Compute Library
 22.08
ClWidthConcatenate4TensorsKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2022 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 
30 #include "arm_compute/core/Utils.h"
31 #include "src/core/CL/CLValidate.h"
34 #include "support/Cast.h"
35 
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 *src1, const ITensorInfo *src2, const ITensorInfo *src3, const ITensorInfo *src4, const ITensorInfo *dst)
47 {
48  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, src3, src4, dst);
50  ARM_COMPUTE_RETURN_ERROR_ON(src1->data_type() == DataType::UNKNOWN);
51  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src1, src2, src3, src4, dst);
52  ARM_COMPUTE_RETURN_ERROR_ON(src1->dimension(0) + src2->dimension(0) + src3->dimension(0) + src4->dimension(0) > dst->dimension(0));
53 
54  for(size_t i = 1; i < Coordinates::num_max_dimensions; ++i)
55  {
56  ARM_COMPUTE_RETURN_ERROR_ON(src1->dimension(i) != dst->dimension(i));
57  ARM_COMPUTE_RETURN_ERROR_ON(src2->dimension(i) != dst->dimension(i));
58  ARM_COMPUTE_RETURN_ERROR_ON(src3->dimension(i) != dst->dimension(i));
59  ARM_COMPUTE_RETURN_ERROR_ON(src4->dimension(i) != dst->dimension(i));
60  }
61  ARM_COMPUTE_RETURN_ERROR_ON(src1->num_dimensions() > 4);
62 
63  return Status{};
64 }
65 } // namespace
66 
68 {
70 }
71 
72 Status ClWidthConcatenate4TensorsKernel::validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *src3, const ITensorInfo *src4, const ITensorInfo *dst)
73 {
74  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src1, src2, src3, src4, dst));
75  return Status{};
76 }
77 
79  ITensorInfo *src1, ITensorInfo *src2,
80  ITensorInfo *src3, ITensorInfo *src4,
81  ITensorInfo *dst)
82 {
83  ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, src3, src4, dst);
84  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src1, src2, src3, src4, dst));
85 
86  auto padding_info = get_padding_info({ src1, src2, src3, src4, dst });
87  const unsigned int min_dimension = std::min(std::min(src1->dimension(0), src2->dimension(0)), std::min(src3->dimension(0), src4->dimension(0)));
88  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(8, min_dimension);
89  const unsigned int vec_size_leftover = dst->dimension(0) % num_elems_processed_per_iteration;
90 
91  // Add build options
92  CLBuildOptions build_opts;
93  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src1->data_type()));
94  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
95  build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_leftover));
96  build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(src1->element_size()));
97  build_opts.add_option("-DINPUT1_ROTATE_N=" + support::cpp11::to_string((src1->dimension(0) - vec_size_leftover) % num_elems_processed_per_iteration));
98  build_opts.add_option("-DINPUT2_ROTATE_N=" + support::cpp11::to_string((src1->dimension(0) + src2->dimension(0) - vec_size_leftover) % num_elems_processed_per_iteration));
99  build_opts.add_option("-DINPUT3_ROTATE_N=" + support::cpp11::to_string((src1->dimension(0) + src2->dimension(0) + src3->dimension(0) - vec_size_leftover) % num_elems_processed_per_iteration));
100 
101  _depth = src1->dimension(2);
102  _input1_width = src1->dimension(0);
103  _input2_width = src2->dimension(0);
104  _input3_width = src3->dimension(0);
105 
106  // If soources have different quantization info set quantization parameters needed for the re-quantization process
107  const bool have_different_qinfo = helpers::tensor_info::tensors_have_different_quantization_info(dst, src1, src2, src3, src4);
108  if(is_data_type_quantized_asymmetric(src1->data_type()) && have_different_qinfo)
109  {
110  const UniformQuantizationInfo iq1_info = src1->quantization_info().uniform();
111  const UniformQuantizationInfo iq2_info = src2->quantization_info().uniform();
112  const UniformQuantizationInfo iq3_info = src3->quantization_info().uniform();
113  const UniformQuantizationInfo iq4_info = src4->quantization_info().uniform();
114  const UniformQuantizationInfo oq_info = dst->quantization_info().uniform();
115 
116  build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq1_info.offset));
117  build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1_info.scale));
118  build_opts.add_option("-DOFFSET_IN2=" + float_to_string_with_full_precision(iq2_info.offset));
119  build_opts.add_option("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2_info.scale));
120  build_opts.add_option("-DOFFSET_IN3=" + float_to_string_with_full_precision(iq3_info.offset));
121  build_opts.add_option("-DSCALE_IN3=" + float_to_string_with_full_precision(iq3_info.scale));
122  build_opts.add_option("-DOFFSET_IN4=" + float_to_string_with_full_precision(iq4_info.offset));
123  build_opts.add_option("-DSCALE_IN4=" + float_to_string_with_full_precision(iq4_info.scale));
124  build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
125  build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oq_info.scale));
126  }
127  std::string kernel_name = "concatenate_width_x4";
128 
129  // A macro guard to compile ONLY the kernel of interest
130  build_opts.add_option("-D" + upper_string(kernel_name));
131 
132  // Create kernel
133  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
134 
135  // Configure kernel window
136  Window win = calculate_max_window(*dst, Steps(num_elems_processed_per_iteration));
137  ICLKernel::configure_internal(win.collapse(win, Window::DimZ));
138 
140 
141  // Set config_id for enabling LWS tuning
142  _config_id = "concatenate_width_x4_";
143  _config_id += lower_string(string_from_data_type(src1->data_type()));
144  _config_id += "_";
145  _config_id += support::cpp11::to_string(src1->dimension(0));
146  _config_id += "_";
147  _config_id += support::cpp11::to_string(src1->dimension(1));
148  _config_id += "_";
149  _config_id += support::cpp11::to_string(src2->dimension(0));
150  _config_id += "_";
151  _config_id += support::cpp11::to_string(src2->dimension(1));
152  _config_id += "_";
153  _config_id += support::cpp11::to_string(src3->dimension(0));
154  _config_id += "_";
155  _config_id += support::cpp11::to_string(src3->dimension(1));
156  _config_id += "_";
157  _config_id += support::cpp11::to_string(src4->dimension(0));
158  _config_id += "_";
159  _config_id += support::cpp11::to_string(src4->dimension(1));
160 }
161 
162 void ClWidthConcatenate4TensorsKernel::run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue)
163 {
166 
167  const auto src0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC));
168  const auto src1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 1));
169  const auto src2 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 2));
170  const auto src3 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 3));
171  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
172 
174 
175  do
176  {
177  unsigned int idx = 0;
178  add_4D_tensor_argument(idx, src0, slice);
179  add_4D_tensor_argument(idx, src1, slice);
180  add_4D_tensor_argument(idx, src2, slice);
181  add_4D_tensor_argument(idx, src3, slice);
182  add_4D_tensor_argument(idx, dst, slice);
183  _kernel.setArg<cl_int>(idx++, _depth);
184  _kernel.setArg<cl_int>(idx++, _input1_width);
185  _kernel.setArg<cl_int>(idx++, _input2_width);
186  _kernel.setArg<cl_int>(idx++, _input3_width);
187  enqueue(queue, *this, window, lws_hint());
188  }
189  while(window.slide_window_slice_4D(slice));
190 }
191 } // namespace kernels
192 } // namespace opencl
193 } // namespace arm_compute
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Definition: CLValidate.h:35
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
#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
Quantization info when assuming per layer quantization.
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context...
Status class.
Definition: Error.h:52
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:351
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
Copyright (c) 2017-2022 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
void add_option(std::string option)
Adds option to the existing build option list.
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
std::string upper_string(const std::string &val)
Raise a given string to upper case.
Definition: Utils.cpp:358
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:404
const std::string & string_from_data_type(DataType dt)
Convert a data type identity into a string.
Definition: Utils.cpp:135
static Status validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *src3, const ITensorInfo *src4, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1124
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
unsigned int num_elems_processed_per_iteration
UniformQuantizationInfo uniform() const
Return per layer quantization info.
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
void configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *src3, ITensorInfo *src4, ITensorInfo *dst)
Initialise the kernel&#39;s sources and destination.
virtual size_t element_size() const =0
Element size in bytes calculated as data_size() * num_channels()
Elementwise CL kernel type.
Definition: CLTypes.h:85
virtual QuantizationInfo quantization_info() const =0
Get the quantization settings (scale and offset) of the tensor.
#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.
bool is_data_type_quantized_asymmetric(DataType dt)
Check if a given data type is of asymmetric quantized type.
Definition: Utils.h:1052
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
bool tensors_have_different_quantization_info(const ITensorInfo *tensor_info_1, const ITensorInfo *tensor_info_2, Ts... tensor_infos)
Checks if the quantization info of given tensors are different.
Definition: tensor_info.h:44
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
void run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue) override
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
Window first_slice_window_4D() const
First 4D slice of the window.
Definition: Window.h:313
bool slide_window_slice_4D(Window &slice) const
Slide the passed 4D window slice.
Definition: Window.h:361
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
static constexpr size_t num_max_dimensions
Number of dimensions the tensor has.
Definition: Dimensions.h:46
std::string kernel_name
void add_4D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 4D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: ICLKernel.h:237
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)