1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
|
//
// Copyright © 2017 Arm Ltd and Contributors. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "ClMultiplicationWorkload.hpp"
#include <aclCommon/ArmComputeUtils.hpp>
#include <armnn/backends/TensorHandle.hpp>
#include <cl/ClTensorHandle.hpp>
#include "ClWorkloadUtils.hpp"
namespace armnn
{
arm_compute::Status ClMultiplicationWorkloadValidate(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
const ActivationDescriptor* activationDescriptor)
{
const arm_compute::TensorInfo aclInput1 = armcomputetensorutils::BuildArmComputeTensorInfo(input0);
const arm_compute::TensorInfo aclInput2 = armcomputetensorutils::BuildArmComputeTensorInfo(input1);
const arm_compute::TensorInfo aclOutput = armcomputetensorutils::BuildArmComputeTensorInfo(output);
auto convertPolicy = (IsQuantizedType(input0.GetDataType()) || IsQuantizedType(input1.GetDataType())) ?
arm_compute::ConvertPolicy::SATURATE :
arm_compute::ConvertPolicy::WRAP;
const arm_compute::ActivationLayerInfo activationInfo = ConvertActivationDescriptorToAclActivationLayerInfo(
activationDescriptor);
// At the time of writing, configure() will fail if a rounding policy other than TO_ZERO is supplied to it,
// when providing a scale of 1.0 for F32 tensors, even though the provided rounding policy appears to be
// ignored for F32 tensors.
return arm_compute::CLPixelWiseMultiplication::validate(&aclInput1,
&aclInput2,
&aclOutput,
1.0f,
convertPolicy,
arm_compute::RoundingPolicy::TO_ZERO,
activationInfo);
}
ClMultiplicationWorkload::ClMultiplicationWorkload(const MultiplicationQueueDescriptor& descriptor,
const WorkloadInfo& info,
const arm_compute::CLCompileContext& clCompileContext)
: ClBaseWorkload<MultiplicationQueueDescriptor>(descriptor, info)
{
m_Data.ValidateInputsOutputs("ClMultiplicationWorkload", 2, 1);
arm_compute::ICLTensor& input0 = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
arm_compute::ICLTensor& input1 = static_cast<IClTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
auto convertPolicy = (IsQuantizedType(info.m_InputTensorInfos[0].GetDataType()) ||
IsQuantizedType(info.m_InputTensorInfos[1].GetDataType())) ?
arm_compute::ConvertPolicy::SATURATE :
arm_compute::ConvertPolicy::WRAP;
const arm_compute::ActivationLayerInfo activationInfo = ConvertAdditionalInfoToAclActivationLayerInfo(descriptor);
{
ARMNN_SCOPED_PROFILING_EVENT(Compute::Undefined, "ClMultiplicationWorkload_configure");
// Construct
m_PixelWiseMultiplication.configure(clCompileContext,
&input0,
&input1,
&output,
1.0f,
convertPolicy,
arm_compute::RoundingPolicy::TO_NEAREST_EVEN,
activationInfo);
}
}
void ClMultiplicationWorkload::Execute() const
{
ARMNN_SCOPED_PROFILING_EVENT_CL_GUID("ClMultiplicationWorkload_Execute", this->GetGuid());
RunClFunction(m_PixelWiseMultiplication, CHECK_LOCATION());
}
} //namespace armnn
|
