From 9f7d55a3566b0f1044110000b033d663b26d3a6c Mon Sep 17 00:00:00 2001 From: Giorgio Arena Date: Mon, 8 Feb 2021 13:20:24 +0000 Subject: Fix CLDepthwiseConvolutionLayer 3x3 QASYMM8 Fix errors when computing tensors with one element only - Replace Tensor3D with raw pointers so to get rid of offset to first element for NCHW layout - Add stronger out of bound constraints for NHWC layout - Set the border size to the input's padding for NHWC - Fill the strides == 0 with the largest stride, so to avoid accessing empty strides and multiplying by 0 Resolve COMPMID-4088 Change-Id: I751a4e6d7094b3c42306ff7f53af848fd35f19ac Signed-off-by: Giorgio Arena Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5024 Tested-by: Arm Jenkins Reviewed-by: Manuel Bottini Reviewed-by: Michele Di Giorgio Comments-Addressed: Arm Jenkins --- src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp | 6 +++--- src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp | 6 +++--- .../CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp | 6 +++--- src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp | 6 +++--- src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp | 6 +++--- src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp | 6 +++--- src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp | 6 +++--- 8 files changed, 23 insertions(+), 23 deletions(-) (limited to 'src/core/CL/kernels') diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp index f553fd1849..43c3ff3bfd 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -233,7 +233,7 @@ void CLDepthwiseConvolutionLayer3x3NHWCKernel::configure(const CLCompileContext if(_is_quantized) { - _border_size = BorderSize(is_stride_1 ? 0 : conv_info.pad_left(), 0, std::max(std::max(conv_info.pad_right(), conv_info.pad_bottom()), conv_info.pad_top()), 0); + _border_size = BorderSize(input->info()->padding()); // If QASYMM8 and the 8 bit dot product is available, force _num_planes_processed_per_iteration to 1 if(is_dot8_supported) diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp index 1f89865908..5633ee5a28 100644 --- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp +++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -284,8 +284,8 @@ void CLGEMMLowpMatrixMultiplyNativeKernel::run(const Window &window, cl::Command if(_input1->info()->num_dimensions() < 3) { - // The stride_z for matrix B must be zero if we do not slice - ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0); + // The stride_w for matrix B must be the same as stride_z if we do not slice + ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != _input1->info()->strides_in_bytes()[2]); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp index ded4b29ae7..3043e01514 100644 --- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp +++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -254,8 +254,8 @@ void CLGEMMLowpMatrixMultiplyReshapedKernel::run(const Window &window, cl::Comma if(_input1->info()->num_dimensions() < 3) { - // The stride_z for matrix B must be zero if we do not slice - ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0); + // The stride_w for matrix B must be the same as stride_z if we do not slice + ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != _input1->info()->strides_in_bytes()[2]); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp index 77cea24829..0122e3ba4b 100644 --- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp +++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -488,8 +488,8 @@ void CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel::run(const Window &window, cl if(_input1->info()->num_dimensions() < 3) { - // The stride_z for matrix B must be zero if we do not slice - ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0); + // The stride_w for matrix B must be the same as stride_z if we do not slice + ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != _input1->info()->strides_in_bytes()[2]); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp index 2419104fba..a653e29f8f 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2020 Arm Limited. + * Copyright (c) 2017-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -484,8 +484,8 @@ void CLGEMMMatrixMultiplyKernel::run(const Window &window, cl::CommandQueue &que if(_input1->info()->num_dimensions() < 3) { - // The stride_z for matrix B must be zero if we do not slice - ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0); + // The stride_w for matrix B must be the same as stride_z if we do not slice + ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != _input1->info()->strides_in_bytes()[2]); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp index 387f1a4ebc..fefcd2f74d 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -348,8 +348,8 @@ void CLGEMMMatrixMultiplyNativeKernel::run(const Window &window, cl::CommandQueu if(_input1->info()->num_dimensions() < 3) { - // The stride_z for matrix B must be zero if we do not slice - ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0); + // The stride_w for matrix B must be the same as stride_z if we do not slice + ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != _input1->info()->strides_in_bytes()[2]); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp index 23e18bac92..8a403555f5 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -359,8 +359,8 @@ void CLGEMMMatrixMultiplyReshapedKernel::run(const Window &window, cl::CommandQu if(_input1->info()->num_dimensions() < 3) { - // The stride_z for matrix B must be zero if we do not slice - ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0); + // The stride_w for matrix B must be the same as stride_z if we do not slice + ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != _input1->info()->strides_in_bytes()[2]); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp index 1f296f8e26..de986de9f6 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -352,8 +352,8 @@ void CLGEMMMatrixMultiplyReshapedOnlyRHSKernel::run(const Window &window, cl::Co if(_input1->info()->num_dimensions() < 3) { - // The stride_z for matrix B must be zero if we do not slice - ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0); + // The stride_w for matrix B must be the same as stride_z if we do not slice + ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != _input1->info()->strides_in_bytes()[2]); } const size_t lhs_idx_batch_size = _reinterpret_input_as_3d && !_has_pad_y ? 3u : 2u; -- cgit v1.2.1