diff options
Diffstat (limited to 'arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h')
| -rw-r--r-- | arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h | 52 |
1 files changed, 51 insertions, 1 deletions
diff --git a/arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h b/arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h index 2f44d19b4f..69d8cc6851 100644 --- a/arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h +++ b/arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h @@ -124,7 +124,7 @@ public: /** Configure the output transform kernel. * - * @param[in] input Input tensor data + * @param[in] input Input tensor data. Data types supported: F32. * @param[in] n_batches Number of batches in input tensor. * @param[in] n_rows Number of rows in input tensor. * @param[in] n_cols Number of columns in input tensor. @@ -152,6 +152,17 @@ public: /** Winograd convolution kernel */ using WinogradConv = typename WinogradBase::template Convolution<T, T>; + /** Static function to check if given info will lead to a valid configuration of @ref NEWinogradLayerTransformInputKernel + * + * @param[in] input First tensor input info. Data types supported: F32. + * @param[in] output Output tensor info. Data types supported: same as @p input. + * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. Currently only unit strides are supported. + * @param[in] kernel_dims Kernel dimensions. Currently only 3x3 and 5x5 kernels are supported + * + * @return a status + */ + static Status validate(const ITensorInfo *input, const ITensorInfo *output, const PadStrideInfo &conv_info, const Size2D &kernel_dims); + private: using InputTransform = typename WinogradBase::template InputTransform<T>; std::unique_ptr<InputTransform> _transform; @@ -301,6 +312,19 @@ public: void run(const Window &window, const ThreadInfo &info) override; bool is_parallelisable() const override; + /** Static function to check if given info will lead to a valid configuration of @ref NEWinogradLayerTransformOutputKernel + * + * @param[in] input Source tensor with shape [C, N, 16, batches] or [C, N, 36, batches]. Data types supported: F32. + * @param[in] bias Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM]. It can be a nullptr. Data type supported: as @p input + * @param[out] output Destination tensor with shape [output_convolved_dims.width, output_convolved_dims.height, C, batches]. Data type supported: same as @p input + * @param[in] kernel_dims Kernel dimensions (Width and height). Currently only supported 3x3 and 5x5 kernels + * @param[in] output_convolved_dims Output dimensions after the convolution (Width and height) + * @param[in] num_tiles Number of tiles of size 2x2 or 4x4 in the output tensor along the X and Y direction + * + * @return a status + */ + static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const Size2D &kernel_dims, const Size2D &output_convolved_dims, const Size2D &num_tiles); + private: using WinogradBase = winograd::WinogradGEMM<OutputTileRows, OutputTileCols, KernelRows, KernelCols>; using WinogradConv = typename WinogradBase::template Convolution<T, T>; @@ -366,6 +390,17 @@ public: return "NEWinogradLayerTransformWeightsKernel"; } + /** Static function to check if given info will lead to a valid configuration of @ref NEWinogradLayerTransformWeightsKernel + * + * @param[in] input Source tensor info. The input is a 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM] (NCHW data layout). + * kernel_x must be 3 and equal to kernel_y. Data types supported: F32. + * @param[in] output Destination tensor info. The output is a 3D tensor with dimensions [OFM, IFM, 16] or [OFM, IFM, 36]. Data type supported: same as @p input + * @param[in] output_tile Output tile. Currently only 2x2 and 4x4 tiles are supported. + * + * @return a status + */ + static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &output_tile); + // Inherited methods overridden: void configure(const ITensor *weights_hwio, T *const output, const int matrix_stride, const int n_output_channels, const int n_input_channels) override; unsigned int get_weight_storage_size(int n_output_channels, int n_input_channels) const override; @@ -496,6 +531,21 @@ public: void run(const Window &window, const ThreadInfo &info) override; + /** Static function to check if given info will lead to a valid configuration of @ref NEWinogradLayerBatchedGEMMKernel. + * + * @param[in] a First input tensor (Matrix or Vector A). Data types supported: F32 + * @param[in] b Second input tensor (Matrix B). Data type supported: same as @p a. + * @param[in] c Third input tensor (Matrix C). It can be a nullptr if just the multiplication between @p a and @p b is needed. Data type supported: same as @p a. + * @param[out] output Output tensor. Data type supported: same as @p a + * @param[in] alpha Weight of the matrix product + * @param[in] beta Weight of matrix C + * @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and + * if the reshape of matrix B should happen only for the first run + * + * @return a status + */ + static Status validate(const ITensorInfo *a, const ITensorInfo *b, const ITensor *c, const ITensorInfo *output, const float alpha, const float beta, const GEMMInfo &gemm_info = GEMMInfo()); + private: static const int _output_tile_rows = OutputTileRows; static const int _output_tile_cols = OutputTileCols; |