diff options
author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2018-07-06 12:59:28 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:54:10 +0000 |
commit | d28b751cf2ba9fcf4ccf294b31bf9d2ec5dfd8bb (patch) | |
tree | 111a96797e6b1cd20a2db7088e5fc4cd1903ff02 /arm_compute/runtime/CL | |
parent | 98f085bf87d55bff3866963e8220cfcb4872709f (diff) | |
download | ComputeLibrary-d28b751cf2ba9fcf4ccf294b31bf9d2ec5dfd8bb.tar.gz |
COMPMID-1340 - Implementing Winograd Convolution Layer 1x5/5x1 on OpenCL NHWC
Change-Id: Id5e0795238f77c049df9c109dafc5ef878c1897d
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/139234
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'arm_compute/runtime/CL')
-rw-r--r-- | arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h | 6 | ||||
-rw-r--r-- | arm_compute/runtime/CL/functions/CLWinogradInputTransform.h | 10 |
2 files changed, 8 insertions, 8 deletions
diff --git a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h index 683aa79788..a24ae31d41 100644 --- a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h @@ -59,8 +59,7 @@ public: CLWinogradConvolutionLayer &operator=(CLWinogradConvolutionLayer &&) = default; /** Set the input and output tensors. * - * @note: This function only works with 3x3,3x1,1x3,5x5,5x1 and 1x5 kernels along with unit strides for NCHW data layout - * @note: This function only works with 3x3, 3x1, 1x3 and 5x5 kernels along with unit strides for NHWC data layout + * @note: This function only works with 3x3,3x1,1x3,5x5,5x1 and 1x5 kernels along with unit strides for both NCHW and NHWC data layout * @note Some Winograd configurations (i.e. F(4x4, 5x5)) are supported only with enable_fast_math = true * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], @@ -79,8 +78,7 @@ public: const ActivationLayerInfo &act_info = ActivationLayerInfo(), bool enable_fast_math = false); /** Static function to check if given info will lead to a valid configuration of @ref CLWinogradConvolutionLayer * - * @note: This function only works with 3x3,3x1,1x3,5x5,5x1 and 1x5 kernels along with unit strides for NCHW data layout - * @note: This function only works with 3x3 and 5x5 kernels along with unit strides for NHWC data layout + * @note: This function only works with 3x3,3x1,1x3,5x5,5x1 and 1x5 kernels along with unit strides for both NCHW and NHWC data layout * @note Some Winograd configurations (i.e. F(4x4, 5x5)) are supported only with enable_fast_math = true * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], diff --git a/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h b/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h index 1f89455aee..8ea25a116a 100644 --- a/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h +++ b/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h @@ -45,8 +45,9 @@ public: * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) * * @note Winograd input transform supports the following configurations for NHWC data layout - * F(output tile, kernel size):F(4x4, 3x3), - * F(4x4, 5x5) + * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), + * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) + * * Strides: only unit strides * * @param[in] input The input tensor to transform. Data types supported: F32 @@ -62,8 +63,9 @@ public: * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) * * @note Winograd input transform supports the following configurations for NHWC data layout - * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3) - * F(4x4, 5x5) + * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), + * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) + * * Strides: only unit strides * * @param[in] input The input tensor to transform. Data types supported: F32 |