Compute Library
 20.05
GCTransposeKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2020 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 
27 #include "arm_compute/core/Error.h"
33 #include "arm_compute/core/Types.h"
34 #include "support/StringSupport.h"
35 
36 #include <set>
37 #include <string>
38 
39 using namespace arm_compute;
40 
42 {
45 
46  TensorShape output_shape{ input->info()->tensor_shape() };
47  const size_t w_out = input->info()->dimension(1);
48  const size_t h_out = input->info()->dimension(0);
49  output_shape.set(0, w_out);
50  output_shape.set(1, h_out);
51 
52  // Output tensor auto inizialitation if not yet initialized
53  auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
54 
57 
58  _input = input;
59  _output = output;
60 
61  // for better performance
62  if(w_out < 512 && h_out < 512)
63  {
64  _lws_hint = gles::NDRange(8U, 1U, 1U);
65  }
66  else
67  {
68  _lws_hint = gles::NDRange(1U, 8U, 1U);
69  }
70 
71  std::set<std::string> build_opts;
72  std::string dt_name = (input->info()->data_type() == DataType::F32) ? "DATA_TYPE_FP32" : "DATA_TYPE_FP16";
73  build_opts.emplace(("#define " + dt_name));
74  build_opts.emplace("#define LOCAL_SIZE_X " + support::cpp11::to_string(_lws_hint[0]));
75  build_opts.emplace("#define LOCAL_SIZE_Y " + support::cpp11::to_string(_lws_hint[1]));
76  build_opts.emplace("#define LOCAL_SIZE_Z " + support::cpp11::to_string(_lws_hint[2]));
77 
78  // Configure kernel window
79  unsigned int num_elems_processed_per_iteration = 4;
80 
81  if(input->info()->data_type() == DataType::F16)
82  {
83 #define TRANSPOSE_8X8
84 
85 #if defined(TRANSPOSE_4X4)
86  build_opts.emplace(("#define TRANSPOSE_4X4"));
88 #elif defined(TRANSPOSE_8X8) /* TRANSPOSE_4X4 */
89  if(w_out != h_out)
90  {
91  build_opts.emplace("#define TRANSPOSE_8X8");
93  }
94  else
95  {
96  build_opts.emplace("#define TRANSPOSE_8X8_SQUARE");
98  }
99 #endif /* TRANSPOSE_4X4 */
100  }
101 
102  // Create kernel
103  _kernel = static_cast<GCKernel>(GCKernelLibrary::get().create_kernel("transpose", build_opts));
104 
105  const unsigned int width_aligned = num_elems_processed_per_iteration * static_cast<unsigned int>(_lws_hint[0]);
106  const unsigned int height_aligned = num_elems_processed_per_iteration * static_cast<unsigned int>(_lws_hint[1]);
107 
108  AccessWindowStatic input_access(input->info(), 0, 0,
109  ceil_to_multiple(input->info()->dimension(0), width_aligned),
110  ceil_to_multiple(input->info()->dimension(1), height_aligned));
111  AccessWindowStatic output_access(output->info(), 0, 0,
112  ceil_to_multiple(output->info()->dimension(0), height_aligned),
113  ceil_to_multiple(output->info()->dimension(1), width_aligned));
114 
115  Window win = calculate_max_window(*input->info(), Steps(width_aligned, height_aligned));
118  update_window_and_padding(win, input_access, output_access);
119  output_access.set_valid_region(win, output->info()->valid_region());
120 
121  IGCKernel::configure(win);
122 }
123 
124 void GCTransposeKernel::run(const Window &window)
125 {
128 
129  _kernel.use();
130 
132 
133  do
134  {
135  unsigned int idx = 0;
136 
137  add_2D_tensor_argument(idx, _input, 1, slice);
138  add_2D_tensor_argument(idx, _output, 2, slice);
139  _kernel.update_shader_params();
140  enqueue(*this, slice, _lws_hint);
141  }
143 }
Window first_slice_window_2D() const
First 2D slice of the window.
Definition: Window.h:281
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
Shape of a tensor.
Definition: TensorShape.h:39
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
#define ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:543
void enqueue(cl::CommandQueue &queue, ICLKernel &kernel, const Window &window, const cl::NDRange &lws_hint=CLKernelLibrary::get().default_ndrange(), bool use_dummy_work_items=false)
Add the kernel to the command queue with the given window.
Definition: ICLKernel.cpp:39
std::string to_string(T &&value)
Convert integer and float values to string.
1 channel, 1 F32 per channel
Interface for GLES Compute tensor.
Definition: IGCTensor.h:35
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps=Steps(), bool skip_border=false, BorderSize border_size=BorderSize())
Calculate the maximum window for a given tensor shape and border setting.
Definition: Helpers.cpp:28
bool slide_window_slice_2D(Window &slice) const
Slide the passed 2D window slice.
Definition: Window.h:321
Copyright (c) 2017-2020 ARM Limited.
bool auto_init_if_empty(ITensorInfo &info, const TensorShape &shape, int num_channels, DataType data_type, QuantizationInfo quantization_info=QuantizationInfo())
Auto initialize the tensor info (shape, number of channels and data type) if the current assignment i...
Definition: Helpers.inl:202
1 channel, 1 F16 per channel
virtual ValidRegion valid_region() const =0
Valid region of the tensor.
Implementation of a static rectangular access pattern.
cl::Kernel create_kernel(const CLCompileContext &ctx, const std::string &kernel_name, const std::set< std::string > &build_opts=std::set< std::string >())
Creates an opencl kernel using a compile context.
Definition: CLHelpers.cpp:387
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: Helpers.h:437
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
auto ceil_to_multiple(S value, T divisor) -> decltype(((value+divisor - 1)/divisor) *divisor)
Computes the smallest number larger or equal to value that is a multiple of divisor.
Definition: Utils.h:67
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor's metadata.
#define ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(...)
Definition: Validate.h:286
void add_2D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window)
Add the passed 2D tensor's parameters to the object's kernel's arguments starting from the index idx.
Definition: IGCKernel.cpp:127
#define ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:790
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
static GCKernelLibrary & get()
Get the static instance of GCKernelLibrary.
void set_dimension_step(size_t dimension, int step)
Set the step of a given dimension.
Definition: Window.inl:167
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
void run(const Window &window) override
Enqueue the OpenGL ES shader to process the given window.
Class interface for specifying NDRange values.
Definition: OpenGLES.h:53
unsigned int num_elems_processed_per_iteration
TensorShape & set(size_t dimension, size_t value, bool apply_dim_correction=true)
Accessor to set the value of one of the dimensions.
Definition: TensorShape.h:78
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
void configure(const IGCTensor *input, IGCTensor *output)
Initialise the kernel's input and output.
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
void set_valid_region(const Window &window, const ValidRegion &input_valid_region)
Set the valid region based on the static access pattern and valid region of the inputs.
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)