1 /*
2 * Copyright (c) 2017-2020 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/runtime/CL/functions/CLDirectConvolutionLayer.h"
25
26 #include "arm_compute/core/CL/ICLTensor.h"
27 #include "arm_compute/core/PixelValue.h"
28 #include "arm_compute/core/Utils.h"
29 #include "arm_compute/core/Validate.h"
30 #include "arm_compute/runtime/CL/CLScheduler.h"
31 #include "src/core/CL/kernels/CLDirectConvolutionLayerKernel.h"
32 #include "src/core/CL/kernels/CLFillBorderKernel.h"
33 #include "support/MemorySupport.h"
34
35 using namespace arm_compute;
36
CLDirectConvolutionLayer()37 CLDirectConvolutionLayer::CLDirectConvolutionLayer()
38 : _direct_conv_kernel(support::cpp14::make_unique<CLDirectConvolutionLayerKernel>()), _input_border_handler(support::cpp14::make_unique<CLFillBorderKernel>()), _activationlayer_function(),
39 _is_activationlayer_enabled(false)
40 {
41 }
42
43 CLDirectConvolutionLayer::~CLDirectConvolutionLayer() = default;
44
configure(ICLTensor * input,const ICLTensor * weights,const ICLTensor * biases,ICLTensor * output,const PadStrideInfo & conv_info,const ActivationLayerInfo & act_info)45 void CLDirectConvolutionLayer::configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info)
46 {
47 configure(CLKernelLibrary::get().get_compile_context(), input, weights, biases, output, conv_info, act_info);
48 }
49
configure(const CLCompileContext & compile_context,ICLTensor * input,const ICLTensor * weights,const ICLTensor * biases,ICLTensor * output,const PadStrideInfo & conv_info,const ActivationLayerInfo & act_info)50 void CLDirectConvolutionLayer::configure(const CLCompileContext &compile_context, ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output,
51 const PadStrideInfo &conv_info,
52 const ActivationLayerInfo &act_info)
53 {
54 // Set GPU target
55 _direct_conv_kernel->set_target(CLScheduler::get().target());
56
57 // Configure direct convolution
58 _direct_conv_kernel->configure(compile_context, input, weights, biases, output, conv_info);
59
60 // Configure border handler
61 PixelValue &&zero_value(0.f);
62 if(is_data_type_quantized_asymmetric(input->info()->data_type()))
63 {
64 zero_value = PixelValue(0, input->info()->data_type(), input->info()->quantization_info());
65 }
66 _input_border_handler->configure(compile_context, input, _direct_conv_kernel->border_size(), BorderMode::CONSTANT, zero_value);
67
68 // Tune kernels
69 CLScheduler::get().tune_kernel_static(*_direct_conv_kernel);
70
71 _is_activationlayer_enabled = act_info.enabled();
72
73 //Configure Activation Layer
74 if(_is_activationlayer_enabled)
75 {
76 _activationlayer_function.configure(compile_context, output, nullptr, act_info);
77 }
78 }
79
validate(const ITensorInfo * input,const ITensorInfo * weights,const ITensorInfo * biases,const ITensorInfo * output,const PadStrideInfo & conv_info,const ActivationLayerInfo & act_info)80 Status CLDirectConvolutionLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
81 const ActivationLayerInfo &act_info)
82 {
83 ARM_COMPUTE_RETURN_ON_ERROR(CLDirectConvolutionLayerKernel::validate(input, weights, biases, output, conv_info, CLScheduler::get().target()));
84 if(act_info.enabled())
85 {
86 ARM_COMPUTE_RETURN_ON_ERROR(CLActivationLayer::validate(output, nullptr, act_info));
87 }
88 return Status{};
89 }
90
run()91 void CLDirectConvolutionLayer::run()
92 {
93 // Run border handler
94 CLScheduler::get().enqueue(*_input_border_handler, false);
95
96 // Run direct convolution
97 CLScheduler::get().enqueue(*_direct_conv_kernel);
98
99 //Run Activation Layer
100 if(_is_activationlayer_enabled)
101 {
102 _activationlayer_function.run();
103 }
104 }
105