Compute Library
 22.11
CpuGemmInterleave4x4Kernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016-2021 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 
32 
33 #include <arm_neon.h>
34 
35 namespace arm_compute
36 {
37 namespace cpu
38 {
39 namespace kernels
40 {
42 
44 {
46 
47  // dst auto inizialitation if not yet initialized
48  auto_init_if_empty(*dst, src->clone()->set_tensor_shape(compute_interleaved_shape(*src)));
49 
50  // Perform validate step
52 
53  Window win = calculate_max_window(*src, Steps(1, 4));
54  ICPPKernel::configure(win);
55 }
56 
58 {
60  //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(src) is not needed here as this kernel doesn't use CPU FP16 instructions.
62 
63  if(dst->total_size() != 0)
64  {
69  }
70 
71  return Status{};
72 }
73 
75 {
76  ARM_COMPUTE_UNUSED(info);
79  ARM_COMPUTE_ERROR_ON(tensors.empty());
80  /*
81  * This kernel puts the values in a 4x4 block of Matrix A on the same row (Interleaved values)
82  * |a00 a01 a02 a03|
83  * |a10 a11 a12 a13|
84  * |a20 a21 a22 a23| = | a00 a10 a20 a30 || a01 a11 a21 a31 || a02 a12 a22 a32 || a03 a13 a23 a33 |
85  * |a30 a31 a32 a33|
86  *
87  * After this operation, the dst matrix will have the following shape: [ height * 4, ceil(width / 4.0f) ]
88  */
91 
92  const size_t window_start_x = window.x().start();
93  const size_t window_end_x = window.x().end();
94 
95  const size_t in_height = src->info()->dimension(1);
96  const size_t in_stride = src->info()->strides_in_bytes()[1];
97 
98  const size_t partial_y = in_height % 4;
99 
100  const size_t element_size = src->info()->element_size();
101 
102  // Set window for the src tensor
103  Window win = window;
104  win.set(Window::DimX, Window::Dimension(0, 1, 1));
105 
106  // Set window for the dst tensor
107  Window win_out(window);
108  win_out.set(Window::DimX, Window::Dimension(0, 1, 1));
109  win_out.scale(Window::DimY, 0.25f);
110 
111  Iterator in(src, win);
112  Iterator out(dst, win_out);
113 
114  execute_window_loop(win, [&](const Coordinates & id)
115  {
116  if(id.y() + 4 <= static_cast<int>(in_height))
117  {
118  for(size_t x = window_start_x; x < window_end_x; ++x)
119  {
120  std::memcpy(out.ptr() + (x * 4 + 0) * element_size, (in.ptr() + 0 * in_stride) + x * element_size, element_size);
121  std::memcpy(out.ptr() + (x * 4 + 1) * element_size, (in.ptr() + 1 * in_stride) + x * element_size, element_size);
122  std::memcpy(out.ptr() + (x * 4 + 2) * element_size, (in.ptr() + 2 * in_stride) + x * element_size, element_size);
123  std::memcpy(out.ptr() + (x * 4 + 3) * element_size, (in.ptr() + 3 * in_stride) + x * element_size, element_size);
124  }
125  }
126  else
127  {
128  for(size_t x = window_start_x; x < window_end_x; ++x)
129  {
130  size_t y = 0;
131  for(; y < partial_y; ++y)
132  {
133  std::memcpy(out.ptr() + (x * 4 + y) * element_size, (in.ptr() + y * in_stride) + x * element_size, element_size);
134  }
135  for(; y < 4; ++y)
136  {
137  std::memset(out.ptr() + (x * 4 + y) * element_size, 0, element_size);
138  }
139  }
140  }
141  },
142  in, out);
143 }
144 
146 {
147  return "CpuGemmInterleave4x4Kernel";
148 }
149 } // namespace kernels
150 } // namespace cpu
151 } // namespace arm_compute
void scale(size_t dimension, float scale_value)
Scale the values of a given dimension by the given scale_value.
Definition: Window.inl:155
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
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_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(...)
Definition: Validate.h:606
bool empty() const
Checks if pack is empty.
Definition: ITensorPack.cpp:80
virtual DataType data_type() const =0
Data type used for each element of the tensor.
#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
Describe one of the image&#39;s dimensions with a start, end and step.
Definition: Window.h:79
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 CPU tensor.
Definition: ITensor.h:36
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(...)
Definition: Validate.h:284
TensorShape compute_interleaved_shape(const ITensorInfo &a, int mult_interleave4x4_height=1, bool reinterpret_input_as_3d=false)
Calculate the interleaved shape of an input tensor.
SimpleTensor< float > src
Definition: DFT.cpp:155
Copyright (c) 2017-2022 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
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
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...
virtual std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
constexpr uint8_t * ptr() const
Return a pointer to the current pixel.
Definition: Helpers.inl:139
const char * name() const override
Name of the kernel.
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
static Status validate(const ITensorInfo *src, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration of CpuGemmInterleave4x4Kern...
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
Information about executing thread and CPU.
Definition: CPPTypes.h:179
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
Tensor packing service.
Definition: ITensorPack.h:39
void configure(const ITensorInfo *src, ITensorInfo *dst)
Initialise the kernel&#39;s src and dst.
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
void execute_window_loop(const Window &w, L &&lambda_function, Ts &&... iterators)
Iterate through the passed window, automatically adjusting the iterators and calling the lambda_funct...
Definition: Helpers.inl:77
void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override
Execute the kernel on the passed window.
constexpr int end() const
Return the end of the dimension.
Definition: Window.h:102
Iterator updated by execute_window_loop for each window element.
Definition: Helpers.h:46
constexpr int start() const
Return the start of the dimension.
Definition: Window.h:97
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201
constexpr const Dimension & x() const
Alias to access the first dimension of the window.
Definition: Window.h:159