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
86
87
88
89
90
91
92
93
94
95
96
97
98
|
//
// Copyright © 2017 Arm Ltd and Contributors. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "ClConvertFp16ToFp32Workload.hpp"
#include <cl/ClTensorHandle.hpp>
#include "ClWorkloadUtils.hpp"
namespace armnn
{
using namespace armcomputetensorutils;
static constexpr arm_compute::ConvertPolicy g_AclConvertPolicy = arm_compute::ConvertPolicy::SATURATE;
ClConvertFp16ToFp32Workload::ClConvertFp16ToFp32Workload(
const ConvertFp16ToFp32QueueDescriptor& descriptor,
const WorkloadInfo& info,
const arm_compute::CLCompileContext& clCompileContext) :
Float16ToFloat32Workload<ConvertFp16ToFp32QueueDescriptor>(descriptor, info)
{
this->m_Data.ValidateInputsOutputs("ClConvertFp16ToFp32Workload", 1, 1);
arm_compute::ICLTensor& input = static_cast<IClTensorHandle*>(this->m_Data.m_Inputs[0])->GetTensor();
arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(this->m_Data.m_Outputs[0])->GetTensor();
{
ARMNN_SCOPED_PROFILING_EVENT(Compute::Undefined, "ClConvertFp16ToFp32Workload_configure");
m_Layer.configure(clCompileContext, &input, &output, g_AclConvertPolicy, 0);
}
}
void ClConvertFp16ToFp32Workload::Execute() const
{
ARMNN_SCOPED_PROFILING_EVENT_CL_GUID("ClConvertFp16ToFp32Workload_Execute", this->GetGuid());
RunClFunction(m_Layer, CHECK_LOCATION());
}
arm_compute::Status ClConvertFp16ToFp32WorkloadValidate(const TensorInfo& input, const TensorInfo& output)
{
if (input.GetDataType() != DataType::Float16)
{
return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, "Input should be Float16");
}
if (output.GetDataType() != DataType::Float32)
{
return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, "Output should be Float32");
}
const arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(input);
const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output);
const arm_compute::Status aclStatus = arm_compute::CLDepthConvertLayer::validate(
&aclInputInfo, &aclOutputInfo, g_AclConvertPolicy, 0);
return aclStatus;
}
void ClConvertFp16ToFp32Workload::ReplaceInputTensorHandle(ITensorHandle* tensorHandle, unsigned int slot)
{
ITensorHandle* backupHandle = this->m_Data.m_Inputs[slot];
this->m_Data.m_Inputs[slot] = tensorHandle;
try
{
Reconfigure();
}
catch(armnn::UnimplementedException& e)
{
// Cannot reconfigure, revert the slot back and throw the exception.
this->m_Data.m_Inputs[slot] = backupHandle;
throw e;
}
}
// Replace output tensor handle with the given TensorHandle
void ClConvertFp16ToFp32Workload::ReplaceOutputTensorHandle(ITensorHandle* tensorHandle, unsigned int slot)
{
ITensorHandle* backupHandle = this->m_Data.m_Inputs[slot];
this->m_Data.m_Inputs[slot] = tensorHandle;
try
{
Reconfigure();
}
catch(armnn::UnimplementedException& e)
{
// Cannot reconfigure, revert the slot back and throw the exception.
this->m_Data.m_Inputs[slot] = backupHandle;
throw e;
}
}
void ClConvertFp16ToFp32Workload::Reconfigure()
{
throw armnn::UnimplementedException("Reconfigure not implemented for this workload");
}
} //namespace armnn
|
