Compute Library
 20.08
CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.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 
28 #include "arm_compute/core/Error.h"
31 #include "arm_compute/core/Types.h"
35 
36 #include "support/StringSupport.h"
37 
38 namespace arm_compute
39 {
40 namespace
41 {
42 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output,
43  int min, int max)
44 {
46  ARM_COMPUTE_RETURN_ERROR_ON(min > max);
47 
48  // Check biases if exist
49  if(bias != nullptr)
50  {
52  ARM_COMPUTE_RETURN_ERROR_ON(bias->num_dimensions() > 1);
53  ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != bias->dimension(0));
54  }
55 
56  if(output->total_size() != 0)
57  {
60  }
61 
62  return Status{};
63 }
64 
65 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *bias, ITensorInfo *output)
66 {
67  constexpr unsigned int num_elems_processed_per_iteration = 4;
68 
69  // Output auto inizialitation if not yet initialized
70  auto_init_if_empty(*output, input->clone()->set_data_type(DataType::QSYMM16));
71 
72  // Configure kernel window
74 
75  AccessWindowHorizontal input_access(input, 0, num_elems_processed_per_iteration);
76 
77  bool window_changed = update_window_and_padding(win, input_access);
78 
79  if(output->total_size() != 0)
80  {
81  Window win_out = calculate_max_window(*output, Steps(num_elems_processed_per_iteration));
82  AccessWindowHorizontal output_result_access(output, 0, num_elems_processed_per_iteration);
83  window_changed = window_changed || update_window_and_padding(win_out, output_result_access);
84 
85  output_result_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
86  }
87 
88  if(bias != nullptr)
89  {
90  AccessWindowStatic bias_access(bias, 0, 0, ceil_to_multiple(bias->dimension(0), num_elems_processed_per_iteration), bias->tensor_shape()[1]);
91  window_changed = window_changed || update_window_and_padding(win, bias_access);
92  }
93 
94  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
95  return std::make_pair(err, win);
96 }
97 } // namespace
98 
100  : _input(nullptr), _bias(nullptr), _output(nullptr)
101 {
102 }
103 
105  int min, int max)
106 {
109  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(),
110  (bias != nullptr) ? bias->clone().get() : nullptr,
111  output->clone().get())
112  .first);
113 
114  return Status{};
115 }
116 
118  int result_fixedpoint_multiplier, int result_shift,
119  int min, int max)
120 {
121  configure(CLKernelLibrary::get().get_compile_context(), input, bias, output, result_fixedpoint_multiplier, result_shift, min, max);
122 }
123 
125  int result_fixedpoint_multiplier, int result_shift,
126  int min, int max)
127 {
128  // Perform validate step
130  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (bias != nullptr) ? bias->info() : nullptr, output->info(),
131  min, max));
132 
133  _input = input;
134  _bias = bias;
135  _output = output;
136 
137  // Set the arguments to pass at compile time
138  CLBuildOptions build_opts;
139  build_opts.add_option("-DRESULT_FIXEDPOINT_MULTIPLIER=" + support::cpp11::to_string(result_fixedpoint_multiplier));
140  build_opts.add_option("-DRESULT_SHIFT=" + support::cpp11::to_string(result_shift));
141  build_opts.add_option_if((min > -32768), "-DMIN_BOUND=" + support::cpp11::to_string(min));
142  build_opts.add_option_if((max < 32767), "-DMAX_BOUND=" + support::cpp11::to_string(max));
143  build_opts.add_option_if(bias != nullptr, "-DADD_BIAS");
144 
145  // Create kernel
146  _kernel = create_kernel(compile_context, "gemmlowp_output_stage_quantize_down_fixedpoint_qsymm16", build_opts.options());
147 
148  // Configure kernel window
149  auto win_config = validate_and_configure_window(input->info(), (bias != nullptr) ? bias->info() : nullptr, output->info());
150  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
151  ICLKernel::configure_internal(win_config.second);
152 }
153 
154 void CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::run(const Window &window, cl::CommandQueue &queue)
155 {
158 
159  // Create input window
161  Window slice = collapsed.first_slice_window_3D();
162 
163  // Setup bias slice
164  unsigned int idx1 = num_arguments_per_3D_tensor();
165  if(_bias != nullptr)
166  {
167  Window biases_slice(slice);
168  biases_slice.set(Window::DimY, Window::Dimension(0, 1, 1));
169  biases_slice.set(Window::DimZ, Window::Dimension(0, 1, 1));
170  add_1D_tensor_argument(idx1, _bias, biases_slice);
171  }
172 
173  do
174  {
175  unsigned int idx = 0;
176  add_3D_tensor_argument(idx, _input, slice);
177  add_3D_tensor_argument(idx1, _output, slice);
178  enqueue(queue, *this, slice, lws_hint());
179  }
180  while(collapsed.slide_window_slice_3D(slice));
181 }
182 } // namespace arm_compute
static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min=0, int max=0)
Static function to check if given info will lead to a valid configuration of CLGEMMLowpQuantizeDownIn...
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
quantized, symmetric fixed-point 16-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.
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.
#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.
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
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
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
void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int min=0, int max=0)
Initialise the kernel's input and output.
virtual std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
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
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
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:289
#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)