Compute Library
 22.05
ClGemmLowpQuantizeDownInt32ScaleKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2020-2021 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 
31 
34 
35 #include "support/Cast.h"
36 #include "support/StringSupport.h"
37 
38 namespace arm_compute
39 {
40 namespace opencl
41 {
42 namespace kernels
43 {
44 namespace
45 {
46 Status validate_arguments(const ITensorInfo *src, const ITensorInfo *bias, const ITensorInfo *dst, const GEMMLowpOutputStageInfo *output_stage)
47 {
49  ARM_COMPUTE_RETURN_ERROR_ON((output_stage->output_data_type != DataType::QASYMM8) && (output_stage->output_data_type != DataType::QASYMM8_SIGNED));
50  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)));
51  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))
52  || output_stage->gemmlowp_min_bound > output_stage->gemmlowp_max_bound);
53 
54  // Check biases if exist
55  if(bias != nullptr)
56  {
58  ARM_COMPUTE_RETURN_ERROR_ON(bias->num_dimensions() > 1);
59  ARM_COMPUTE_RETURN_ERROR_ON(src->dimension(0) != bias->dimension(0));
60  }
61 
62  if(dst->total_size() != 0)
63  {
64  ARM_COMPUTE_RETURN_ERROR_ON_MSG(dst->data_type() != output_stage->output_data_type, "Mismatching output data type");
66  }
67 
68  return Status{};
69 }
70 } //namespace
71 
73 {
75 }
76 
78 {
80  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, bias, dst, output_stage));
81 
82  return Status{};
83 }
84 
86  const GEMMLowpOutputStageInfo *output_stage)
87 {
88  // Perform validate step
90  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, bias, dst, output_stage));
91 
92  auto padding_info = get_padding_info({ src, bias, dst });
93 
94  // Output auto inizialitation if not yet initialized
95  auto_init_if_empty(*dst, src->clone()->set_data_type(output_stage->output_data_type));
96 
97  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(4, src->dimension(0));
98 
99  // Set the arguments to pass at compile time
100  auto min = output_stage->gemmlowp_min_bound;
101  auto max = output_stage->gemmlowp_max_bound;
102  CLBuildOptions build_opts;
103  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
104  build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(src->dimension(0) % num_elems_processed_per_iteration));
105  build_opts.add_option("-DRESULT_OFFSET=" + support::cpp11::to_string(output_stage->gemmlowp_offset));
106  build_opts.add_option("-DRESULT_MULT_INT=" + support::cpp11::to_string(output_stage->gemmlowp_multiplier));
107  build_opts.add_option("-DRESULT_SHIFT=" + support::cpp11::to_string(output_stage->gemmlowp_shift));
108  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),
109  "-DMIN_BOUND=" + support::cpp11::to_string(min));
110  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),
111  "-DMAX_BOUND=" + support::cpp11::to_string(max));
112  build_opts.add_option("-DOUTPUT_DATA_TYPE=" + get_cl_type_from_data_type(dst->data_type()));
113  build_opts.add_option_if(bias != nullptr, "-DADD_BIAS");
114 
115  // Create kernel
116  _kernel = create_kernel(compile_context, "gemmlowp_output_stage_quantize_down", build_opts.options());
117 
118  // Configure kernel window
119  Window win = calculate_max_window(*src, Steps(num_elems_processed_per_iteration));
120  ICLKernel::configure_internal(win);
121 
123 }
124 
125 void ClGemmLowpQuantizeDownInt32ScaleKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
126 {
129 
130  const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
131  const auto bias = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_BIAS));
132  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
133 
135  Window slice = collapsed.first_slice_window_3D();
136 
137  unsigned int idx1 = num_arguments_per_3D_tensor();
138  if(bias != nullptr)
139  {
140  Window biases_slice(slice);
141  biases_slice.set(Window::DimY, Window::Dimension(0, 1, 1));
142  biases_slice.set(Window::DimZ, Window::Dimension(0, 1, 1));
143  add_1D_tensor_argument(idx1, bias, biases_slice);
144  }
145 
146  do
147  {
148  unsigned int idx = 0;
149  add_3D_tensor_argument(idx, src, slice);
150  add_3D_tensor_argument(idx1, dst, slice);
151  enqueue(queue, *this, slice, lws_hint());
152  }
153  while(collapsed.slide_window_slice_3D(slice));
154 }
155 } // namespace kernels
156 } // namespace opencl
157 } // namespace arm_compute
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:2042
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
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:32
const StringSet & options() const
Gets the current options list set.
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:384
void configure(const CLCompileContext &compile_context, const ITensorInfo *src, const ITensorInfo *bias, ITensorInfo *dst, const GEMMLowpOutputStageInfo *output_stage)
Initialise the kernel&#39;s source and destination.
#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
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
int32_t gemmlowp_offset
GEMMLowp output stage offset used for quantizing to QASYMM8.
Definition: Types.h:2041
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:2045
Status validate_arguments(const ITensorInfo *src, const ITensorInfo *weights, const ITensorInfo *dst, const PadStrideInfo &conv_info)
#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:227
SimpleTensor< float > src
Definition: DFT.cpp:155
Copyright (c) 2017-2022 Arm Limited.
1 channel, 1 S32 per channel
void add_option(std::string option)
Adds option to the existing build option list.
const OutputStage & output_stage
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
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:391
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:314
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
static Status validate(const ITensorInfo *src, const ITensorInfo *bias, const ITensorInfo *dst, const GEMMLowpOutputStageInfo *output_stage)
Static function to check if given info will lead to a valid configuration.
unsigned int num_elems_processed_per_iteration
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp:39
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:2038
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
Elementeise CL kernel type.
Definition: CLTypes.h:84
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
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:601
CLCompileContext class.
int32_t gemmlowp_shift
GEMMLowp output stage shift used for quantizing to uint8.
Definition: Types.h:2043
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:439
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:541
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:788
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:586
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
Tensor packing service.
Definition: ITensorPack.h:39
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
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:1222
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:2044
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:179
DataType output_data_type
Output tensor data type to use if the output is not initialized.
Definition: Types.h:2050
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)
const int32_t * bias