diff options
author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2017-07-04 16:46:32 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-09-17 14:15:39 +0100 |
commit | 2bbd96457e3740fd9df5556607514b5e80a25720 (patch) | |
tree | 679935dd849bdac044769dfff67516962493dd51 /arm_compute | |
parent | 8a383694445dfebb84732b19d5b3299961e8ffe3 (diff) | |
download | ComputeLibrary-2bbd96457e3740fd9df5556607514b5e80a25720.tar.gz |
COMPMID-436, COMPMID-437 - Port NEConvolutionLayer & NEFullyConnectedLayer to support 16 bit fixed point
Change-Id: I69edf2dac242f941bac95c8479d921e7be6abca7
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/79725
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Diffstat (limited to 'arm_compute')
6 files changed, 9 insertions, 9 deletions
diff --git a/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h b/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h index c0ecafcd39..1eed4e7a84 100644 --- a/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h +++ b/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h @@ -47,7 +47,7 @@ public: ~NEGEMMMatrixAccumulateBiasesKernel() = default; /** Set the accumulate buffer and the biases of the kernel. * - * @param[in, out] accum The accumulate tensor to convert. Data type supported: QS8/F32 + * @param[in, out] accum The accumulate tensor to convert. Data type supported: QS8/QS16/F32 * @param[in] biases The shared biases tensor to append. It must be 1D Tensor. Data type supported: Same as @p input */ void configure(ITensor *accum, const ITensor *biases); diff --git a/arm_compute/core/NEON/kernels/NEIm2ColKernel.h b/arm_compute/core/NEON/kernels/NEIm2ColKernel.h index 9b8b98b388..87d7cc0a8b 100644 --- a/arm_compute/core/NEON/kernels/NEIm2ColKernel.h +++ b/arm_compute/core/NEON/kernels/NEIm2ColKernel.h @@ -73,7 +73,7 @@ public: /** Set the input and output of the kernel. * * @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], - * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/F16/F32 + * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QS16/F16/F32 * @param[out] output The output tensor. Data types supported: Same as @p input * @param[in] kernel_dims The kernel dimensions (width and height). * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. diff --git a/arm_compute/core/NEON/kernels/NETransposeKernel.h b/arm_compute/core/NEON/kernels/NETransposeKernel.h index ac9449ff92..2f757f18eb 100644 --- a/arm_compute/core/NEON/kernels/NETransposeKernel.h +++ b/arm_compute/core/NEON/kernels/NETransposeKernel.h @@ -53,7 +53,7 @@ public: /** Initialise the kernel's input and output. * - * @param[in] input Input tensor. Data types supported: U8/S8/QS8/U16/S16/F16/U32/S32/F32 + * @param[in] input Input tensor. Data types supported: U8/S8/QS8/U16/S16/QS16/F16/U32/S32/F32 * @param[out] output Output tensor. Data type supported: Same as @p input */ void configure(const ITensor *input, ITensor *output); @@ -64,7 +64,7 @@ public: private: /** Common signature for all the transpose functions * - * @param[in] input An input tensor. Data types supported: U8/S8/QS8/U16/S16/F16/U32/S32/F32 + * @param[in] input An input tensor. Data types supported: U8/S8/QS8/U16/S16/QS16/F16/U32/S32/F32 * @param[out] output The output tensor. Data type supported: same as @p input * @param[in] window Region on which to execute the kernel. */ diff --git a/arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h b/arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h index cad2d00b1f..6b76d19314 100644 --- a/arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h +++ b/arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h @@ -71,7 +71,7 @@ public: /** Set the input and output of the kernel. * * @param[in] input The input tensor to convert. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM] if shared, - * and 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches] if unshared. Data types supported: QS8/F32 + * and 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches] if unshared. Data types supported: QS8/QS16/F32 * @param[in] bias The shared biases tensor to append. Bias is 1D tensor with dimensions [OFM] if shared and 2D tensor with * dimensions [OFM, num_patches] if unshared. Data types supported: Same as @p input * @param[out] output The output tensor. Data types supported: Same as @p input diff --git a/arm_compute/runtime/NEON/functions/NEConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEConvolutionLayer.h index a8fff8d047..1bd7e6a95f 100644 --- a/arm_compute/runtime/NEON/functions/NEConvolutionLayer.h +++ b/arm_compute/runtime/NEON/functions/NEConvolutionLayer.h @@ -51,7 +51,7 @@ public: NEConvolutionLayerReshapeWeights(); /** Set the input and output tensors. * - * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: QS8/F32. + * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: QS8/QS16/F32. * @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p weights. * @param[out] output Destination tensor. Data types supported: Same as @p weights. * @param[in] transpose1xW True if the weights are to undergo a 1xW transposition after reshaping (in case of GEMM operation), false otherwise. @@ -84,7 +84,7 @@ public: * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], * while every optional dimension from 4 and above represent a batch of inputs. - * Data types supported: QS8/F32. + * Data types supported: QS8/QS16/F32. * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: Same as @p input. * @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. * @param[out] output Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs. diff --git a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h index 33ec4ef721..af571d1057 100644 --- a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h +++ b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h @@ -50,7 +50,7 @@ public: NEFullyConnectedLayerReshapeWeights(); /** Set the input and output tensors. * - * @param[in] input Weights tensor. The weights must be 2 dimensional. Data types supported: QS8/F32. + * @param[in] input Weights tensor. The weights must be 2 dimensional. Data types supported: QS8/QS16/F32. * @param[out] output Destination tensor. Data type supported: Same as @p input. * @param[in] transpose_weights True if the weights must be transposed. Data types supported: Same as @p weights. * @param[in] is_batched_fc_layer True if it is a batched fully connected layer @@ -84,7 +84,7 @@ public: NEFullyConnectedLayer(); /** Set the input and output tensors. * - * @param[in] input Source tensor. Data type supported: QS8/F32. + * @param[in] input Source tensor. Data type supported: QS8/QS16/F32. * @param[in] weights Weights tensor. The weights must be 2 dimensional. Data type supported: Same as @p input. * @param[in] biases Bias tensor. Can be nullptr. Data type supported:Same as @p input. * @param[out] output Destination tensor. Data type supported: Same as @p input. |