Compute Library
 21.02
NEFFT1D.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2019-2020 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  */
25 
33 
34 namespace arm_compute
35 {
36 NEFFT1D::~NEFFT1D() = default;
37 
38 NEFFT1D::NEFFT1D(std::shared_ptr<IMemoryManager> memory_manager)
39  : _memory_group(std::move(memory_manager)), _digit_reverse_kernel(), _fft_kernels(), _scale_kernel(), _digit_reversed_input(), _digit_reverse_indices(), _num_ffts(0), _axis(0), _run_scale(false)
40 {
41 }
42 
43 void NEFFT1D::configure(const ITensor *input, ITensor *output, const FFT1DInfo &config)
44 {
45  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
46  ARM_COMPUTE_ERROR_THROW_ON(NEFFT1D::validate(input->info(), output->info(), config));
47 
48  // Decompose size to radix factors
49  const auto supported_radix = NEFFTRadixStageKernel::supported_radix();
50  const unsigned int N = input->info()->tensor_shape()[config.axis];
51  const auto decomposed_vector = arm_compute::helpers::fft::decompose_stages(N, supported_radix);
52  ARM_COMPUTE_ERROR_ON(decomposed_vector.empty());
53 
54  // Flags
55  _run_scale = config.direction == FFTDirection::Inverse;
56 
57  const bool is_c2r = input->info()->num_channels() == 2 && output->info()->num_channels() == 1;
58 
59  // Configure digit reverse
60  FFTDigitReverseKernelInfo digit_reverse_config;
61  digit_reverse_config.axis = config.axis;
62  digit_reverse_config.conjugate = config.direction == FFTDirection::Inverse;
63  TensorInfo digit_reverse_indices_info(TensorShape(input->info()->tensor_shape()[config.axis]), 1, DataType::U32);
64  _digit_reverse_indices.allocator()->init(digit_reverse_indices_info);
65  _memory_group.manage(&_digit_reversed_input);
66  _digit_reverse_kernel = std::make_unique<NEFFTDigitReverseKernel>();
67  _digit_reverse_kernel->configure(input, &_digit_reversed_input, &_digit_reverse_indices, digit_reverse_config);
68 
69  // Create and configure FFT kernels
70  unsigned int Nx = 1;
71  _num_ffts = decomposed_vector.size();
72  _fft_kernels.resize(_num_ffts);
73  _axis = config.axis;
74 
75  for(unsigned int i = 0; i < _num_ffts; ++i)
76  {
77  const unsigned int radix_for_stage = decomposed_vector.at(i);
78 
79  FFTRadixStageKernelInfo fft_kernel_info;
80  fft_kernel_info.axis = config.axis;
81  fft_kernel_info.radix = radix_for_stage;
82  fft_kernel_info.Nx = Nx;
83  fft_kernel_info.is_first_stage = (i == 0);
84  _fft_kernels[i] = std::make_unique<NEFFTRadixStageKernel>();
85  _fft_kernels[i]->configure(&_digit_reversed_input, ((i == (_num_ffts - 1)) && !is_c2r) ? output : nullptr, fft_kernel_info);
86 
87  Nx *= radix_for_stage;
88  }
89 
90  // Configure scale kernel
91  if(_run_scale)
92  {
93  FFTScaleKernelInfo scale_config;
94  scale_config.scale = static_cast<float>(N);
95  scale_config.conjugate = config.direction == FFTDirection::Inverse;
96  _scale_kernel = std::make_unique<NEFFTScaleKernel>();
97  is_c2r ? _scale_kernel->configure(&_digit_reversed_input, output, scale_config) : _scale_kernel->configure(output, nullptr, scale_config);
98  }
99 
100  // Allocate tensors
101  _digit_reversed_input.allocator()->allocate();
102  _digit_reverse_indices.allocator()->allocate();
103 
104  // Init digit reverse indices
105  const auto digit_reverse_cpu = arm_compute::helpers::fft::digit_reverse_indices(N, decomposed_vector);
106  std::copy_n(digit_reverse_cpu.data(), N, reinterpret_cast<unsigned int *>(_digit_reverse_indices.buffer()));
107 }
108 
109 Status NEFFT1D::validate(const ITensorInfo *input, const ITensorInfo *output, const FFT1DInfo &config)
110 {
114  ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({ 0, 1 }).count(config.axis) == 0);
115 
116  // Check if FFT is decomposable
117  const auto supported_radix = NEFFTRadixStageKernel::supported_radix();
118  const unsigned int N = input->tensor_shape()[config.axis];
119  const auto decomposed_vector = arm_compute::helpers::fft::decompose_stages(N, supported_radix);
120  ARM_COMPUTE_RETURN_ERROR_ON(decomposed_vector.empty());
121 
122  // Checks performed when output is configured
123  if((output != nullptr) && (output->total_size() != 0))
124  {
125  // All combinations are supported except real input with real output (i.e., both input channels set to 1)
126  ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() == 1 && input->num_channels() == 1);
130  }
131 
132  return Status{};
133 }
134 
136 {
137  MemoryGroupResourceScope scope_mg(_memory_group);
138 
139  NEScheduler::get().schedule(_digit_reverse_kernel.get(), (_axis == 0 ? Window::DimY : Window::DimZ));
140 
141  for(unsigned int i = 0; i < _num_ffts; ++i)
142  {
143  NEScheduler::get().schedule(_fft_kernels[i].get(), (_axis == 0 ? Window::DimY : Window::DimX));
144  }
145 
146  // Run output scaling
147  if(_run_scale)
148  {
149  NEScheduler::get().schedule(_scale_kernel.get(), Window::DimY);
150  }
151 }
152 } // namespace arm_compute
~NEFFT1D()
Default destructor.
unsigned int axis
Axis to run the FFT on.
Shape of a tensor.
Definition: TensorShape.h:39
void init(const TensorAllocator &allocator, const Coordinates &coords, TensorInfo &sub_info)
Shares the same backing memory with another tensor allocator, while the tensor info might be differen...
std::vector< unsigned int > decompose_stages(unsigned int N, const std::set< unsigned int > &supported_factors)
Decompose a given 1D input size using the provided supported factors.
Definition: fft.cpp:34
Descriptor used by the FFT1D function.
std::vector< unsigned int > digit_reverse_indices(unsigned int N, const std::vector< unsigned int > &fft_stages)
Calculate digit reverse index vector given fft size and the decomposed stages.
Definition: fft.cpp:79
virtual DataType data_type() const =0
Data type used for each element of the tensor.
1 channel, 1 F32 per channel
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
Definition: Error.h:466
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
unsigned int axis
Axis to run the kernel on.
Status class.
Definition: Error.h:52
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
Interface for Neon tensor.
Definition: ITensor.h:36
unsigned int N
Copyright (c) 2017-2021 Arm Limited.
TensorAllocator * allocator()
Return a pointer to the tensor&#39;s allocator.
Definition: Tensor.cpp:48
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
void manage(IMemoryManageable *obj) override
Sets a object to be managed by the given memory group.
Definition: MemoryGroup.h:79
void run() override
Run the kernels contained in the function.
Definition: NEFFT1D.cpp:135
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
1 channel, 1 U32 per channel
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
void configure(const ITensor *input, ITensor *output, const FFT1DInfo &config)
Initialise the function&#39;s source and destinations.
Definition: NEFFT1D.cpp:43
bool conjugate
Flag to conjugate the output/.
void allocate() override
Allocate size specified by TensorInfo of CPU memory.
static std::set< unsigned int > supported_radix()
Returns the radix that are support by the FFT kernel.
Descriptor used by the FFT core kernels.
Descriptor for FFT scale kernels.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
unsigned int axis
Axis to perform the kernel on.
unsigned int radix
Radix to use.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
Memory group resources scope handling class.
Definition: IMemoryGroup.h:82
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
virtual void schedule(ICPPKernel *kernel, const Hints &hints)=0
Runs the kernel in the same thread as the caller synchronously.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:443
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
bool is_first_stage
Flags if the FFT kernels is the first stage of a decomposed FFT.
Descriptor for FFT digit reverse kernels.
unsigned int Nx
Nx coefficient.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
uint8_t * buffer() const override
Interface to be implemented by the child class to return a pointer to CPU memory. ...
Definition: Tensor.cpp:43
static Status validate(const ITensorInfo *input, const ITensorInfo *output, const FFT1DInfo &config)
Static function to check if given info will lead to a valid configuration of NEFFT1D.
Definition: NEFFT1D.cpp:109
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
Store the tensor&#39;s metadata.
Definition: TensorInfo.h:45
FFTDirection direction
Direction of the FFT.
float scale
Axis to perform the kernel on.
NEFFT1D(std::shared_ptr< IMemoryManager > memory_manager=nullptr)
Default Constructor.
Definition: NEFFT1D.cpp:38
virtual size_t num_channels() const =0
The number of channels for each tensor element.
bool conjugate
Flag to conjugate the output/.
static IScheduler & get()
Access the scheduler singleton.
Definition: Scheduler.cpp:94