diff options
author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2018-07-03 12:22:09 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:54:10 +0000 |
commit | 876be2a0d11874d871860dbd22481f831d6878f6 (patch) | |
tree | 907afac009513f882b39b2770b187635cebd1664 /arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h | |
parent | 09daf4ddf5940d18ce95e7dd0859d1dace3b133e (diff) | |
download | ComputeLibrary-876be2a0d11874d871860dbd22481f831d6878f6.tar.gz |
COMPMID-1339 - Implementing Winograd Convolution Layer 1x5 and 5x1 kernels on OpenCL NCHW
Change-Id: Ia293cd89651146a0e27e5f7c74ca9c924807e83c
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/138707
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Diffstat (limited to 'arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h')
-rw-r--r-- | arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h | 10 |
1 files changed, 6 insertions, 4 deletions
diff --git a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h index 594d6028e1..683aa79788 100644 --- a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h @@ -59,8 +59,9 @@ public: CLWinogradConvolutionLayer &operator=(CLWinogradConvolutionLayer &&) = default; /** Set the input and output tensors. * - * @note: This function only works with 3x3 and 5x5 kernels along with unit strides - * @note Some Winograd configurations (i.e. F(4x4, 3x3) and F(4x4, 5x5)) are supported only with enable_fast_math = true + * @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 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], * while every optional dimension from 4 and above represent a batch of inputs. @@ -78,8 +79,9 @@ 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 and 5x5 kernels along with unit strides - * @note Some Winograd configurations (i.e. F(4x4, 3x3) and F(4x4, 5x5)) are supported only with enable_fast_math = true + * @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 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], * while every optional dimension from 4 and above represent a batch of inputs. |