Compute Library
 21.02
CpuConcatenateBatchKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2019-2021 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 
26 #include "arm_compute/core/Error.h"
31 #include "arm_compute/core/Utils.h"
34 #include "src/core/NEON/NEAsymm.h"
38 
39 namespace arm_compute
40 {
41 namespace cpu
42 {
43 namespace kernels
44 {
45 namespace
46 {
47 template <typename T>
48 void batch_concat(const ITensor *src, ITensor *dst, unsigned int batch_offset, const Window &window)
49 {
50  // Offset src
51  uint8_t *src_ptr = src->buffer() + src->info()->offset_first_element_in_bytes();
52 
53  // Offset dst
54  uint8_t *dst_ptr = dst->buffer() + dst->info()->offset_first_element_in_bytes() + batch_offset * dst->info()->strides_in_bytes()[3];
55 
56  const auto window_start_x = static_cast<int>(window.x().start());
57  const auto window_end_x = static_cast<int>(window.x().end());
58  const int window_step_x = 16 / dst->info()->element_size();
59 
60  Window win{ window };
61  win.set(Window::DimX, Window::Dimension(0, 1, 1));
62  win.set(3, Window::Dimension(0, src->info()->tensor_shape()[3], 1));
63 
64  Iterator src_it(src, win);
65  Iterator dst_it(dst, win);
66 
67  const DataType dt = src->info()->data_type();
68  const UniformQuantizationInfo src_qinfo = src->info()->quantization_info().uniform();
69  const UniformQuantizationInfo dst_qinfo = dst->info()->quantization_info().uniform();
70  if(dt == DataType::QASYMM8 && src_qinfo != dst_qinfo)
71  {
72  execute_window_loop(win, [&](const Coordinates &)
73  {
74  const auto in_ptr = reinterpret_cast<const uint8_t *>(src_ptr + src_it.offset());
75  const auto out_ptr = reinterpret_cast<uint8_t *>(dst_ptr + dst_it.offset());
76 
77  int x = window_start_x;
78  for(; x <= (window_end_x - window_step_x); x += window_step_x)
79  {
80  wrapper::vstore(out_ptr, vquantize(vdequantize(wrapper::vloadq(in_ptr), src_qinfo), dst_qinfo));
81  }
82 
83  // Compute left-over elements
84  for(; x < window_end_x; ++x)
85  {
86  *(out_ptr + x) = quantize_qasymm8(dequantize_qasymm8(*(in_ptr + x), src_qinfo), dst_qinfo);
87  }
88  },
89  src_it, dst_it);
90  }
91  else if(dt == DataType::QASYMM8_SIGNED && src_qinfo != dst_qinfo)
92  {
93  execute_window_loop(win, [&](const Coordinates &)
94  {
95  const auto in_ptr = reinterpret_cast<const int8_t *>(src_ptr + src_it.offset());
96  const auto out_ptr = reinterpret_cast<int8_t *>(dst_ptr + dst_it.offset());
97  int x = window_start_x;
98  for(; x <= (window_end_x - window_step_x); x += window_step_x)
99  {
100  wrapper::vstore(out_ptr, vquantize_signed(vdequantize(wrapper::vloadq(in_ptr), src_qinfo), dst_qinfo));
101  }
102  // Compute left-over elements
103  for(; x < window_end_x; ++x)
104  {
105  *(out_ptr + x) = quantize_qasymm8_signed(dequantize_qasymm8_signed(*(in_ptr + x), src_qinfo), dst_qinfo);
106  }
107  },
108  src_it, dst_it);
109  }
110  else
111  {
112  execute_window_loop(win, [&](const Coordinates &)
113  {
114  const auto in_ptr = reinterpret_cast<const T *>(src_ptr + src_it.offset());
115  const auto out_ptr = reinterpret_cast<T *>(dst_ptr + dst_it.offset());
116 
117  int x = window_start_x;
118  for(; x <= (window_end_x - window_step_x); x += window_step_x)
119  {
120  wrapper::vstore(out_ptr + x, wrapper::vloadq(in_ptr + x));
121  }
122 
123  // Compute left-over elements
124  for(; x < window_end_x; ++x)
125  {
126  *(out_ptr + x) = *(in_ptr + x);
127  }
128  },
129  src_it, dst_it);
130  }
131 }
132 
133 Status validate_arguments(const ITensorInfo *src, unsigned int batch_offset, const ITensorInfo *dst)
134 {
136  //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(src) is not needed here as this kernel doesn't use Neon FP16 instructions.
137  ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
139 
140  ARM_COMPUTE_RETURN_ERROR_ON(src->dimension(Window::DimX) != dst->dimension(Window::DimX));
141  ARM_COMPUTE_RETURN_ERROR_ON(src->dimension(Window::DimY) != dst->dimension(Window::DimY));
142  ARM_COMPUTE_RETURN_ERROR_ON(src->dimension(Window::DimZ) != dst->dimension(Window::DimZ));
143  ARM_COMPUTE_RETURN_ERROR_ON(src->dimension(3) + batch_offset > dst->dimension(3));
145 
146  return Status{};
147 }
148 } // namespace
149 
151  : _func(nullptr), _batch_offset(0)
152 {
153 }
154 
155 void CpuConcatenateBatchKernel::configure(const ITensorInfo *src, unsigned int batch_offset, ITensorInfo *dst)
156 {
158  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, batch_offset, dst));
159 
160  _func = nullptr;
161  _batch_offset = batch_offset;
162 
163  switch(src->data_type())
164  {
165  case DataType::S8:
166  case DataType::U8:
167  case DataType::QASYMM8:
169  _func = &batch_concat<uint8_t>;
170  break;
171  case DataType::S16:
172  case DataType::U16:
173  case DataType::F16:
174  _func = &batch_concat<uint16_t>;
175  break;
176  case DataType::S32:
177  case DataType::U32:
178  case DataType::F32:
179  _func = &batch_concat<uint32_t>;
180  break;
181  default:
182  ARM_COMPUTE_ERROR("Unsupported data type.");
183  }
184 
185  // Configure kernel window
186  Window win = calculate_max_window(*dst, Steps());
187  Coordinates coord;
188  coord.set_num_dimensions(dst->num_dimensions());
189  dst->set_valid_region(ValidRegion(coord, dst->tensor_shape()));
190  ICpuKernel::configure(win);
191 }
192 
194  unsigned int batch_offset,
195  const arm_compute::ITensorInfo *dst)
196 {
197  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, batch_offset, dst));
198  return Status{};
199 }
200 
202 {
203  ARM_COMPUTE_UNUSED(info);
206  ARM_COMPUTE_ERROR_ON(_func == nullptr);
207 
208  (*_func)(tensors.get_const_tensor(TensorType::ACL_SRC),
210  _batch_offset,
211  window);
212 }
213 
215 {
216  return "CpuConcatenateBatchKernel";
217 }
218 } // namespace kernels
219 } // namespace cpu
220 } // namespace arm_compute
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
float dequantize_qasymm8(uint8_t value, const INFO_TYPE &qinfo)
Dequantize a value given an unsigned 8-bit asymmetric quantization scheme.
uint8_t quantize_qasymm8(float value, const INFO_TYPE &qinfo, RoundingPolicy rounding_policy=RoundingPolicy::TO_NEAREST_UP)
Quantize a value given an unsigned 8-bit asymmetric quantization scheme.
#define ARM_COMPUTE_ERROR(msg)
Print the given message then throw an std::runtime_error.
Definition: Error.h:352
float32x4x2_t vdequantize(const uint8x8_t &qv, const UniformQuantizationInfo &qi)
Dequantize a neon vector holding 8 quantized values.
Definition: NEAsymm.h:415
1 channel, 1 U8 per channel
uint8x16_t vloadq(const uint8_t *ptr)
Definition: load.h:58
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
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
1 channel, 1 U16 per channel
Status class.
Definition: Error.h:52
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
SimpleTensor< float > src
Definition: DFT.cpp:155
Copyright (c) 2017-2021 Arm Limited.
virtual void set_valid_region(const ValidRegion &valid_region)=0
Set the valid region of the tensor.
1 channel, 1 F16 per channel
DataType dt
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
1 channel, 1 S32 per channel
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:40
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
1 channel, 1 U32 per channel
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
const char * name() const override
Name of the kernel.
int8_t quantize_qasymm8_signed(float value, const INFO_TYPE &qinfo, RoundingPolicy rounding_policy=RoundingPolicy::TO_NEAREST_UP)
Quantize a value given a signed 8-bit asymmetric quantization scheme.
quantized, asymmetric fixed-point 8-bit number unsigned
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
Coordinates of an item.
Definition: Coordinates.h:37
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
1 channel, 1 S16 per channel
void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override
Execute the kernel on the passed window.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:50
Information about executing thread and CPU.
Definition: CPPTypes.h:235
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:443
static Status validate(const ITensorInfo *src, unsigned int batch_offset, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration of CpuConcatenateBatchKerne...
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:545
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
uint8x8_t vquantize(const float32x4x2_t &qv, const UniformQuantizationInfo &qi)
Quantize a neon vector holding 8 floating point values.
Definition: NEAsymm.h:602
void vstore(uint8_t *ptr, uint8x8_t val)
Definition: store.h:39
float dequantize_qasymm8_signed(int8_t value, const INFO_TYPE &qinfo)
Dequantize a value given a signed 8-bit asymmetric quantization scheme.
Tensor packing service.
Definition: ITensorPack.h:37
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
void execute_window_loop(const Window &w, L &&lambda_function, Ts &&... iterators)
Iterate through the passed window, automatically adjusting the iterators and calling the lambda_funct...
Definition: Helpers.inl:77
void set_num_dimensions(size_t num_dimensions)
Set number of dimensions.
Definition: Dimensions.h:149
quantized, asymmetric fixed-point 8-bit number signed
Includes all wrapper headers at once.
int8x8_t vquantize_signed(const float32x4x2_t &qv, const UniformQuantizationInfo &qi)
Quantize a neon vector holding 8 floating point values.
Definition: NEAsymm.h:630
Container for valid region of a window.
Definition: Types.h:188
DataType
Available data types.
Definition: Types.h:77
signed 8-bit number
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
void configure(const ITensorInfo *src, unsigned int batch_offset, ITensorInfo *dst)
Configure kernel for a given list of arguments.