Compute Library
 21.02
CLGEMMLowpQuantizeDownInt32ScaleKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 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 
34 #include "support/StringSupport.h"
35 
36 namespace arm_compute
37 {
38 namespace
39 {
40 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
41 {
43  ARM_COMPUTE_RETURN_ERROR_ON((output_stage->output_data_type != DataType::QASYMM8) && (output_stage->output_data_type != DataType::QASYMM8_SIGNED));
44  ARM_COMPUTE_RETURN_ERROR_ON(output_stage->gemmlowp_max_bound > std::get<1>(quantization::get_min_max_values_from_quantized_data_type(output_stage->output_data_type)));
45  ARM_COMPUTE_RETURN_ERROR_ON(output_stage->gemmlowp_min_bound < std::get<0>(quantization::get_min_max_values_from_quantized_data_type(output_stage->output_data_type))
46  || output_stage->gemmlowp_min_bound > output_stage->gemmlowp_max_bound);
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  {
58  ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->data_type() != output_stage->output_data_type, "Mismatching output data type");
60  }
61 
62  return Status{};
63 }
64 } //namespace
65 
67  : _input(nullptr), _bias(nullptr), _output(nullptr)
68 {
69 }
70 
72 {
73  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
74  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, bias, output, output_stage));
75 
76  return Status{};
77 }
78 
79 void CLGEMMLowpQuantizeDownInt32ScaleKernel::configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo *output_stage)
80 {
81  configure(CLKernelLibrary::get().get_compile_context(), input, bias, output, output_stage);
82 }
83 
84 void CLGEMMLowpQuantizeDownInt32ScaleKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output,
85  const GEMMLowpOutputStageInfo *output_stage)
86 {
87  // Perform validate step
88  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
89 
91  (bias != nullptr) ? bias->info() : nullptr,
92  output->info(),
93  output_stage));
94 
95  auto padding_info = get_padding_info({ input, bias, output });
96 
97  // Output auto inizialitation if not yet initialized
98  auto_init_if_empty(*output->info(), input->info()->clone()->set_data_type(output_stage->output_data_type));
99 
100  _input = input;
101  _bias = bias;
102  _output = output;
103 
104  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(4, input->info()->dimension(0));
105 
106  // Set the arguments to pass at compile time
107  auto min = output_stage->gemmlowp_min_bound;
108  auto max = output_stage->gemmlowp_max_bound;
109  CLBuildOptions build_opts;
110  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
111  build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(input->info()->dimension(0) % num_elems_processed_per_iteration));
112  build_opts.add_option("-DRESULT_OFFSET=" + support::cpp11::to_string(output_stage->gemmlowp_offset));
113  build_opts.add_option("-DRESULT_MULT_INT=" + support::cpp11::to_string(output_stage->gemmlowp_multiplier));
114  build_opts.add_option("-DRESULT_SHIFT=" + support::cpp11::to_string(output_stage->gemmlowp_shift));
115  build_opts.add_option_if((min > std::get<0>(quantization::get_min_max_values_from_quantized_data_type(output_stage->output_data_type))) && (min != max),
116  "-DMIN_BOUND=" + support::cpp11::to_string(min));
117  build_opts.add_option_if((max < std::get<1>(quantization::get_min_max_values_from_quantized_data_type(output_stage->output_data_type))) && (min != max),
118  "-DMAX_BOUND=" + support::cpp11::to_string(max));
119  build_opts.add_option("-DOUTPUT_DATA_TYPE=" + get_cl_type_from_data_type(output->info()->data_type()));
120  build_opts.add_option_if(bias != nullptr, "-DADD_BIAS");
121 
122  // Create kernel
123  _kernel = create_kernel(compile_context, "gemmlowp_output_stage_quantize_down", build_opts.options());
124 
125  // Configure kernel window
126  Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration));
127  ICLKernel::configure_internal(win);
128 
130 }
131 
132 void CLGEMMLowpQuantizeDownInt32ScaleKernel::run(const Window &window, cl::CommandQueue &queue)
133 {
136 
138  Window slice = collapsed.first_slice_window_3D();
139 
140  unsigned int idx1 = num_arguments_per_3D_tensor();
141  if(_bias != nullptr)
142  {
143  Window biases_slice(slice);
144  biases_slice.set(Window::DimY, Window::Dimension(0, 1, 1));
145  biases_slice.set(Window::DimZ, Window::Dimension(0, 1, 1));
146  add_1D_tensor_argument(idx1, _bias, biases_slice);
147  }
148 
149  do
150  {
151  unsigned int idx = 0;
152  add_3D_tensor_argument(idx, _input, slice);
153  add_3D_tensor_argument(idx1, _output, slice);
154  enqueue(queue, *this, slice, lws_hint());
155  }
156  while(collapsed.slide_window_slice_3D(slice));
157 }
158 }
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
int32_t gemmlowp_multiplier
GEMMLowp output stage multiplier used for quantizing to QASYMM8.
Definition: Types.h:1956
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
Static function to check if given info will lead to a valid configuration of CLGEMMLowpQuantizeDownIn...
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
#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_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&#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:77
int32_t gemmlowp_offset
GEMMLowp output stage offset used for quantizing to QASYMM8.
Definition: Types.h:1955
Status class.
Definition: Error.h:52
int32_t gemmlowp_max_bound
GEMMLowp max value used to saturate down the output result before converting back to QASYMM8...
Definition: Types.h:1959
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
void add_3D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 3D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: ICLKernel.h:172
Copyright (c) 2017-2021 Arm Limited.
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
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.
static constexpr unsigned int num_arguments_per_3D_tensor()
Returns the number of arguments enqueued per 3D tensor object.
Definition: ICLKernel.h:214
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
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
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 std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
GEMMLowp output stage info.
Definition: Types.h:1952
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
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:335
#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
CLCompileContext class.
int32_t gemmlowp_shift
GEMMLowp output stage shift used for quantizing to uint8.
Definition: Types.h:1957
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:443
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
#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
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
#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
int32_t gemmlowp_min_bound
GEMMLowp min value used to saturate down the output result before converting back to QASYMM8...
Definition: Types.h:1958
void add_1D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, 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: ICLKernel.h:124
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:291
DataType output_data_type
Output tensor data type to use if the output is not initialized.
Definition: Types.h:1964
void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo *output_stage)
Initialise the kernel&#39;s input and output.
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)