diff options
Diffstat (limited to 'src/runtime/CL/functions/CLFFT1D.cpp')
-rw-r--r-- | src/runtime/CL/functions/CLFFT1D.cpp | 31 |
1 files changed, 20 insertions, 11 deletions
diff --git a/src/runtime/CL/functions/CLFFT1D.cpp b/src/runtime/CL/functions/CLFFT1D.cpp index cf136dc75e..48e9ae824a 100644 --- a/src/runtime/CL/functions/CLFFT1D.cpp +++ b/src/runtime/CL/functions/CLFFT1D.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -26,6 +26,8 @@ #include "arm_compute/core/CL/ICLTensor.h" #include "arm_compute/core/Validate.h" #include "arm_compute/runtime/CL/CLScheduler.h" + +#include "src/common/utils/Log.h" #include "src/core/CL/kernels/CLFFTDigitReverseKernel.h" #include "src/core/CL/kernels/CLFFTRadixStageKernel.h" #include "src/core/CL/kernels/CLFFTScaleKernel.h" @@ -52,10 +54,14 @@ void CLFFT1D::configure(const ICLTensor *input, ICLTensor *output, const FFT1DIn configure(CLKernelLibrary::get().get_compile_context(), input, output, config); } -void CLFFT1D::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const FFT1DInfo &config) +void CLFFT1D::configure(const CLCompileContext &compile_context, + const ICLTensor *input, + ICLTensor *output, + const FFT1DInfo &config) { ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); ARM_COMPUTE_ERROR_THROW_ON(CLFFT1D::validate(input->info(), output->info(), config)); + ARM_COMPUTE_LOG_PARAMS(input, output, config); // Decompose size to radix factors const auto supported_radix = CLFFTRadixStageKernel::supported_radix(); @@ -74,13 +80,14 @@ void CLFFT1D::configure(const CLCompileContext &compile_context, const ICLTensor TensorInfo digit_reverse_indices_info(TensorShape(input->info()->tensor_shape()[config.axis]), 1, DataType::U32); _digit_reverse_indices.allocator()->init(digit_reverse_indices_info); _memory_group.manage(&_digit_reversed_input); - _digit_reverse_kernel->configure(compile_context, input, &_digit_reversed_input, &_digit_reverse_indices, digit_reverse_config); + _digit_reverse_kernel->configure(compile_context, input, &_digit_reversed_input, &_digit_reverse_indices, + digit_reverse_config); // Create and configure FFT kernels unsigned int Nx = 1; _num_ffts = decomposed_vector.size(); _fft_kernels.reserve(_num_ffts); - for(unsigned int i = 0; i < _num_ffts; ++i) + for (unsigned int i = 0; i < _num_ffts; ++i) { const unsigned int radix_for_stage = decomposed_vector.at(i); @@ -90,18 +97,20 @@ void CLFFT1D::configure(const CLCompileContext &compile_context, const ICLTensor fft_kernel_info.Nx = Nx; fft_kernel_info.is_first_stage = (i == 0); _fft_kernels.emplace_back(std::make_unique<CLFFTRadixStageKernel>()); - _fft_kernels.back()->configure(compile_context, &_digit_reversed_input, ((i == (_num_ffts - 1)) && !is_c2r) ? output : nullptr, fft_kernel_info); + _fft_kernels.back()->configure(compile_context, &_digit_reversed_input, + ((i == (_num_ffts - 1)) && !is_c2r) ? output : nullptr, fft_kernel_info); Nx *= radix_for_stage; } // Configure scale kernel - if(_run_scale) + if (_run_scale) { FFTScaleKernelInfo scale_config; scale_config.scale = static_cast<float>(N); scale_config.conjugate = config.direction == FFTDirection::Inverse; - is_c2r ? _scale_kernel->configure(compile_context, &_digit_reversed_input, output, scale_config) : _scale_kernel->configure(output, nullptr, scale_config); + is_c2r ? _scale_kernel->configure(compile_context, &_digit_reversed_input, output, scale_config) + : _scale_kernel->configure(output, nullptr, scale_config); } // Allocate tensors @@ -120,7 +129,7 @@ Status CLFFT1D::validate(const ITensorInfo *input, const ITensorInfo *output, co ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output); ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(input, DataType::F16, DataType::F32); ARM_COMPUTE_RETURN_ERROR_ON(input->num_channels() != 1 && input->num_channels() != 2); - ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({ 0, 1 }).count(config.axis) == 0); + ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({0, 1}).count(config.axis) == 0); // Check if FFT is decomposable const auto supported_radix = CLFFTRadixStageKernel::supported_radix(); @@ -129,7 +138,7 @@ Status CLFFT1D::validate(const ITensorInfo *input, const ITensorInfo *output, co ARM_COMPUTE_RETURN_ERROR_ON(decomposed_vector.empty()); // Checks performed when output is configured - if((output != nullptr) && (output->total_size() != 0)) + if ((output != nullptr) && (output->total_size() != 0)) { ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() == 1 && input->num_channels() == 1); ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() != 1 && output->num_channels() != 2); @@ -148,13 +157,13 @@ void CLFFT1D::run() CLScheduler::get().enqueue(*_digit_reverse_kernel, false); // Run radix kernels - for(unsigned int i = 0; i < _num_ffts; ++i) + for (unsigned int i = 0; i < _num_ffts; ++i) { CLScheduler::get().enqueue(*_fft_kernels[i], i == (_num_ffts - 1) && !_run_scale); } // Run output scaling - if(_run_scale) + if (_run_scale) { CLScheduler::get().enqueue(*_scale_kernel, true); } |