diff options
Diffstat (limited to 'libs/ARMComputeEx/src/core/CL/kernels/CLArithmeticSubtractionExKernel.cpp')
-rw-r--r-- | libs/ARMComputeEx/src/core/CL/kernels/CLArithmeticSubtractionExKernel.cpp | 216 |
1 files changed, 0 insertions, 216 deletions
diff --git a/libs/ARMComputeEx/src/core/CL/kernels/CLArithmeticSubtractionExKernel.cpp b/libs/ARMComputeEx/src/core/CL/kernels/CLArithmeticSubtractionExKernel.cpp deleted file mode 100644 index 1c505b4d5..000000000 --- a/libs/ARMComputeEx/src/core/CL/kernels/CLArithmeticSubtractionExKernel.cpp +++ /dev/null @@ -1,216 +0,0 @@ -/* - * Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved - * Copyright (c) 2016-2018 ARM Limited. - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ -#include "arm_compute/core/CL/kernels/CLArithmeticSubtractionExKernel.h" - -#include "arm_compute/core/CL/CLHelpers.h" -#include "arm_compute/core/CL/CLKernelLibraryEx.h" -#include "arm_compute/core/CL/ICLTensor.h" - -using namespace arm_compute; - -namespace -{ -constexpr unsigned int num_elems_processed_per_iteration = 16; - -Status validate_arguments(const ITensorInfo *input1, const ITensorInfo *input2, - const ITensorInfo *output, ConvertPolicy policy) -{ - ARM_COMPUTE_UNUSED(policy); - - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::U8, DataType::S16, - DataType::F16, DataType::F32); - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input2, 1, DataType::U8, DataType::S16, - DataType::F16, DataType::F32); - - const TensorShape &out_shape = - TensorShape::broadcast_shape(input1->tensor_shape(), input2->tensor_shape()); - - ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, - "Inputs are not broadcast compatible"); - - // Validate in case of configured output - if (output->total_size() > 0) - { - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8, DataType::S16, - DataType::F16, DataType::F32); - ARM_COMPUTE_RETURN_ERROR_ON_MSG( - output->data_type() == DataType::U8 && - (input1->data_type() != DataType::U8 || input2->data_type() != DataType::U8), - "Output can only be U8 if both inputs are U8"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG( - detail::have_different_dimensions(out_shape, output->tensor_shape(), 0), - "Wrong shape for output"); - } - - return Status{}; -} - -std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input1, ITensorInfo *input2, - ITensorInfo *output) -{ - const std::pair<TensorShape, ValidRegion> broadcast_pair = - ITensorInfo::broadcast_shape_and_valid_region(*input1, *input2); - const TensorShape &out_shape = broadcast_pair.first; - const ValidRegion &valid_region = broadcast_pair.second; - - // Auto initialize output if not initialized - { - set_shape_if_empty(*output, out_shape); - - if (input1->data_type() == DataType::S16 || input2->data_type() == DataType::S16) - { - set_format_if_unknown(*output, Format::S16); - } - else if (input1->data_type() == DataType::F16 && input2->data_type() == DataType::F16) - { - set_format_if_unknown(*output, Format::F16); - } - else if (input1->data_type() == DataType::F32 || input2->data_type() == DataType::F32) - { - set_format_if_unknown(*output, Format::F32); - } - } - - Window win = calculate_max_window(valid_region, Steps(num_elems_processed_per_iteration)); - Window win_input1 = win.broadcast_if_dimension_le_one(*input1); - Window win_input2 = win.broadcast_if_dimension_le_one(*input2); - - AccessWindowHorizontal input1_access(input1, 0, num_elems_processed_per_iteration); - AccessWindowHorizontal input2_access(input2, 0, num_elems_processed_per_iteration); - AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration); - - bool window_changed = update_window_and_padding(win_input1, input1_access) || - update_window_and_padding(win_input2, input2_access) || - update_window_and_padding(win, output_access); - - output_access.set_valid_region(win, valid_region); - - Status err = (window_changed) - ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") - : Status{}; - return std::make_pair(err, win); -} -} // namespace - -CLArithmeticSubtractionExKernel::CLArithmeticSubtractionExKernel() - : _input1(nullptr), _input2(nullptr), _output(nullptr) -{ -} - -void CLArithmeticSubtractionExKernel::configure(const ICLTensor *input1, const ICLTensor *input2, - ICLTensor *output, ConvertPolicy policy) -{ - ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output); - ARM_COMPUTE_ERROR_THROW_ON( - validate_arguments(input1->info(), input2->info(), output->info(), policy)); - - // Configure kernel window - auto win_config = validate_and_configure_window(input1->info(), input2->info(), output->info()); - ARM_COMPUTE_ERROR_THROW_ON(win_config.first); - - _input1 = input1; - _input2 = input2; - _output = output; - - const bool has_float_out = is_data_type_float(output->info()->data_type()); - - // Set kernel build options - std::set<std::string> build_opts; - build_opts.emplace((policy == ConvertPolicy::WRAP || has_float_out) ? "-DWRAP" : "-DSATURATE"); - build_opts.emplace("-DDATA_TYPE_IN1=" + get_cl_type_from_data_type(input1->info()->data_type())); - build_opts.emplace("-DDATA_TYPE_IN2=" + get_cl_type_from_data_type(input2->info()->data_type())); - build_opts.emplace("-DDATA_TYPE_OUT=" + get_cl_type_from_data_type(output->info()->data_type())); - - // Create kernel - _kernel = static_cast<cl::Kernel>( - CLKernelLibraryEx::get().create_kernel("arithmetic_sub_ex", build_opts)); - - ICLKernel::configure_internal(win_config.second); -} - -Status CLArithmeticSubtractionExKernel::validate(const ITensorInfo *input1, - const ITensorInfo *input2, - const ITensorInfo *output, ConvertPolicy policy) -{ - ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output); - ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input1, input2, output, policy)); - ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input1->clone().get(), - input2->clone().get(), - output->clone().get()) - .first); - - return Status{}; -} - -void CLArithmeticSubtractionExKernel::run(const Window &window, cl::CommandQueue &queue) -{ - ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); - ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window); - - const TensorShape &in_shape1 = _input1->info()->tensor_shape(); - const TensorShape &in_shape2 = _input2->info()->tensor_shape(); - const TensorShape &out_shape = _output->info()->tensor_shape(); - - bool can_collapse = true; - if (std::min(in_shape1.total_size(), in_shape2.total_size()) > 1) - { - can_collapse = - (std::min(in_shape1.num_dimensions(), in_shape2.num_dimensions()) > Window::DimZ); - for (size_t d = Window::DimZ; can_collapse && (d < out_shape.num_dimensions()); d++) - { - can_collapse = (in_shape1[d] == in_shape2[d]); - } - } - - bool has_collapsed = false; - Window collapsed = - can_collapse ? window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &has_collapsed) - : window; - - const TensorShape &in_shape1_collapsed = - has_collapsed ? in_shape1.collapsed_from(Window::DimZ) : in_shape1; - const TensorShape &in_shape2_collapsed = - has_collapsed ? in_shape2.collapsed_from(Window::DimZ) : in_shape2; - - Window slice = collapsed.first_slice_window_3D(); - Window slice_input1 = slice.broadcast_if_dimension_le_one(in_shape1_collapsed); - Window slice_input2 = slice.broadcast_if_dimension_le_one(in_shape2_collapsed); - - do - { - unsigned int idx = 0; - - add_3D_tensor_argument(idx, _input1, slice_input1); - add_3D_tensor_argument(idx, _input2, slice_input2); - add_3D_tensor_argument(idx, _output, slice); - - enqueue(queue, *this, slice); - - collapsed.slide_window_slice_3D(slice_input1); - collapsed.slide_window_slice_3D(slice_input2); - } while (collapsed.slide_window_slice_3D(slice)); -} - -BorderSize CLArithmeticSubtractionExKernel::border_size() const -{ - const unsigned int replicateSize = - _output->info()->dimension(0) - - std::min(_input1->info()->dimension(0), _input2->info()->dimension(0)); - const unsigned int border = - std::min<unsigned int>(num_elems_processed_per_iteration - 1U, replicateSize); - return BorderSize(0, border, 0, 0); -} |