Compute Library
 21.11
ClDequantizeKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2021 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"
32 
33 #include "src/core/CL/CLValidate.h"
36 
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 *src, const ITensorInfo *dst)
49 {
52 
53  if(dst->tensor_shape().total_size() > 0)
54  {
58  }
59 
60  return Status{};
61 }
62 } // namespace
63 
65 {
67 }
68 
70 {
72 
73  // Output tensor auto initialization if not yet initialized
75 
76  auto padding_info = get_padding_info({ src, dst });
77 
78  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst));
79 
80  const int vec_size_x = 16 / dst->element_size();
81  const int output_width_x = dst->tensor_shape().x();
82  const bool multi_access_x = (output_width_x / vec_size_x > 0);
83 
84  const bool is_quantized_per_channel = is_data_type_quantized_per_channel(src->data_type());
85  std::string kernel_name = "dequantization_layer";
86 
87  // Create kernel
88  CLBuildOptions build_opts;
89  if(!is_quantized_per_channel)
90  {
92  const int qoffset = is_data_type_quantized_asymmetric(src->data_type()) ? qinfo.offset : 0;
93  build_opts.add_option("-DSCALE=" + float_to_string_with_full_precision(qinfo.scale));
94  build_opts.add_option("-DOFFSET=" + support::cpp11::to_string(qoffset));
95  }
96  else
97  {
98  kernel_name += "_per_channel";
99  kernel_name += src->data_layout() == DataLayout::NCHW ? "_nchw" : "_nhwc";
100  }
101 
102  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
103  build_opts.add_option("-DDATA_TYPE_SRC=" + get_cl_type_from_data_type(src->data_type()));
104  build_opts.add_option("-DDATA_TYPE_DST=" + get_cl_type_from_data_type(dst->data_type()));
105  build_opts.add_option_if(multi_access_x, "-DLAST_ACCESSED_X=" + support::cpp11::to_string(std::max<int>(output_width_x - vec_size_x, 0)));
106 
107  // Create kernel name
108  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
109 
110  // Configure kernel window
111  Window win = calculate_max_window(*dst);
112  if(multi_access_x)
113  {
114  win.set(Window::DimX, Window::Dimension(win.x().start(), ceil_to_multiple(win.x().end(), vec_size_x), vec_size_x));
115  }
116  ICLKernel::configure_internal(win);
117 
119 }
120 
122 {
123  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst));
124  return Status{};
125 }
126 
127 void ClDequantizeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
128 {
131 
132  auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
133  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
134 
135  const bool is_quantized_per_channel = is_data_type_quantized_per_channel(src->info()->data_type());
136 
137  // Collapse windo
138  Window new_window = is_quantized_per_channel ? window.collapse_if_possible(ICLKernel::window(), 4) : window.collapse_if_possible(ICLKernel::window(), 3);
139  Window slice = new_window.first_slice_window_3D();
140 
141  if(is_quantized_per_channel)
142  {
143  unsigned int idx = num_arguments_per_3D_tensor() * 2; //Skip the input and output parameters
144  _kernel.setArg(idx++, src->quantization().scale->cl_buffer());
145  }
146 
147  do
148  {
149  unsigned int idx = 0;
150  add_3D_tensor_argument(idx, src, slice);
151  add_3D_tensor_argument(idx, dst, slice);
152  enqueue(queue, *this, slice, lws_hint());
153  }
154  while(new_window.slide_window_slice_3D(slice));
155 }
156 } // namespace kernels
157 } // namespace opencl
158 } // 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
quantized, symmetric fixed-point 16-bit number
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:318
#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.
1 channel, 1 F32 per channel
#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.
Describe one of the image&#39;s dimensions with a start, end and step.
Definition: Window.h:77
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context...
Status class.
Definition: Error.h:52
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:214
SimpleTensor< float > src
Definition: DFT.cpp:155
Copyright (c) 2017-2021 Arm Limited.
1 channel, 1 F16 per channel
#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.
T x() const
Alias to access the size of the first dimension.
Definition: Dimensions.h:87
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
static Status validate(const ITensorInfo *src, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
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
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
static constexpr unsigned int num_arguments_per_3D_tensor()
Returns the number of arguments enqueued per 3D tensor object.
Definition: ICLKernel.h:256
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
bool is_data_type_quantized_per_channel(DataType dt)
Check if a given data type is of per channel type.
Definition: Utils.h:1058
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1075
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
auto ceil_to_multiple(S value, T divisor) -> decltype(((value+divisor - 1)/divisor) *divisor)
Computes the smallest number larger or equal to value that is a multiple of divisor.
Definition: Utils.h:71
quantized, asymmetric fixed-point 8-bit number unsigned
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
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...
void add_option_if(bool cond, std::string option)
Adds option if a given condition is true;.
virtual size_t element_size() const =0
Element size in bytes calculated as data_size() * num_channels()
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...
Elementeise CL kernel type.
Definition: CLTypes.h:84
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
void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst)
Initialise the kernel&#39;s input and output.
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:533
quantized, symmetric fixed-point 8-bit number
Num samples, channels, height, width.
CLCompileContext class.
bool is_data_type_quantized_asymmetric(DataType dt)
Check if a given data type is of asymmetric quantized type.
Definition: Utils.h:1003
quantized, symmetric per channel fixed-point 8-bit number
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:439
const QuantizationInfo qinfo
Definition: Im2Col.cpp:155
#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:518
Tensor packing service.
Definition: ITensorPack.h:39
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
quantized, asymmetric fixed-point 8-bit number signed
std::string kernel_name
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)
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.