From edc524ef7ed38e0521c874f28bb9a1f2407b44c6 Mon Sep 17 00:00:00 2001 From: Giorgio Arena Date: Wed, 10 Feb 2021 11:54:47 +0000 Subject: Revert changes on tensor's strides and fix CLDepthwiseConvolution 3x3 Quantized - Revert changes in strides > num_dimensions. Set them to 0 - Fix offset calculcation in depthwise 3x3 quantized using select and stride_y for max offset Resolve COMPMID-4254 Change-Id: Ia99b9637f18b99b1fa3d4b7b4892046027d3e7e5 Signed-off-by: Giorgio Arena Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5040 Tested-by: Arm Jenkins Reviewed-by: Georgios Pinitas Comments-Addressed: Arm Jenkins --- src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp | 4 ++-- src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp | 4 ++-- 8 files changed, 16 insertions(+), 16 deletions(-) (limited to 'src/core/CL/kernels') diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp index 43c3ff3bfd..d13afd2010 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp @@ -438,8 +438,8 @@ void CLDepthwiseConvolutionLayer3x3NHWCKernel::run(const Window &window, cl::Com // |__________________| // | pad_bottom | // |******************| - const int max_offset = _input->info()->strides_in_bytes().z() * _input->info()->dimension(2) - (_input->info()->padding().bottom + _input->info()->padding().top) * - _input->info()->strides_in_bytes().y(); + const int max_offset = ((_input->info()->dimension(1) * _input->info()->dimension(2)) + (_input->info()->padding().bottom + _input->info()->padding().top) * (_input->info()->dimension( + 2) - 1)) * _input->info()->strides_in_bytes().y(); _kernel.setArg(idx, max_offset); } diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp index 5633ee5a28..9215fd602d 100644 --- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp +++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.cpp @@ -284,8 +284,8 @@ void CLGEMMLowpMatrixMultiplyNativeKernel::run(const Window &window, cl::Command if(_input1->info()->num_dimensions() < 3) { - // 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]); + // 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); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp index 3043e01514..848f272e50 100644 --- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp +++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp @@ -254,8 +254,8 @@ void CLGEMMLowpMatrixMultiplyReshapedKernel::run(const Window &window, cl::Comma if(_input1->info()->num_dimensions() < 3) { - // 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]); + // 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); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp index 0122e3ba4b..eba52b08bd 100644 --- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp +++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.cpp @@ -488,8 +488,8 @@ void CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel::run(const Window &window, cl if(_input1->info()->num_dimensions() < 3) { - // 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]); + // 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); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp index a653e29f8f..6d3b1e5897 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp @@ -484,8 +484,8 @@ void CLGEMMMatrixMultiplyKernel::run(const Window &window, cl::CommandQueue &que if(_input1->info()->num_dimensions() < 3) { - // 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]); + // 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); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp index fefcd2f74d..f07166e4bb 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.cpp @@ -348,8 +348,8 @@ void CLGEMMMatrixMultiplyNativeKernel::run(const Window &window, cl::CommandQueu if(_input1->info()->num_dimensions() < 3) { - // 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]); + // 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); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp index 8a403555f5..9f1ffa48eb 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.cpp @@ -359,8 +359,8 @@ void CLGEMMMatrixMultiplyReshapedKernel::run(const Window &window, cl::CommandQu if(_input1->info()->num_dimensions() < 3) { - // 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]); + // 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); } Window slice = window.first_slice_window_3D(); diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp index de986de9f6..3dee4f24cd 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp @@ -352,8 +352,8 @@ void CLGEMMMatrixMultiplyReshapedOnlyRHSKernel::run(const Window &window, cl::Co if(_input1->info()->num_dimensions() < 3) { - // 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]); + // 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); } const size_t lhs_idx_batch_size = _reinterpret_input_as_3d && !_has_pad_y ? 3u : 2u; -- cgit v1.2.1