Compute Library
 22.05
unpooling_layer.cl
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2020-2021 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  */
24 
25 #include "helpers.h"
26 
27 /** Performs max unpooling function with pool size equal to 2.
28  *
29  * @note Datatype must be passed using -DDATA_TYPE e.g. -DDATA_TYPE=float. Supported data types are F16/F32
30  * @note The width of the output tensor must be passed using -DWIDTH_DST e.g. -DWIDTH_DST=24
31  * @note The height of the output tensor must be passed using -DHEIGHT_DST e.g. -DHEIGHT_DST=54
32  * @note The depth of the output tensor must be passed using -DDEPTH_DST e.g. -DDEPTH_DST=32
33  *
34  * @param[in] input_ptr Pointer to the source tensor. Supported data types: F16/F32
35  * @param[in] input_stride_x Stride of the source tensor in X dimension (in bytes)
36  * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes)
37  * @param[in] input_stride_y Stride of the source tensor in Y dimension (in bytes)
38  * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes)
39  * @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes)
40  * @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes)
41  * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source tensor
42  * @param[out] output_ptr Pointer to the output tensor. Supported data types: same as @p input_ptr
43  * @param[in] output_stride_x Stride of the output tensor in X dimension (in bytes)
44  * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes)
45  * @param[in] output_stride_y Stride of the output tensor in Y dimension (in bytes)
46  * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes)
47  * @param[in] output_stride_z Stride of the output tensor in Z dimension (in bytes)
48  * @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes)
49  * @param[in] output_offset_first_element_in_bytes The offset of the first element in the output tensor
50  * @param[in] indices_ptr Pointer to the indices tensor. Supported data types: U32
51  * @param[in] indices_stride_x Stride of the indices tensor in X dimension (in bytes)
52  * @param[in] indices_step_x indices_stride_x * number of elements along X processed per workitem(in bytes)
53  * @param[in] indices_stride_y Stride of the indices tensor in Y dimension (in bytes)
54  * @param[in] indices_step_y indices_stride_y * number of elements along Y processed per workitem(in bytes)
55  * @param[in] indices_stride_z Stride of the indices tensor in Z dimension (in bytes)
56  * @param[in] indices_step_z indices_stride_z * number of elements along Z processed per workitem(in bytes)
57  * @param[in] indices_offset_first_element_in_bytes The offset of the first element in the indices tensor
58  */
59 __kernel void max_unpooling_layer_2(
61  TENSOR3D_DECLARATION(output),
62  TENSOR3D_DECLARATION(indices))
63 {
66  Tensor3D indices = CONVERT_TO_TENSOR3D_STRUCT(indices);
67 
68  unsigned int index = *((__global unsigned int *)indices.ptr);
69  DATA_TYPE value = *((__global DATA_TYPE *)input.ptr);
70 
71  *((__global DATA_TYPE *)tensor3D_index2ptr(&output, WIDTH_DST, HEIGHT_DST, DEPTH_DST, index)) = value;
72 }
__global const uchar * tensor3D_index2ptr(const Tensor3D *tensor, uint width, uint height, uint depth, uint index)
Get the offset for a given linear index of a Tensor3D.
Definition: helpers.h:1121
#define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name)
Definition: helpers.h:883
Structure to hold 3D tensor information.
Definition: helpers.h:905
#define CONVERT_TO_TENSOR3D_STRUCT(name)
Definition: helpers.h:869
__kernel void max_unpooling_layer_2(__global uchar *input_ptr, uint input_stride_x, uint input_step_x, uint input_stride_y, uint input_step_y, uint input_stride_z, uint input_step_z, uint input_offset_first_element_in_bytes, __global uchar *output_ptr, uint output_stride_x, uint output_step_x, uint output_stride_y, uint output_step_y, uint output_stride_z, uint output_step_z, uint output_offset_first_element_in_bytes, __global uchar *indices_ptr, uint indices_stride_x, uint indices_step_x, uint indices_stride_y, uint indices_step_y, uint indices_stride_z, uint indices_step_z, uint indices_offset_first_element_in_bytes)
Performs max unpooling function with pool size equal to 2.
__global uchar * ptr
Pointer to the starting postion of the buffer.
Definition: helpers.h:907
#define TENSOR3D_DECLARATION(name)
Definition: helpers.h:812