Compute Library
 20.08
CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2020 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 
29 #include "arm_compute/core/Error.h"
32 #include "arm_compute/core/Types.h"
37 #include "support/StringSupport.h"
38 
39 namespace arm_compute
40 {
41 namespace
42 {
43 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output,
44  const GEMMLowpOutputStageInfo *info)
45 {
47  ARM_COMPUTE_RETURN_ERROR_ON((info->output_data_type != DataType::QASYMM8) && (info->output_data_type != DataType::QASYMM8_SIGNED));
48  ARM_COMPUTE_RETURN_ERROR_ON(info->gemmlowp_max_bound > std::get<1>(quantization::get_min_max_values_from_quantized_data_type(info->output_data_type)));
49  ARM_COMPUTE_RETURN_ERROR_ON(info->gemmlowp_min_bound < std::get<0>(quantization::get_min_max_values_from_quantized_data_type(info->output_data_type))
50  || info->gemmlowp_min_bound > info->gemmlowp_max_bound);
51 
52  // Check biases if exist
53  if(bias != nullptr)
54  {
56  ARM_COMPUTE_RETURN_ERROR_ON(bias->num_dimensions() > 1);
57  ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != bias->dimension(0));
58  }
59 
60  if(output->total_size() != 0)
61  {
62  ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->data_type() != info->output_data_type, "Mismatching output data type");
64  }
65 
66  return Status{};
67 }
68 
69 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *bias, ITensorInfo *output, DataType output_data_type)
70 {
71  // Output auto inizialitation if not yet initialized
72  auto_init_if_empty(*output, input->clone()->set_data_type(output_data_type));
73 
74  constexpr unsigned int num_elems_processed_per_iteration = 4;
75 
76  // Output auto inizialitation if not yet initialized
77  auto_init_if_empty(*output, input->clone()->set_data_type(DataType::QASYMM8));
78 
79  // Configure kernel window
81 
82  AccessWindowHorizontal input_access(input, 0, num_elems_processed_per_iteration);
83 
84  bool window_changed = update_window_and_padding(win,
85  input_access);
86 
87  AccessWindowHorizontal output_result_access(output, 0, num_elems_processed_per_iteration);
88  window_changed = window_changed || update_window_and_padding(win, output_result_access);
89  output_result_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
90 
91  if(bias != nullptr)
92  {
93  AccessWindowStatic bias_access(bias, 0, 0, ceil_to_multiple(bias->dimension(0), num_elems_processed_per_iteration), bias->tensor_shape()[1]);
94  window_changed = window_changed || update_window_and_padding(win, bias_access);
95  }
96 
97  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
98  return std::make_pair(err, win);
99 }
100 } // namespace
101 
102 class Coordinates;
104  : _input(nullptr), _bias(nullptr), _output(nullptr)
105 {
106 }
107 
110 {
113 
114  return Status{};
115 }
116 
119 {
120  configure(CLKernelLibrary::get().get_compile_context(), input, bias, output, info);
121 }
122 
125 {
126  // Perform validate step
128  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (bias != nullptr) ? bias->info() : nullptr, output->info(), info));
129 
130  _input = input;
131  _bias = bias;
132  _output = output;
133 
134  auto min = info->gemmlowp_min_bound;
135  auto max = info->gemmlowp_max_bound;
136 
137  // Set the arguments to pass at compile time
138  CLBuildOptions build_opts;
139  build_opts.add_option("-DREAL_MULTIPLIER=" + float_to_string_with_full_precision(info->gemmlowp_real_multiplier));
140  build_opts.add_option("-DOUTPUT_OFFSET=" + support::cpp11::to_string(info->gemmlowp_offset));
141  build_opts.add_option("-DOUTPUT_DATA_TYPE=" + get_cl_type_from_data_type(output->info()->data_type()));
142  build_opts.add_option_if((min > 0), "-DMIN_BOUND=" + support::cpp11::to_string(min));
143  build_opts.add_option_if((max < 255), "-DMAX_BOUND=" + support::cpp11::to_string(max));
144  build_opts.add_option_if(bias != nullptr, "-DADD_BIAS");
145 
146  // Create kernel
147  _kernel = create_kernel(compile_context, "gemmlowp_output_stage_quantize_down_float", build_opts.options());
148 
149  // Configure kernel window
150  auto win_config = validate_and_configure_window(input->info(), (bias != nullptr) ? bias->info() : nullptr, output->info(), info->output_data_type);
151  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
152  ICLKernel::configure_internal(win_config.second);
153 }
154 
155 void CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel::run(const Window &window, cl::CommandQueue &queue)
156 {
159 
160  // Create input window
162  Window slice = collapsed.first_slice_window_3D();
163 
164  // Setup bias slice
165  unsigned int idx1 = num_arguments_per_3D_tensor();
166  if(_bias != nullptr)
167  {
168  Window biases_slice(slice);
169  biases_slice.set(Window::DimY, Window::Dimension(0, 1, 1));
170  biases_slice.set(Window::DimZ, Window::Dimension(0, 1, 1));
171  add_1D_tensor_argument(idx1, _bias, biases_slice);
172  }
173 
174  do
175  {
176  unsigned int idx = 0;
177  add_3D_tensor_argument(idx, _input, slice);
178  add_3D_tensor_argument(idx1, _output, slice);
179  enqueue(queue, *this, slice, lws_hint());
180  }
181  while(collapsed.slide_window_slice_3D(slice));
182 }
183 } // namespace arm_compute
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
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
const StringSet & options() const
Gets the current options list set.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:263
#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.
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:792
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:75
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
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
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:159
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:207
ITensorInfo * info() const override
Interface to be implemented by the child class to return the tensor's metadata.
Definition: Tensor.cpp:33
1 channel, 1 S32 per channel
void add_option(std::string option)
Adds option to the existing build option list.
static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *info)
Static function to check if given info will lead to a valid configuration of CLGEMMLowpQuantizeDownIn...
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
std::pair< int, int > get_min_max_values_from_quantized_data_type(DataType data_type)
Get minimum and maximum values for the input quantized data type.
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: Helpers.h:437
static constexpr unsigned int num_arguments_per_3D_tensor()
Returns the number of arguments enqueued per 3D tensor object.
Definition: ICLKernel.h:201
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
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:443
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1215
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
quantized, asymmetric fixed-point 8-bit number unsigned
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
GEMMLowp output stage info.
Definition: Types.h:1881
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor'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
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:333
CLCompileContext class.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
#define ARM_COMPUTE_CREATE_ERROR(error_code, msg)
Creates an error with a given message.
Definition: Error.h:159
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
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
quantized, asymmetric fixed-point 8-bit number signed
void add_1D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 1D tensor's parameters to the object's kernel's arguments starting from the index idx.
Definition: ICLKernel.h:111
void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo *info)
Initialise the kernel's input and output.
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:289
DataType
Available data types.
Definition: Types.h:77
#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
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue.