits are implementation defined,

prebuilt-sdk/x86_64_linux/include/VX/vx_nodes.h

@@ -963,6 +963,40 @@ VX_API_ENTRY vx_node VX_API_CALL vxBatchGemmNode(vx_graph graph,
                                                  vx_scalar trans_c,
                                                  vx_tensor output);
 
+typedef struct _vx_lut_params_s
+{
+    vx_enum         lut_function;     /*!< \brief Set VX_NN_ACTIVATION_NONE to disable lut table or set VX_NN_ACTIVATION_CUSTOM to customize lut table or set others to use fixed lut table */
+    vx_float32      float_values[4];  /*!< \brief Float parameters of fixed lut table */
+    vx_uint32       fvalues_count;    /*!< \brief Count of float_values */
+    vx_int32        int_values[4];    /*!< \brief Int parameters of fixed lut table */
+    vx_uint32       ivalues_count;    /*!< \brief Count of int_values */
+    vx_lut          in_lut;           /*!< \brief Only valid when lut_function is VX_NN_ACTIVATION_CUSTOM */
+    vx_lut          out_lut;          /*!< \brief Only valid when lut_function is VX_NN_ACTIVATION_CUSTOM */
+} vx_lut_params_s, * vx_lut_params;
+
+/*! \brief Create a stream processor node.
+ * \param [in] graph The reference to the graph.
+ * \param [in] input_list The input tensor list.
+ * \param [in] input_count The input tensor count.
+ * \param [in] output_list The output tensor list.
+ * \param [in] output_count The output tensor count.
+ * \param [in] spinst_obj The stream processor instrunction object. Use vxCreateSPINST() to create.
+ * \param [in] lut_params The lut parameters. Refer to vx_lut_params_s.
+ * \return <tt>\ref vx_node</tt>.
+ * \retval vx_node A node reference. Any possible errors preventing a successful creation
+ * should be checked using <tt>\ref vxGetStatus</tt>
+ * \ingroup group_vision_function_sp
+ */
+VX_API_ENTRY vx_node VX_API_CALL vxStreamProcessorNode(
+    vx_graph                    graph,
+    vx_tensor*                  input_list,
+    vx_uint32                   input_count,
+    vx_tensor*                  output_list,
+    vx_uint32                   output_count,
+    vx_spinst                   spinst_obj,
+    vx_lut_params               lut_params
+    );
+
 #ifdef __cplusplus
 }
 #endif