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
|
#pragma once
#include <array>
#include <emmintrin.h>
#include "../../D3D/enums.h"
namespace Tracing
{
using DirectCompute::eDataType;
// File header of the trace file
struct sFileHeader
{
static constexpr uint32_t correctMagic = 0xE6B4A12Du; // random.org
uint32_t magic;
uint8_t formatVersion;
uint8_t zzPadding;
uint16_t cbItem;
uint32_t countItems;
uint32_t zzPadding2;
uint64_t bytesPayload;
uint32_t countStrings, bytesStrings;
};
// Payload data starts immediately after the header, bytesPayload bytes in total.
// Then `bytesStrings` with string names, first countStrings * 4 of them are offsets, then ( bytesStrings - countStrings * 4 ) bytes with the string data.
// The strings in the file are null-terminated.
// Immediately after the strings, the next `cbItem` * `countItems` bytes are actual items (tensors and vectors) saved in the trace.
// The format is weird because optimized for streaming.
// These traces can grow large, we can’t afford memory keeping the payload data in memory.
// Metadata is tiny compared to payload, we accumulate that in memory, and write to the end of the file when closed.
enum struct eItemType : uint8_t
{
Buffer = 1,
Tensor = 2,
};
struct sTraceItem
{
uint64_t payloadOffset;
uint64_t payloadSize;
std::array<uint32_t, 4> size;
std::array<uint32_t, 4> stride;
std::array<uint32_t, 4> formatArgs;
eItemType itemType;
eDataType dataType;
uint8_t countFormatArgs = 0;
uint8_t zzPadding = 0;
uint32_t stringIndex;
uint64_t buffer( uint64_t off, size_t length, eDataType type );
uint64_t tensor( uint64_t off, __m128i ne, __m128i nb, eDataType type );
};
}
|
