Compute Library
 20.05
GCGEMMInterleave4x4Kernel.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 
26 #include "arm_compute/core/Error.h"
32 #include "arm_compute/core/Types.h"
33 #include "arm_compute/core/Utils.h"
36 #include "support/StringSupport.h"
37 
38 using namespace arm_compute;
39 
41  : _input(nullptr), _output(nullptr)
42 {
43 }
44 
46 {
49 
50  TensorShape output_shape = input->info()->tensor_shape();
51  output_shape.set(0, input->info()->dimension(0) * 4);
52  output_shape.set(1, std::ceil(input->info()->dimension(1) / 4.0f));
53 
54  // Output auto inizialitation if not yet initialized
55  auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
56 
59 
60  _input = input;
61  _output = output;
62 
63  std::set<std::string> build_opts;
64  std::string dt_name = (input->info()->data_type() == DataType::F32) ? "DATA_TYPE_FP32" : "DATA_TYPE_FP16";
65  build_opts.emplace(("#define " + dt_name));
66  build_opts.emplace("#define LOCAL_SIZE_X " + support::cpp11::to_string(1));
67  build_opts.emplace("#define LOCAL_SIZE_Y " + support::cpp11::to_string(1));
68  build_opts.emplace("#define LOCAL_SIZE_Z " + support::cpp11::to_string(1));
69 
70  // Create kernel
71  build_opts.emplace("#define GEMM_INTERLEAVE4x4");
72  _kernel = static_cast<GCKernel>(GCKernelLibrary::get().create_kernel("gemm_interleave4x4", build_opts));
73 
74  // Configure kernel window
75  const unsigned int num_elems_processed_per_iteration_x = max_gc_vector_width / data_size_from_type(input->info()->data_type());
76  constexpr unsigned int num_elems_processed_per_iteration_y = 4;
77  const unsigned int num_elems_written_per_iteration = num_elems_processed_per_iteration_x * num_elems_processed_per_iteration_y;
78 
79  Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
80 
81  AccessWindowRectangle input_access(input->info(), 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
82  AccessWindowRectangle output_access(output->info(), 0, 0, num_elems_written_per_iteration, 1, 4.f, 0.25f);
83 
84  update_window_and_padding(win, input_access, output_access);
85 
86  output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
87 
88  IGCKernel::configure(win);
89 }
90 
92 {
95 
96  _kernel.use();
97 
98  /*
99  * This kernel puts the values in a 4x4 block of Matrix A on the same row (Interleaved values)
100  * |a00 a01 a02 a03|
101  * |a10 a11 a12 a13|
102  * |a20 a21 a22 a23| = | a00 a10 a20 a30 || a01 a11 a21 a31 || a02 a12 a22 a32 || a03 a13 a23 a33 |
103  * |a30 a31 a32 a33|
104  *
105  * After this operation, the output matrix will have the following shape: [ height * 4, width / 4 ]
106  */
107  Window in_slice = window.first_slice_window_2D();
108  Window out_slice = window.first_slice_window_2D();
109 
110  // Change x and y steps for the slide of output tensor
111  out_slice.scale(Window::DimX, 4.f);
112  out_slice.scale(Window::DimY, 0.25f);
113 
114  do
115  {
116  unsigned int idx = 0;
117  add_2D_tensor_argument(idx, _input, 1, in_slice);
118  add_2D_tensor_argument(idx, _output, 2, out_slice);
119 
120  _kernel.update_shader_params();
121 
122  enqueue(*this, in_slice);
123  }
124  while(window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(out_slice));
125 }
Window first_slice_window_2D() const
First 2D slice of the window.
Definition: Window.h:281
void scale(size_t dimension, float scale_value)
Scale the values of a given dimension by the given scale_value.
Definition: Window.inl:155
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
#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.
virtual void set_valid_region(const ValidRegion &valid_region)=0
Set the valid region of the tensor.
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
Implementation of a 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.
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
Coordinates of an item.
Definition: Coordinates.h:37
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor's metadata.
size_t data_size_from_type(DataType data_type)
The size in bytes of the data type.
Definition: Utils.h:102
#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
void run(const Window &window) override
Enqueue the OpenGL ES shader to process the given window.
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.
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
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
Container for valid region of a window.
Definition: Types.h:187
void configure(const IGCTensor *input, IGCTensor *output)
Initialise the kernel's input and output.
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941