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
|
// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#pragma once
#include "tensor_store.h"
#include "empty_subspace.h"
#include "vector_bundle.h"
#include <vespa/eval/eval/value_type.h>
#include <vespa/eval/eval/typed_cells.h>
#include <vespa/vespalib/datastore/datastore.h>
namespace vespalib::eval { struct Value; }
namespace search::tensor {
/**
* Class for storing dense tensors with known bounds in memory, used
* by DenseTensorAttribute.
*/
class DenseTensorStore : public TensorStore
{
public:
// 4 Ki buffers of 256 MiB each is 1 TiB.
using RefType = vespalib::datastore::EntryRefT<20>;
using DataStoreType = vespalib::datastore::DataStoreT<RefType>;
using ValueType = vespalib::eval::ValueType;
static constexpr size_t max_dense_tensor_buffer_size = 256_Mi;
struct TensorSizeCalc
{
size_t _numCells; // product of dimension sizes
vespalib::eval::CellType _cell_type;
size_t _aligned_size;
TensorSizeCalc(const ValueType &type);
size_t bufSize() const {
return vespalib::eval::CellTypeUtils::mem_size(_cell_type, _numCells);
}
size_t alignedSize() const noexcept { return _aligned_size; }
};
class BufferType : public vespalib::datastore::BufferType<char>
{
using CleanContext = vespalib::datastore::BufferType<char>::CleanContext;
std::shared_ptr<vespalib::alloc::MemoryAllocator> _allocator;
public:
BufferType(const TensorSizeCalc &tensorSizeCalc, std::shared_ptr<vespalib::alloc::MemoryAllocator> allocator);
~BufferType() override;
void clean_hold(void *buffer, size_t offset, EntryCount num_entries, CleanContext cleanCtx) override;
const vespalib::alloc::MemoryAllocator* get_memory_allocator() const override;
};
private:
DataStoreType _concreteStore;
TensorSizeCalc _tensorSizeCalc;
BufferType _bufferType;
ValueType _type; // type of dense tensor
SubspaceType _subspace_type;
EmptySubspace _empty;
public:
DenseTensorStore(const ValueType &type, std::shared_ptr<vespalib::alloc::MemoryAllocator> allocator);
~DenseTensorStore() override;
const ValueType &type() const { return _type; }
size_t getNumCells() const { return _tensorSizeCalc._numCells; }
size_t getBufSize() const { return _tensorSizeCalc.bufSize(); }
const void *getRawBuffer(RefType ref) const {
return _store.getEntryArray<char>(ref, _bufferType.getArraySize());
}
vespalib::datastore::Handle<char> allocRawBuffer();
void holdTensor(EntryRef ref) override;
EntryRef move_on_compact(EntryRef ref) override;
vespalib::MemoryUsage update_stat(const vespalib::datastore::CompactionStrategy& compaction_strategy) override;
std::unique_ptr<vespalib::datastore::ICompactionContext> start_compact(const vespalib::datastore::CompactionStrategy& compaction_strategy) override;
EntryRef store_tensor(const vespalib::eval::Value &tensor) override;
EntryRef store_encoded_tensor(vespalib::nbostream &encoded) override;
std::unique_ptr<vespalib::eval::Value> get_tensor(EntryRef ref) const override;
bool encode_stored_tensor(EntryRef ref, vespalib::nbostream &target) const override;
const DenseTensorStore* as_dense() const override;
DenseTensorStore* as_dense() override;
vespalib::eval::TypedCells get_typed_cells(EntryRef ref) const {
if (!ref.valid()) {
return _empty.cells();
}
return vespalib::eval::TypedCells(getRawBuffer(ref),
_type.cell_type(), getNumCells());
}
VectorBundle get_vectors(EntryRef ref) const {
if (!ref.valid()) {
return VectorBundle();
}
return VectorBundle(getRawBuffer(ref), 1, _subspace_type);
}
const SubspaceType& get_subspace_type() const noexcept { return _subspace_type; }
// The following methods are meant to be used only for unit tests.
uint32_t getArraySize() const { return _bufferType.getArraySize(); }
uint32_t get_max_buffer_entries() const noexcept { return _bufferType.get_max_entries(); }
};
}
|