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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529
|
/*
* Copyright (c) 2018-2019, NVIDIA CORPORATION. All rights reserved.
*
* All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#pragma once
#include <stdint.h>
#include <array>
#include <fstream>
#include <iomanip>
#include <sstream>
#include <vector>
#include "TimestampConverter.h"
namespace NvTraceFormat {
struct GlobalData
{
int64_t cpuTimestampTicksPerSecond; // On x86, frequency of RDTSC ticks
int64_t earliestTimestamp; // Min of all captured timestamps for any device
int64_t latestTimestamp; // Max of all captured timestamps for any device
int32_t totalEventCount; // Total number of records captured for all devices
uint8_t boolGpuTimestampsAlreadyConvertedToCpu; // 1 if true, 0 if false
uint8_t padding[3];
};
using nvtrcMagic_t = std::array<char, 8>;
static const nvtrcMagic_t nvtrcVersionMagic{"nvtrc02"};
struct FileHeader
{
nvtrcMagic_t magic;
GlobalData global;
};
struct ArrayHeader
{
int32_t count;
int32_t elementSize;
};
enum class GpuCtxSwTraceError8 : uint8_t
{
None = 0,
UnsupportedGpu = 1,
UnsupportedDriver = 2,
NeedRoot = 3,
Unknown = 255
};
struct DeviceDesc
{
int64_t cpuTimestampStart; // On x86, RDTSC
int64_t gpuTimestampStart; // NVIDIA GPU globaltimer
int64_t cpuTimestampEnd; // On x86, RDTSC
int64_t gpuTimestampEnd; // NVIDIA GPU globaltimer
int64_t earliestTimestamp; // Min of all captured timestamps for this device
int64_t latestTimestamp; // Max of all captured timestamps for this device
uint8_t uuid[16]; // As in nvidia-smi and VkPhysicalDeviceIDProperties::deviceUUID
char name[190]; // Null-terminated string in fixed-size buffer
GpuCtxSwTraceError8 gpuCtxSwTraceError;
uint8_t boolTimestampsInSortedOrder; // 1 if true, 0 if false
};
enum class Category16 : uint16_t
{
Invalid = 0,
GpuContextSwitch = 1,
Reserved0 = 2
};
enum class TypeGpuCtxSw16 : uint16_t
{
Invalid = 0,
ContextSwitchedIn = 1,
ContextSwitchedOut = 2
};
struct RecordGpuCtxSw
{
Category16 category;
TypeGpuCtxSw16 type;
uint32_t processId;
int64_t timestamp;
uint64_t contextHandle;
};
struct FileData
{
GlobalData global;
std::vector<DeviceDesc> deviceDescs;
std::vector<std::vector<RecordGpuCtxSw>> perDeviceData; // First index is device, second is record
};
template <class T>
inline void Read(std::ifstream& ifs, T& value, int sizeOfValue = sizeof(T))
{
ifs.read(reinterpret_cast<char*>(&value), sizeOfValue);
}
template <class T, class Alloc>
inline bool ReadVector(std::ifstream& ifs, std::vector<T, Alloc>& buffer)
{
ArrayHeader header;
Read(ifs, header);
if (!ifs) return false;
buffer.resize(header.count);
if (header.elementSize == sizeof(T))
{
ifs.read(reinterpret_cast<char*>(&buffer[0]), header.count * header.elementSize);
}
else if (header.elementSize > sizeof(T))
{
for (T& elem : buffer)
{
ifs.read(reinterpret_cast<char*>(&elem), sizeof(T));
}
}
else
{
// File has older version than expected.
// Could attempt to upconvert, but for now simply fail.
return false;
}
if (!ifs) return false;
return true;
}
template <class T>
inline void Write(std::ofstream& ofs, T& value) noexcept
{
ofs.write(reinterpret_cast<char*>(&value), sizeof(T));
}
template <class T, class Alloc>
inline void WriteVector(std::ofstream& ofs, std::vector<T, Alloc> const& buffer) noexcept
{
ArrayHeader header{(int32_t)buffer.size(), (int32_t)sizeof(T)};
Write(ofs, header);
ofs.write(reinterpret_cast<char const*>(&buffer[0]), header.count * header.elementSize);
}
inline bool ReadFileDataVersion(char const* inputFile, FileData& fileData, nvtrcMagic_t const& magic)
{
fileData = FileData();
std::ifstream ifs(inputFile, std::ios::in | std::ios::binary);
if (!ifs) return false;
FileHeader header;
Read(ifs, header);
if (!ifs) return false;
if (header.magic != magic) return false;
fileData.global = header.global;
bool success = ReadVector(ifs, fileData.deviceDescs);
if (!success) return false;
fileData.perDeviceData.resize(fileData.deviceDescs.size());
for (auto& deviceData : fileData.perDeviceData)
{
bool success = ReadVector(ifs, deviceData);
if (!success) return false;
}
return true;
}
inline bool WriteFileDataVersion(char const* outputFile, FileData const& fileData, nvtrcMagic_t const& magic) noexcept
{
if (fileData.deviceDescs.size() != fileData.perDeviceData.size()) return false;
try
{
std::ofstream ofs(outputFile, std::ios::out | std::ios::binary);
if (!ofs) return false;
FileHeader header{magic, fileData.global};
Write(ofs, header);
WriteVector(ofs, fileData.deviceDescs);
for (auto const& deviceData : fileData.perDeviceData)
{
WriteVector(ofs, deviceData);
}
if (!ofs) return false;
}
catch(...)
{
return false;
}
return true;
}
// Helper functions for reading/writing current version, but factored so
// tools can read other versions using the same header.
inline bool ReadFileData(char const* inputFile, FileData& fileData)
{
return ReadFileDataVersion(inputFile, fileData, nvtrcVersionMagic);
}
inline bool WriteFileData(char const* outputFile, FileData const& fileData) noexcept
{
return WriteFileDataVersion(outputFile, fileData, nvtrcVersionMagic);
}
// Helper functions for converting GPU to CPU timestamps, based on the two sync
// points for a given device. If boolGpuTimestampsAlreadyConvertedToCpu in the
// global data is set to 0, use this helper to construct a conversion function
// for a given device.
inline TimestampConverter GpuToCpuTimestampConverter(DeviceDesc const& desc)
{
// Source is GPU time, destination is CPU time
return CreateTimestampConverter(
desc.gpuTimestampStart,
desc.gpuTimestampEnd,
desc.cpuTimestampStart,
desc.cpuTimestampEnd);
}
inline double SecondsElapsed(
int64_t cpuTimestamp,
int64_t earliestCpuTimestamp,
int64_t cpuTimestampTicksPerSecond)
{
return static_cast<double>(cpuTimestamp - earliestCpuTimestamp)
/ static_cast<double>(cpuTimestampTicksPerSecond);
};
inline double ToSeconds(
int64_t cpuTimestamp,
FileData const& fileData)
{
return SecondsElapsed(
cpuTimestamp,
fileData.global.earliestTimestamp,
fileData.global.cpuTimestampTicksPerSecond);
};
// Note that timestamps are automatically converted to CPU time unless raw GPU
// timestamps were explicitly requested. The automatic conversion effectively
// works like this:
// for (size_t deviceIndex = 0; deviceIndex < deviceDescs.size(); ++deviceIndex)
// {
// auto const& deviceDesc = fileData.deviceDescs[deviceIndex];
// auto& records = fileData.perDeviceData[deviceIndex];
//
// auto convertToCpuTime = GpuToCpuTimestampConverter(deviceDesc);
//
// for (auto& record : records)
// record.timestamp = convertToCpuTime(record.timestamp);
// }
//
// In the case of merging multiple FileData objects onto a single timeline, it
// is most accurate to leave all the timestamps in GPU time, and then convert
// them all afterwards using a single conversion factor. Create this common
// converter using the start time of the earliest capture and the end time of
// the latest capture (remembering to handle this separately for each device).
inline void SetName(DeviceDesc& desc, std::string const& name)
{
// Avoid min() macro trouble
auto quickMin = [](size_t a, size_t b) { return a < b ? a : b; };
// Truncate name if too long, ensuring there's a null terminator
size_t indexOfNull = quickMin(name.size(), sizeof(desc.name) - 1);
memcpy(&desc.name[0], name.c_str(), indexOfNull);
desc.name[indexOfNull] = '\0';
}
// Assume uuid points to 16-byte array
inline std::string PrintableUuid(uint8_t const* uuid)
{
std::ostringstream oss;
oss << std::hex;
int offset = 0;
auto printBytes = [&](int count)
{
int end = offset + count;
for (int i = offset; i < end; ++i)
{
// Cast to avoid potentially printing a char type as ASCII
oss << std::setw(2) << std::setfill('0') << static_cast<uint16_t>(uuid[i]);
}
offset = end;
};
printBytes(4);
oss << '-';
printBytes(2);
oss << '-';
printBytes(2);
oss << '-';
printBytes(2);
oss << '-';
printBytes(6);
return oss.str();
}
inline const char* GpuCtxSwTraceErrorToString(GpuCtxSwTraceError8 code, const char* defaultVal = nullptr)
{
switch (code)
{
case GpuCtxSwTraceError8::None:
return defaultVal;
case GpuCtxSwTraceError8::UnsupportedGpu:
return "GPU must be Pascal or newer to use GPU context switch trace";
break;
case GpuCtxSwTraceError8::UnsupportedDriver:
return "Installed NVIDIA display driver does not support GPU context switch trace";
break;
case GpuCtxSwTraceError8::NeedRoot:
return "Process must be running as root/Administrator to use GPU context switch trace";
break;
case GpuCtxSwTraceError8::Unknown: [[fallthrough]];
default:
return "Internal error occurred, please report to NVIDIA";
break;
}
}
// Use these ostream manipulators to concisely stream out hex values, prefixed with 0x
// and padded with leading zeros to be fixed width. For example:
// std::cout << Hex64(0xFFFF'0000'FFFF) << " " << Hex32(0xFFFF);
// writes 0x0000FFFF0000FFFF 0x0000FFFF.
template <typename Value>
class Hex
{
Value n;
public:
template <typename InputValue>
Hex(InputValue n_) : n(static_cast<Value>(n_)) {}
friend std::ostream& operator<<(std::ostream& os, Hex const& value)
{
// 2 digits per byte + 2 chars for "0x" prefix
int digits = 2 * sizeof(Value) + 2;
auto oldFlags = os.flags();
auto oldFill = os.fill();
auto oldWidth = os.width();
os << std::hex << std::showbase << std::internal << std::setw(digits) << std::setfill('0');
os << value.n;
os << std::setw(oldWidth) << std::setfill(oldFill);
os.flags(oldFlags);
return os;
}
};
using Hex64 = Hex<uint64_t>;
using Hex32 = Hex<uint32_t>;
using Hex16 = Hex<uint16_t>;
using Hex8 = Hex<uint8_t>;
// Stream out a textual representation of a FileData's Global data
inline void PrettyPrintFileDataGlobal(
std::ostream& os,
FileData const& fileData)
{
using std::endl;
auto toSec = [&fileData](int64_t cpuTimestamp){ return ToSeconds(cpuTimestamp, fileData); };
os <<
"CPU timestamp ticks per second: " <<
fileData.global.cpuTimestampTicksPerSecond << endl <<
"GPU timestamps pre-converted to CPU equivalents: " <<
(fileData.global.boolGpuTimestampsAlreadyConvertedToCpu ? "Yes" : "No") << endl <<
"Total number of events captured: " << fileData.global.totalEventCount << endl;
if (fileData.global.totalEventCount > 0)
{
os <<
"Range of timestamps captured:" << endl <<
" Earliest: " << Hex64(fileData.global.earliestTimestamp) << endl <<
" Latest: " << Hex64(fileData.global.latestTimestamp) << endl;
if (fileData.global.boolGpuTimestampsAlreadyConvertedToCpu)
{
os <<
"Total duration of captured events: " <<
toSec(fileData.global.latestTimestamp) << " seconds" << endl;
}
}
}
// Stream out a textual representation of a FileData's DeviceDesc list
inline void PrettyPrintFileDataDeviceDescs(
std::ostream& os,
FileData const& fileData)
{
using std::endl;
auto toSec = [&fileData](int64_t cpuTimestamp){ return ToSeconds(cpuTimestamp, fileData); };
for (int d = 0; d < fileData.deviceDescs.size(); ++d)
{
auto& desc = fileData.deviceDescs[d];
auto& records = fileData.perDeviceData[d];
os <<
"Device " << d << ":" << endl <<
"\tName: " << &desc.name[0] << endl <<
"\tUUID: {" << PrintableUuid(&desc.uuid[0]) << "}" << endl <<
"\tSupports GPU context-switch trace: ";
const char* ctxswError = GpuCtxSwTraceErrorToString(desc.gpuCtxSwTraceError);
if (!ctxswError)
{
os << "Yes" << endl;
}
else
{
os << "No -- " << ctxswError << endl;
}
os <<
"\tTimestamps for synchronization:\n" <<
"\t CPU start: " << Hex64(desc.cpuTimestampStart) <<
" GPU start: " << Hex64(desc.gpuTimestampStart) << endl <<
"\t CPU end: " << Hex64(desc.cpuTimestampEnd) <<
" GPU end: " << Hex64(desc.gpuTimestampEnd) << endl <<
"\tNumber of events captured: " << records.size() << endl;
if (records.size() > 0)
{
if (fileData.global.boolGpuTimestampsAlreadyConvertedToCpu)
{
os <<
"\tRange of timestamps captured:" << endl <<
"\t Earliest: " << Hex64(desc.earliestTimestamp) <<
" (" << toSec(desc.earliestTimestamp) << " seconds)" << endl <<
"\t Latest: " << Hex64(desc.latestTimestamp) <<
" (" << toSec(desc.latestTimestamp) << " seconds)" << endl;
}
else
{
os <<
"\tRange of timestamps captured:" << endl <<
"\t Earliest: " << Hex64(desc.earliestTimestamp) << endl <<
"\t Latest: " << Hex64(desc.latestTimestamp) << endl;
}
}
}
}
// Stream out a textual representation of a FileData's record list
inline void PrettyPrintFileDataRecords(
std::ostream& os,
FileData const& fileData)
{
using std::endl;
for (int d = 0; d < fileData.deviceDescs.size(); ++d)
{
auto& records = fileData.perDeviceData[d];
os << "Device " << d << " records:" << endl;
for (auto& record : records)
{
if (record.category == Category16::GpuContextSwitch)
{
char const* type =
(record.type == TypeGpuCtxSw16::ContextSwitchedIn) ? "Context Start" :
(record.type == TypeGpuCtxSw16::ContextSwitchedOut) ? "Context Stop" :
"<Other>";
os <<
"\tTimestamp: " << Hex64(record.timestamp) <<
" | Event: " << std::left << std::setw(13) << type <<
" | PID: " << std::right << std::setw(5) << record.processId <<
" | ContextID: " << Hex32(record.contextHandle) << endl;
}
}
}
}
// Stream out a textual representation of a complete FileData structure
inline void PrettyPrintFileData(
std::ostream& os,
FileData const& fileData,
bool showDeviceDescs = true,
bool showRecords = true)
{
PrettyPrintFileDataGlobal(os, fileData);
if (showDeviceDescs)
{
PrettyPrintFileDataDeviceDescs(os, fileData);
}
if (showRecords)
{
os << std::endl;
PrettyPrintFileDataRecords(os, fileData);
}
}
} // namespace
|