Compute Library
 21.02
CLGEMMLowpReductionKernel.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 
32 #include "support/StringSupport.h"
33 
34 namespace arm_compute
35 {
36 namespace
37 {
38 Status validate_arguments_matrix_a_reduction(const ITensorInfo *input, const ITensorInfo *output)
39 {
42 
43  if(output->total_size() > 0)
44  {
46  ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->dimension(0) != input->dimension(1), "Output vector must have length equal to the number of rows of the input matrix");
47  }
48  return Status{};
49 }
50 
51 Status validate_arguments_matrix_b_reduction(const ITensorInfo *input, const ITensorInfo *output)
52 {
55 
56  if(output->total_size() > 0)
57  {
59  ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->dimension(0) != input->dimension(0), "Output vector must have length equal to the number of columns of the input matrix");
60  }
61  return Status{};
62 }
63 } // namespace
64 
66  : _input(), _output()
67 {
68 }
69 
71 {
72  configure(CLKernelLibrary::get().get_compile_context(), mtx_a, vector_sum_row, info);
73 }
74 
75 void CLGEMMLowpMatrixAReductionKernel::configure(const CLCompileContext &compile_context, const ICLTensor *mtx_a, ICLTensor *vector_sum_row, const GEMMLowpReductionKernelInfo &info)
76 {
77  // Perform validate step
78  ARM_COMPUTE_ERROR_ON_NULLPTR(mtx_a, vector_sum_row);
79  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_matrix_a_reduction(mtx_a->info(), vector_sum_row->info()));
80 
81  // Output auto initialization if not yet initialized
82  auto_init_if_empty(*vector_sum_row->info(), TensorShape(mtx_a->info()->dimension(1)), 1, DataType::S32);
83 
84  auto padding_info = get_padding_info({ mtx_a, vector_sum_row });
85 
86  _input = mtx_a;
87  _output = vector_sum_row;
88 
89  // Set the arguments to pass at compile time
90  CLBuildOptions build_opts;
91  build_opts.add_option("-DCOLS_A=" + support::cpp11::to_string(mtx_a->info()->dimension(0)));
92  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(mtx_a->info()->data_type()));
93  build_opts.add_option("-DACC_DATA_TYPE=" + get_cl_dot8_acc_type_from_data_type(mtx_a->info()->data_type()));
94  build_opts.add_option_if(info.mul_by_scalar, "-DSCALAR=" + support::cpp11::to_string(info.scalar));
95 
96  const bool is_dot8_supported = dot8_supported(CLKernelLibrary::get().get_device());
97 
98  std::string kernel_name = "gemmlowp_matrix_a_reduction" + std::string(is_dot8_supported ? "_dot8" : "");
99 
100  // Create kernel
101  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
102 
103  // Configure kernel window
104  // This kernel does not need padding
105  Window win = calculate_max_window(*vector_sum_row->info(), Steps());
106  ICLKernel::configure_internal(win);
107 
108  _config_id = kernel_name;
109  _config_id += "_";
110  _config_id += support::cpp11::to_string(_input->info()->dimension(0));
111  _config_id += "_";
112  _config_id += support::cpp11::to_string(_input->info()->dimension(1));
113  _config_id += "_";
114  _config_id += support::cpp11::to_string(_input->info()->dimension(2));
115 
117 }
118 
120 {
121  ARM_COMPUTE_UNUSED(info);
122  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_matrix_a_reduction(mtx_a, vector_sum_row));
123 
124  return Status{};
125 }
126 
127 void CLGEMMLowpMatrixAReductionKernel::run(const Window &window, cl::CommandQueue &queue)
128 {
131 
133  Window slice_in = collapsed.first_slice_window_2D();
134  Window slice_out = collapsed.first_slice_window_2D();
135 
136  // Setup input slice. Its dimensions are increased in the cl kernel.
137  slice_in.set(Window::DimX, Window::Dimension(0, 0, 0));
138  slice_in.set(Window::DimY, Window::Dimension(0, 0, 0));
139  slice_in.set(Window::DimZ, Window::Dimension(0, 0, 0));
140 
141  do
142  {
143  unsigned int idx = 0;
144  add_3D_tensor_argument(idx, _input, slice_in);
145  add_2D_tensor_argument(idx, _output, slice_out);
146  enqueue(queue, *this, slice_out, lws_hint());
147  }
148  while(collapsed.slide_window_slice_2D(slice_out));
149 }
150 
152 {
153  configure(CLKernelLibrary::get().get_compile_context(), mtx_b, vector_sum_col, info);
154 }
155 
156 void CLGEMMLowpMatrixBReductionKernel::configure(const CLCompileContext &compile_context, const ICLTensor *mtx_b, ICLTensor *vector_sum_col, const GEMMLowpReductionKernelInfo &info)
157 {
158  ARM_COMPUTE_ERROR_ON_NULLPTR(mtx_b, vector_sum_col);
159  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_matrix_b_reduction(mtx_b->info(), vector_sum_col->info()));
160 
161  _input = mtx_b;
162  _output = vector_sum_col;
163 
164  // Output auto initialization if not yet initialized
165  auto_init_if_empty(*_output->info(), TensorShape(mtx_b->info()->dimension(0)), 1, DataType::S32);
166 
167  auto padding_info = get_padding_info({ mtx_b, vector_sum_col });
168 
169  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(16, mtx_b->info()->dimension(0));
170 
171  // Set the arguments to pass at compile time
172  CLBuildOptions build_opts;
173  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
174  build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(mtx_b->info()->dimension(0) % num_elems_processed_per_iteration));
175  build_opts.add_option("-DCOLS_B=" + support::cpp11::to_string(mtx_b->info()->dimension(0)));
176  build_opts.add_option("-DROWS_B=" + support::cpp11::to_string(mtx_b->info()->dimension(1)));
177  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(mtx_b->info()->data_type()));
178  build_opts.add_option("-DACC_DATA_TYPE=" + get_cl_dot8_acc_type_from_data_type(mtx_b->info()->data_type()));
179  build_opts.add_option_if(info.mul_by_scalar, "-DSCALAR=" + support::cpp11::to_string(info.scalar));
180 
181  // Create kernel
182  _kernel = create_kernel(compile_context, "gemmlowp_matrix_b_reduction", build_opts.options());
183 
184  // Configure kernel window
185  Window win = calculate_max_window(*_output->info(), Steps(num_elems_processed_per_iteration));
186  ICLKernel::configure_internal(win);
187 
189 }
190 
192 {
193  ARM_COMPUTE_UNUSED(info);
194  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_matrix_b_reduction(mtx_b, vector_sum_col));
195 
196  return Status{};
197 }
198 
199 void CLGEMMLowpMatrixBReductionKernel::run(const Window &window, cl::CommandQueue &queue)
200 {
203 
205 
206  Window slice_out = collapsed.first_slice_window_2D();
207  Window slice_in = slice_out;
208 
209  slice_in.set(Window::DimY, Window::Dimension(0, 0, 0));
210  slice_in.set(Window::DimZ, Window::Dimension(0, 0, 0));
211 
212  do
213  {
214  unsigned int idx = 0;
215  add_3D_tensor_argument(idx, _input, slice_in);
216  add_2D_tensor_argument(idx, _output, slice_out);
217  enqueue(queue, *this, slice_out, lws_hint());
218  }
219  while(collapsed.slide_window_slice_2D(slice_out));
220 }
221 } // namespace arm_compute
Window first_slice_window_2D() const
First 2D slice of the window.
Definition: Window.h:283
void configure(const ICLTensor *mtx_b, ICLTensor *vector_sum_col, const GEMMLowpReductionKernelInfo &info) override
Initialise the kernel's input and output.
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
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
bool dot8_supported(const cl::Device &device)
Helper function to check whether the cl_arm_integer_dot_product_int8 extension is supported...
Definition: CLHelpers.cpp:239
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.
const StringSet & options() const
Gets the current options list set.
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:276
std::string get_cl_dot8_acc_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL dot8 accumulator type.
Definition: CLHelpers.cpp:173
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
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.
static Status validate(const ITensorInfo *mtx_b, const ITensorInfo *vector_sum_col, const GEMMLowpReductionKernelInfo &info)
Static function to check if given info will lead to a valid configuration of CLGEMMLowpMatrixBReducti...
#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
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Describe one of the image's dimensions with a start, end and step.
Definition: Window.h:77
Status class.
Definition: Error.h:52
int32_t scalar
Scalar value to multiply each reduced column/row by.
void add_3D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 3D tensor's parameters to the object's kernel's arguments starting from the index idx...
Definition: ICLKernel.h:172
bool slide_window_slice_2D(Window &slice) const
Slide the passed 2D window slice.
Definition: Window.h:323
Copyright (c) 2017-2021 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
1 channel, 1 S32 per channel
void add_option(std::string option)
Adds option to the existing build option list.
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:403
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
Window collapse_if_possible(const Window &full_window, size_t first, size_t last, bool *has_collapsed=nullptr) const
Collapse the dimensions between first and last if possible.
Definition: Window.inl:68
quantized, asymmetric fixed-point 8-bit number unsigned
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
std::string kernel_name
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp: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 ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
void add_option_if(bool cond, std::string option)
Adds option if a given condition is true;.
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
void configure(const ICLTensor *mtx_a, ICLTensor *vector_sum_row, const GEMMLowpReductionKernelInfo &info) override
Initialise the kernel&#39;s input and output.
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
bool has_padding_changed(const std::unordered_map< const ITensorInfo *, PaddingSize > &padding_map)
Check if the previously stored padding info has changed after configuring a kernel.
Definition: Utils.cpp:528
quantized, symmetric fixed-point 8-bit number
CLCompileContext class.
quantized, symmetric per channel fixed-point 8-bit number
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
void add_2D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, 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: ICLKernel.h:148
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:792
std::unordered_map< const ITensorInfo *, PaddingSize > get_padding_info(std::initializer_list< const ITensorInfo *> infos)
Stores padding information before configuring a kernel.
Definition: Utils.cpp:513
static Status validate(const ITensorInfo *mtx_a, const ITensorInfo *vector_sum_row, const GEMMLowpReductionKernelInfo &info)
Static function to check if given info will lead to a valid configuration of CLGEMMLowpMatrixAReducti...
virtual void configure(const ICLTensor *input, ICLTensor *output, const GEMMLowpReductionKernelInfo &info)=0
Initialise the kernel&#39;s input and output.
unsigned int num_elems_processed_per_iteration
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
unsigned int adjust_vec_size(unsigned int vec_size, size_t dim0)
Returns the adjusted vector size in case it is less than the input&#39;s first dimension, getting rounded down to its closest valid vector size.
Definition: Utils.h:1358
quantized, asymmetric fixed-point 8-bit number signed
bool mul_by_scalar
True if each column/row reduction has to be multiplied by a scalar value.
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205