Georgios Pinitas | 0bc7849 | 2019-03-18 20:07:37 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2019 ARM Limited. |
| 3 | * |
| 4 | * SPDX-License-Identifier: MIT |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | * of this software and associated documentation files (the "Software"), to |
| 8 | * deal in the Software without restriction, including without limitation the |
| 9 | * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| 10 | * sell copies of the Software, and to permit persons to whom the Software is |
| 11 | * furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all |
| 14 | * copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 | * SOFTWARE. |
| 23 | */ |
| 24 | #include "arm_compute/runtime/CL/functions/CLFFT1D.h" |
| 25 | |
| 26 | #include "arm_compute/core/CL/ICLTensor.h" |
| 27 | #include "arm_compute/core/Validate.h" |
| 28 | #include "arm_compute/core/utils/helpers/fft.h" |
| 29 | #include "arm_compute/runtime/CL/CLScheduler.h" |
| 30 | |
| 31 | namespace arm_compute |
| 32 | { |
| 33 | CLFFT1D::CLFFT1D(std::shared_ptr<IMemoryManager> memory_manager) |
| 34 | : _memory_group(std::move(memory_manager)), _digit_reversed_input(), _digit_reverse_indices(), _digit_reverse_kernel(), _fft_kernels(), _num_ffts(0) |
| 35 | { |
| 36 | } |
| 37 | |
| 38 | void CLFFT1D::configure(const ICLTensor *input, ICLTensor *output, const FFT1DInfo &config) |
| 39 | { |
| 40 | ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); |
| 41 | ARM_COMPUTE_ERROR_THROW_ON(CLFFT1D::validate(input->info(), output->info(), config)); |
| 42 | |
| 43 | // Decompose size to radix factors |
| 44 | const auto supported_radix = CLFFTRadixStageKernel::supported_radix(); |
| 45 | const unsigned int N = input->info()->tensor_shape()[config.axis]; |
| 46 | const auto decomposed_vector = arm_compute::helpers::fft::decompose_stages(N, supported_radix); |
| 47 | ARM_COMPUTE_ERROR_ON(decomposed_vector.empty()); |
| 48 | |
| 49 | // Configure digit reverse |
| 50 | TensorInfo digit_reverse_indices_info(TensorShape(input->info()->tensor_shape()[config.axis]), 1, DataType::U32); |
| 51 | _digit_reverse_indices.allocator()->init(digit_reverse_indices_info); |
| 52 | _memory_group.manage(&_digit_reversed_input); |
| 53 | _digit_reverse_kernel.configure(input, &_digit_reversed_input, &_digit_reverse_indices, config.axis); |
| 54 | |
| 55 | // Create and configure FFT kernels |
| 56 | unsigned int Nx = 1; |
| 57 | _num_ffts = decomposed_vector.size(); |
| 58 | _fft_kernels = arm_compute::support::cpp14::make_unique<CLFFTRadixStageKernel[]>(_num_ffts); |
| 59 | for(unsigned int i = 0; i < _num_ffts; ++i) |
| 60 | { |
| 61 | const unsigned int radix_for_stage = decomposed_vector.at(i); |
| 62 | |
| 63 | FFTRadixStageKernelDescriptor fft_kernel_desc; |
| 64 | fft_kernel_desc.axis = config.axis; |
| 65 | fft_kernel_desc.radix = radix_for_stage; |
| 66 | fft_kernel_desc.Nx = Nx; |
| 67 | fft_kernel_desc.is_first_stage = (i == 0); |
| 68 | _fft_kernels[i].configure(&_digit_reversed_input, i == (_num_ffts - 1) ? output : nullptr, fft_kernel_desc); |
| 69 | |
| 70 | Nx *= radix_for_stage; |
| 71 | } |
| 72 | |
| 73 | // Allocate tensors |
| 74 | _digit_reversed_input.allocator()->allocate(); |
| 75 | _digit_reverse_indices.allocator()->allocate(); |
| 76 | |
| 77 | // Init digit reverse indices |
| 78 | const auto digit_reverse_cpu = arm_compute::helpers::fft::digit_reverse_indices(N, decomposed_vector); |
| 79 | _digit_reverse_indices.map(CLScheduler::get().queue(), true); |
| 80 | std::copy_n(digit_reverse_cpu.data(), N, reinterpret_cast<unsigned int *>(_digit_reverse_indices.buffer())); |
| 81 | _digit_reverse_indices.unmap(CLScheduler::get().queue()); |
| 82 | } |
| 83 | |
| 84 | Status CLFFT1D::validate(const ITensorInfo *input, const ITensorInfo *output, const FFT1DInfo &config) |
| 85 | { |
| 86 | ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output); |
| 87 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F32); |
| 88 | ARM_COMPUTE_RETURN_ERROR_ON(config.axis != 0); |
| 89 | |
| 90 | // Check if FFT is decomposable |
| 91 | const auto supported_radix = CLFFTRadixStageKernel::supported_radix(); |
| 92 | const unsigned int N = input->tensor_shape()[config.axis]; |
| 93 | const auto decomposed_vector = arm_compute::helpers::fft::decompose_stages(N, supported_radix); |
| 94 | ARM_COMPUTE_RETURN_ERROR_ON(decomposed_vector.empty()); |
| 95 | |
| 96 | // Checks performed when output is configured |
| 97 | if((output != nullptr) && (output->total_size() != 0)) |
| 98 | { |
| 99 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output); |
| 100 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); |
| 101 | } |
| 102 | |
| 103 | return Status{}; |
| 104 | } |
| 105 | |
| 106 | void CLFFT1D::run() |
| 107 | { |
| 108 | _memory_group.acquire(); |
| 109 | |
| 110 | CLScheduler::get().enqueue(_digit_reverse_kernel, false); |
| 111 | |
| 112 | for(unsigned int i = 0; i < _num_ffts; ++i) |
| 113 | { |
| 114 | CLScheduler::get().enqueue(_fft_kernels[i], i == (_num_ffts - 1)); |
| 115 | } |
| 116 | |
| 117 | _memory_group.release(); |
| 118 | } |
| 119 | } // namespace arm_compute |