Compute Library
 20.08
CLFFT1D.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 
30 
31 namespace arm_compute
32 {
33 CLFFT1D::CLFFT1D(std::shared_ptr<IMemoryManager> memory_manager)
34  : _memory_group(std::move(memory_manager)), _digit_reverse_kernel(), _fft_kernels(), _scale_kernel(), _digit_reversed_input(), _digit_reverse_indices(), _num_ffts(0), _run_scale(false)
35 {
36 }
37 
38 void CLFFT1D::configure(const ICLTensor *input, ICLTensor *output, const FFT1DInfo &config)
39 {
40  configure(CLKernelLibrary::get().get_compile_context(), input, output, config);
41 }
42 
43 void CLFFT1D::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const FFT1DInfo &config)
44 {
46  ARM_COMPUTE_ERROR_THROW_ON(CLFFT1D::validate(input->info(), output->info(), config));
47 
48  // Decompose size to radix factors
49  const auto supported_radix = CLFFTRadixStageKernel::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  const bool is_c2r = input->info()->num_channels() == 2 && output->info()->num_channels() == 1;
57 
58  // Configure digit reverse
59  FFTDigitReverseKernelInfo digit_reverse_config;
60  digit_reverse_config.axis = config.axis;
61  digit_reverse_config.conjugate = config.direction == FFTDirection::Inverse;
62  TensorInfo digit_reverse_indices_info(TensorShape(input->info()->tensor_shape()[config.axis]), 1, DataType::U32);
63  _digit_reverse_indices.allocator()->init(digit_reverse_indices_info);
64  _memory_group.manage(&_digit_reversed_input);
65  _digit_reverse_kernel.configure(compile_context, input, &_digit_reversed_input, &_digit_reverse_indices, digit_reverse_config);
66 
67  // Create and configure FFT kernels
68  unsigned int Nx = 1;
69  _num_ffts = decomposed_vector.size();
70  _fft_kernels.resize(_num_ffts);
71  for(unsigned int i = 0; i < _num_ffts; ++i)
72  {
73  const unsigned int radix_for_stage = decomposed_vector.at(i);
74 
75  FFTRadixStageKernelInfo fft_kernel_info;
76  fft_kernel_info.axis = config.axis;
77  fft_kernel_info.radix = radix_for_stage;
78  fft_kernel_info.Nx = Nx;
79  fft_kernel_info.is_first_stage = (i == 0);
80  _fft_kernels[i].configure(compile_context, &_digit_reversed_input, ((i == (_num_ffts - 1)) && !is_c2r) ? output : nullptr, fft_kernel_info);
81 
82  Nx *= radix_for_stage;
83  }
84 
85  // Configure scale kernel
86  if(_run_scale)
87  {
88  FFTScaleKernelInfo scale_config;
89  scale_config.scale = static_cast<float>(N);
90  scale_config.conjugate = config.direction == FFTDirection::Inverse;
91  is_c2r ? _scale_kernel.configure(compile_context, &_digit_reversed_input, output, scale_config) : _scale_kernel.configure(output, nullptr, scale_config);
92  }
93 
94  // Allocate tensors
95  _digit_reversed_input.allocator()->allocate();
96  _digit_reverse_indices.allocator()->allocate();
97 
98  // Init digit reverse indices
99  const auto digit_reverse_cpu = arm_compute::helpers::fft::digit_reverse_indices(N, decomposed_vector);
100  _digit_reverse_indices.map(CLScheduler::get().queue(), true);
101  std::copy_n(digit_reverse_cpu.data(), N, reinterpret_cast<unsigned int *>(_digit_reverse_indices.buffer()));
102  _digit_reverse_indices.unmap(CLScheduler::get().queue());
103 }
104 
105 Status CLFFT1D::validate(const ITensorInfo *input, const ITensorInfo *output, const FFT1DInfo &config)
106 {
109  ARM_COMPUTE_RETURN_ERROR_ON(input->num_channels() != 1 && input->num_channels() != 2);
110  ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({ 0, 1 }).count(config.axis) == 0);
111 
112  // Check if FFT is decomposable
113  const auto supported_radix = CLFFTRadixStageKernel::supported_radix();
114  const unsigned int N = input->tensor_shape()[config.axis];
115  const auto decomposed_vector = arm_compute::helpers::fft::decompose_stages(N, supported_radix);
116  ARM_COMPUTE_RETURN_ERROR_ON(decomposed_vector.empty());
117 
118  // Checks performed when output is configured
119  if((output != nullptr) && (output->total_size() != 0))
120  {
121  ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() == 1 && input->num_channels() == 1);
122  ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() != 1 && output->num_channels() != 2);
125  }
126 
127  return Status{};
128 }
129 
131 {
132  MemoryGroupResourceScope scope_mg(_memory_group);
133 
134  // Run digit reverse
135  CLScheduler::get().enqueue(_digit_reverse_kernel, false);
136 
137  // Run radix kernels
138  for(unsigned int i = 0; i < _num_ffts; ++i)
139  {
140  CLScheduler::get().enqueue(_fft_kernels[i], i == (_num_ffts - 1) && !_run_scale);
141  }
142 
143  // Run output scaling
144  if(_run_scale)
145  {
146  CLScheduler::get().enqueue(_scale_kernel, true);
147  }
148 }
149 } // namespace arm_compute
unsigned int axis
Axis to run the FFT on.
Shape of a tensor.
Definition: TensorShape.h:39
void configure(const ICLTensor *input, ICLTensor *output, const FFT1DInfo &config)
Initialise the function's source, destinations and border mode.
Definition: CLFFT1D.cpp:38
static CLScheduler & get()
Access the scheduler singleton.
Definition: CLScheduler.cpp:99
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
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
Descriptor used by the FFT1D function.
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 CLFFT1D.
Definition: CLFFT1D.cpp:105
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
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
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
Store the tensor's metadata.
Definition: ITensorInfo.h:40
CLTensorAllocator * allocator()
Return a pointer to the tensor's allocator.
Definition: CLTensor.cpp:61
#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
void configure(ICLTensor *input, ICLTensor *output, const FFTScaleKernelInfo &config)
Set the input and output tensors.
void init(const TensorInfo &input, size_t alignment=0)
Initialize a tensor based on the passed TensorInfo.
Copyright (c) 2017-2020 Arm Limited.
void map(bool blocking=true)
Enqueue a map operation of the allocated buffer.
Definition: CLTensor.cpp:66
uint8_t * buffer() const override
Interface to be implemented by the child class to return a pointer to CPU memory.
Definition: ICLTensor.cpp:53
void manage(IMemoryManageable *obj) override
Sets a object to be managed by the given memory group.
Definition: MemoryGroup.h:79
1 channel, 1 U32 per channel
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:443
bool conjugate
Flag to conjugate the output/.
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's metadata.
void configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *idx, const FFTDigitReverseKernelInfo &config)
Set the input and output tensors.
void enqueue(ICLKernel &kernel, bool flush=true)
Schedule the execution of the passed kernel if possible.
CLCompileContext class.
static std::set< unsigned int > supported_radix()
Returns the radix that are support by the FFT kernel.
unsigned int axis
Axis to perform the kernel on.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
void allocate() override
Allocate size specified by TensorInfo of OpenCL memory.
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
Memory group resources scope handling class.
Definition: IMemoryGroup.h:82
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
CLFFT1D(std::shared_ptr< IMemoryManager > memory_manager=nullptr)
Default Constructor.
Definition: CLFFT1D.cpp:33
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
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.
Store the tensor's metadata.
Definition: TensorInfo.h:45
FFTDirection direction
Direction of the FFT.
float scale
Axis to perform the kernel on.
void run() override
Run the kernels contained in the function.
Definition: CLFFT1D.cpp:130
void unmap()
Enqueue an unmap operation of the allocated and mapped buffer.
Definition: CLTensor.cpp:71
virtual size_t num_channels() const =0
The number of channels for each tensor element.
bool conjugate
Flag to conjugate the output/.