xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/cl_kernels/common/activation_layer.cl (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1/*
2 * Copyright (c) 2016-2021 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#if defined(ACT) && defined(DATA_TYPE) && defined(VEC_SIZE)
25
26#define TYPE VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
27
28#include "activation_float_helpers.h"
29
30/** This performs an activation function floating point inputs.
31 *
32 * @note In order to perform the activation function "in-place", the pre-processor -DIN_PLACE must be passed at compile time
33 *
34 * @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=short
35 * @note Vector size should be given as a preprocessor argument using -DVEC_SIZE=size. e.g. -DVEC_SIZE=16
36 * @note Leftover vector size has to be passed at compile time using -DVEC_SIZE_LEFTOVER. e.g. -DVEC_SIZE_LEFTOVER=3. It is defined as the remainder between the input's first dimension and VEC_SIZE
37 * @note Activation function should be given as a preprocessor argument using -DACT=name. e.g. -DACT=TANH
38 * @note A, B variables required by some activation functions are set using -DA_VAL= and -DB_VAL= respectively.
39 *
40 * @param[in]  input_ptr                            Pointer to the source image. Supported data types: F16/F32
41 * @param[in]  input_stride_x                       Stride of the source image in X dimension (in bytes)
42 * @param[in]  input_step_x                         input_stride_x * number of elements along X processed per workitem(in bytes)
43 * @param[in]  input_stride_y                       Stride of the source image in Y dimension (in bytes)
44 * @param[in]  input_step_y                         input_stride_y * number of elements along Y processed per workitem(in bytes)
45 * @param[in]  input_stride_z                       Stride of the source tensor in Z dimension (in bytes)
46 * @param[in]  input_step_z                         input_stride_z * number of elements along Z processed per workitem(in bytes)
47 * @param[in]  input_offset_first_element_in_bytes  The offset of the first element in the source image
48 * @param[out] output_ptr                           Pointer to the destination image. Supported data types: same as @p input_ptr
49 * @param[in]  output_stride_x                      Stride of the destination image in X dimension (in bytes)
50 * @param[in]  output_step_x                        output_stride_x * number of elements along X processed per workitem(in bytes)
51 * @param[in]  output_stride_y                      Stride of the destination image in Y dimension (in bytes)
52 * @param[in]  output_step_y                        output_stride_y * number of elements along Y processed per workitem(in bytes)
53 * @param[in]  output_stride_z                      Stride of the source tensor in Z dimension (in bytes)
54 * @param[in]  output_step_z                        output_stride_z * number of elements along Z processed per workitem(in bytes)
55 * @param[in]  output_offset_first_element_in_bytes The offset of the first element in the destination image
56 */
57__kernel void activation_layer(
58    TENSOR3D_DECLARATION(input)
59#ifndef IN_PLACE
60    ,
61    TENSOR3D_DECLARATION(output)
62#endif /* not IN_PLACE */
63)
64{
65    uint x_offs = max((int)(get_global_id(0) * VEC_SIZE * sizeof(DATA_TYPE) - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE * sizeof(DATA_TYPE)), 0);
66
67    // Get pixels pointer
68    __global uchar *input_addr = input_ptr + input_offset_first_element_in_bytes + x_offs + get_global_id(1) * input_stride_y + get_global_id(2) * input_stride_z;
69#ifdef IN_PLACE
70    __global uchar *output_addr = input_addr;
71#else  /* IN_PLACE */
72    __global uchar *output_addr = output_ptr + output_offset_first_element_in_bytes + x_offs + get_global_id(1) * output_stride_y + get_global_id(2) * output_stride_z;
73#endif /* IN_PLACE */
74
75    // Load data
76    TYPE data0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)input_addr);
77
78    // Perform activation
79    data0 = ACTIVATION(ACT, DATA_TYPE, VEC_SIZE, data0, A_VAL, B_VAL);
80
81    // Store result
82    STORE_VECTOR_SELECT(data, DATA_TYPE, output_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
83}
84
85#endif /* defined(ACT) */
86