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
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
|
// Copyright 2019 Oath Inc. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include "tensor_modify_update.h"
#include <vespa/document/base/exceptions.h>
#include <vespa/document/base/field.h>
#include <vespa/document/datatype/tensor_data_type.h>
#include <vespa/document/fieldvalue/document.h>
#include <vespa/document/fieldvalue/tensorfieldvalue.h>
#include <vespa/document/serialization/vespadocumentdeserializer.h>
#include <vespa/document/util/serializableexceptions.h>
#include <vespa/eval/eval/operation.h>
#include <vespa/eval/eval/value.h>
#include <vespa/eval/eval/fast_value.h>
#include <vespa/eval/tensor/partial_update.h>
#include <vespa/vespalib/objects/nbostream.h>
#include <vespa/vespalib/stllike/asciistream.h>
#include <vespa/vespalib/util/stringfmt.h>
#include <vespa/vespalib/util/xmlstream.h>
#include <ostream>
using vespalib::IllegalArgumentException;
using vespalib::IllegalStateException;
using vespalib::make_string;
using vespalib::eval::ValueType;
using vespalib::eval::FastValueBuilderFactory;
using vespalib::tensor::TensorPartialUpdate;
using join_fun_t = double (*)(double, double);
namespace document {
namespace {
double
replace(double, double b)
{
return b;
}
join_fun_t
getJoinFunction(TensorModifyUpdate::Operation operation)
{
using Operation = TensorModifyUpdate::Operation;
switch (operation) {
case Operation::REPLACE:
return replace;
case Operation::ADD:
return vespalib::eval::operation::Add::f;
case Operation::MULTIPLY:
return vespalib::eval::operation::Mul::f;
default:
throw IllegalArgumentException("Bad operation", VESPA_STRLOC);
}
}
vespalib::string
getJoinFunctionName(TensorModifyUpdate::Operation operation)
{
using Operation = TensorModifyUpdate::Operation;
switch (operation) {
case Operation::REPLACE:
return "replace";
case Operation::ADD:
return "add";
case Operation::MULTIPLY:
return "multiply";
default:
throw IllegalArgumentException("Bad operation", VESPA_STRLOC);
}
}
std::unique_ptr<const TensorDataType>
convertToCompatibleType(const TensorDataType &tensorType)
{
std::vector<ValueType::Dimension> list;
for (const auto &dim : tensorType.getTensorType().dimensions()) {
list.emplace_back(dim.name);
}
return std::make_unique<const TensorDataType>(ValueType::tensor_type(std::move(list), tensorType.getTensorType().cell_type()));
}
}
IMPLEMENT_IDENTIFIABLE(TensorModifyUpdate, ValueUpdate);
TensorModifyUpdate::TensorModifyUpdate()
: _operation(Operation::MAX_NUM_OPERATIONS),
_tensorType(),
_tensor()
{
}
TensorModifyUpdate::TensorModifyUpdate(const TensorModifyUpdate &rhs)
: _operation(rhs._operation),
_tensorType(rhs._tensorType->clone()),
_tensor(Identifiable::cast<TensorFieldValue *>(_tensorType->createFieldValue().release()))
{
*_tensor = *rhs._tensor;
}
TensorModifyUpdate::TensorModifyUpdate(Operation operation, std::unique_ptr<TensorFieldValue> tensor)
: _operation(operation),
_tensorType(Identifiable::cast<const TensorDataType &>(*tensor->getDataType()).clone()),
_tensor(Identifiable::cast<TensorFieldValue *>(_tensorType->createFieldValue().release()))
{
*_tensor = *tensor;
}
TensorModifyUpdate::~TensorModifyUpdate() = default;
TensorModifyUpdate &
TensorModifyUpdate::operator=(const TensorModifyUpdate &rhs)
{
if (&rhs != this) {
_operation = rhs._operation;
_tensor.reset();
_tensorType.reset(rhs._tensorType->clone());
_tensor.reset(Identifiable::cast<TensorFieldValue *>(_tensorType->createFieldValue().release()));
*_tensor = *rhs._tensor;
}
return *this;
}
TensorModifyUpdate &
TensorModifyUpdate::operator=(TensorModifyUpdate &&rhs)
{
_operation = rhs._operation;
_tensorType = std::move(rhs._tensorType);
_tensor = std::move(rhs._tensor);
return *this;
}
bool
TensorModifyUpdate::operator==(const ValueUpdate &other) const
{
if (other.getClass().id() != TensorModifyUpdate::classId) {
return false;
}
const TensorModifyUpdate& o(static_cast<const TensorModifyUpdate&>(other));
if (_operation != o._operation) {
return false;
}
if (*_tensor != *o._tensor) {
return false;
}
return true;
}
void
TensorModifyUpdate::checkCompatibility(const Field& field) const
{
if (field.getDataType().getClass().id() != TensorDataType::classId) {
throw IllegalArgumentException(make_string("Cannot perform tensor modify update on non-tensor field '%s'",
field.getName().data()), VESPA_STRLOC);
}
}
std::unique_ptr<vespalib::eval::Value>
TensorModifyUpdate::applyTo(const vespalib::eval::Value &tensor) const
{
auto cellsTensor = _tensor->getAsTensorPtr();
if (cellsTensor) {
const auto &factory = FastValueBuilderFactory::get();
return TensorPartialUpdate::modify(tensor, getJoinFunction(_operation), *cellsTensor, factory);
}
return {};
}
bool
TensorModifyUpdate::applyTo(FieldValue& value) const
{
if (value.inherits(TensorFieldValue::classId)) {
TensorFieldValue &tensorFieldValue = static_cast<TensorFieldValue &>(value);
auto oldTensor = tensorFieldValue.getAsTensorPtr();
if (oldTensor) {
auto newTensor = applyTo(*oldTensor);
if (newTensor) {
tensorFieldValue = std::move(newTensor);
}
}
} else {
vespalib::string err = make_string("Unable to perform a tensor modify update on a '%s' field value",
value.getClass().name());
throw IllegalStateException(err, VESPA_STRLOC);
}
return true;
}
void
TensorModifyUpdate::printXml(XmlOutputStream& xos) const
{
xos << "{TensorModifyUpdate::printXml not yet implemented}";
}
void
TensorModifyUpdate::print(std::ostream& out, bool verbose, const std::string& indent) const
{
out << indent << "TensorModifyUpdate(" << getJoinFunctionName(_operation) << ",";
if (_tensor) {
_tensor->print(out, verbose, indent);
}
out << ")";
}
namespace {
void
verifyCellsTensorIsSparse(const vespalib::eval::Value *cellsTensor)
{
if (cellsTensor == nullptr) {
return;
}
if (cellsTensor->type().is_sparse()) {
return;
}
vespalib::string err = make_string("Expected cells tensor to be sparse, but has type '%s'",
cellsTensor->type().to_spec().c_str());
throw IllegalStateException(err, VESPA_STRLOC);
}
}
void
TensorModifyUpdate::deserialize(const DocumentTypeRepo &repo, const DataType &type, nbostream & stream)
{
uint8_t op;
stream >> op;
if (op >= static_cast<uint8_t>(Operation::MAX_NUM_OPERATIONS)) {
vespalib::asciistream msg;
msg << "Unrecognized tensor modify update operation " << static_cast<uint32_t>(op);
throw DeserializeException(msg.str(), VESPA_STRLOC);
}
_operation = static_cast<Operation>(op);
_tensorType = convertToCompatibleType(Identifiable::cast<const TensorDataType &>(type));
auto tensor = _tensorType->createFieldValue();
if (tensor->inherits(TensorFieldValue::classId)) {
_tensor.reset(static_cast<TensorFieldValue *>(tensor.release()));
} else {
vespalib::string err = make_string("Expected tensor field value, got a '%s' field value",
tensor->getClass().name());
throw IllegalStateException(err, VESPA_STRLOC);
}
VespaDocumentDeserializer deserializer(repo, stream, Document::getNewestSerializationVersion());
deserializer.read(*_tensor);
verifyCellsTensorIsSparse(_tensor->getAsTensorPtr());
}
TensorModifyUpdate*
TensorModifyUpdate::clone() const
{
return new TensorModifyUpdate(*this);
}
}
|
