1 /*
2 * Copyright (c) 2022 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 "arm_compute/core/CL/CLHelpers.h"
25 #include "arm_compute/core/TensorInfo.h"
26 #include "arm_compute/core/TensorShape.h"
27
28 namespace arm_compute
29 {
30 namespace cl_dwc
31 {
use_cl_image_for_weights(const ITensorInfo * weights,unsigned int depth_multiplier)32 bool use_cl_image_for_weights(const ITensorInfo *weights, unsigned int depth_multiplier)
33 {
34 // Check whether we can use the cl image with the weights.
35 if(!export_to_cl_image(weights))
36 {
37 return false;
38 }
39
40 const size_t idx_w = get_data_layout_dimension_index(weights->data_layout(), DataLayoutDimension::WIDTH);
41 const size_t idx_h = get_data_layout_dimension_index(weights->data_layout(), DataLayoutDimension::HEIGHT);
42 const size_t kernel_w = weights->tensor_shape()[idx_w];
43 const size_t kernel_h = weights->tensor_shape()[idx_h];
44
45 // If we can use the cl image storage with the weights, we prefer to use the cl buffer storage in the following cases for performance reasons:
46 // 1- When the kernel size is 1x1
47 // 2- When the depth multiplier is greater than 1 and not multiple of 4.
48 if((kernel_w == 1) && (kernel_h == 1))
49 {
50 return false;
51 }
52
53 if((depth_multiplier > 1) && (depth_multiplier % 4) != 0)
54 {
55 return false;
56 }
57
58 return true;
59 }
60 } // namespace cl_dwc
61 } // namespace arm_compute