Compute Library
 21.02
ClWidthConcatenate4TensorsKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2021 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 
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 {
69 }
70 
71 Status ClWidthConcatenate4TensorsKernel::validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *src3, const ITensorInfo *src4, const ITensorInfo *dst)
72 {
73  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src1, src2, src3, src4, dst));
74  return Status{};
75 }
76 
78  ITensorInfo *src1, ITensorInfo *src2,
79  ITensorInfo *src3, ITensorInfo *src4,
80  ITensorInfo *dst)
81 {
82  ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, src3, src4, dst);
83  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src1, src2, src3, src4, dst));
84 
85  auto padding_info = get_padding_info({ src1, src2, src3, src4, dst });
86  const unsigned int min_dimension = std::min(std::min(src1->dimension(0), src2->dimension(0)), std::min(src3->dimension(0), src4->dimension(0)));
87  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(8, min_dimension);
88  const unsigned int vec_size_leftover = dst->dimension(0) % num_elems_processed_per_iteration;
89 
90  // Add build options
91  CLBuildOptions build_opts;
92  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src1->data_type()));
93  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
94  build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_leftover));
95  build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(src1->dimension(2)));
96  build_opts.add_option("-DINPUT1_WIDTH=" + support::cpp11::to_string(src1->dimension(0)));
97  build_opts.add_option("-DINPUT2_WIDTH=" + support::cpp11::to_string(src2->dimension(0)));
98  build_opts.add_option("-DINPUT3_WIDTH=" + support::cpp11::to_string(src3->dimension(0)));
99  build_opts.add_option("-DINPUT4_WIDTH=" + support::cpp11::to_string(src4->dimension(0)));
100  build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(src1->element_size()));
101  build_opts.add_option("-DINPUT1_ROTATE_N=" + support::cpp11::to_string((src1->dimension(0) - vec_size_leftover) % num_elems_processed_per_iteration));
102  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));
103  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));
104 
105  // If soources have different quantization info set quantization parameters needed for the re-quantization process
106  const bool have_different_qinfo = helpers::tensor_info::tensors_have_different_quantization_info(dst, src1, src2, src3, src4);
107  if(is_data_type_quantized_asymmetric(src1->data_type()) && have_different_qinfo)
108  {
109  const UniformQuantizationInfo iq1_info = src1->quantization_info().uniform();
110  const UniformQuantizationInfo iq2_info = src2->quantization_info().uniform();
111  const UniformQuantizationInfo iq3_info = src3->quantization_info().uniform();
112  const UniformQuantizationInfo iq4_info = src4->quantization_info().uniform();
113  const UniformQuantizationInfo oq_info = dst->quantization_info().uniform();
114 
115  build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq1_info.offset));
116  build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1_info.scale));
117  build_opts.add_option("-DOFFSET_IN2=" + float_to_string_with_full_precision(iq2_info.offset));
118  build_opts.add_option("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2_info.scale));
119  build_opts.add_option("-DOFFSET_IN3=" + float_to_string_with_full_precision(iq3_info.offset));
120  build_opts.add_option("-DSCALE_IN3=" + float_to_string_with_full_precision(iq3_info.scale));
121  build_opts.add_option("-DOFFSET_IN4=" + float_to_string_with_full_precision(iq4_info.offset));
122  build_opts.add_option("-DSCALE_IN4=" + float_to_string_with_full_precision(iq4_info.scale));
123  build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
124  build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oq_info.scale));
125  }
126 
127  // Create kernel
128  _kernel = create_kernel(compile_context, "concatenate_width_x4", build_opts.options());
129 
130  // Configure kernel window
131  Window win = calculate_max_window(*dst, Steps(num_elems_processed_per_iteration));
132  ICLKernel::configure_internal(win.collapse(win, Window::DimZ));
133 
134  // Set dst valid region
137 
138  // Set config_id for enabling LWS tuning
139  _config_id = "concatenate_width_x4_";
140  _config_id += lower_string(string_from_data_type(src1->data_type()));
141  _config_id += "_";
142  _config_id += support::cpp11::to_string(src1->dimension(0));
143  _config_id += "_";
144  _config_id += support::cpp11::to_string(src1->dimension(1));
145  _config_id += "_";
146  _config_id += support::cpp11::to_string(src2->dimension(0));
147  _config_id += "_";
148  _config_id += support::cpp11::to_string(src2->dimension(1));
149  _config_id += "_";
150  _config_id += support::cpp11::to_string(src3->dimension(0));
151  _config_id += "_";
152  _config_id += support::cpp11::to_string(src3->dimension(1));
153  _config_id += "_";
154  _config_id += support::cpp11::to_string(src4->dimension(0));
155  _config_id += "_";
156  _config_id += support::cpp11::to_string(src4->dimension(1));
157 }
158 
159 void ClWidthConcatenate4TensorsKernel::run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue)
160 {
163 
164  const auto src0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC));
165  const auto src1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 1));
166  const auto src2 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 2));
167  const auto src3 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_VEC + 3));
168  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
169 
171 
172  do
173  {
174  unsigned int idx = 0;
175  add_4D_tensor_argument(idx, src0, slice);
176  add_4D_tensor_argument(idx, src1, slice);
177  add_4D_tensor_argument(idx, src2, slice);
178  add_4D_tensor_argument(idx, src3, slice);
179  add_4D_tensor_argument(idx, dst, slice);
180  enqueue(queue, *this, window, lws_hint());
181  }
182  while(window.slide_window_slice_4D(slice));
183 }
184 } // namespace kernels
185 } // namespace opencl
186 } // 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
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
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.
Status class.
Definition: Error.h:52
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:350
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
Copyright (c) 2017-2021 Arm Limited.
virtual void set_valid_region(const ValidRegion &valid_region)=0
Set the valid region of the tensor.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
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:40
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
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 of ClWidthConcatenate4Tenso...
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1262
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
Coordinates of an item.
Definition: Coordinates.h:37
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:37
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()
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: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.
bool is_data_type_quantized_asymmetric(DataType dt)
Check if a given data type is of asymmetric quantized type.
Definition: Utils.h:1190
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:50
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
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context...
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
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:513
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
Window first_slice_window_4D() const
First 4D slice of the window.
Definition: Window.h:299
bool slide_window_slice_4D(Window &slice) const
Slide the passed 4D window slice.
Definition: Window.h:347
unsigned int num_elems_processed_per_iteration
Tensor packing service.
Definition: ITensorPack.h:37
#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
Container for valid region of a window.
Definition: Types.h:188
static constexpr size_t num_max_dimensions
Number of dimensions the tensor has.
Definition: Dimensions.h:46
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:182
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)