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
|
#include "stdafx.h"
#include "TempBuffers.h"
#include "../D3D/createBuffer.h"
#include "../D3D/MappedResource.h"
#include "../D3D/shaders.h"
using namespace DirectCompute;
#define CHECK( hr ) { const HRESULT __hr = ( hr ); if( FAILED( __hr ) ) return __hr; }
HRESULT TempBuffers::Buffer::resize( DXGI_FORMAT format, size_t elements, size_t cbElement, bool zeroMemory, CComPtr<ID3D11Buffer>& cb )
{
if( elements <= capacity )
{
if( zeroMemory )
TempBuffers::zeroMemory( *this, (uint32_t)elements, cb );
return S_OK;
}
clear();
CComPtr<ID3D11Buffer> buffer;
const size_t totalBytes = elements * cbElement;
CHECK( createBuffer( eBufferUse::ReadWrite, totalBytes, &buffer, nullptr, nullptr ) );
CHECK( TensorGpuViews::create( buffer, format, elements, true ) );
capacity = elements;
return S_OK;
}
void TempBuffers::zeroMemory( ID3D11UnorderedAccessView* uav, uint32_t length, CComPtr<ID3D11Buffer>& cb )
{
const __m128i cbData = _mm_cvtsi32_si128( (int)length );
if( cb )
{
MappedResource mapped;
check( mapped.map( cb, false ) );
store16( mapped.data(), cbData );
}
else
{
CD3D11_BUFFER_DESC desc{ 16, D3D11_BIND_CONSTANT_BUFFER, D3D11_USAGE_DYNAMIC, D3D11_CPU_ACCESS_WRITE };
std::array<uint32_t, 4> cbBuffer;
store( cbBuffer, cbData );
D3D11_SUBRESOURCE_DATA srd{ cbBuffer.data(), 0, 0 };
check( device()->CreateBuffer( &desc, &srd, &cb ) );
}
ID3D11DeviceContext* ctx = context();
ctx->CSSetUnorderedAccessViews( 0, 1, &uav, nullptr );
csSetCB( cb );
constexpr uint32_t THREADS = 512;
constexpr uint32_t ITERATIONS = 128;
constexpr uint32_t elementsPerGroup = THREADS * ITERATIONS;
const uint32_t countGroups = ( length + elementsPerGroup - 1 ) / elementsPerGroup;
bindShader( eComputeShader::zeroMemory );
ctx->Dispatch( countGroups, 1, 1 );
}
const TensorGpuViews& TempBuffers::fp16( size_t countElements, bool zeroMemory )
{
HRESULT hr = m_fp16.resize( DXGI_FORMAT_R16_FLOAT, countElements, 2, zeroMemory, smallCb );
if( FAILED( hr ) )
throw hr;
return m_fp16;
}
const TensorGpuViews& TempBuffers::fp16_2( size_t countElements, bool zeroMemory )
{
HRESULT hr = m_fp16_2.resize( DXGI_FORMAT_R16_FLOAT, countElements, 2, zeroMemory, smallCb );
if( FAILED( hr ) )
throw hr;
return m_fp16_2;
}
const TensorGpuViews& TempBuffers::fp32( size_t countElements, bool zeroMemory )
{
HRESULT hr = m_fp32.resize( DXGI_FORMAT_R32_FLOAT, countElements, 4, zeroMemory, smallCb );
if( FAILED( hr ) )
throw hr;
return m_fp32;
}
__m128i TempBuffers::getMemoryUse() const
{
size_t cb = m_fp16.getCapacity() * 2;
cb += m_fp16_2.getCapacity() * 2;
cb += m_fp32.getCapacity() * 4;
return setHigh_size( cb );
}
|
