From ff2719299ea76a95f20a35a7900875a8152e293a Mon Sep 17 00:00:00 2001 From: Michele Di Giorgio Date: Wed, 17 Jul 2019 15:59:32 +0100 Subject: COMPMID-2336: Fix validation for quantized NEDepthwiseConvolutionLayer Assertions from calculate_quantized_multiplier_less_than_one were not captured resulting in incorrect computation being performed on invalid inputs. This patch also fixes a discrepancy between the interfaces of validate and configure methods of NEDirectConvolutionLayerOutputStageKernel. Change-Id: I011822f63b4062b5b3346ef047adc684e0bffa3d Signed-off-by: Michele Di Giorgio Reviewed-on: https://review.mlplatform.org/c/1552 Reviewed-by: Giorgio Arena Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins --- .../NEDirectConvolutionLayerOutputStageKernel.h | 25 +++++++++++++--------- 1 file changed, 15 insertions(+), 10 deletions(-) (limited to 'arm_compute/core') diff --git a/arm_compute/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.h b/arm_compute/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.h index 9af3de5ffe..6c9002a3c5 100644 --- a/arm_compute/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.h +++ b/arm_compute/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -59,22 +59,27 @@ public: * @param[in] bias (Optional) The shared bias tensor to add. It must be 1D Tensor. Data type supported: Same as @p input * @param[out] output (Optional) If the output tensor is specified the accumulation is done out-of-place. (Defaults to nullptr) * Data type supported: F16/F32 - * @param[in] result_fixedpoint_multiplier (Optional)Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add - * @param[in] result_shift (Optional)Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication - * @param[in] result_offset_after_shift (Optional)Offset to be applied to result before converting it back to QASYMM8 + * @param[in] result_fixedpoint_multiplier (Optional) Fixed point value to be multiplied to each element of the input matrix once the result_offset has been added + * @param[in] result_shift (Optional) Integer value used to round the result of the fixed point multiplication to nearest division by a power-of-two + * @param[in] result_offset_after_shift (Optional) Offset to be applied to result before converting it back to QASYMM8 */ void configure(ITensor *input, const ITensor *bias = nullptr, ITensor *output = nullptr, int result_fixedpoint_multiplier = 0, int result_shift = 0, int result_offset_after_shift = 0); /** Static function to check if given info will lead to a valid configuration of @ref NEDirectConvolutionLayerOutputStageKernel * - * @param[in] input Input to add the bias to. If @p output is not specified then accumulation is done in-place. - * Data type supported: F16/F32 - * @param[in] bias (Optional) The shared bias tensor to add. It must be 1D Tensor. Data type supported: Same as @p input - * @param[in] output (Optional) If the output tensor is specified the accumulation is done out-of-place. (Defaults to nullptr) - * Data type supported: F16/F32 + * @param[in] input Input to add the bias to. If @p output is not specified then accumulation is done in-place. + * Data type supported: F16/F32 + * @param[in] bias (Optional) The shared bias tensor to add. It must be 1D Tensor. Data type supported: Same as @p input + * @param[in] output (Optional) If the output tensor is specified the accumulation is done out-of-place. (Defaults to nullptr) + * Data type supported: F16/F32 + * @param[in] result_fixedpoint_multiplier (Optional) Fixed point value to be multiplied to each element of the input matrix once the result_offset has been added + * @param[in] result_shift (Optional) Integer value used to round the result of the fixed point multiplication to nearest division by a power-of-two + * @param[in] result_offset_after_shift (Optional) Offset to be applied to result before converting it back to QASYMM8 + * * @return a status */ - static Status validate(const ITensorInfo *input, const ITensorInfo *bias = nullptr, const ITensorInfo *output = nullptr); + static Status validate(const ITensorInfo *input, const ITensorInfo *bias = nullptr, const ITensorInfo *output = nullptr, + int result_fixedpoint_multiplier = 0, int result_shift = 0, int result_offset_after_shift = 0); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; -- cgit v1.2.1