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
|
#include "stdafx.h"
#include "TensorEx.h"
#include "../D3D/createBuffer.h"
#include "../source/ggml.h"
#include "../D3D/MappedResource.h"
using namespace DirectCompute;
HRESULT TensorEx::create( const ggml_tensor& ggml, eBufferUse usage, bool uploadData )
{
TensorGpuViews::clear();
buffer = nullptr;
stagingBuffer = nullptr;
CHECK( TensorShape::create( ggml ) );
const ggml_type dataType = ggml.type;
const uint32_t cbElement = (uint32_t)ggml_type_size( dataType );
const size_t totalBytes = ggml_nbytes( &ggml );
if( totalBytes > INT_MAX )
return DISP_E_OVERFLOW;
const uint32_t countElements = (uint32_t)( totalBytes / cbElement );
{
const void* const rsi = uploadData ? ggml.data : nullptr;
ID3D11Buffer** ppStagingBuffer = ( usage == eBufferUse::ReadWriteDownload ) ? &stagingBuffer : nullptr;
CHECK( createBuffer( usage, totalBytes, &buffer, rsi, ppStagingBuffer ) );
}
DXGI_FORMAT format;
switch( dataType )
{
case GGML_TYPE_F16:
format = DXGI_FORMAT_R16_FLOAT;
break;
case GGML_TYPE_F32:
format = DXGI_FORMAT_R32_FLOAT;
break;
default:
return E_NOTIMPL;
}
const bool makeUav = usage == eBufferUse::ReadWrite || usage == eBufferUse::ReadWriteDownload;
return TensorGpuViews::create( buffer, format, totalBytes / cbElement, makeUav );
}
HRESULT TensorEx::create( eDataType type, eBufferUse usage, const std::array<uint32_t, 4>& sizeElements )
{
TensorGpuViews::clear();
buffer = nullptr;
stagingBuffer = nullptr;
std::initializer_list<uint32_t> il( sizeElements.data(), sizeElements.data() + 4 );
ID3D11Buffer** ppStaging = ( usage == eBufferUse::ReadWriteDownload ) ? &stagingBuffer : nullptr;
return Tensor::create( type, il, usage, buffer, nullptr, ppStaging );
}
HRESULT TensorEx::getViewSize( uint32_t& cbElement, uint32_t& countElements ) const
{
ID3D11ShaderResourceView* const srv = *this;
if( nullptr == srv )
return OLE_E_BLANK;
D3D11_SHADER_RESOURCE_VIEW_DESC viewDesc;
srv->GetDesc( &viewDesc );
cbElement = dxgiSizeof( viewDesc.Format );
assert( viewDesc.ViewDimension == D3D_SRV_DIMENSION_BUFFER );
assert( viewDesc.Buffer.FirstElement == 0 );
countElements = viewDesc.Buffer.NumElements;
return S_OK;
}
HRESULT TensorEx::download( void* rdi, size_t cb ) const
{
if( nullptr == stagingBuffer )
return HRESULT_FROM_WIN32( ERROR_GPIO_OPERATION_DENIED ); // The requested operation is not supported for the specified handle.
ID3D11DeviceContext* const ctx = context();
ctx->CopyResource( stagingBuffer, buffer );
MappedResource mapped;
CHECK( mapped.map( stagingBuffer, true ) );
memcpy( rdi, mapped.data(), cb );
return S_OK;
}
HRESULT TensorEx::download( void* rdi ) const
{
uint32_t cbElement, numElements;
CHECK( getViewSize( cbElement, numElements ) );
size_t cb = (size_t)cbElement * numElements;
return download( rdi, cb );
}
|
