Compute Library
 23.11
ClWidthConcatenate4TensorsKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2023 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  */
24 #include "src/gpu/cl/kernels/ClWidthConcatenate4TensorsKernel.h"
25 
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/CLKernelLibrary.h"
28 #include "arm_compute/core/CL/ICLTensor.h"
29 #include "arm_compute/core/Helpers.h"
30 #include "arm_compute/core/Utils.h"
31 #include "arm_compute/core/utils/helpers/AdjustVecSize.h"
32 #include "arm_compute/core/utils/StringUtils.h"
33 
34 #include "src/core/CL/CLValidate.h"
35 #include "src/core/helpers/WindowHelpers.h"
36 #include "src/core/utils/helpers/tensor_info.h"
37 #include "support/Cast.h"
38 #include "support/StringSupport.h"
39 
40 namespace arm_compute
41 {
42 namespace opencl
43 {
44 namespace kernels
45 {
46 namespace
47 {
48 Status validate_arguments(const ITensorInfo *src1,
49  const ITensorInfo *src2,
50  const ITensorInfo *src3,
51  const ITensorInfo *src4,
52  const ITensorInfo *dst)
53 {
54  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, src3, src4, dst);
55  ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src1);
56  ARM_COMPUTE_RETURN_ERROR_ON(src1->data_type() == DataType::UNKNOWN);
57  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src1, src2, src3, src4, dst);
58  ARM_COMPUTE_RETURN_ERROR_ON(src1->dimension(0) + src2->dimension(0) + src3->dimension(0) + src4->dimension(0) >
59  dst->dimension(0));
60 
61  for (size_t i = 1; i < Coordinates::num_max_dimensions; ++i)
62  {
63  ARM_COMPUTE_RETURN_ERROR_ON(src1->dimension(i) != dst->dimension(i));
64  ARM_COMPUTE_RETURN_ERROR_ON(src2->dimension(i) != dst->dimension(i));
65  ARM_COMPUTE_RETURN_ERROR_ON(src3->dimension(i) != dst->dimension(i));
66  ARM_COMPUTE_RETURN_ERROR_ON(src4->dimension(i) != dst->dimension(i));
67  }
68  ARM_COMPUTE_RETURN_ERROR_ON(src1->num_dimensions() > 4);
69 
70  return Status{};
71 }
72 } // namespace
73 
74 ClWidthConcatenate4TensorsKernel::ClWidthConcatenate4TensorsKernel()
75 {
76  _type = CLKernelType::ELEMENTWISE;
77 }
78 
79 Status ClWidthConcatenate4TensorsKernel::validate(const ITensorInfo *src1,
80  const ITensorInfo *src2,
81  const ITensorInfo *src3,
82  const ITensorInfo *src4,
83  const ITensorInfo *dst)
84 {
85  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src1, src2, src3, src4, dst));
86  return Status{};
87 }
88 
89 void ClWidthConcatenate4TensorsKernel::configure(const CLCompileContext &compile_context,
90  ITensorInfo *src1,
91  ITensorInfo *src2,
92  ITensorInfo *src3,
93  ITensorInfo *src4,
94  ITensorInfo *dst)
95 {
96  ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, src3, src4, dst);
97  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src1, src2, src3, src4, dst));
98 
99  auto padding_info = get_padding_info({src1, src2, src3, src4, dst});
100  const unsigned int min_dimension =
101  std::min(std::min(src1->dimension(0), src2->dimension(0)), std::min(src3->dimension(0), src4->dimension(0)));
102  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(8, min_dimension);
103  const unsigned int vec_size_leftover = dst->dimension(0) % num_elems_processed_per_iteration;
104 
105  // Add build options
106  CLBuildOptions build_opts;
107  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src1->data_type()));
108  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
109  build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_leftover));
110  build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(src1->element_size()));
111  build_opts.add_option("-DINPUT1_ROTATE_N=" + support::cpp11::to_string((src1->dimension(0) - vec_size_leftover) %
112  num_elems_processed_per_iteration));
113  build_opts.add_option("-DINPUT2_ROTATE_N=" +
114  support::cpp11::to_string((src1->dimension(0) + src2->dimension(0) - vec_size_leftover) %
115  num_elems_processed_per_iteration));
116  build_opts.add_option("-DINPUT3_ROTATE_N=" + support::cpp11::to_string((src1->dimension(0) + src2->dimension(0) +
117  src3->dimension(0) - vec_size_leftover) %
118  num_elems_processed_per_iteration));
119 
120  _depth = src1->dimension(2);
121  _input1_width = src1->dimension(0);
122  _input2_width = src2->dimension(0);
123  _input3_width = src3->dimension(0);
124 
125  // If soources have different quantization info set quantization parameters needed for the re-quantization process
126  const bool have_different_qinfo =
127  helpers::tensor_info::tensors_have_different_quantization_info(dst, src1, src2, src3, src4);
128  if (is_data_type_quantized_asymmetric(src1->data_type()) && have_different_qinfo)
129  {
130  const UniformQuantizationInfo iq1_info = src1->quantization_info().uniform();
131  const UniformQuantizationInfo iq2_info = src2->quantization_info().uniform();
132  const UniformQuantizationInfo iq3_info = src3->quantization_info().uniform();
133  const UniformQuantizationInfo iq4_info = src4->quantization_info().uniform();
134  const UniformQuantizationInfo oq_info = dst->quantization_info().uniform();
135 
136  build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq1_info.offset));
137  build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1_info.scale));
138  build_opts.add_option("-DOFFSET_IN2=" + float_to_string_with_full_precision(iq2_info.offset));
139  build_opts.add_option("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2_info.scale));
140  build_opts.add_option("-DOFFSET_IN3=" + float_to_string_with_full_precision(iq3_info.offset));
141  build_opts.add_option("-DSCALE_IN3=" + float_to_string_with_full_precision(iq3_info.scale));
142  build_opts.add_option("-DOFFSET_IN4=" + float_to_string_with_full_precision(iq4_info.offset));
143  build_opts.add_option("-DSCALE_IN4=" + float_to_string_with_full_precision(iq4_info.scale));
144  build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
145  build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oq_info.scale));
146  }
147  std::string kernel_name = "concatenate_width_x4";
148 
149  // A macro guard to compile ONLY the kernel of interest
150  build_opts.add_option("-D" + upper_string(kernel_name));
151 
152  // Create kernel
153  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
154 
155  // Configure kernel window
156  Window win = calculate_max_window(*dst, Steps(num_elems_processed_per_iteration));
157  ICLKernel::configure_internal(win.collapse(win, Window::DimZ));
158 
159  ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
160 
161  // Set config_id for enabling LWS tuning
162  _config_id = "concatenate_width_x4_";
163  _config_id += lower_string(string_from_data_type(src1->data_type()));
164  _config_id += "_";
165  _config_id += support::cpp11::to_string(src1->dimension(0));
166  _config_id += "_";
167  _config_id += support::cpp11::to_string(src1->dimension(1));
168  _config_id += "_";
169  _config_id += support::cpp11::to_string(src2->dimension(0));
170  _config_id += "_";
171  _config_id += support::cpp11::to_string(src2->dimension(1));
172  _config_id += "_";
173  _config_id += support::cpp11::to_string(src3->dimension(0));
174  _config_id += "_";
175  _config_id += support::cpp11::to_string(src3->dimension(1));
176  _config_id += "_";
177  _config_id += support::cpp11::to_string(src4->dimension(0));
178  _config_id += "_";
179  _config_id += support::cpp11::to_string(src4->dimension(1));
180 }
181 
182 void ClWidthConcatenate4TensorsKernel::run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue)
183 {
184  ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
185  ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
186 
187  const auto src0 =
188  utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC));
189  const auto src1 =
190  utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 1));
191  const auto src2 =
192  utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 2));
193  const auto src3 =
194  utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 3));
195  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
196 
197  Window slice = window.first_slice_window_4D();
198 
199  do
200  {
201  unsigned int idx = 0;
202  add_4D_tensor_argument(idx, src0, slice);
203  add_4D_tensor_argument(idx, src1, slice);
204  add_4D_tensor_argument(idx, src2, slice);
205  add_4D_tensor_argument(idx, src3, slice);
206  add_4D_tensor_argument(idx, dst, slice);
207  _kernel.setArg<cl_int>(idx++, _depth);
208  _kernel.setArg<cl_int>(idx++, _input1_width);
209  _kernel.setArg<cl_int>(idx++, _input2_width);
210  _kernel.setArg<cl_int>(idx++, _input3_width);
211  enqueue(queue, *this, window, lws_hint());
212  } while (window.slide_window_slice_4D(slice));
213 }
214 } // namespace kernels
215 } // namespace opencl
216 } // namespace arm_compute
arm_compute::Steps
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
arm_compute::support::cpp11::to_string
std::string to_string(T &&value)
Convert integer and float values to string.
Definition: StringSupport.h:168
Cast.h
StringSupport.h
arm_compute::UniformQuantizationInfo::offset
int32_t offset
Definition: QuantizationInfo.h:63
AdjustVecSize.h
ICLTensor.h
arm_compute::opencl::kernels::ClWidthConcatenate4TensorsKernel::ClWidthConcatenate4TensorsKernel
ClWidthConcatenate4TensorsKernel()
Definition: ClWidthConcatenate4TensorsKernel.cpp:74
Helpers.h
arm_compute::CLBuildOptions::options
const StringSet & options() const
Gets the current options list set.
Definition: CLCompileContext.cpp:72
arm_compute::calculate_max_window
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
Definition: WindowHelpers.cpp:29
ClWidthConcatenate4TensorsKernel.h
arm_compute::test::validation::dst
auto dst
Definition: DFT.cpp:170
arm_compute::lower_string
std::string lower_string(const std::string &val)
Lower a given string.
Definition: StringUtils.cpp:38
arm_compute::cpu::kernels::validate_arguments
Status validate_arguments(const ITensorInfo *src, const ITensorInfo *weights, const ITensorInfo *dst, const PadStrideInfo &conv_info)
Definition: CpuDirectConv2dKernel.cpp:57
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:1079
arm_compute::ITensorInfo::element_size
virtual size_t element_size() const =0
Element size in bytes calculated as data_size() * num_channels()
arm_compute::ITensorPack::get_tensor
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:63
arm_compute::string_from_data_type
const std::string & string_from_data_type(DataType dt)
Convert a data type identity into a string.
Definition: DataTypeUtils.cpp:31
arm_compute::upper_string
std::string upper_string(const std::string &val)
Raise a given string to upper case.
Definition: StringUtils.cpp:45
arm_compute::UniformQuantizationInfo
Quantization info when assuming per layer quantization.
Definition: QuantizationInfo.h:42
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:677
CLKernelLibrary.h
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context.
arm_compute::helpers::tensor_info::tensors_have_different_quantization_info
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
StringUtils.h
ARM_COMPUTE_RETURN_ON_ERROR
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:205
arm_compute::ITensorInfo::dimension
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
ARM_COMPUTE_ERROR_ON_NULLPTR
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
ARM_COMPUTE_ERROR_ON
#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
arm_compute::ITensorPack::get_const_tensor
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:53
ARM_COMPUTE_ERROR_THROW_ON
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
tensor_info.h
arm_compute::ITensorPack
Tensor packing service.
Definition: ITensorPack.h:39
arm_compute::CLBuildOptions::add_option
void add_option(std::string option)
Adds option to the existing build option list.
Definition: CLCompileContext.cpp:41
arm_compute::CLCompileContext
CLCompileContext class.
Definition: CLCompileContext.h:204
arm_compute::Window::collapse
Window collapse(const Window &full_window, size_t first, size_t last=Coordinates::num_max_dimensions) const
Collapse the dimensions between first and last.
Definition: Window.inl:125
ARM_COMPUTE_RETURN_ERROR_ON
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:298
arm_compute::ACL_DST
@ ACL_DST
Definition: Types.h:55
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Definition: CLValidate.h:36
arm_compute::opencl::kernels::ClWidthConcatenate4TensorsKernel::validate
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.
Definition: ClWidthConcatenate4TensorsKernel.cpp:79
arm_compute::create_kernel
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:409
arm_compute::Status
Status class.
Definition: Error.h:52
WindowHelpers.h
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:203
arm_compute::float_to_string_with_full_precision
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: StringUtils.cpp:52
arm_compute::ITensorInfo::data_type
virtual DataType data_type() const =0
Data type used for each element of the tensor.
arm_compute::opencl::kernels::ClWidthConcatenate4TensorsKernel::configure
void configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *src3, ITensorInfo *src4, ITensorInfo *dst)
Initialise the kernel's sources and destination.
Definition: ClWidthConcatenate4TensorsKernel.cpp:89
arm_compute::QuantizationInfo::uniform
UniformQuantizationInfo uniform() const
Return per layer quantization info.
Definition: QuantizationInfo.h:140
arm_compute::opencl::kernels::ClWidthConcatenate4TensorsKernel::run_op
void run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue) override
Definition: ClWidthConcatenate4TensorsKernel.cpp:182
arm_compute::ICLKernel::add_4D_tensor_argument
void add_4D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 4D tensor's parameters to the object's kernel's arguments starting from the index idx.
Definition: ICLKernel.h:244
arm_compute::ACL_SRC_VEC
@ ACL_SRC_VEC
Definition: Types.h:68
arm_compute::ITensorInfo::quantization_info
virtual QuantizationInfo quantization_info() const =0
Get the quantization settings (scale and offset) of the tensor.
CLValidate.h
arm_compute::IKernel::window
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
arm_compute::get_cl_type_from_data_type
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp:41
Utils.h
arm_compute::UniformQuantizationInfo::scale
float scale
Definition: QuantizationInfo.h:62
arm_compute::Window
Describe a multidimensional execution window.
Definition: Window.h:39
arm_compute::ELEMENTWISE
@ ELEMENTWISE
Elementwise CL kernel type.
Definition: CLTypes.h:83
arm_compute
Copyright (c) 2017-2023 Arm Limited.
Definition: introduction.dox:24
arm_compute::adjust_vec_size
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's first dimension,...
Definition: AdjustVecSize.h:38
arm_compute::has_padding_changed
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:491
arm_compute::Window::DimZ
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
arm_compute::ICLKernel::lws_hint
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:383
arm_compute::is_data_type_quantized_asymmetric
bool is_data_type_quantized_asymmetric(DataType dt)
Check if a given data type is of asymmetric quantized type.
Definition: DataTypeUtils.h:346
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
arm_compute::ITensorInfo
Store the tensor's metadata.
Definition: ITensorInfo.h:44
arm_compute::Window::first_slice_window_4D
Window first_slice_window_4D() const
First 4D slice of the window.
Definition: Window.h:314
arm_compute::DataType::UNKNOWN
@ UNKNOWN
Unknown data type.
arm_compute::CLBuildOptions
Build options.
Definition: CLCompileContext.h:38
arm_compute::Window::slide_window_slice_4D
bool slide_window_slice_4D(Window &slice) const
Slide the passed 4D window slice.
Definition: Window.h:362
num_elems_processed_per_iteration
unsigned int num_elems_processed_per_iteration
Definition: ClIm2ColKernel.cpp:60
arm_compute::get_padding_info
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:476
arm_compute::test::validation::reference::slice
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)
Definition: SliceOperations.cpp:38
arm_compute::Dimensions< int >::num_max_dimensions
static constexpr size_t num_max_dimensions
Number of dimensions the tensor has.
Definition: Dimensions.h:46
CLHelpers.h
kernel_name
std::string kernel_name
Definition: ClIm2ColKernel.cpp:58
arm_compute::enqueue
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:33