From 9637b2e4fc33b2264aa5586dd6b2ed1045db5075 Mon Sep 17 00:00:00 2001 From: Michele Di Giorgio Date: Mon, 23 Sep 2019 16:49:49 +0100 Subject: COMPMID-2671: Change ArgMinMax NEON/CL output type to Signed32 Change-Id: I718f3884928271c5b0afb259d5bfe9df284f18e6 Signed-off-by: Michele Di Giorgio Reviewed-on: https://review.mlplatform.org/c/1995 Reviewed-by: Georgios Pinitas Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins --- arm_compute/core/NEON/kernels/NEReductionOperationKernel.h | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) (limited to 'arm_compute/core/NEON/kernels/NEReductionOperationKernel.h') diff --git a/arm_compute/core/NEON/kernels/NEReductionOperationKernel.h b/arm_compute/core/NEON/kernels/NEReductionOperationKernel.h index a4cb330445..4b28b8dbcd 100644 --- a/arm_compute/core/NEON/kernels/NEReductionOperationKernel.h +++ b/arm_compute/core/NEON/kernels/NEReductionOperationKernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -30,7 +30,13 @@ namespace arm_compute { class ITensor; -/** NEON kernel to perform a reduction operation */ +/** NEON kernel to perform a reduction operation + * + * @note For ARG_MIN/ARG_MAX reduction, the indices are computed in unsigned + * 32-bit (U32). It is the user's responsibility to check that the + * results do not overflow in case the output data type is set to signed + * 32-bit integer (S32). + */ class NEReductionOperationKernel : public INEKernel { public: @@ -54,7 +60,7 @@ public: /** Set the source, destination of the kernel * * @param[in] input Source tensor. Data type supported: QASYMM8/F16/F32. Data layouts supported: NCHW. - * @param[out] output Destination tensor.Data types and data layouts supported: same as @p input. + * @param[out] output Destination tensor.Data types and data layouts supported: same as @p input, S32 for ARG_MIX/ARG_MAX. * Output will have the same number of dimensions as input. * @param[in] axis Axis along which to reduce. Supported reduction axis : 0 * @param[in] op Reduction operation to perform. @@ -64,7 +70,7 @@ public: /** Static function to check if given info will lead to a valid configuration of @ref NEReductionOperationKernel. * * @param[in] input Source tensor info. Data type supported: QASYMM8/F16/F32. Data layouts supported: NCHW. - * @param[in] output Destination tensor info.Data types and data layouts supported: same as @p input. + * @param[in] output Destination tensor info.Data types and data layouts supported: same as @p input, S32 for ARG_MIX/ARG_MAX. * Output will have the same number of dimensions as input. * @param[in] axis Axis along which to reduce. Supported reduction axis : 0 * @param[in] op Reduction operation to perform. -- cgit v1.2.1