From 0c34fe29c298057091d48cde332cb60bb14efee1 Mon Sep 17 00:00:00 2001 From: Pablo Tello Date: Mon, 26 Jun 2017 17:17:42 +0100 Subject: COMPMID-421: Added FP16 support in Pooling Layer Change-Id: I6b6119c8770051c1656da40aa073c539c15b493e Reviewed-on: http://mpd-gerrit.cambridge.arm.com/78985 Reviewed-by: Moritz Pflanzer Tested-by: Kaizen --- arm_compute/core/NEON/kernels/NEPoolingLayerKernel.h | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-) (limited to 'arm_compute/core/NEON/kernels/NEPoolingLayerKernel.h') diff --git a/arm_compute/core/NEON/kernels/NEPoolingLayerKernel.h b/arm_compute/core/NEON/kernels/NEPoolingLayerKernel.h index bf06fdd639..a5de81137b 100644 --- a/arm_compute/core/NEON/kernels/NEPoolingLayerKernel.h +++ b/arm_compute/core/NEON/kernels/NEPoolingLayerKernel.h @@ -48,7 +48,7 @@ public: ~NEPoolingLayerKernel() = default; /** Set the input and output tensors. * - * @param[in] input Source tensor. Data types supported: QS8/F32. + * @param[in] input Source tensor. Data types supported: QS8/F16/F32. * @param[out] output Destination tensor. Data types supported: Same as @p input. * @param[in] pool_info Contains pooling operation information described in @ref PoolingLayerInfo. */ @@ -66,6 +66,14 @@ private: */ template void pooling2_f32(const Window &window_input, const Window &window); + /** Function to perform 2x2 pooling for float16_t. + * + * @param[in] window_input Input region on which to execute the kernel. + * @param[in] window Output region on which to execute the kernel. + */ + template + void pooling2_f16(const Window &window_input, const Window &window); + /** Function to perform 2x2 pooling for 8bit fixed point. * * @param[in] window_input Input region on which to execute the kernel. @@ -80,6 +88,13 @@ private: */ template void pooling3_f32(const Window &window_input, const Window &window); + /** Function to perform 3x3 pooling. + * + * @param[in] window_input Input region on which to execute the kernel. + * @param[in] window Output region on which to execute the kernel. + */ + template + void pooling3_f16(const Window &window_input, const Window &window); /** Function to perform 3x3 pooling for 8bit fixed point. * * @param[in] window_input Input region on which to execute the kernel. -- cgit v1.2.1