Compute Library
 21.02
GCGEMMMatrixAccumulateBiasesKernel.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"
38 #include "support/StringSupport.h"
39 
40 using namespace arm_compute;
41 
43  : _accum(nullptr), _biases(nullptr), _lws(gles::NDRange(1U, 1U, 1U))
44 {
45 }
46 
48 {
51  ARM_COMPUTE_ERROR_ON(biases->info()->num_dimensions() != 1);
52 
53  _biases = biases;
54  _accum = accum;
55 
56  std::set<std::string> build_opts;
57  build_opts.emplace("#define LOCAL_SIZE_X " + support::cpp11::to_string(_lws[0]));
58  build_opts.emplace("#define LOCAL_SIZE_Y " + support::cpp11::to_string(_lws[1]));
59  build_opts.emplace("#define LOCAL_SIZE_Z " + support::cpp11::to_string(_lws[2]));
60 
61  // Create kernel
62  build_opts.emplace("#define GEMM_ACCUMULATE_BIASES");
63 
64 #define ACCUM_PROCESS_4X
65 
66 #if defined(ACCUM_PROCESS_4X)
67  build_opts.emplace("#define ACCUM_PROCESS_4X");
68 #elif defined(ACCUM_PROCESS_8X) /* ACCUM_PROCESS_4X */
69  build_opts.emplace("#define ACCUM_PROCESS_8X");
70 #endif /* ACCUM_PROCESS_4X */
71  std::string dt_name = (accum->info()->data_type() == DataType::F32) ? "DATA_TYPE_FP32" : "DATA_TYPE_FP16";
72  build_opts.emplace(("#define " + dt_name));
73 
74  _kernel = GCKernelLibrary::get().create_kernel("gemm_accumulate_biases", build_opts);
75 
76  // Configure kernel window
77  unsigned int num_elems_processed_per_iteration = 1;
78 
79  if(_accum->info()->data_type() == DataType::F32)
80  {
81  num_elems_processed_per_iteration = 16;
82  }
83  else if(_accum->info()->data_type() == DataType::F16)
84  {
85 #if defined(ACCUM_PROCESS_4X)
86  num_elems_processed_per_iteration = 4;
87 #elif defined(ACCUM_PROCESS_8X) /* ACCUM_PROCESS_4X */
88  num_elems_processed_per_iteration = 8;
89 #endif /* ACCUM_PROCESS_4X */
90  }
91 
92  const int accum_width = accum->info()->dimension(0);
93  const int accum_padding_right = ceil_to_multiple(accum_width, num_elems_processed_per_iteration * _lws[0]) - accum_width;
94  BorderSize border = BorderSize(0, accum_padding_right, 0, 0);
95 
96  Window win = calculate_max_enlarged_window(*_accum->info(), Steps(num_elems_processed_per_iteration), border);
97 
98  AccessWindowStatic biases_access(biases->info(), 0, 0, ceil_to_multiple(biases->info()->dimension(0), num_elems_processed_per_iteration * _lws[0]), biases->info()->dimension(1));
99  AccessWindowStatic accum_access(_accum->info(), 0, 0, accum_width + accum_padding_right, _accum->info()->dimension(1));
100 
101  update_window_and_padding(win, biases_access, accum_access);
102 
103  IGCKernel::configure(win);
104 }
105 
107 {
110 
111  _kernel.use();
112 
113  Window accum_slice = window.first_slice_window_2D();
114 
115  Window biases_slice(accum_slice);
116  biases_slice.set(Window::DimY, Window::Dimension(0, 1, 1));
117 
118  // Run kernel
119  do
120  {
121  // Set arguments
122  unsigned int idx = 0;
123 
124  add_2D_tensor_argument(idx, _accum, 1, accum_slice);
125  add_1D_tensor_argument(idx, _biases, 2, biases_slice);
126  _kernel.update_shader_params();
127 
128  enqueue(*this, accum_slice, _lws);
129  }
130  while(window.slide_window_slice_2D(accum_slice));
131 }
Window first_slice_window_2D() const
First 2D slice of the window.
Definition: Window.h:283
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
void enqueue(IGCKernel &kernel, const Window &window, const gles::NDRange &lws=gles::NDRange(1U, 1U, 1U))
Add the kernel to the command queue with the given window.
Definition: IGCKernel.cpp:41
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
Container for 2D border size.
Definition: Types.h:273
std::string to_string(T &&value)
Convert integer and float values to string.
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
Interface for GLES Compute tensor.
Definition: IGCTensor.h:35
Describe one of the image&#39;s dimensions with a start, end and step.
Definition: Window.h:77
void run(const Window &window) override
Enqueue the OpenGL ES shader to process the given window.
bool slide_window_slice_2D(Window &slice) const
Slide the passed 2D window slice.
Definition: Window.h:323
Copyright (c) 2017-2021 Arm Limited.
1 channel, 1 F16 per channel
Implementation of a static rectangular access pattern.
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: WindowHelpers.h:46
#define ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(f, w)
Definition: Validate.h:183
#define ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:543
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:71
Manages all the GLES kernels compilation and caching, provides accessors for the GLES Context...
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&#39;s metadata.
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
void add_2D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window)
Add the passed 2D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: IGCKernel.cpp:127
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
#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
Wrapper to configure the Khronos EGL and OpenGL ES C header.
static GCKernelLibrary & get()
Get the static instance of GCKernelLibrary.
Window calculate_max_enlarged_window(const ValidRegion &valid_region, const Steps &steps, BorderSize border_size)
GCKernel create_kernel(const std::string &shader_name, const StringSet &build_options_set={}) const
Creates a kernel from the kernel library.
unsigned int num_elems_processed_per_iteration
void configure(IGCTensor *accum, const IGCTensor *biases)
Set the accumulate buffer and the biases of the kernel.
Describe a multidimensional execution window.
Definition: Window.h:39
void add_1D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window)
Add the passed 1D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: IGCKernel.cpp:122