diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-05-17 18:14:40 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-05-21 11:28:01 +0000 |
commit | 09f24975437e2e141ba51a07055a9372b0d173a2 (patch) | |
tree | fe565e4b9abd379cb1f467e5d9e36d68fcfbacef /src/runtime/CL/functions | |
parent | f24411ffc842970609a1fb6ba2f9527cfb681dbd (diff) | |
download | ComputeLibrary-09f24975437e2e141ba51a07055a9372b0d173a2.tar.gz |
COMPMID-2109: Remove CL/NE Width/Depth ConcatenateLayer functions.
Change-Id: Icbda771abffbb45d4ed0958933c60ff9ace01314
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1178
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/runtime/CL/functions')
-rw-r--r-- | src/runtime/CL/functions/CLConcatenateLayer.cpp | 6 | ||||
-rw-r--r-- | src/runtime/CL/functions/CLDepthConcatenateLayer.cpp | 107 | ||||
-rw-r--r-- | src/runtime/CL/functions/CLLSTMLayer.cpp | 4 | ||||
-rw-r--r-- | src/runtime/CL/functions/CLWidthConcatenateLayer.cpp | 143 |
4 files changed, 6 insertions, 254 deletions
diff --git a/src/runtime/CL/functions/CLConcatenateLayer.cpp b/src/runtime/CL/functions/CLConcatenateLayer.cpp index b8224d2cce..0594a17a7a 100644 --- a/src/runtime/CL/functions/CLConcatenateLayer.cpp +++ b/src/runtime/CL/functions/CLConcatenateLayer.cpp @@ -23,11 +23,13 @@ */ #include "arm_compute/runtime/CL/functions/CLConcatenateLayer.h" +#include "arm_compute/core/CL/kernels/CLDepthConcatenateLayerKernel.h" #include "arm_compute/core/CL/kernels/CLHeightConcatenateLayerKernel.h" +#include "arm_compute/core/CL/kernels/CLWidthConcatenate2TensorsKernel.h" +#include "arm_compute/core/CL/kernels/CLWidthConcatenate4TensorsKernel.h" +#include "arm_compute/core/CL/kernels/CLWidthConcatenateLayerKernel.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" #include "arm_compute/runtime/CL/CLScheduler.h" -#include "arm_compute/runtime/CL/functions/CLDepthConcatenateLayer.h" -#include "arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h" #include "arm_compute/core/CL/ICLTensor.h" #include "arm_compute/core/Error.h" diff --git a/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp b/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp deleted file mode 100644 index f687e54552..0000000000 --- a/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp +++ /dev/null @@ -1,107 +0,0 @@ -/* - * Copyright (c) 2017-2019 ARM Limited. - * - * SPDX-License-Identifier: MIT - * - * Permission is hereby granted, free of charge, to any person obtaining a copy - * of this software and associated documentation files (the "Software"), to - * deal in the Software without restriction, including without limitation the - * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or - * sell copies of the Software, and to permit persons to whom the Software is - * furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in all - * copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE - * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, - * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE - * SOFTWARE. - */ -#include "arm_compute/runtime/CL/functions/CLDepthConcatenateLayer.h" - -#include "arm_compute/core/CL/ICLTensor.h" -#include "arm_compute/core/Error.h" -#include "arm_compute/core/Helpers.h" -#include "arm_compute/core/PixelValue.h" -#include "arm_compute/core/TensorInfo.h" -#include "arm_compute/core/Types.h" -#include "arm_compute/core/utils/misc/ShapeCalculator.h" -#include "arm_compute/runtime/CL/CLScheduler.h" -#include "support/ToolchainSupport.h" - -using namespace arm_compute; - -CLDepthConcatenateLayer::CLDepthConcatenateLayer() // NOLINT - : _concat_kernels_vector(), - _border_handlers_vector(), - _num_inputs(0) -{ -} - -void CLDepthConcatenateLayer::configure(const std::vector<ICLTensor *> &inputs_vector, ICLTensor *output) // NOLINT -{ - _num_inputs = inputs_vector.size(); - - std::vector<ITensorInfo *> inputs_vector_info; - for(unsigned int i = 0; i < _num_inputs; i++) - { - inputs_vector_info.emplace_back(inputs_vector.at(i)->info()); - } - - _concat_kernels_vector.resize(_num_inputs); - _border_handlers_vector.resize(_num_inputs); - - TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector_info, Window::DimZ); - - // Output auto inizialitation if not yet initialized - auto_init_if_empty(*output->info(), output_shape, 1, inputs_vector[0]->info()->data_type()); - ARM_COMPUTE_ERROR_THROW_ON(CLDepthConcatenateLayer::validate(inputs_vector_info, output->info())); - - unsigned int depth_offset = 0; - for(unsigned int i = 0; i < _num_inputs; i++) - { - _concat_kernels_vector[i].configure(inputs_vector.at(i), depth_offset, output); - _border_handlers_vector[i].configure(inputs_vector.at(i), _concat_kernels_vector[i].border_size(), BorderMode::CONSTANT, PixelValue()); - - depth_offset += inputs_vector.at(i)->info()->dimension(2); - } - - // Set valid region from shape - output->info()->set_valid_region(ValidRegion(Coordinates(), output_shape)); -} - -Status CLDepthConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vector, const ITensorInfo *output) -{ - ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output); - ARM_COMPUTE_RETURN_ERROR_ON(inputs_vector.size() < 2); - - // Output auto inizialitation if not yet initialized - TensorInfo tmp_output_info = *output->clone(); - TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, Window::DimZ); - auto_init_if_empty(tmp_output_info, output_shape, 1, inputs_vector[0]->data_type()); - - unsigned int depth_offset = 0; - for(const auto &input : inputs_vector) - { - ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input); - ARM_COMPUTE_RETURN_ON_ERROR(CLDepthConcatenateLayerKernel::validate(input, depth_offset, &tmp_output_info)); - depth_offset += input->dimension(2); - } - - return Status{}; -} - -void CLDepthConcatenateLayer::run() -{ - cl::CommandQueue q = CLScheduler::get().queue(); - - for(unsigned i = 0; i < _num_inputs; i++) - { - CLScheduler::get().enqueue(_border_handlers_vector[i], false); - CLScheduler::get().enqueue(_concat_kernels_vector[i], true); - } -} diff --git a/src/runtime/CL/functions/CLLSTMLayer.cpp b/src/runtime/CL/functions/CLLSTMLayer.cpp index 4606a66bf2..85a81a8cd4 100644 --- a/src/runtime/CL/functions/CLLSTMLayer.cpp +++ b/src/runtime/CL/functions/CLLSTMLayer.cpp @@ -316,7 +316,7 @@ void CLLSTMLayer::configure(const ICLTensor *input, scratch_inputs.emplace_back(&_cell_state_out1); scratch_inputs.emplace_back(forget_gate_out); scratch_inputs.emplace_back(output_gate_out); - _concat_scratch_buffer.configure(scratch_inputs, scratch_buffer); + _concat_scratch_buffer.configure(scratch_inputs, scratch_buffer, Window::DimX); input_gate_out->allocator()->allocate(); _cell_state_out1.allocator()->allocate(); forget_gate_out->allocator()->allocate(); @@ -497,7 +497,7 @@ Status CLLSTMLayer::validate(const ITensorInfo *input, inputs_vector_info_raw.push_back(&forget_gate); inputs_vector_info_raw.push_back(&output_gate_tmp); - ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenateLayer::validate(inputs_vector_info_raw, scratch_buffer)); + ARM_COMPUTE_RETURN_ON_ERROR(CLConcatenateLayer::validate(inputs_vector_info_raw, scratch_buffer, Window::DimX)); return Status{}; } diff --git a/src/runtime/CL/functions/CLWidthConcatenateLayer.cpp b/src/runtime/CL/functions/CLWidthConcatenateLayer.cpp deleted file mode 100644 index a8667c3138..0000000000 --- a/src/runtime/CL/functions/CLWidthConcatenateLayer.cpp +++ /dev/null @@ -1,143 +0,0 @@ -/* - * Copyright (c) 2018-2019 ARM Limited. - * - * SPDX-License-Identifier: MIT - * - * Permission is hereby granted, free of charge, to any person obtaining a copy - * of this software and associated documentation files (the "Software"), to - * deal in the Software without restriction, including without limitation the - * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or - * sell copies of the Software, and to permit persons to whom the Software is - * furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in all - * copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE - * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, - * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE - * SOFTWARE. - */ -#include "arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h" - -#include "arm_compute/core/CL/ICLTensor.h" -#include "arm_compute/core/Error.h" -#include "arm_compute/core/Helpers.h" -#include "arm_compute/core/TensorInfo.h" -#include "arm_compute/core/Types.h" -#include "arm_compute/core/utils/misc/ShapeCalculator.h" -#include "arm_compute/runtime/CL/CLScheduler.h" -#include "support/ToolchainSupport.h" - -using namespace arm_compute; - -CLWidthConcatenateLayer::CLWidthConcatenateLayer() // NOLINT - : _concat_kernels_vector(), - _concat_x2_kernel(), - _concat_x4_kernel(), - _num_inputs(0) -{ -} - -Status CLWidthConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vector, const ITensorInfo *output) // NOLINT -{ - const unsigned int num_inputs = inputs_vector.size(); - - ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output); - ARM_COMPUTE_RETURN_ERROR_ON(num_inputs < 2); - - // Output auto inizialitation if not yet initialized - TensorInfo tmp_output_info = *output->clone(); - const TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, Window::DimX); - auto_init_if_empty(tmp_output_info, output_shape, 1, inputs_vector[0]->data_type()); - - switch(num_inputs) - { - case 2: - // Validate WidthConcatenate2Tensors kernels if there are 2 inputs - ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(inputs_vector[0], inputs_vector[1]); - ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenate2TensorsKernel::validate(inputs_vector[0], inputs_vector[1], &tmp_output_info)); - break; - case 4: - // Validate WidthConcatenate4Tensors kernels if there are 4 inputs - ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(inputs_vector[0], inputs_vector[1], inputs_vector[2], inputs_vector[3]); - ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenate4TensorsKernel::validate(inputs_vector[0], inputs_vector[1], inputs_vector[2], inputs_vector[3], &tmp_output_info)); - break; - default: - unsigned int width_offset = 0; - // Validate generic case of WidthConcatenate kernel - for(const auto &input : inputs_vector) - { - ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input); - ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenateLayerKernel::validate(input, width_offset, &tmp_output_info)); - width_offset += input->dimension(0); - } - break; - } - - return Status{}; -} - -void CLWidthConcatenateLayer::configure(std::vector<ICLTensor *> inputs_vector, ICLTensor *output) // NOLINT -{ - _num_inputs = inputs_vector.size(); - - std::vector<ITensorInfo *> inputs_vector_info; - for(unsigned int i = 0; i < _num_inputs; i++) - { - inputs_vector_info.emplace_back(inputs_vector.at(i)->info()); - } - const TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, Window::DimX); - - // Output auto inizialitation if not yet initialized - auto_init_if_empty(*output->info(), output_shape, 1, inputs_vector[0]->info()->data_type()); - - ARM_COMPUTE_ERROR_THROW_ON(CLWidthConcatenateLayer::validate(inputs_vector_info, output->info())); - - switch(_num_inputs) - { - case 2: - // Configure WidthConcatenate2Tensors kernel - _concat_x2_kernel.configure(inputs_vector.at(0), inputs_vector.at(1), output); - break; - case 4: - // Configure WidthConcatenate4Tensors kernel - _concat_x4_kernel.configure(inputs_vector.at(0), inputs_vector.at(1), inputs_vector.at(2), inputs_vector.at(3), output); - break; - default: - // Configure generic case WidthConcatenate kernels - _concat_kernels_vector.resize(_num_inputs); - - unsigned int width_offset = 0; - for(unsigned int i = 0; i < _num_inputs; ++i) - { - _concat_kernels_vector[i].configure(inputs_vector.at(i), width_offset, output); - width_offset += inputs_vector.at(i)->info()->dimension(0); - } - break; - } -} - -void CLWidthConcatenateLayer::run() -{ - cl::CommandQueue q = CLScheduler::get().queue(); - - switch(_num_inputs) - { - case 2: - CLScheduler::get().enqueue(_concat_x2_kernel, true); - break; - case 4: - CLScheduler::get().enqueue(_concat_x4_kernel, true); - break; - default: - for(unsigned int i = 0; i < _num_inputs; ++i) - { - CLScheduler::get().enqueue(_concat_kernels_vector[i], true); - } - break; - } -} |