xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/cl_kernels/common/pixelwise_mul_float.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#include "helpers.h"
25
26#ifdef SATURATE
27#define CONVERT_OP_FLOAT_STR(x, type, round) (convert_##type##_sat##round(x))
28#else /* SATURATE */
29#define CONVERT_OP_FLOAT_STR(x, type, round) (convert_##type##round(x))
30#endif /* SATURATE */
31#define CONVERT_OP_FLOAT(x, type, round) CONVERT_OP_FLOAT_STR(x, type, round)
32
33#if defined(DATA_TYPE_IN1) && defined(DATA_TYPE_IN2) && defined(ACC_DATA_TYPE) && defined(DATA_TYPE_OUT)
34
35#if defined(ACTIVATION_TYPE)
36#include "activation_float_helpers.h"
37#endif // defined(ACTIVATION_TYPE)
38
39#define VEC_ACC_TYPE VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE_OUT)
40#define VEC_OUT_TYPE VEC_DATA_TYPE(DATA_TYPE_OUT, VEC_SIZE_OUT)
41#define VEC_FLOAT VEC_DATA_TYPE(float, VEC_SIZE_OUT)
42
43/** Performs a pixelwise multiplication with float scale of either integer or float inputs.
44 *
45 * @attention The inputs and output data types need to be passed at compile time using -DDATA_TYPE_IN1, -DDATA_TYPE_IN2 and -DDATA_TYPE_OUT:
46 * e.g. -DDATA_TYPE_IN1=uchar -DDATA_TYPE_IN2=ushort -DDATA_TYPE_OUT=short
47 * @attention The data type of the intermediate result of the multiplication should passed as well using -DACC_DATA_TYPE.
48 * e.g. If one of inputs is S16 -DACC_DATA_TYPE=int should be passed else -DACC_DATA_TYPE=short.
49 * @attention -DDATA_TYPE_FLOAT must be passed if floating point inputs are provided.
50 *
51 * @param[in]  in1_ptr                           Pointer to the source image. Supported data types: U8, S16, F16, F32
52 * @param[in]  in1_stride_x                      Stride of the source image in X dimension (in bytes)
53 * @param[in]  in1_step_x                        in1_stride_x * number of elements along X processed per workitem(in bytes)
54 * @param[in]  in1_stride_y                      Stride of the source image in Y dimension (in bytes)
55 * @param[in]  in1_step_y                        in1_stride_y * number of elements along Y processed per workitem(in bytes)
56 * @param[in]  in1_stride_z                      Stride of the source image in Y dimension (in bytes)
57 * @param[in]  in1_step_z                        in1_stride_z * number of elements along Y processed per workitem(in bytes)
58 * @param[in]  in1_offset_first_element_in_bytes The offset of the first element in the source image
59 * @param[in]  in2_ptr                           Pointer to the source image. Supported data types: U8, S16, F16, F32
60 * @param[in]  in2_stride_x                      Stride of the source image in X dimension (in bytes)
61 * @param[in]  in2_step_x                        in2_stride_x * number of elements along X processed per workitem(in bytes)
62 * @param[in]  in2_stride_y                      Stride of the source image in Y dimension (in bytes)
63 * @param[in]  in2_step_y                        in2_stride_y * number of elements along Y processed per workitem(in bytes)
64 * @param[in]  in2_stride_z                      Stride of the source image in Y dimension (in bytes)
65 * @param[in]  in2_step_z                        in2_stride_z * number of elements along Y processed per workitem(in bytes)
66 * @param[in]  in2_offset_first_element_in_bytes The offset of the first element in the source image
67 * @param[out] out_ptr                           Pointer to the destination image. Supported data types: U8, S16, F16, F32
68 * @param[in]  out_stride_x                      Stride of the destination image in X dimension (in bytes)
69 * @param[in]  out_step_x                        out_stride_x * number of elements along X processed per workitem(in bytes)
70 * @param[in]  out_stride_y                      Stride of the destination image in Y dimension (in bytes)
71 * @param[in]  out_step_y                        out_stride_y * number of elements along Y processed per workitem(in bytes)
72 * @param[in]  out_stride_z                      Stride of the destination image in Y dimension (in bytes)
73 * @param[in]  out_step_z                        out_stride_z * number of elements along Y processed per workitem(in bytes)
74 * @param[in]  out_offset_first_element_in_bytes The offset of the first element in the destination image
75 * @param[in]  scale                             Float scaling factor. Supported data types: F32
76 */
77__kernel void pixelwise_mul_float(
78    TENSOR3D_DECLARATION(in1),
79    TENSOR3D_DECLARATION(in2),
80#if !defined(IN_PLACE)
81    TENSOR3D_DECLARATION(out),
82#endif // !defined(IN_PLACE)
83    const float scale)
84{
85    // Get pixels pointer
86    size_t x = max((int)(get_global_id(0) * VEC_SIZE_OUT - (VEC_SIZE_OUT - VEC_SIZE_LEFTOVER) % VEC_SIZE_OUT), 0);
87    size_t y = get_global_id(1);
88    size_t z = get_global_id(2);
89
90    __global uchar *in1_addr = in1_ptr + in1_offset_first_element_in_bytes + x * in1_stride_x + y * in1_stride_y + z * in1_stride_z;
91    __global uchar *in2_addr = in2_ptr + in2_offset_first_element_in_bytes + x * in2_stride_x + y * in2_stride_y + z * in2_stride_z;
92    __global        uchar *
93#if !defined(IN_PLACE)
94    out_addr = out_ptr + out_offset_first_element_in_bytes + x * out_stride_x + y * out_stride_y + z * out_stride_z;
95#else // !defined(IN_PLACE)
96#if defined(SRC1_IN_PLACE)
97    out_addr      = in1_addr;
98#else  //defined(SRC1_IN_PLACE)
99    out_addr = in2_addr;
100#endif //defined(SRC1_IN_PLACE)
101#endif // !defined(IN_PLACE)
102
103    // Load data
104    VEC_ACC_TYPE in1_data = CONVERT((VEC_DATA_TYPE(DATA_TYPE_IN1, VEC_SIZE_OUT))(VLOAD(VEC_SIZE_IN1)(0, (__global DATA_TYPE_IN1 *)in1_addr)), VEC_ACC_TYPE);
105    VEC_ACC_TYPE in2_data = CONVERT((VEC_DATA_TYPE(DATA_TYPE_IN2, VEC_SIZE_OUT))(VLOAD(VEC_SIZE_IN2)(0, (__global DATA_TYPE_IN2 *)in2_addr)), VEC_ACC_TYPE);
106
107    // Perform multiplication
108#ifdef DATA_TYPE_FLOAT
109    VEC_OUT_TYPE res0 = CONVERT(in1_data * in2_data * (ACC_DATA_TYPE)scale, VEC_OUT_TYPE);
110#else  /* DATA_TYPE_FLOAT */
111    VEC_OUT_TYPE res0 = CONVERT_OP_FLOAT(CONVERT_OP_FLOAT((CONVERT(in1_data * in2_data, VEC_FLOAT) * scale), VEC_ACC_TYPE, ROUND), VEC_OUT_TYPE, ROUND);
112#endif /* DATA_TYPE_FLOAT */
113
114#if defined(ACTIVATION_TYPE)
115    res0 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE_OUT, VEC_SIZE_OUT, res0, A_VAL, B_VAL);
116#endif // defined(ACTIVATION_TYPE)
117
118    STORE_VECTOR_SELECT(res, DATA_TYPE_OUT, out_addr, VEC_SIZE_OUT, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0);
119}
120#endif /* defined(DATA_TYPE_IN1) && defined(DATA_TYPE_IN2) && defined(ACC_DATA_TYPE) && defined(DATA_TYPE_OUT) */
121
122#if defined(DATA_TYPE)
123
124/** Performs a pixelwise multiplication of complex float values
125 *
126 * @param[in]  in1_ptr                           Pointer to the source image. Supported data types: F16/F32
127 * @param[in]  in1_stride_x                      Stride of the source image in X dimension (in bytes)
128 * @param[in]  in1_step_x                        in1_stride_x * number of elements along X processed per workitem(in bytes)
129 * @param[in]  in1_stride_y                      Stride of the source image in Y dimension (in bytes)
130 * @param[in]  in1_step_y                        in1_stride_y * number of elements along Y processed per workitem(in bytes)
131 * @param[in]  in1_stride_z                      Stride of the source image in Y dimension (in bytes)
132 * @param[in]  in1_step_z                        in1_stride_z * number of elements along Y processed per workitem(in bytes)
133 * @param[in]  in1_offset_first_element_in_bytes The offset of the first element in the source image
134 * @param[in]  in2_ptr                           Pointer to the source image. Supported data types: same as @p in1_ptr
135 * @param[in]  in2_stride_x                      Stride of the source image in X dimension (in bytes)
136 * @param[in]  in2_step_x                        in2_stride_x * number of elements along X processed per workitem(in bytes)
137 * @param[in]  in2_stride_y                      Stride of the source image in Y dimension (in bytes)
138 * @param[in]  in2_step_y                        in2_stride_y * number of elements along Y processed per workitem(in bytes)
139 * @param[in]  in2_stride_z                      Stride of the source image in Y dimension (in bytes)
140 * @param[in]  in2_step_z                        in2_stride_z * number of elements along Y processed per workitem(in bytes)
141 * @param[in]  in2_offset_first_element_in_bytes The offset of the first element in the source image
142 * @param[out] out_ptr                           Pointer to the destination image. Supported data types: same as @p in1_ptr
143 * @param[in]  out_stride_x                      Stride of the destination image in X dimension (in bytes)
144 * @param[in]  out_step_x                        out_stride_x * number of elements along X processed per workitem(in bytes)
145 * @param[in]  out_stride_y                      Stride of the destination image in Y dimension (in bytes)
146 * @param[in]  out_step_y                        out_stride_y * number of elements along Y processed per workitem(in bytes)
147 * @param[in]  out_stride_z                      Stride of the destination image in Y dimension (in bytes)
148 * @param[in]  out_step_z                        out_stride_z * number of elements along Y processed per workitem(in bytes)
149 * @param[in]  out_offset_first_element_in_bytes The offset of the first element in the destination image
150 */
151__kernel void pixelwise_mul_complex(
152    TENSOR3D_DECLARATION(in1),
153    TENSOR3D_DECLARATION(in2),
154    TENSOR3D_DECLARATION(out))
155{
156    // Get pixels pointer
157    Tensor3D in1 = CONVERT_TO_TENSOR3D_STRUCT(in1);
158    Tensor3D in2 = CONVERT_TO_TENSOR3D_STRUCT(in2);
159    Tensor3D out = CONVERT_TO_TENSOR3D_STRUCT(out);
160
161    // Load data
162    VEC_DATA_TYPE(DATA_TYPE, 2)
163    vin1 = vload2(0, (__global DATA_TYPE *)in1.ptr);
164    VEC_DATA_TYPE(DATA_TYPE, 2)
165    vin2 = vload2(0, (__global DATA_TYPE *)in2.ptr);
166
167    // Perform complex multiplication
168    VEC_DATA_TYPE(DATA_TYPE, 2)
169    res = { vin1.x *vin2.x - vin1.y * vin2.y, vin1.x *vin2.y + vin2.x * vin1.y };
170
171#if defined(ACTIVATION_TYPE)
172    vstore2(ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE_OUT, res, A_VAL, B_VAL), 0, (__global DATA_TYPE *)out.ptr);
173#else  // defined(ACTIVATION_TYPE)
174    // Store result
175    vstore2(res, 0, (__global DATA_TYPE *)out.ptr);
176#endif // defined(ACTIVATION_TYPE)
177}
178
179#endif // defined(DATA_TYPE)