diff options
Diffstat (limited to 'arm_compute/runtime')
-rw-r--r-- | arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h | 82 | ||||
-rw-r--r-- | arm_compute/runtime/NEON/functions/NEDepthwiseConvolutionLayer.h | 65 |
2 files changed, 0 insertions, 147 deletions
diff --git a/arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h index 63c359e68c..f15271b63c 100644 --- a/arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h @@ -338,87 +338,5 @@ private: CLDepthwiseConvolutionLayerInternal3x3 _func_3x3; CLDepthwiseConvolutionLayerGeneric _func_generic; }; - -/** Basic function to execute a depthwise convolution for kernel size 3x3xC (when data layout NCHW) or Cx3x3 (when data layout NHWC). This function calls the following OpenCL kernels: - * - * -# @ref CLDepthwiseConvolutionLayer3x3NCHWKernel (if data_layout == NCHW) - * -# @ref CLDepthwiseConvolutionLayer3x3NHWCKernel (if data_layout == NHWC) - * -# @ref CLDepthwiseConvolutionLayerReshapeWeightsKernel (if data_layout == NHWC) - * -# @ref CLFillBorderKernel (if pad_x or pad_y > 0) - * - */ -class CLDepthwiseConvolutionLayer3x3 : public IFunction -{ -public: - /** Default constructor */ - CLDepthwiseConvolutionLayer3x3(std::shared_ptr<IMemoryManager> memory_manager = nullptr); - /** Prevent instances of this class from being copied (As this class contains pointers) */ - CLDepthwiseConvolutionLayer3x3(const CLDepthwiseConvolutionLayer3x3 &) = delete; - /** Default move constructor */ - CLDepthwiseConvolutionLayer3x3(CLDepthwiseConvolutionLayer3x3 &&) = default; - /** Prevent instances of this class from being copied (As this class contains pointers) */ - CLDepthwiseConvolutionLayer3x3 &operator=(const CLDepthwiseConvolutionLayer3x3 &) = delete; - /** Default move assignment operator */ - CLDepthwiseConvolutionLayer3x3 &operator=(CLDepthwiseConvolutionLayer3x3 &&) = default; - /** Initialize the function's source, destination, conv and border_size. - * - * @param[in, out] input Source tensor. Data type supported: QASYMM8/F16/F32. (Written to only for border filling). - * @param[in] weights Weights tensor. A 3D tensor with shape [3, 3, IFM]. - * Data type supported: Same as @p input or QASYMM8/QSYMM8_PER_CHANNEL when @p input is QASYMM8. - * @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed. - * Data type supported: Same as @p input. - * @param[out] output Destination tensor. Data type supported: same as @p input. - * @param[in] conv_info Padding and stride information to use for the convolution. - * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU for 3x3 QASYMM8 supported. - * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). - */ - ARM_COMPUTE_DEPRECATED_REL_REPLACE(20.02, CLDepthwiseConvolutionLayer) - void configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier = 1, - ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U)); - /** Initialize the function's source, destination, conv and border_size. - * - * @param[in] compile_context The compile context to be used. - * @param[in, out] input Source tensor. Data type supported: QASYMM8/F16/F32. (Written to only for border filling). - * @param[in] weights Weights tensor. A 3D tensor with shape [3, 3, IFM]. - * Data type supported: Same as @p input or QASYMM8/QSYMM8_PER_CHANNEL when @p input is QASYMM8. - * @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed. - * Data type supported: Same as @p input. - * @param[out] output Destination tensor. Data type supported: same as @p input. - * @param[in] conv_info Padding and stride information to use for the convolution. - * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU for 3x3 QASYMM8 supported. - * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). - */ - ARM_COMPUTE_DEPRECATED_REL_REPLACE(20.02, CLDepthwiseConvolutionLayer) - void configure(const CLCompileContext &compile_context, ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, - unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U)); - - /** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseConvolutionLayer3x3 - * - * @param[in] input Source tensor info. Data type supported: QASYMM8 for all layouts, F16/F32 for NCHW. - * @param[in] weights Weights tensor info. A 3D tensor with shape [3, 3, IFM]. - * Data type supported: Same as @p input or QASYMM8/QSYMM8_PER_CHANNEL when @p input is QASYMM8. - * @param[in] biases Biases tensor info. A 1D tensor with shape [IFM]. Must be nullptr if not needed. - * Data type supported: Same as @p input, S32 when input is QASYMM8. - * @param[in] output Destination tensor. Data type supported: same as @p input. - * @param[in] conv_info Padding and stride information to use for the convolution. - * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU for 3x3 QASYMM8 supported. - * @param[in] gpu_target (Optional) GPU target to validate the kernel for. Defaults to midgard. - * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). - * - * @return a status - */ - static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier = 1, - ActivationLayerInfo act_info = ActivationLayerInfo(), GPUTarget gpu_target = GPUTarget::MIDGARD, const Size2D &dilation = Size2D(1U, 1U)); - - // Inherited methods overriden: - void run() override; - void prepare() override; - -private: - CLDepthwiseConvolutionLayer _func; -}; } // namespace arm_compute #endif /*ARM_COMPUTE_CLDEPTHWISECONVOLUTION_H */ diff --git a/arm_compute/runtime/NEON/functions/NEDepthwiseConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDepthwiseConvolutionLayer.h index ccab671644..811dc82843 100644 --- a/arm_compute/runtime/NEON/functions/NEDepthwiseConvolutionLayer.h +++ b/arm_compute/runtime/NEON/functions/NEDepthwiseConvolutionLayer.h @@ -303,70 +303,5 @@ private: NEDepthwiseConvolutionLayerOptimizedInternal _func_optimized; NEDepthwiseConvolutionLayerGeneric _func_generic; }; - -/** Basic function to execute optimized depthwise convolution routines. This function calls the following NEON kernels: - * - * @note At the moment 3x3 and 5x5 convolution of stride 1, 2 are supported - * - * -# @ref NEFillBorderKernel (if pad_x or pad_y > 0) and no assembly kernel implementation is present - * -# @ref NEDepthwiseConvolutionLayer3x3Kernel if 3x3 and no assembly kernel implementation is present - * -# @ref NEDepthwiseConvolutionAssemblyDispatch if assembly kernel implementation is present - * -# @ref NEDirectConvolutionLayerOutputStageKernel if re-quantization of output is required - * -# @ref NEActivationLayer if fused activation is required - * - */ -class NEDepthwiseConvolutionLayerOptimized : public IFunction -{ -public: - /** Default constructor */ - NEDepthwiseConvolutionLayerOptimized(std::shared_ptr<IMemoryManager> memory_manager = nullptr); - /** Prevent instances of this class from being copied (As this class contains pointers) */ - NEDepthwiseConvolutionLayerOptimized(const NEDepthwiseConvolutionLayerOptimized &) = delete; - /** Default move constructor */ - NEDepthwiseConvolutionLayerOptimized(NEDepthwiseConvolutionLayerOptimized &&) = default; - /** Prevent instances of this class from being copied (As this class contains pointers) */ - NEDepthwiseConvolutionLayerOptimized &operator=(const NEDepthwiseConvolutionLayerOptimized &) = delete; - /** Default move assignment operator */ - NEDepthwiseConvolutionLayerOptimized &operator=(NEDepthwiseConvolutionLayerOptimized &&) = default; - /** Initialize the function's source, destination, kernels and border_size. - * - * @param[in, out] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. (Written to only for border filling). - * @param[in] weights Weights tensor. These are 3D tensors with shape [W, H, IFM]. Data type supported: Same as @p input. - * @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed. - * Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED. - * @param[out] output Destination tensor. Data type supported: same as @p input. - * @param[in] conv_info Padding and stride information to use for the convolution. - * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. - * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). - */ - ARM_COMPUTE_DEPRECATED_REL_REPLACE(20.02, NEDepthwiseConvolutionLayer) - void configure(ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info, - unsigned int depth_multiplier = 1, const ActivationLayerInfo &act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U)); - - /** Static function to check if given info will lead to a valid configuration of @ref NEDepthwiseConvolutionLayerOptimized - * - * @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. (Written to only for border filling). - * @param[in] weights Weights tensor. These are 3D tensors with shape [W, H, IFM]. Data type supported: Same as @p input. - * @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed. - * Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED. - * @param[in] output Destination tensor. Data type supported: same as @p input. - * @param[in] conv_info Padding and stride information to use for the convolution. - * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. - * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). - * - * @return a status - */ - static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, - unsigned int depth_multiplier = 1, const ActivationLayerInfo &act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U)); - - // Inherited methods overriden: - void run() override; - void prepare() override; - -private: - NEDepthwiseConvolutionLayer _func; -}; } // namespace arm_compute #endif /* ARM_COMPUTE_NEDEPTHWISECONVOLUTION_H */
\ No newline at end of file |