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
|
#pragma once
#include "Tensor.h"
namespace DirectCompute
{
using pfnNewCapacity = uint32_t( * )( uint32_t current, uint32_t requested );
uint32_t defaultNewCapacity( uint32_t current, uint32_t requested );
class PooledTensor
{
TensorGpuViews views;
uint32_t capacity = 0;
public:
Tensor tensor( eDataType type, const std::array<uint32_t, 4>& ne, pfnNewCapacity pfnNewCap );
size_t getCapacity() const { return capacity; }
void clear()
{
views.clear();
capacity = 0;
}
HRESULT zeroMemory( CComPtr<ID3D11Buffer>& cb );
};
__interface iTensorArena
{
Tensor tensor( eDataType type, const std::array<uint32_t, 4>& ne );
void reset();
};
class TensorsArena: public iTensorArena
{
public:
struct sArenaConfig
{
pfnNewCapacity pfnCapInner;
size_t initialCapOuter;
};
struct sArenaConfigs
{
sArenaConfig fp16, fp32;
};
TensorsArena( const sArenaConfigs& configs );
Tensor tensor( eDataType type, const std::array<uint32_t, 4>& ne ) override final;
void reset() override final;
void clear();
__m128i getMemoryUse() const;
HRESULT zeroMemory( CComPtr<ID3D11Buffer>& cb );
private:
struct ArenaImpl
{
ArenaImpl( eDataType dataType, const sArenaConfig& config );
void reset()
{
index = 0;
}
void clear()
{
index = 0;
pool.clear();
}
Tensor tensor( const std::array<uint32_t, 4>& ne );
__m128i getMemoryUse() const;
HRESULT zeroMemory( CComPtr<ID3D11Buffer>& cb );
private:
const eDataType type;
const pfnNewCapacity pfnNewCap;
std::vector<PooledTensor> pool;
size_t index = 0;
};
static constexpr size_t countTypes = 2;
std::array<ArenaImpl, countTypes> arenas;
};
}
|
