diff options
Diffstat (limited to 'src/runtime/CL')
25 files changed, 124 insertions, 119 deletions
diff --git a/src/runtime/CL/CLHelpers.cpp b/src/runtime/CL/CLHelpers.cpp index 533e6fabfa..8bc7b8eb7b 100644 --- a/src/runtime/CL/CLHelpers.cpp +++ b/src/runtime/CL/CLHelpers.cpp @@ -47,7 +47,7 @@ void printf_callback(const char *buffer, unsigned int len, size_t complete, void * @return A pointer to the context properties which can be used to create an opencl context */ -void initialise_context_properties(const cl::Platform &platform, const cl::Device &device, cl_context_properties prop[7]) +void initialise_context_properties(const cl::Platform &platform, const cl::Device &device, std::array<cl_context_properties, 7> &prop) { ARM_COMPUTE_UNUSED(device); #if defined(ARM_COMPUTE_ASSERTS_ENABLED) @@ -55,7 +55,7 @@ void initialise_context_properties(const cl::Platform &platform, const cl::Devic if(arm_compute::device_supports_extension(device, "cl_arm_printf")) { // Create a cl_context with a printf_callback and user specified buffer size. - cl_context_properties properties_printf[] = + std::array<cl_context_properties, 7> properties_printf = { CL_CONTEXT_PLATFORM, reinterpret_cast<cl_context_properties>(platform()), // Enable a printf callback function for this context. @@ -65,17 +65,17 @@ void initialise_context_properties(const cl::Platform &platform, const cl::Devic CL_PRINTF_BUFFERSIZE_ARM, 0x1000, 0 }; - std::copy_n(properties_printf, 7, prop); + prop = properties_printf; } else #endif // defined(ARM_COMPUTE_ASSERTS_ENABLED) { - cl_context_properties properties[] = + std::array<cl_context_properties, 3> properties = { CL_CONTEXT_PLATFORM, reinterpret_cast<cl_context_properties>(platform()), 0 }; - std::copy_n(properties, 3, prop); + std::copy(properties.begin(), properties.end(), prop.begin()); }; } } //namespace @@ -94,11 +94,11 @@ create_opencl_context_and_device() std::vector<cl::Device> platform_devices; p.getDevices(CL_DEVICE_TYPE_DEFAULT, &platform_devices); ARM_COMPUTE_ERROR_ON_MSG(platform_devices.size() == 0, "Couldn't find any OpenCL device"); - device = platform_devices[0]; - cl_int err = CL_SUCCESS; - cl_context_properties properties[7] = { 0, 0, 0, 0, 0, 0, 0 }; + device = platform_devices[0]; + cl_int err = CL_SUCCESS; + std::array<cl_context_properties, 7> properties = { 0, 0, 0, 0, 0, 0, 0 }; initialise_context_properties(p, device, properties); - cl::Context cl_context = cl::Context(device, properties, nullptr, nullptr, &err); + cl::Context cl_context = cl::Context(device, properties.data(), nullptr, nullptr, &err); ARM_COMPUTE_ERROR_ON_MSG(err != CL_SUCCESS, "Failed to create OpenCL context"); return std::make_tuple(cl_context, device, err); } diff --git a/src/runtime/CL/CLMemory.cpp b/src/runtime/CL/CLMemory.cpp index 5bea85cfae..557378b6f1 100644 --- a/src/runtime/CL/CLMemory.cpp +++ b/src/runtime/CL/CLMemory.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018 ARM Limited. + * Copyright (c) 2018-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -33,8 +33,8 @@ CLMemory::CLMemory() { } -CLMemory::CLMemory(std::shared_ptr<ICLMemoryRegion> memory) - : _region(nullptr), _region_owned(std::move(memory)) +CLMemory::CLMemory(const std::shared_ptr<ICLMemoryRegion> &memory) + : _region(nullptr), _region_owned(memory) { _region_owned = memory; _region = _region_owned.get(); diff --git a/src/runtime/CL/CLMultiHOG.cpp b/src/runtime/CL/CLMultiHOG.cpp index 88d45acd12..2577ec08ac 100644 --- a/src/runtime/CL/CLMultiHOG.cpp +++ b/src/runtime/CL/CLMultiHOG.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017 ARM Limited. + * Copyright (c) 2017-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -30,8 +30,9 @@ using namespace arm_compute; CLMultiHOG::CLMultiHOG(size_t num_models) - : _num_models(num_models), _model(arm_compute::support::cpp14::make_unique<CLHOG[]>(_num_models)) + : _num_models(num_models), _model() { + _model.resize(_num_models); } size_t CLMultiHOG::num_models() const @@ -42,11 +43,11 @@ size_t CLMultiHOG::num_models() const ICLHOG *CLMultiHOG::cl_model(size_t index) { ARM_COMPUTE_ERROR_ON(index >= _num_models); - return (_model.get() + index); + return (&_model[index]); } const ICLHOG *CLMultiHOG::cl_model(size_t index) const { ARM_COMPUTE_ERROR_ON(index >= _num_models); - return (_model.get() + index); + return (&_model[index]); }
\ No newline at end of file diff --git a/src/runtime/CL/CLPyramid.cpp b/src/runtime/CL/CLPyramid.cpp index 865f389f7f..6d5dba0031 100644 --- a/src/runtime/CL/CLPyramid.cpp +++ b/src/runtime/CL/CLPyramid.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016, 2017 ARM Limited. + * Copyright (c) 2016-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -35,7 +35,7 @@ using namespace arm_compute; CLPyramid::CLPyramid() - : _info(), _pyramid(nullptr) + : _info(), _pyramid() { } @@ -51,8 +51,8 @@ void CLPyramid::init_auto_padding(const PyramidInfo &info) void CLPyramid::internal_init(const PyramidInfo &info, bool auto_padding) { - _info = info; - _pyramid = arm_compute::support::cpp14::make_unique<CLTensor[]>(_info.num_levels()); + _info = info; + _pyramid.resize(_info.num_levels()); size_t w = _info.width(); size_t h = _info.height(); @@ -109,11 +109,9 @@ void CLPyramid::internal_init(const PyramidInfo &info, bool auto_padding) void CLPyramid::allocate() { - ARM_COMPUTE_ERROR_ON(_pyramid == nullptr); - for(size_t i = 0; i < _info.num_levels(); ++i) { - (_pyramid.get() + i)->allocator()->allocate(); + _pyramid[i].allocator()->allocate(); } } @@ -126,5 +124,5 @@ CLTensor *CLPyramid::get_pyramid_level(size_t index) const { ARM_COMPUTE_ERROR_ON(index >= _info.num_levels()); - return (_pyramid.get() + index); + return &_pyramid[index]; } diff --git a/src/runtime/CL/CLTensorAllocator.cpp b/src/runtime/CL/CLTensorAllocator.cpp index 2ce64551ae..101e4f1cd4 100644 --- a/src/runtime/CL/CLTensorAllocator.cpp +++ b/src/runtime/CL/CLTensorAllocator.cpp @@ -34,7 +34,7 @@ const cl::Buffer CLTensorAllocator::_empty_buffer = cl::Buffer(); namespace { -std::unique_ptr<ICLMemoryRegion> allocate_region(cl::Context context, size_t size, cl_uint alignment) +std::unique_ptr<ICLMemoryRegion> allocate_region(const cl::Context &context, size_t size, cl_uint alignment) { // Try fine-grain SVM std::unique_ptr<ICLMemoryRegion> region = support::cpp14::make_unique<CLFineSVMMemoryRegion>(context, diff --git a/src/runtime/CL/CLTuner.cpp b/src/runtime/CL/CLTuner.cpp index 8f8d3e7c3a..929def24cc 100644 --- a/src/runtime/CL/CLTuner.cpp +++ b/src/runtime/CL/CLTuner.cpp @@ -275,7 +275,7 @@ void CLTuner::save_to_file(const std::string &filename) const std::ofstream fs; fs.exceptions(std::ifstream::failbit | std::ifstream::badbit); fs.open(filename, std::ios::out); - for(auto kernel_data : _lws_table) + for(auto const &kernel_data : _lws_table) { fs << kernel_data.first << ";" << kernel_data.second[0] << ";" << kernel_data.second[1] << ";" << kernel_data.second[2] << std::endl; } diff --git a/src/runtime/CL/functions/CLConvolution.cpp b/src/runtime/CL/functions/CLConvolution.cpp index 2f43ce1974..f09585e7ec 100644 --- a/src/runtime/CL/functions/CLConvolution.cpp +++ b/src/runtime/CL/functions/CLConvolution.cpp @@ -58,13 +58,13 @@ void CLConvolutionSquare<matrix_size>::configure(ICLTensor *input, ICLTensor *ou { ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8); ARM_COMPUTE_ERROR_ON(conv == nullptr); - int16_t conv_col[matrix_size]; - int16_t conv_row[matrix_size]; - _is_separable = separate_matrix(conv, conv_col, conv_row, matrix_size); + std::array<int16_t, matrix_size> conv_col{ 0 }; + std::array<int16_t, matrix_size> conv_row{ 0 }; + _is_separable = separate_matrix(conv, conv_col.data(), conv_row.data(), matrix_size); if(_is_separable) { - std::pair<DataType, DataType> type_pair = data_type_for_convolution(conv_col, conv_row, matrix_size); + std::pair<DataType, DataType> type_pair = data_type_for_convolution(conv_col.data(), conv_row.data(), matrix_size); _tmp.allocator()->init(TensorInfo(input->info()->tensor_shape(), 1, type_pair.first)); // Manage intermediate buffers @@ -75,8 +75,8 @@ void CLConvolutionSquare<matrix_size>::configure(ICLTensor *input, ICLTensor *ou scale = calculate_matrix_scale(conv, matrix_size); } - _kernel_hor.configure(input, &_tmp, conv_row, border_mode == BorderMode::UNDEFINED); - _kernel_vert.configure(&_tmp, output, conv_col, scale, border_mode == BorderMode::UNDEFINED, type_pair.second); + _kernel_hor.configure(input, &_tmp, conv_row.data(), border_mode == BorderMode::UNDEFINED); + _kernel_vert.configure(&_tmp, output, conv_col.data(), scale, border_mode == BorderMode::UNDEFINED, type_pair.second); _border_handler.configure(input, _kernel_hor.border_size(), border_mode, PixelValue(constant_border_value)); // Allocate intermediate buffer diff --git a/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp b/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp index 4a5f845631..f687e54552 100644 --- a/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp +++ b/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp @@ -36,8 +36,7 @@ using namespace arm_compute; CLDepthConcatenateLayer::CLDepthConcatenateLayer() // NOLINT - : _inputs_vector(), - _concat_kernels_vector(), + : _concat_kernels_vector(), _border_handlers_vector(), _num_inputs(0) { @@ -53,8 +52,8 @@ void CLDepthConcatenateLayer::configure(const std::vector<ICLTensor *> &inputs_v inputs_vector_info.emplace_back(inputs_vector.at(i)->info()); } - _concat_kernels_vector = arm_compute::support::cpp14::make_unique<CLDepthConcatenateLayerKernel[]>(_num_inputs); - _border_handlers_vector = arm_compute::support::cpp14::make_unique<CLFillBorderKernel[]>(_num_inputs); + _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); diff --git a/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp b/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp index 8211104bda..97b0a01331 100644 --- a/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp @@ -322,7 +322,8 @@ void CLDepthwiseConvolutionLayer::configure(ICLTensor *input, const ICLTensor *w const QuantizationInfo output_quant_info = (output->info()->total_size() == 0) ? input->info()->quantization_info() : output->info()->quantization_info(); float multiplier = input->info()->quantization_info().scale * weights->info()->quantization_info().scale / output_quant_info.scale; - int output_multiplier, output_shift; + int output_multiplier; + int output_shift; quantization::calculate_quantized_multiplier_less_than_one(multiplier, &output_multiplier, &output_shift); _output_stage_kernel.configure(&_output_reshaped, biases, output, output_multiplier, output_shift, output_quant_info.offset); _output_reshaped.allocator()->allocate(); diff --git a/src/runtime/CL/functions/CLFFT1D.cpp b/src/runtime/CL/functions/CLFFT1D.cpp index 67111e7e5c..49b5a2a2e6 100644 --- a/src/runtime/CL/functions/CLFFT1D.cpp +++ b/src/runtime/CL/functions/CLFFT1D.cpp @@ -62,7 +62,7 @@ void CLFFT1D::configure(const ICLTensor *input, ICLTensor *output, const FFT1DIn // Create and configure FFT kernels unsigned int Nx = 1; _num_ffts = decomposed_vector.size(); - _fft_kernels = arm_compute::support::cpp14::make_unique<CLFFTRadixStageKernel[]>(_num_ffts); + _fft_kernels.resize(_num_ffts); for(unsigned int i = 0; i < _num_ffts; ++i) { const unsigned int radix_for_stage = decomposed_vector.at(i); diff --git a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp index 40ce6b4e0f..03d516f703 100644 --- a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp @@ -372,7 +372,9 @@ Status CLGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI const unsigned int kernel_width = weights->dimension(idx_width); const unsigned int kernel_height = weights->dimension(idx_height); - TensorInfo im2col_reshaped_info, info_gemm, weights_reshaped_info; + TensorInfo im2col_reshaped_info{}; + TensorInfo info_gemm{}; + TensorInfo weights_reshaped_info{}; const ITensorInfo *gemm_input_to_use = input; const ITensorInfo *gemm_output_to_use = output; const ITensorInfo *weights_to_use = weights; diff --git a/src/runtime/CL/functions/CLGaussianPyramid.cpp b/src/runtime/CL/functions/CLGaussianPyramid.cpp index fd82769004..b671b23c87 100644 --- a/src/runtime/CL/functions/CLGaussianPyramid.cpp +++ b/src/runtime/CL/functions/CLGaussianPyramid.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -76,10 +76,10 @@ void CLGaussianPyramidHalf::configure(ICLTensor *input, CLPyramid *pyramid, Bord if(num_levels > 1) { - _horizontal_border_handler = arm_compute::support::cpp14::make_unique<CLFillBorderKernel[]>(num_levels - 1); - _vertical_border_handler = arm_compute::support::cpp14::make_unique<CLFillBorderKernel[]>(num_levels - 1); - _horizontal_reduction = arm_compute::support::cpp14::make_unique<CLGaussianPyramidHorKernel[]>(num_levels - 1); - _vertical_reduction = arm_compute::support::cpp14::make_unique<CLGaussianPyramidVertKernel[]>(num_levels - 1); + _horizontal_border_handler.resize(num_levels - 1); + _vertical_border_handler.resize(num_levels - 1); + _horizontal_reduction.resize(num_levels - 1); + _vertical_reduction.resize(num_levels - 1); // Apply half scale to the X dimension of the tensor shape TensorShape tensor_shape = pyramid->info()->tensor_shape(); @@ -153,8 +153,8 @@ void CLGaussianPyramidOrb::configure(ICLTensor *input, CLPyramid *pyramid, Borde if(num_levels > 1) { - _gauss5x5 = arm_compute::support::cpp14::make_unique<CLGaussian5x5[]>(num_levels - 1); - _scale_nearest = arm_compute::support::cpp14::make_unique<CLScaleKernel[]>(num_levels - 1); + _gauss5x5.resize(num_levels - 1); + _scale_nearest.resize(num_levels - 1); PyramidInfo pyramid_info(num_levels - 1, SCALE_PYRAMID_ORB, pyramid->info()->tensor_shape(), Format::U8); diff --git a/src/runtime/CL/functions/CLHOGMultiDetection.cpp b/src/runtime/CL/functions/CLHOGMultiDetection.cpp index 0865f50fd5..f799d61b16 100644 --- a/src/runtime/CL/functions/CLHOGMultiDetection.cpp +++ b/src/runtime/CL/functions/CLHOGMultiDetection.cpp @@ -128,12 +128,11 @@ void CLHOGMultiDetection::configure(ICLTensor *input, const ICLMultiHOG *multi_h _num_block_norm_kernel = input_block_norm.size(); // Number of CLHOGBlockNormalizationKernel kernels to compute _num_hog_detect_kernel = input_hog_detect.size(); // Number of CLHOGDetector functions to compute - _orient_bin_kernel = arm_compute::support::cpp14::make_unique<CLHOGOrientationBinningKernel[]>(_num_orient_bin_kernel); - _block_norm_kernel = arm_compute::support::cpp14::make_unique<CLHOGBlockNormalizationKernel[]>(_num_block_norm_kernel); - _hog_detect_kernel = arm_compute::support::cpp14::make_unique<CLHOGDetector[]>(_num_hog_detect_kernel); - _non_maxima_kernel = arm_compute::support::cpp14::make_unique<CPPDetectionWindowNonMaximaSuppressionKernel>(); - _hog_space = arm_compute::support::cpp14::make_unique<CLTensor[]>(_num_orient_bin_kernel); - _hog_norm_space = arm_compute::support::cpp14::make_unique<CLTensor[]>(_num_block_norm_kernel); + _orient_bin_kernel.resize(_num_orient_bin_kernel); + _block_norm_kernel.resize(_num_block_norm_kernel); + _hog_detect_kernel.resize(_num_hog_detect_kernel); + _hog_space.resize(_num_orient_bin_kernel); + _hog_norm_space.resize(_num_block_norm_kernel); // Allocate tensors for magnitude and phase TensorInfo info_mag(shape_img, Format::S16); @@ -172,10 +171,10 @@ void CLHOGMultiDetection::configure(ICLTensor *input, const ICLMultiHOG *multi_h _hog_space[i].allocator()->init(info_space); // Manage intermediate buffers - _memory_group.manage(_hog_space.get() + i); + _memory_group.manage(&_hog_space[i]); // Initialise orientation binning kernel - _orient_bin_kernel[i].configure(&_mag, &_phase, _hog_space.get() + i, multi_hog->model(idx_multi_hog)->info()); + _orient_bin_kernel[i].configure(&_mag, &_phase, &_hog_space[i], multi_hog->model(idx_multi_hog)->info()); } // Allocate intermediate tensors @@ -193,10 +192,10 @@ void CLHOGMultiDetection::configure(ICLTensor *input, const ICLMultiHOG *multi_h _hog_norm_space[i].allocator()->init(tensor_info); // Manage intermediate buffers - _memory_group.manage(_hog_norm_space.get() + i); + _memory_group.manage(&_hog_norm_space[i]); // Initialize block normalization kernel - _block_norm_kernel[i].configure(_hog_space.get() + idx_orient_bin, _hog_norm_space.get() + i, multi_hog->model(idx_multi_hog)->info()); + _block_norm_kernel[i].configure(&_hog_space[idx_orient_bin], &_hog_norm_space[i], multi_hog->model(idx_multi_hog)->info()); } // Allocate intermediate tensors @@ -212,13 +211,13 @@ void CLHOGMultiDetection::configure(ICLTensor *input, const ICLMultiHOG *multi_h { const size_t idx_block_norm = input_hog_detect[i]; - _hog_detect_kernel[i].configure(_hog_norm_space.get() + idx_block_norm, multi_hog->cl_model(i), detection_windows, detection_window_strides->at(i), threshold, i); + _hog_detect_kernel[i].configure(&_hog_norm_space[idx_block_norm], multi_hog->cl_model(i), detection_windows, detection_window_strides->at(i), threshold, i); } detection_window_strides->unmap(CLScheduler::get().queue()); // Configure non maxima suppression kernel - _non_maxima_kernel->configure(_detection_windows, min_distance); + _non_maxima_kernel.configure(_detection_windows, min_distance); // Allocate intermediate tensors for(size_t i = 0; i < _num_block_norm_kernel; ++i) @@ -242,13 +241,13 @@ void CLHOGMultiDetection::run() // Run orientation binning kernel for(size_t i = 0; i < _num_orient_bin_kernel; ++i) { - CLScheduler::get().enqueue(*(_orient_bin_kernel.get() + i), false); + CLScheduler::get().enqueue(_orient_bin_kernel[i], false); } // Run block normalization kernel for(size_t i = 0; i < _num_block_norm_kernel; ++i) { - CLScheduler::get().enqueue(*(_block_norm_kernel.get() + i), false); + CLScheduler::get().enqueue(_block_norm_kernel[i], false); } // Run HOG detector kernel @@ -262,7 +261,7 @@ void CLHOGMultiDetection::run() { // Map detection windows array before computing non maxima suppression _detection_windows->map(CLScheduler::get().queue(), true); - Scheduler::get().schedule(_non_maxima_kernel.get(), Window::DimY); + Scheduler::get().schedule(&_non_maxima_kernel, Window::DimY); _detection_windows->unmap(CLScheduler::get().queue()); } } diff --git a/src/runtime/CL/functions/CLHarrisCorners.cpp b/src/runtime/CL/functions/CLHarrisCorners.cpp index 342d1cad49..67f550d318 100644 --- a/src/runtime/CL/functions/CLHarrisCorners.cpp +++ b/src/runtime/CL/functions/CLHarrisCorners.cpp @@ -55,7 +55,7 @@ CLHarrisCorners::CLHarrisCorners(std::shared_ptr<IMemoryManager> memory_manager) _gy(), _score(), _nonmax(), - _corners_list(nullptr), + _corners_list(), _num_corner_candidates(0), _corners(nullptr) { @@ -84,7 +84,7 @@ void CLHarrisCorners::configure(ICLImage *input, float threshold, float min_dist _score.allocator()->init(info_f32); _nonmax.allocator()->init(info_f32); - _corners_list = arm_compute::support::cpp14::make_unique<InternalKeypoint[]>(shape.x() * shape.y()); + _corners_list.resize(shape.x() * shape.y()); // Manage intermediate buffers _memory_group.manage(&_gx); @@ -146,13 +146,13 @@ void CLHarrisCorners::configure(ICLImage *input, float threshold, float min_dist _score.allocator()->allocate(); // Init corner candidates kernel - _candidates.configure(&_nonmax, _corners_list.get(), &_num_corner_candidates); + _candidates.configure(&_nonmax, _corners_list.data(), &_num_corner_candidates); // Allocate intermediate buffers _nonmax.allocator()->allocate(); // Init euclidean distance - _sort_euclidean.configure(_corners_list.get(), _corners, &_num_corner_candidates, min_dist); + _sort_euclidean.configure(_corners_list.data(), _corners, &_num_corner_candidates, min_dist); } void CLHarrisCorners::run() diff --git a/src/runtime/CL/functions/CLLaplacianPyramid.cpp b/src/runtime/CL/functions/CLLaplacianPyramid.cpp index 559b57fd8d..a11851898c 100644 --- a/src/runtime/CL/functions/CLLaplacianPyramid.cpp +++ b/src/runtime/CL/functions/CLLaplacianPyramid.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -70,8 +70,8 @@ void CLLaplacianPyramid::configure(ICLTensor *input, CLPyramid *pyramid, ICLTens // Create Gaussian Pyramid function _gaussian_pyr_function.configure(input, &_gauss_pyr, border_mode, constant_border_value); - _convf = arm_compute::support::cpp14::make_unique<CLGaussian5x5[]>(_num_levels); - _subf = arm_compute::support::cpp14::make_unique<CLArithmeticSubtraction[]>(_num_levels); + _convf.resize(_num_levels); + _subf.resize(_num_levels); for(unsigned int i = 0; i < _num_levels; ++i) { diff --git a/src/runtime/CL/functions/CLLaplacianReconstruct.cpp b/src/runtime/CL/functions/CLLaplacianReconstruct.cpp index 911c9b3b27..13116bf08d 100644 --- a/src/runtime/CL/functions/CLLaplacianReconstruct.cpp +++ b/src/runtime/CL/functions/CLLaplacianReconstruct.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -63,8 +63,8 @@ void CLLaplacianReconstruct::configure(const CLPyramid *pyramid, ICLTensor *inpu _tmp_pyr.init(pyramid_info); // Allocate add and scale functions. Level 0 does not need to be scaled. - _addf = arm_compute::support::cpp14::make_unique<CLArithmeticAddition[]>(num_levels); - _scalef = arm_compute::support::cpp14::make_unique<CLScale[]>(num_levels - 1); + _addf.resize(num_levels); + _scalef.resize(num_levels - 1); const size_t last_level = num_levels - 1; @@ -85,7 +85,7 @@ void CLLaplacianReconstruct::configure(const CLPyramid *pyramid, ICLTensor *inpu void CLLaplacianReconstruct::run() { - ARM_COMPUTE_ERROR_ON_MSG(_addf == nullptr, "Unconfigured function"); + ARM_COMPUTE_ERROR_ON_MSG(_addf.empty(), "Unconfigured function"); const size_t last_level = _tmp_pyr.info()->num_levels() - 1; diff --git a/src/runtime/CL/functions/CLOpticalFlow.cpp b/src/runtime/CL/functions/CLOpticalFlow.cpp index 7ef1c83d04..a013a1fe19 100644 --- a/src/runtime/CL/functions/CLOpticalFlow.cpp +++ b/src/runtime/CL/functions/CLOpticalFlow.cpp @@ -84,12 +84,12 @@ void CLOpticalFlow::configure(const CLPyramid *old_pyramid, const CLPyramid *new const int old_values_list_length = list_length * window_dimension * window_dimension; // Create kernels and tensors - _tracker_init_kernel = arm_compute::support::cpp14::make_unique<CLLKTrackerInitKernel[]>(_num_levels); - _tracker_stage0_kernel = arm_compute::support::cpp14::make_unique<CLLKTrackerStage0Kernel[]>(_num_levels); - _tracker_stage1_kernel = arm_compute::support::cpp14::make_unique<CLLKTrackerStage1Kernel[]>(_num_levels); - _func_scharr = arm_compute::support::cpp14::make_unique<CLScharr3x3[]>(_num_levels); - _scharr_gx = arm_compute::support::cpp14::make_unique<CLTensor[]>(_num_levels); - _scharr_gy = arm_compute::support::cpp14::make_unique<CLTensor[]>(_num_levels); + _tracker_init_kernel.resize(_num_levels); + _tracker_stage0_kernel.resize(_num_levels); + _tracker_stage1_kernel.resize(_num_levels); + _func_scharr.resize(_num_levels); + _scharr_gx.resize(_num_levels); + _scharr_gy.resize(_num_levels); // Create internal keypoint arrays _old_points_internal = arm_compute::support::cpp14::make_unique<CLLKInternalKeypointArray>(list_length); @@ -118,8 +118,8 @@ void CLOpticalFlow::configure(const CLPyramid *old_pyramid, const CLPyramid *new _scharr_gy[i].allocator()->init(tensor_info); // Manage intermediate buffers - _memory_group.manage(_scharr_gx.get() + i); - _memory_group.manage(_scharr_gy.get() + i); + _memory_group.manage(&_scharr_gx[i]); + _memory_group.manage(&_scharr_gy[i]); // Init Scharr kernel _func_scharr[i].configure(old_ith_input, &_scharr_gx[i], &_scharr_gy[i], border_mode, constant_border_value); diff --git a/src/runtime/CL/functions/CLPadLayer.cpp b/src/runtime/CL/functions/CLPadLayer.cpp index dba7f23f3b..99e312183a 100644 --- a/src/runtime/CL/functions/CLPadLayer.cpp +++ b/src/runtime/CL/functions/CLPadLayer.cpp @@ -31,7 +31,7 @@ namespace arm_compute { CLPadLayer::CLPadLayer() - : _copy_kernel(), _mode(), _padding(), _memset_kernel(), _num_dimensions(0), _slice_functions(nullptr), _concat_functions(nullptr), _slice_results(nullptr), _concat_results(nullptr) + : _copy_kernel(), _mode(), _padding(), _memset_kernel(), _num_dimensions(0), _slice_functions(), _concat_functions(), _slice_results(), _concat_results() { } @@ -67,11 +67,16 @@ void CLPadLayer::configure_reflect_symmetric_mode(ICLTensor *input, ICLTensor *o // Two strided slice functions will be required for each dimension padded as well as a // concatenate function and the tensors to hold the temporary results. - _slice_functions = arm_compute::support::cpp14::make_unique<CLStridedSlice[]>(2 * _num_dimensions); - _slice_results = arm_compute::support::cpp14::make_unique<CLTensor[]>(2 * _num_dimensions); - _concat_functions = arm_compute::support::cpp14::make_unique<CLConcatenateLayer[]>(_num_dimensions); - _concat_results = arm_compute::support::cpp14::make_unique<CLTensor[]>(_num_dimensions - 1); - Coordinates starts_before, ends_before, starts_after, ends_after, strides; + _slice_functions.resize(2 * _num_dimensions); + _slice_results.resize(2 * _num_dimensions); + _concat_functions.resize(_num_dimensions); + _concat_results.resize(_num_dimensions - 1); + + Coordinates starts_before{}; + Coordinates ends_before{}; + Coordinates starts_after{}; + Coordinates ends_after{}; + Coordinates strides{}; ICLTensor *prev = input; for(uint32_t i = 0; i < _num_dimensions; ++i) { diff --git a/src/runtime/CL/functions/CLReduceMean.cpp b/src/runtime/CL/functions/CLReduceMean.cpp index 702ce34a4d..15091f9066 100644 --- a/src/runtime/CL/functions/CLReduceMean.cpp +++ b/src/runtime/CL/functions/CLReduceMean.cpp @@ -40,10 +40,10 @@ void CLReduceMean::configure(ICLTensor *input, const Coordinates &reduction_axis { ARM_COMPUTE_ERROR_ON_NULLPTR(input); - _reduction_ops = reduction_axis.num_dimensions(); - _reduction_kernels = arm_compute::support::cpp14::make_unique<CLReductionOperation[]>(_reduction_ops); - _reduced_outs = arm_compute::support::cpp14::make_unique<CLTensor[]>(_reduction_ops - (keep_dims ? 1 : 0)); - _keep_dims = keep_dims; + _reduction_ops = reduction_axis.num_dimensions(); + _reduction_kernels.resize(_reduction_ops); + _reduced_outs.resize(_reduction_ops - (keep_dims ? 1 : 0)); + _keep_dims = keep_dims; Coordinates axis_local = reduction_axis; const int input_dims = input->info()->num_dimensions(); @@ -57,9 +57,9 @@ void CLReduceMean::configure(ICLTensor *input, const Coordinates &reduction_axis // Perform reduction for every axis for(unsigned int i = 0; i < _reduction_ops; ++i) { - TensorShape out_shape = i == 0 ? input->info()->tensor_shape() : (_reduced_outs.get() + i - 1)->info()->tensor_shape(); + TensorShape out_shape = i == 0 ? input->info()->tensor_shape() : (&_reduced_outs[i - 1])->info()->tensor_shape(); out_shape.set(axis_local[i], 1); - auto in = (i == 0) ? input : (_reduced_outs.get() + i - 1); + auto in = (i == 0) ? input : (&_reduced_outs[i - 1]); if(i == _reduction_ops - 1 && keep_dims) { @@ -68,8 +68,8 @@ void CLReduceMean::configure(ICLTensor *input, const Coordinates &reduction_axis else { _reduced_outs[i].allocator()->init(TensorInfo(out_shape, input->info()->num_channels(), input->info()->data_type(), input->info()->quantization_info())); - _memory_group.manage(_reduced_outs.get() + i); - _reduction_kernels[i].configure(in, _reduced_outs.get() + i, axis_local[i], ReductionOperation::MEAN_SUM); + _memory_group.manage(&_reduced_outs[i]); + _reduction_kernels[i].configure(in, &_reduced_outs[i], axis_local[i], ReductionOperation::MEAN_SUM); } } @@ -92,7 +92,7 @@ void CLReduceMean::configure(ICLTensor *input, const Coordinates &reduction_axis out_shape.remove_dimension(axis_local[i] - i); } auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(out_shape)); - _reshape.configure(_reduced_outs.get() + _reduction_ops - 1, output); + _reshape.configure(&_reduced_outs[_reduction_ops - 1], output); } } diff --git a/src/runtime/CL/functions/CLReductionOperation.cpp b/src/runtime/CL/functions/CLReductionOperation.cpp index bb285d7cc8..9f99d2db6f 100644 --- a/src/runtime/CL/functions/CLReductionOperation.cpp +++ b/src/runtime/CL/functions/CLReductionOperation.cpp @@ -71,7 +71,7 @@ Status CLReductionOperation::validate(const ITensorInfo *input, const ITensorInf else { // Create temporary tensor infos - auto sums_vector = arm_compute::support::cpp14::make_unique<TensorInfo[]>(num_of_stages - 1); + std::vector<TensorInfo> sums_vector(num_of_stages - 1); // Create intermediate tensor info TensorShape shape{ input->tensor_shape() }; @@ -110,17 +110,17 @@ Status CLReductionOperation::validate(const ITensorInfo *input, const ITensorInf } // Validate ReductionOperation only on first kernel - ARM_COMPUTE_RETURN_ON_ERROR(CLReductionOperationKernel::validate(input, sums_vector.get(), axis, first_kernel_op)); + ARM_COMPUTE_RETURN_ON_ERROR(CLReductionOperationKernel::validate(input, &sums_vector[0], axis, first_kernel_op)); // Validate ReductionOperation on intermediate stages for(unsigned int i = 1; i < num_of_stages - 1; ++i) { - ARM_COMPUTE_RETURN_ON_ERROR(CLReductionOperationKernel::validate(sums_vector.get() + i - 1, sums_vector.get() + i, axis, intermediate_kernel_op)); + ARM_COMPUTE_RETURN_ON_ERROR(CLReductionOperationKernel::validate(&sums_vector[i - 1], &sums_vector[i], axis, intermediate_kernel_op)); } // Validate ReductionOperation on the last stage const unsigned int last_stage = num_of_stages - 1; - ARM_COMPUTE_RETURN_ON_ERROR(CLReductionOperationKernel::validate(sums_vector.get() + last_stage - 1, output, axis, last_kernel_op, input->dimension(0))); + ARM_COMPUTE_RETURN_ON_ERROR(CLReductionOperationKernel::validate(&sums_vector[last_stage - 1], output, axis, last_kernel_op, input->dimension(0))); } return Status{}; @@ -133,7 +133,7 @@ void CLReductionOperation::configure(ICLTensor *input, ICLTensor *output, unsign _is_serial = is_data_type_quantized(input->info()->data_type()) || axis != 0; // Configure reduction operation kernels - _reduction_kernels_vector = arm_compute::support::cpp14::make_unique<CLReductionOperationKernel[]>(_num_of_stages); + _reduction_kernels_vector.resize(_num_of_stages); // Create temporary tensors if(_is_serial) @@ -142,8 +142,8 @@ void CLReductionOperation::configure(ICLTensor *input, ICLTensor *output, unsign } else { - _border_handlers_vector = arm_compute::support::cpp14::make_unique<CLFillBorderKernel[]>(_num_of_stages); - _results_vector = arm_compute::support::cpp14::make_unique<CLTensor[]>(_num_of_stages - 1); + _border_handlers_vector.resize(_num_of_stages); + _results_vector.resize(_num_of_stages - 1); TensorShape shape{ input->info()->tensor_shape() }; for(unsigned int i = 0; i < _num_of_stages - 1; i++) { @@ -152,7 +152,7 @@ void CLReductionOperation::configure(ICLTensor *input, ICLTensor *output, unsign } // Apply ReductionOperation only on first kernel - _memory_group.manage(_results_vector.get()); + _memory_group.manage(&_results_vector[0]); ReductionOperation first_kernel_op; ReductionOperation intermediate_kernel_op; @@ -183,23 +183,23 @@ void CLReductionOperation::configure(ICLTensor *input, ICLTensor *output, unsign ARM_COMPUTE_ERROR("Not supported"); } - _reduction_kernels_vector[0].configure(input, _results_vector.get(), axis, first_kernel_op); + _reduction_kernels_vector[0].configure(input, &_results_vector[0], axis, first_kernel_op); _border_handlers_vector[0].configure(input, _reduction_kernels_vector[0].border_size(), BorderMode::CONSTANT, pixelValue); // Apply ReductionOperation on intermediate stages for(unsigned int i = 1; i < _num_of_stages - 1; ++i) { - _memory_group.manage(_results_vector.get() + i); - _reduction_kernels_vector[i].configure(_results_vector.get() + i - 1, _results_vector.get() + i, axis, intermediate_kernel_op); - _border_handlers_vector[i].configure(_results_vector.get() + i - 1, _reduction_kernels_vector[i].border_size(), BorderMode::CONSTANT, pixelValue); + _memory_group.manage(&_results_vector[i]); + _reduction_kernels_vector[i].configure(&_results_vector[i - 1], &_results_vector[i], axis, intermediate_kernel_op); + _border_handlers_vector[i].configure(&_results_vector[i - 1], _reduction_kernels_vector[i].border_size(), BorderMode::CONSTANT, pixelValue); _results_vector[i - 1].allocator()->allocate(); } // Apply ReductionOperation on the last stage const unsigned int last_stage = _num_of_stages - 1; const unsigned int input_width = input->info()->dimension(0); - _reduction_kernels_vector[last_stage].configure(_results_vector.get() + last_stage - 1, output, axis, last_kernel_op, input_width); - _border_handlers_vector[last_stage].configure(_results_vector.get() + last_stage - 1, _reduction_kernels_vector[last_stage].border_size(), BorderMode::CONSTANT, pixelValue); + _reduction_kernels_vector[last_stage].configure(&_results_vector[last_stage - 1], output, axis, last_kernel_op, input_width); + _border_handlers_vector[last_stage].configure(&_results_vector[last_stage - 1], _reduction_kernels_vector[last_stage].border_size(), BorderMode::CONSTANT, pixelValue); _results_vector[last_stage - 1].allocator()->allocate(); } } diff --git a/src/runtime/CL/functions/CLSplit.cpp b/src/runtime/CL/functions/CLSplit.cpp index f0843517e7..8d37d538c8 100644 --- a/src/runtime/CL/functions/CLSplit.cpp +++ b/src/runtime/CL/functions/CLSplit.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018 ARM Limited. + * Copyright (c) 2018-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -42,8 +42,8 @@ CLSplit::CLSplit() void CLSplit::configure(const ICLTensor *input, const std::vector<ICLTensor *> &outputs, unsigned int axis) { // Create Slice functions - _num_outputs = outputs.size(); - _slice_functions = arm_compute::support::cpp14::make_unique<CLSlice[]>(_num_outputs); + _num_outputs = outputs.size(); + _slice_functions.resize(_num_outputs); // Get output shape const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_split_shape(input->info(), axis, _num_outputs); diff --git a/src/runtime/CL/functions/CLStackLayer.cpp b/src/runtime/CL/functions/CLStackLayer.cpp index 71327fead4..2700b49272 100644 --- a/src/runtime/CL/functions/CLStackLayer.cpp +++ b/src/runtime/CL/functions/CLStackLayer.cpp @@ -46,8 +46,8 @@ CLStackLayer::CLStackLayer() // NOLINT void CLStackLayer::configure(const std::vector<ICLTensor *> &input, int axis, ICLTensor *output) { - _num_inputs = input.size(); - _stack_kernels = arm_compute::support::cpp14::make_unique<CLStackLayerKernel[]>(_num_inputs); + _num_inputs = input.size(); + _stack_kernels.resize(_num_inputs); // Wrap around negative values const unsigned int axis_u = wrap_around(axis, static_cast<int>(input[0]->info()->num_dimensions() + 1)); diff --git a/src/runtime/CL/functions/CLUnstack.cpp b/src/runtime/CL/functions/CLUnstack.cpp index 428d09148b..eb1dd8cd44 100644 --- a/src/runtime/CL/functions/CLUnstack.cpp +++ b/src/runtime/CL/functions/CLUnstack.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018 ARM Limited. + * Copyright (c) 2018-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -74,7 +74,7 @@ void CLUnstack::configure(const ICLTensor *input, const std::vector<ICLTensor *> // Wrap around negative values const unsigned int axis_u = wrap_axis(axis, input->info()); _num_slices = std::min(outputs_vector_info.size(), input->info()->dimension(axis_u)); - _strided_slice_vector = arm_compute::support::cpp14::make_unique<CLStridedSlice[]>(_num_slices); + _strided_slice_vector.resize(_num_slices); Coordinates slice_start; int32_t slice_end_mask; diff --git a/src/runtime/CL/functions/CLWidthConcatenateLayer.cpp b/src/runtime/CL/functions/CLWidthConcatenateLayer.cpp index 6e42377a07..a8667c3138 100644 --- a/src/runtime/CL/functions/CLWidthConcatenateLayer.cpp +++ b/src/runtime/CL/functions/CLWidthConcatenateLayer.cpp @@ -109,7 +109,7 @@ void CLWidthConcatenateLayer::configure(std::vector<ICLTensor *> inputs_vector, break; default: // Configure generic case WidthConcatenate kernels - _concat_kernels_vector = arm_compute::support::cpp14::make_unique<CLWidthConcatenateLayerKernel[]>(_num_inputs); + _concat_kernels_vector.resize(_num_inputs); unsigned int width_offset = 0; for(unsigned int i = 0; i < _num_inputs; ++i) diff --git a/src/runtime/CL/tuners/CLLWSList.cpp b/src/runtime/CL/tuners/CLLWSList.cpp index 97134b1b2c..6eb251420c 100644 --- a/src/runtime/CL/tuners/CLLWSList.cpp +++ b/src/runtime/CL/tuners/CLLWSList.cpp @@ -36,7 +36,7 @@ cl::NDRange CLLWSListExhaustive::operator[](size_t index) { ARM_COMPUTE_ERROR_ON(index >= size()); auto coords = index2coords(search_space_shape, index); - return cl::NDRange(coords[0] + 1, coords[1] + 1, coords[2] + 1); + return cl::NDRange{ coords[0] + 1U, coords[1] + 1U, coords[2] + 1U }; } CLLWSListExhaustive::CLLWSListExhaustive(const cl::NDRange &gws) @@ -49,7 +49,7 @@ cl::NDRange CLLWSListNormal::operator[](size_t index) { ARM_COMPUTE_ERROR_ON(index >= size()); auto coords = index2coords(search_space_shape, index); - return cl::NDRange(_lws_x[coords[0]], _lws_y[coords[1]], _lws_z[coords[2]]); + return cl::NDRange{ _lws_x[coords[0]], _lws_y[coords[1]], _lws_z[coords[2]] }; } CLLWSListNormal::CLLWSListNormal(const cl::NDRange &gws) |