Hash :
333da2cf
Author :
Date :
2022-05-03T16:21:41
Vulkan: Limit the total bytes of suballocation garbages in flight This CL tracks the total number of bytes of suballocation garbage. It checks against the limit so that when there are excessive garbages in flight, we will wait for GPU to finish and free up some of these memory before continue. That way we will ensure we do not end up accumulating too much memory and end up with low memory kill on mobile devices. Bug: b/230538246 Change-Id: Ic8292db5617bcee4ec3abe8632f54edfd249cfaa Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3617226 Reviewed-by: Amirali Abdolrashidi <abdolrashidi@google.com> Commit-Queue: Charlie Lao <cclao@google.com> Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510
//
// Copyright 2020 The ANGLE Project Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
//
// CommandProcessor.cpp:
// Implements the class methods for CommandProcessor.
//
#include "libANGLE/renderer/vulkan/CommandProcessor.h"
#include "libANGLE/renderer/vulkan/RendererVk.h"
#include "libANGLE/trace.h"
namespace rx
{
namespace vk
{
namespace
{
constexpr size_t kInFlightCommandsLimit = 50u;
constexpr bool kOutputVmaStatsString = false;
// When suballocation garbages is more than this, we may wait for GPU to finish and free up some
// memory for allocation.
constexpr VkDeviceSize kMaxBufferSuballocationGarbageSize = 64 * 1024 * 1024;
void InitializeSubmitInfo(VkSubmitInfo *submitInfo,
const vk::PrimaryCommandBuffer &commandBuffer,
const std::vector<VkSemaphore> &waitSemaphores,
const std::vector<VkPipelineStageFlags> &waitSemaphoreStageMasks,
const vk::Semaphore *signalSemaphore)
{
// Verify that the submitInfo has been zero'd out.
ASSERT(submitInfo->signalSemaphoreCount == 0);
ASSERT(waitSemaphores.size() == waitSemaphoreStageMasks.size());
submitInfo->sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
submitInfo->commandBufferCount = commandBuffer.valid() ? 1 : 0;
submitInfo->pCommandBuffers = commandBuffer.ptr();
submitInfo->waitSemaphoreCount = static_cast<uint32_t>(waitSemaphores.size());
submitInfo->pWaitSemaphores = waitSemaphores.empty() ? nullptr : waitSemaphores.data();
submitInfo->pWaitDstStageMask = waitSemaphoreStageMasks.data();
if (signalSemaphore)
{
submitInfo->signalSemaphoreCount = 1;
submitInfo->pSignalSemaphores = signalSemaphore->ptr();
}
}
bool CommandsHaveValidOrdering(const std::vector<vk::CommandBatch> &commands)
{
Serial currentSerial;
for (const vk::CommandBatch &commandBatch : commands)
{
if (commandBatch.serial <= currentSerial)
{
return false;
}
currentSerial = commandBatch.serial;
}
return true;
}
template <typename SecondaryCommandBufferListT>
void ResetSecondaryCommandBuffers(VkDevice device,
vk::CommandPool *commandPool,
SecondaryCommandBufferListT *commandBuffers)
{
// Nothing to do when using ANGLE secondary command buffers.
}
template <>
ANGLE_MAYBE_UNUSED void ResetSecondaryCommandBuffers<std::vector<VulkanSecondaryCommandBuffer>>(
VkDevice device,
vk::CommandPool *commandPool,
std::vector<VulkanSecondaryCommandBuffer> *commandBuffers)
{
// Note: we currently free the command buffers individually, but we could potentially reset the
// entire command pool. https://issuetracker.google.com/issues/166793850
for (VulkanSecondaryCommandBuffer &secondary : *commandBuffers)
{
commandPool->freeCommandBuffers(device, 1, secondary.ptr());
secondary.releaseHandle();
}
commandBuffers->clear();
}
// Count the number of batches with serial <= given serial. A reference to the fence of the last
// batch with a valid fence is returned for waiting purposes. Note that due to empty submissions
// being optimized out, there may not be a fence associated with every batch.
size_t GetBatchCountUpToSerial(std::vector<CommandBatch> &inFlightCommands,
Serial serial,
Shared<Fence> **fenceToWaitOnOut)
{
size_t batchCount = 0;
while (batchCount < inFlightCommands.size() && inFlightCommands[batchCount].serial <= serial)
{
if (inFlightCommands[batchCount].fence.isReferenced())
{
*fenceToWaitOnOut = &inFlightCommands[batchCount].fence;
}
batchCount++;
}
return batchCount;
}
} // namespace
angle::Result FenceRecycler::newSharedFence(vk::Context *context,
vk::Shared<vk::Fence> *sharedFenceOut)
{
bool gotRecycledFence = false;
vk::Fence fence;
{
std::lock_guard<std::mutex> lock(mMutex);
if (!mRecyler.empty())
{
mRecyler.fetch(&fence);
gotRecycledFence = true;
}
}
VkDevice device(context->getDevice());
if (gotRecycledFence)
{
ANGLE_VK_TRY(context, fence.reset(device));
}
else
{
VkFenceCreateInfo fenceCreateInfo = {};
fenceCreateInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
fenceCreateInfo.flags = 0;
ANGLE_VK_TRY(context, fence.init(device, fenceCreateInfo));
}
sharedFenceOut->assign(device, std::move(fence));
return angle::Result::Continue;
}
void FenceRecycler::destroy(vk::Context *context)
{
std::lock_guard<std::mutex> lock(mMutex);
mRecyler.destroy(context->getDevice());
}
// CommandProcessorTask implementation
void CommandProcessorTask::initTask()
{
mTask = CustomTask::Invalid;
mOutsideRenderPassCommandBuffer = nullptr;
mRenderPassCommandBuffer = nullptr;
mRenderPass = nullptr;
mSemaphore = nullptr;
mCommandPools = nullptr;
mOneOffWaitSemaphore = nullptr;
mOneOffWaitSemaphoreStageMask = 0;
mOneOffFence = nullptr;
mPresentInfo = {};
mPresentInfo.pResults = nullptr;
mPresentInfo.pSwapchains = nullptr;
mPresentInfo.pImageIndices = nullptr;
mPresentInfo.pNext = nullptr;
mPresentInfo.pWaitSemaphores = nullptr;
mOneOffCommandBufferVk = VK_NULL_HANDLE;
mPriority = egl::ContextPriority::Medium;
mHasProtectedContent = false;
}
void CommandProcessorTask::initOutsideRenderPassProcessCommands(
bool hasProtectedContent,
OutsideRenderPassCommandBufferHelper *commandBuffer)
{
mTask = CustomTask::ProcessOutsideRenderPassCommands;
mOutsideRenderPassCommandBuffer = commandBuffer;
mHasProtectedContent = hasProtectedContent;
}
void CommandProcessorTask::initRenderPassProcessCommands(
bool hasProtectedContent,
RenderPassCommandBufferHelper *commandBuffer,
const RenderPass *renderPass)
{
mTask = CustomTask::ProcessRenderPassCommands;
mRenderPassCommandBuffer = commandBuffer;
mRenderPass = renderPass;
mHasProtectedContent = hasProtectedContent;
}
void CommandProcessorTask::copyPresentInfo(const VkPresentInfoKHR &other)
{
if (other.sType == 0)
{
return;
}
mPresentInfo.sType = other.sType;
mPresentInfo.pNext = other.pNext;
if (other.swapchainCount > 0)
{
ASSERT(other.swapchainCount == 1);
mPresentInfo.swapchainCount = 1;
mSwapchain = other.pSwapchains[0];
mPresentInfo.pSwapchains = &mSwapchain;
mImageIndex = other.pImageIndices[0];
mPresentInfo.pImageIndices = &mImageIndex;
}
if (other.waitSemaphoreCount > 0)
{
ASSERT(other.waitSemaphoreCount == 1);
mPresentInfo.waitSemaphoreCount = 1;
mWaitSemaphore = other.pWaitSemaphores[0];
mPresentInfo.pWaitSemaphores = &mWaitSemaphore;
}
mPresentInfo.pResults = other.pResults;
void *pNext = const_cast<void *>(other.pNext);
while (pNext != nullptr)
{
VkStructureType sType = *reinterpret_cast<VkStructureType *>(pNext);
switch (sType)
{
case VK_STRUCTURE_TYPE_PRESENT_REGIONS_KHR:
{
const VkPresentRegionsKHR *presentRegions =
reinterpret_cast<VkPresentRegionsKHR *>(pNext);
mPresentRegion = *presentRegions->pRegions;
mRects.resize(mPresentRegion.rectangleCount);
for (uint32_t i = 0; i < mPresentRegion.rectangleCount; i++)
{
mRects[i] = presentRegions->pRegions->pRectangles[i];
}
mPresentRegion.pRectangles = mRects.data();
mPresentRegions.sType = VK_STRUCTURE_TYPE_PRESENT_REGIONS_KHR;
mPresentRegions.pNext = presentRegions->pNext;
mPresentRegions.swapchainCount = 1;
mPresentRegions.pRegions = &mPresentRegion;
mPresentInfo.pNext = &mPresentRegions;
pNext = const_cast<void *>(presentRegions->pNext);
break;
}
default:
ERR() << "Unknown sType: " << sType << " in VkPresentInfoKHR.pNext chain";
UNREACHABLE();
break;
}
}
}
void CommandProcessorTask::initPresent(egl::ContextPriority priority,
const VkPresentInfoKHR &presentInfo)
{
mTask = CustomTask::Present;
mPriority = priority;
copyPresentInfo(presentInfo);
}
void CommandProcessorTask::initFinishToSerial(Serial serial)
{
// Note: sometimes the serial is not valid and that's okay, the finish will early exit in the
// TaskProcessor::finishToSerial
mTask = CustomTask::FinishToSerial;
mSerial = serial;
}
void CommandProcessorTask::initWaitIdle()
{
mTask = CustomTask::WaitIdle;
}
void CommandProcessorTask::initFlushAndQueueSubmit(
const std::vector<VkSemaphore> &waitSemaphores,
const std::vector<VkPipelineStageFlags> &waitSemaphoreStageMasks,
const Semaphore *semaphore,
bool hasProtectedContent,
egl::ContextPriority priority,
SecondaryCommandPools *commandPools,
GarbageList &¤tGarbage,
SecondaryCommandBufferList &&commandBuffersToReset,
Serial submitQueueSerial)
{
mTask = CustomTask::FlushAndQueueSubmit;
mWaitSemaphores = waitSemaphores;
mWaitSemaphoreStageMasks = waitSemaphoreStageMasks;
mSemaphore = semaphore;
mCommandPools = commandPools;
mGarbage = std::move(currentGarbage);
mCommandBuffersToReset = std::move(commandBuffersToReset);
mPriority = priority;
mHasProtectedContent = hasProtectedContent;
mSerial = submitQueueSerial;
}
void CommandProcessorTask::initOneOffQueueSubmit(VkCommandBuffer commandBufferHandle,
bool hasProtectedContent,
egl::ContextPriority priority,
const Semaphore *waitSemaphore,
VkPipelineStageFlags waitSemaphoreStageMask,
const Fence *fence,
Serial submitQueueSerial)
{
mTask = CustomTask::OneOffQueueSubmit;
mOneOffCommandBufferVk = commandBufferHandle;
mOneOffWaitSemaphore = waitSemaphore;
mOneOffWaitSemaphoreStageMask = waitSemaphoreStageMask;
mOneOffFence = fence;
mPriority = priority;
mHasProtectedContent = hasProtectedContent;
mSerial = submitQueueSerial;
}
CommandProcessorTask &CommandProcessorTask::operator=(CommandProcessorTask &&rhs)
{
if (this == &rhs)
{
return *this;
}
std::swap(mRenderPass, rhs.mRenderPass);
std::swap(mOutsideRenderPassCommandBuffer, rhs.mOutsideRenderPassCommandBuffer);
std::swap(mRenderPassCommandBuffer, rhs.mRenderPassCommandBuffer);
std::swap(mTask, rhs.mTask);
std::swap(mWaitSemaphores, rhs.mWaitSemaphores);
std::swap(mWaitSemaphoreStageMasks, rhs.mWaitSemaphoreStageMasks);
std::swap(mSemaphore, rhs.mSemaphore);
std::swap(mOneOffWaitSemaphore, rhs.mOneOffWaitSemaphore);
std::swap(mOneOffWaitSemaphoreStageMask, rhs.mOneOffWaitSemaphoreStageMask);
std::swap(mOneOffFence, rhs.mOneOffFence);
std::swap(mCommandPools, rhs.mCommandPools);
std::swap(mGarbage, rhs.mGarbage);
std::swap(mCommandBuffersToReset, rhs.mCommandBuffersToReset);
std::swap(mSerial, rhs.mSerial);
std::swap(mPriority, rhs.mPriority);
std::swap(mHasProtectedContent, rhs.mHasProtectedContent);
std::swap(mOneOffCommandBufferVk, rhs.mOneOffCommandBufferVk);
copyPresentInfo(rhs.mPresentInfo);
// clear rhs now that everything has moved.
rhs.initTask();
return *this;
}
// CommandBatch implementation.
CommandBatch::CommandBatch() : commandPools(nullptr), hasProtectedContent(false) {}
CommandBatch::~CommandBatch() = default;
CommandBatch::CommandBatch(CommandBatch &&other) : CommandBatch()
{
*this = std::move(other);
}
CommandBatch &CommandBatch::operator=(CommandBatch &&other)
{
std::swap(primaryCommands, other.primaryCommands);
std::swap(commandPools, other.commandPools);
std::swap(commandBuffersToReset, other.commandBuffersToReset);
std::swap(fence, other.fence);
std::swap(serial, other.serial);
std::swap(hasProtectedContent, other.hasProtectedContent);
return *this;
}
void CommandBatch::destroy(VkDevice device)
{
primaryCommands.destroy(device);
fence.reset(device);
hasProtectedContent = false;
}
void CommandBatch::resetSecondaryCommandBuffers(VkDevice device)
{
ResetSecondaryCommandBuffers(device, &commandPools->outsideRenderPassPool,
&commandBuffersToReset.outsideRenderPassCommandBuffers);
ResetSecondaryCommandBuffers(device, &commandPools->renderPassPool,
&commandBuffersToReset.renderPassCommandBuffers);
}
// CommandProcessor implementation.
void CommandProcessor::handleError(VkResult errorCode,
const char *file,
const char *function,
unsigned int line)
{
ASSERT(errorCode != VK_SUCCESS);
std::stringstream errorStream;
errorStream << "Internal Vulkan error (" << errorCode << "): " << VulkanResultString(errorCode)
<< ".";
if (errorCode == VK_ERROR_DEVICE_LOST)
{
WARN() << errorStream.str();
handleDeviceLost(mRenderer);
}
std::lock_guard<std::mutex> queueLock(mErrorMutex);
Error error = {errorCode, file, function, line};
mErrors.emplace(error);
}
CommandProcessor::CommandProcessor(RendererVk *renderer)
: Context(renderer), mWorkerThreadIdle(false)
{
std::lock_guard<std::mutex> queueLock(mErrorMutex);
while (!mErrors.empty())
{
mErrors.pop();
}
}
CommandProcessor::~CommandProcessor() = default;
angle::Result CommandProcessor::checkAndPopPendingError(Context *errorHandlingContext)
{
std::lock_guard<std::mutex> queueLock(mErrorMutex);
if (mErrors.empty())
{
return angle::Result::Continue;
}
else
{
Error err = mErrors.front();
mErrors.pop();
errorHandlingContext->handleError(err.errorCode, err.file, err.function, err.line);
return angle::Result::Stop;
}
}
void CommandProcessor::queueCommand(CommandProcessorTask &&task)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandProcessor::queueCommand");
// Grab the worker mutex so that we put things on the queue in the same order as we give out
// serials.
std::lock_guard<std::mutex> queueLock(mWorkerMutex);
mTasks.emplace(std::move(task));
mWorkAvailableCondition.notify_one();
}
void CommandProcessor::processTasks()
{
while (true)
{
bool exitThread = false;
angle::Result result = processTasksImpl(&exitThread);
if (exitThread)
{
// We are doing a controlled exit of the thread, break out of the while loop.
break;
}
if (result != angle::Result::Continue)
{
// TODO: https://issuetracker.google.com/issues/170311829 - follow-up on error handling
// ContextVk::commandProcessorSyncErrorsAndQueueCommand and WindowSurfaceVk::destroy
// do error processing, is anything required here? Don't think so, mostly need to
// continue the worker thread until it's been told to exit.
UNREACHABLE();
}
}
}
angle::Result CommandProcessor::processTasksImpl(bool *exitThread)
{
while (true)
{
std::unique_lock<std::mutex> lock(mWorkerMutex);
if (mTasks.empty())
{
mWorkerThreadIdle = true;
mWorkerIdleCondition.notify_all();
// Only wake if notified and command queue is not empty
mWorkAvailableCondition.wait(lock, [this] { return !mTasks.empty(); });
}
mWorkerThreadIdle = false;
CommandProcessorTask task(std::move(mTasks.front()));
mTasks.pop();
lock.unlock();
ANGLE_TRY(processTask(&task));
if (task.getTaskCommand() == CustomTask::Exit)
{
*exitThread = true;
lock.lock();
mWorkerThreadIdle = true;
mWorkerIdleCondition.notify_one();
return angle::Result::Continue;
}
}
UNREACHABLE();
return angle::Result::Stop;
}
angle::Result CommandProcessor::processTask(CommandProcessorTask *task)
{
switch (task->getTaskCommand())
{
case CustomTask::Exit:
{
ANGLE_TRY(mCommandQueue.finishToSerial(this, Serial::Infinite(),
mRenderer->getMaxFenceWaitTimeNs()));
// Shutting down so cleanup
mCommandQueue.destroy(this);
break;
}
case CustomTask::FlushAndQueueSubmit:
{
ANGLE_TRACE_EVENT0("gpu.angle", "processTask::FlushAndQueueSubmit");
// End command buffer
// Call submitFrame()
ANGLE_TRY(mCommandQueue.submitFrame(
this, task->hasProtectedContent(), task->getPriority(), task->getWaitSemaphores(),
task->getWaitSemaphoreStageMasks(), task->getSemaphore(),
std::move(task->getGarbage()), std::move(task->getCommandBuffersToReset()),
task->getCommandPools(), task->getQueueSerial()));
ASSERT(task->getGarbage().empty());
break;
}
case CustomTask::OneOffQueueSubmit:
{
ANGLE_TRACE_EVENT0("gpu.angle", "processTask::OneOffQueueSubmit");
ANGLE_TRY(mCommandQueue.queueSubmitOneOff(
this, task->hasProtectedContent(), task->getPriority(),
task->getOneOffCommandBufferVk(), task->getOneOffWaitSemaphore(),
task->getOneOffWaitSemaphoreStageMask(), task->getOneOffFence(),
SubmitPolicy::EnsureSubmitted, task->getQueueSerial()));
ANGLE_TRY(mCommandQueue.checkCompletedCommands(this));
break;
}
case CustomTask::FinishToSerial:
{
ANGLE_TRY(mCommandQueue.finishToSerial(this, task->getQueueSerial(),
mRenderer->getMaxFenceWaitTimeNs()));
break;
}
case CustomTask::WaitIdle:
{
ANGLE_TRY(mCommandQueue.waitIdle(this, mRenderer->getMaxFenceWaitTimeNs()));
break;
}
case CustomTask::Present:
{
VkResult result = present(task->getPriority(), task->getPresentInfo());
if (ANGLE_UNLIKELY(result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_SUBOPTIMAL_KHR))
{
// We get to ignore these as they are not fatal
}
else if (ANGLE_UNLIKELY(result != VK_SUCCESS))
{
// Save the error so that we can handle it.
// Don't leave processing loop, don't consider errors from present to be fatal.
// TODO: https://issuetracker.google.com/issues/170329600 - This needs to improve to
// properly parallelize present
handleError(result, __FILE__, __FUNCTION__, __LINE__);
}
break;
}
case CustomTask::ProcessOutsideRenderPassCommands:
{
OutsideRenderPassCommandBufferHelper *commandBuffer =
task->getOutsideRenderPassCommandBuffer();
ANGLE_TRY(mCommandQueue.flushOutsideRPCommands(this, task->hasProtectedContent(),
&commandBuffer));
OutsideRenderPassCommandBufferHelper *originalCommandBuffer =
task->getOutsideRenderPassCommandBuffer();
mRenderer->recycleOutsideRenderPassCommandBufferHelper(mRenderer->getDevice(),
&originalCommandBuffer);
break;
}
case CustomTask::ProcessRenderPassCommands:
{
RenderPassCommandBufferHelper *commandBuffer = task->getRenderPassCommandBuffer();
ANGLE_TRY(mCommandQueue.flushRenderPassCommands(
this, task->hasProtectedContent(), *task->getRenderPass(), &commandBuffer));
RenderPassCommandBufferHelper *originalCommandBuffer =
task->getRenderPassCommandBuffer();
mRenderer->recycleRenderPassCommandBufferHelper(mRenderer->getDevice(),
&originalCommandBuffer);
break;
}
case CustomTask::CheckCompletedCommands:
{
ANGLE_TRY(mCommandQueue.checkCompletedCommands(this));
break;
}
default:
UNREACHABLE();
break;
}
return angle::Result::Continue;
}
angle::Result CommandProcessor::checkCompletedCommands(Context *context)
{
ANGLE_TRY(checkAndPopPendingError(context));
CommandProcessorTask checkCompletedTask;
checkCompletedTask.initTask(CustomTask::CheckCompletedCommands);
queueCommand(std::move(checkCompletedTask));
return angle::Result::Continue;
}
angle::Result CommandProcessor::waitForWorkComplete(Context *context)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandProcessor::waitForWorkComplete");
std::unique_lock<std::mutex> lock(mWorkerMutex);
mWorkerIdleCondition.wait(lock, [this] { return (mTasks.empty() && mWorkerThreadIdle); });
// Worker thread is idle and command queue is empty so good to continue
// Sync any errors to the context
bool shouldStop = hasPendingError();
while (hasPendingError())
{
(void)checkAndPopPendingError(context);
}
return shouldStop ? angle::Result::Stop : angle::Result::Continue;
}
angle::Result CommandProcessor::init(Context *context, const DeviceQueueMap &queueMap)
{
ANGLE_TRY(mCommandQueue.init(context, queueMap));
mTaskThread = std::thread(&CommandProcessor::processTasks, this);
return angle::Result::Continue;
}
void CommandProcessor::destroy(Context *context)
{
CommandProcessorTask endTask;
endTask.initTask(CustomTask::Exit);
queueCommand(std::move(endTask));
(void)waitForWorkComplete(context);
if (mTaskThread.joinable())
{
mTaskThread.join();
}
}
bool CommandProcessor::isBusy() const
{
std::lock_guard<std::mutex> serialLock(mQueueSerialMutex);
std::lock_guard<std::mutex> workerLock(mWorkerMutex);
return !mTasks.empty() || mCommandQueue.isBusy();
}
Serial CommandProcessor::reserveSubmitSerial()
{
std::lock_guard<std::mutex> lock(mQueueSerialMutex);
return mCommandQueue.reserveSubmitSerial();
}
// Wait until all commands up to and including serial have been processed
angle::Result CommandProcessor::finishToSerial(Context *context, Serial serial, uint64_t timeout)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandProcessor::finishToSerial");
ANGLE_TRY(checkAndPopPendingError(context));
CommandProcessorTask task;
task.initFinishToSerial(serial);
queueCommand(std::move(task));
// Wait until the worker is idle. At that point we know that the finishToSerial command has
// completed executing, including any associated state cleanup.
return waitForWorkComplete(context);
}
angle::Result CommandProcessor::waitIdle(Context *context, uint64_t timeout)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandProcessor::waitIdle");
CommandProcessorTask task;
task.initWaitIdle();
queueCommand(std::move(task));
return waitForWorkComplete(context);
}
void CommandProcessor::handleDeviceLost(RendererVk *renderer)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandProcessor::handleDeviceLost");
std::unique_lock<std::mutex> lock(mWorkerMutex);
mWorkerIdleCondition.wait(lock, [this] { return (mTasks.empty() && mWorkerThreadIdle); });
// Worker thread is idle and command queue is empty so good to continue
mCommandQueue.handleDeviceLost(renderer);
}
VkResult CommandProcessor::getLastAndClearPresentResult(VkSwapchainKHR swapchain)
{
std::unique_lock<std::mutex> lock(mSwapchainStatusMutex);
if (mSwapchainStatus.find(swapchain) == mSwapchainStatus.end())
{
// Wake when required swapchain status becomes available
mSwapchainStatusCondition.wait(lock, [this, swapchain] {
return mSwapchainStatus.find(swapchain) != mSwapchainStatus.end();
});
}
VkResult result = mSwapchainStatus[swapchain];
mSwapchainStatus.erase(swapchain);
return result;
}
VkResult CommandProcessor::present(egl::ContextPriority priority,
const VkPresentInfoKHR &presentInfo)
{
std::lock_guard<std::mutex> lock(mSwapchainStatusMutex);
ANGLE_TRACE_EVENT0("gpu.angle", "vkQueuePresentKHR");
VkResult result = mCommandQueue.queuePresent(priority, presentInfo);
// Verify that we are presenting one and only one swapchain
ASSERT(presentInfo.swapchainCount == 1);
ASSERT(presentInfo.pResults == nullptr);
mSwapchainStatus[presentInfo.pSwapchains[0]] = result;
mSwapchainStatusCondition.notify_all();
return result;
}
angle::Result CommandProcessor::submitFrame(
Context *context,
bool hasProtectedContent,
egl::ContextPriority priority,
const std::vector<VkSemaphore> &waitSemaphores,
const std::vector<VkPipelineStageFlags> &waitSemaphoreStageMasks,
const Semaphore *signalSemaphore,
GarbageList &¤tGarbage,
SecondaryCommandBufferList &&commandBuffersToReset,
SecondaryCommandPools *commandPools,
Serial submitQueueSerial)
{
ANGLE_TRY(checkAndPopPendingError(context));
CommandProcessorTask task;
task.initFlushAndQueueSubmit(waitSemaphores, waitSemaphoreStageMasks, signalSemaphore,
hasProtectedContent, priority, commandPools,
std::move(currentGarbage), std::move(commandBuffersToReset),
submitQueueSerial);
queueCommand(std::move(task));
return angle::Result::Continue;
}
angle::Result CommandProcessor::queueSubmitOneOff(Context *context,
bool hasProtectedContent,
egl::ContextPriority contextPriority,
VkCommandBuffer commandBufferHandle,
const Semaphore *waitSemaphore,
VkPipelineStageFlags waitSemaphoreStageMask,
const Fence *fence,
SubmitPolicy submitPolicy,
Serial submitQueueSerial)
{
ANGLE_TRY(checkAndPopPendingError(context));
CommandProcessorTask task;
task.initOneOffQueueSubmit(commandBufferHandle, hasProtectedContent, contextPriority,
waitSemaphore, waitSemaphoreStageMask, fence, submitQueueSerial);
queueCommand(std::move(task));
if (submitPolicy == SubmitPolicy::EnsureSubmitted)
{
// Caller has synchronization requirement to have work in GPU pipe when returning from this
// function.
ANGLE_TRY(waitForWorkComplete(context));
}
return angle::Result::Continue;
}
VkResult CommandProcessor::queuePresent(egl::ContextPriority contextPriority,
const VkPresentInfoKHR &presentInfo)
{
CommandProcessorTask task;
task.initPresent(contextPriority, presentInfo);
ANGLE_TRACE_EVENT0("gpu.angle", "CommandProcessor::queuePresent");
queueCommand(std::move(task));
// Always return success, when we call acquireNextImage we'll check the return code. This
// allows the app to continue working until we really need to know the return code from
// present.
return VK_SUCCESS;
}
angle::Result CommandProcessor::waitForSerialWithUserTimeout(vk::Context *context,
Serial serial,
uint64_t timeout,
VkResult *result)
{
// If finishToSerial times out we generate an error. Therefore we a large timeout.
// TODO: https://issuetracker.google.com/170312581 - Wait with timeout.
return finishToSerial(context, serial, mRenderer->getMaxFenceWaitTimeNs());
}
angle::Result CommandProcessor::flushOutsideRPCommands(
Context *context,
bool hasProtectedContent,
OutsideRenderPassCommandBufferHelper **outsideRPCommands)
{
ANGLE_TRY(checkAndPopPendingError(context));
(*outsideRPCommands)->markClosed();
CommandProcessorTask task;
task.initOutsideRenderPassProcessCommands(hasProtectedContent, *outsideRPCommands);
queueCommand(std::move(task));
return mRenderer->getOutsideRenderPassCommandBufferHelper(
context, (*outsideRPCommands)->getCommandPool(), outsideRPCommands);
}
angle::Result CommandProcessor::flushRenderPassCommands(
Context *context,
bool hasProtectedContent,
const RenderPass &renderPass,
RenderPassCommandBufferHelper **renderPassCommands)
{
ANGLE_TRY(checkAndPopPendingError(context));
(*renderPassCommands)->markClosed();
CommandProcessorTask task;
task.initRenderPassProcessCommands(hasProtectedContent, *renderPassCommands, &renderPass);
queueCommand(std::move(task));
return mRenderer->getRenderPassCommandBufferHelper(
context, (*renderPassCommands)->getCommandPool(), renderPassCommands);
}
angle::Result CommandProcessor::ensureNoPendingWork(Context *context)
{
return waitForWorkComplete(context);
}
// CommandQueue implementation.
CommandQueue::CommandQueue() : mCurrentQueueSerial(mQueueSerialFactory.generate()), mPerfCounters{}
{}
CommandQueue::~CommandQueue() = default;
void CommandQueue::destroy(Context *context)
{
// Force all commands to finish by flushing all queues.
for (VkQueue queue : mQueueMap)
{
if (queue != VK_NULL_HANDLE)
{
vkQueueWaitIdle(queue);
}
}
RendererVk *renderer = context->getRenderer();
mLastCompletedQueueSerial = Serial::Infinite();
(void)clearAllGarbage(renderer);
mPrimaryCommands.destroy(renderer->getDevice());
mPrimaryCommandPool.destroy(renderer->getDevice());
if (mProtectedPrimaryCommandPool.valid())
{
mProtectedPrimaryCommands.destroy(renderer->getDevice());
mProtectedPrimaryCommandPool.destroy(renderer->getDevice());
}
mFenceRecycler.destroy(context);
ASSERT(mInFlightCommands.empty() && mGarbageQueue.empty());
}
angle::Result CommandQueue::init(Context *context, const vk::DeviceQueueMap &queueMap)
{
// Initialize the command pool now that we know the queue family index.
ANGLE_TRY(mPrimaryCommandPool.init(context, false, queueMap.getIndex()));
mQueueMap = queueMap;
if (queueMap.isProtected())
{
ANGLE_TRY(mProtectedPrimaryCommandPool.init(context, true, queueMap.getIndex()));
}
return angle::Result::Continue;
}
angle::Result CommandQueue::checkCompletedCommands(Context *context)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandQueue::checkCompletedCommandsNoLock");
RendererVk *renderer = context->getRenderer();
VkDevice device = renderer->getDevice();
int finishedCount = 0;
for (CommandBatch &batch : mInFlightCommands)
{
// For empty submissions, fence is not set but there may be garbage to be collected. In
// such a case, the empty submission is "completed" at the same time as the last submission
// that actually happened.
if (batch.fence.isReferenced())
{
VkResult result = batch.fence.get().getStatus(device);
if (result == VK_NOT_READY)
{
break;
}
ANGLE_VK_TRY(context, result);
}
++finishedCount;
}
if (finishedCount == 0)
{
return angle::Result::Continue;
}
return retireFinishedCommands(context, finishedCount);
}
angle::Result CommandQueue::retireFinishedCommands(Context *context, size_t finishedCount)
{
ASSERT(finishedCount > 0);
RendererVk *renderer = context->getRenderer();
VkDevice device = renderer->getDevice();
// First store the last completed queue serial value into a local variable and then update
// mLastCompletedQueueSerial once in the end.
Serial lastCompletedQueueSerial;
for (size_t commandIndex = 0; commandIndex < finishedCount; ++commandIndex)
{
CommandBatch &batch = mInFlightCommands[commandIndex];
lastCompletedQueueSerial = batch.serial;
if (batch.fence.isReferenced())
{
mFenceRecycler.resetSharedFence(&batch.fence);
}
if (batch.primaryCommands.valid())
{
ANGLE_TRACE_EVENT0("gpu.angle", "Primary command buffer recycling");
PersistentCommandPool &commandPool = getCommandPool(batch.hasProtectedContent);
ANGLE_TRY(commandPool.collect(context, std::move(batch.primaryCommands)));
}
ANGLE_TRACE_EVENT0("gpu.angle", "Secondary command buffer recycling");
batch.resetSecondaryCommandBuffers(device);
}
mLastCompletedQueueSerial = lastCompletedQueueSerial;
auto beginIter = mInFlightCommands.begin();
mInFlightCommands.erase(beginIter, beginIter + finishedCount);
while (!mGarbageQueue.empty())
{
GarbageAndSerial &garbageList = mGarbageQueue.front();
if (garbageList.getSerial() < lastCompletedQueueSerial)
{
for (GarbageObject &garbage : garbageList.get())
{
garbage.destroy(renderer);
}
mGarbageQueue.pop();
}
else
{
break;
}
}
// Now clean up RendererVk garbage
renderer->cleanupGarbage(getLastCompletedQueueSerial());
return angle::Result::Continue;
}
void CommandQueue::releaseToCommandBatch(bool hasProtectedContent,
PrimaryCommandBuffer &&commandBuffer,
SecondaryCommandPools *commandPools,
CommandBatch *batch)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandQueue::releaseToCommandBatch");
batch->primaryCommands = std::move(commandBuffer);
batch->commandPools = commandPools;
batch->hasProtectedContent = hasProtectedContent;
}
void CommandQueue::clearAllGarbage(RendererVk *renderer)
{
while (!mGarbageQueue.empty())
{
GarbageAndSerial &garbageList = mGarbageQueue.front();
for (GarbageObject &garbage : garbageList.get())
{
garbage.destroy(renderer);
}
mGarbageQueue.pop();
}
}
void CommandQueue::handleDeviceLost(RendererVk *renderer)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandQueue::handleDeviceLost");
VkDevice device = renderer->getDevice();
for (CommandBatch &batch : mInFlightCommands)
{
// On device loss we need to wait for fence to be signaled before destroying it
if (batch.fence.isReferenced())
{
VkResult status = batch.fence.get().wait(device, renderer->getMaxFenceWaitTimeNs());
// If the wait times out, it is probably not possible to recover from lost device
ASSERT(status == VK_SUCCESS || status == VK_ERROR_DEVICE_LOST);
batch.fence.reset(device);
}
// On device lost, here simply destroy the CommandBuffer, it will fully cleared later
// by CommandPool::destroy
if (batch.primaryCommands.valid())
{
batch.primaryCommands.destroy(device);
}
batch.resetSecondaryCommandBuffers(device);
}
mInFlightCommands.clear();
}
bool CommandQueue::allInFlightCommandsAreAfterSerial(Serial serial)
{
return mInFlightCommands.empty() || mInFlightCommands[0].serial > serial;
}
angle::Result CommandQueue::finishToSerial(Context *context, Serial finishSerial, uint64_t timeout)
{
if (mInFlightCommands.empty())
{
return angle::Result::Continue;
}
ANGLE_TRACE_EVENT0("gpu.angle", "CommandQueue::finishToSerial");
// Find the serial in the the list. The serials should be in order.
ASSERT(CommandsHaveValidOrdering(mInFlightCommands));
Shared<Fence> *fenceToWaitOn = nullptr;
size_t finishCount = GetBatchCountUpToSerial(mInFlightCommands, finishSerial, &fenceToWaitOn);
if (finishCount == 0)
{
return angle::Result::Continue;
}
// Wait for it finish. If no fence, the serial is already finished, it might just have garbage
// to clean up.
if (fenceToWaitOn != nullptr)
{
VkDevice device = context->getDevice();
VkResult status = fenceToWaitOn->get().wait(device, timeout);
ANGLE_VK_TRY(context, status);
}
// Clean up finished batches.
ANGLE_TRY(retireFinishedCommands(context, finishCount));
ASSERT(allInFlightCommandsAreAfterSerial(finishSerial));
return angle::Result::Continue;
}
angle::Result CommandQueue::waitIdle(Context *context, uint64_t timeout)
{
return finishToSerial(context, mLastSubmittedQueueSerial, timeout);
}
Serial CommandQueue::reserveSubmitSerial()
{
Serial returnSerial = mCurrentQueueSerial;
mCurrentQueueSerial = mQueueSerialFactory.generate();
return returnSerial;
}
angle::Result CommandQueue::submitFrame(
Context *context,
bool hasProtectedContent,
egl::ContextPriority priority,
const std::vector<VkSemaphore> &waitSemaphores,
const std::vector<VkPipelineStageFlags> &waitSemaphoreStageMasks,
const Semaphore *signalSemaphore,
GarbageList &¤tGarbage,
SecondaryCommandBufferList &&commandBuffersToReset,
SecondaryCommandPools *commandPools,
Serial submitQueueSerial)
{
RendererVk *renderer = context->getRenderer();
VkDevice device = renderer->getDevice();
++mPerfCounters.commandQueueSubmitCallsTotal;
++mPerfCounters.commandQueueSubmitCallsPerFrame;
DeviceScoped<CommandBatch> scopedBatch(device);
CommandBatch &batch = scopedBatch.get();
batch.serial = submitQueueSerial;
batch.hasProtectedContent = hasProtectedContent;
batch.commandBuffersToReset = std::move(commandBuffersToReset);
// Don't make a submission if there is nothing to submit.
PrimaryCommandBuffer &commandBuffer = getCommandBuffer(hasProtectedContent);
const bool hasAnyPendingCommands = commandBuffer.valid();
if (hasAnyPendingCommands || signalSemaphore != nullptr || !waitSemaphores.empty())
{
if (commandBuffer.valid())
{
ANGLE_VK_TRY(context, commandBuffer.end());
}
VkSubmitInfo submitInfo = {};
InitializeSubmitInfo(&submitInfo, commandBuffer, waitSemaphores, waitSemaphoreStageMasks,
signalSemaphore);
VkProtectedSubmitInfo protectedSubmitInfo = {};
if (hasProtectedContent)
{
protectedSubmitInfo.sType = VK_STRUCTURE_TYPE_PROTECTED_SUBMIT_INFO;
protectedSubmitInfo.pNext = nullptr;
protectedSubmitInfo.protectedSubmit = true;
submitInfo.pNext = &protectedSubmitInfo;
}
ANGLE_TRACE_EVENT0("gpu.angle", "CommandQueue::submitFrame");
ANGLE_TRY(mFenceRecycler.newSharedFence(context, &batch.fence));
ANGLE_TRY(queueSubmit(context, priority, submitInfo, &batch.fence.get(), batch.serial));
}
else
{
mLastSubmittedQueueSerial = batch.serial;
}
if (!currentGarbage.empty())
{
mGarbageQueue.emplace(std::move(currentGarbage), batch.serial);
}
// Store the primary CommandBuffer and command pool used for secondary CommandBuffers
// in the in-flight list.
if (hasProtectedContent)
{
releaseToCommandBatch(hasProtectedContent, std::move(mProtectedPrimaryCommands),
commandPools, &batch);
}
else
{
releaseToCommandBatch(hasProtectedContent, std::move(mPrimaryCommands), commandPools,
&batch);
}
mInFlightCommands.emplace_back(scopedBatch.release());
ANGLE_TRY(checkCompletedCommands(context));
// CPU should be throttled to avoid mInFlightCommands from growing too fast. Important for
// off-screen scenarios.
if (mInFlightCommands.size() > kInFlightCommandsLimit)
{
size_t numCommandsToFinish = mInFlightCommands.size() - kInFlightCommandsLimit;
Serial finishSerial = mInFlightCommands[numCommandsToFinish].serial;
ANGLE_TRY(finishToSerial(context, finishSerial, renderer->getMaxFenceWaitTimeNs()));
}
// CPU should be throttled to avoid accumulating too much memory garbage waiting to be
// destroyed. This is important to keep peak memory usage at check when game launched and a lot
// of staging buffers used for textures upload and then gets released. But if there is only one
// command buffer in flight, we do not wait here to ensure we keep GPU busy.
VkDeviceSize suballocationGarbageSize = renderer->getSuballocationGarbageSize();
while (suballocationGarbageSize > kMaxBufferSuballocationGarbageSize &&
mInFlightCommands.size() > 1)
{
Serial finishSerial = mInFlightCommands.back().serial;
ANGLE_TRY(finishToSerial(context, finishSerial, renderer->getMaxFenceWaitTimeNs()));
suballocationGarbageSize = renderer->getSuballocationGarbageSize();
}
return angle::Result::Continue;
}
angle::Result CommandQueue::waitForSerialWithUserTimeout(vk::Context *context,
Serial serial,
uint64_t timeout,
VkResult *result)
{
Shared<Fence> *fenceToWaitOn = nullptr;
size_t finishCount = GetBatchCountUpToSerial(mInFlightCommands, serial, &fenceToWaitOn);
// The serial is already complete if:
//
// - There is no in-flight work (i.e. mInFlightCommands is empty), or
// - The given serial is smaller than the smallest serial, or
// - Every batch up to this serial is a garbage-clean-up-only batch (i.e. empty submission
// that's optimized out)
if (finishCount == 0 || fenceToWaitOn == nullptr)
{
*result = VK_SUCCESS;
return angle::Result::Continue;
}
const CommandBatch &batch = mInFlightCommands[finishCount - 1];
// Serial is not yet submitted. This is undefined behaviour, so we can do anything.
if (serial > batch.serial)
{
ASSERT(finishCount == mInFlightCommands.size());
WARN() << "Waiting on an unsubmitted serial.";
*result = VK_TIMEOUT;
return angle::Result::Continue;
}
ASSERT(serial == batch.serial);
*result = fenceToWaitOn->get().wait(context->getDevice(), timeout);
// Don't trigger an error on timeout.
if (*result != VK_TIMEOUT)
{
ANGLE_VK_TRY(context, *result);
}
return angle::Result::Continue;
}
angle::Result CommandQueue::ensurePrimaryCommandBufferValid(Context *context,
bool hasProtectedContent)
{
PersistentCommandPool &commandPool = getCommandPool(hasProtectedContent);
PrimaryCommandBuffer &commandBuffer = getCommandBuffer(hasProtectedContent);
if (commandBuffer.valid())
{
return angle::Result::Continue;
}
ANGLE_TRY(commandPool.allocate(context, &commandBuffer));
VkCommandBufferBeginInfo beginInfo = {};
beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
beginInfo.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
beginInfo.pInheritanceInfo = nullptr;
ANGLE_VK_TRY(context, commandBuffer.begin(beginInfo));
return angle::Result::Continue;
}
angle::Result CommandQueue::flushOutsideRPCommands(
Context *context,
bool hasProtectedContent,
OutsideRenderPassCommandBufferHelper **outsideRPCommands)
{
ANGLE_TRY(ensurePrimaryCommandBufferValid(context, hasProtectedContent));
PrimaryCommandBuffer &commandBuffer = getCommandBuffer(hasProtectedContent);
return (*outsideRPCommands)->flushToPrimary(context, &commandBuffer);
}
angle::Result CommandQueue::flushRenderPassCommands(
Context *context,
bool hasProtectedContent,
const RenderPass &renderPass,
RenderPassCommandBufferHelper **renderPassCommands)
{
ANGLE_TRY(ensurePrimaryCommandBufferValid(context, hasProtectedContent));
PrimaryCommandBuffer &commandBuffer = getCommandBuffer(hasProtectedContent);
return (*renderPassCommands)->flushToPrimary(context, &commandBuffer, &renderPass);
}
angle::Result CommandQueue::queueSubmitOneOff(Context *context,
bool hasProtectedContent,
egl::ContextPriority contextPriority,
VkCommandBuffer commandBufferHandle,
const Semaphore *waitSemaphore,
VkPipelineStageFlags waitSemaphoreStageMask,
const Fence *fence,
SubmitPolicy submitPolicy,
Serial submitQueueSerial)
{
VkSubmitInfo submitInfo = {};
submitInfo.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
VkProtectedSubmitInfo protectedSubmitInfo = {};
if (hasProtectedContent)
{
protectedSubmitInfo.sType = VK_STRUCTURE_TYPE_PROTECTED_SUBMIT_INFO;
protectedSubmitInfo.pNext = nullptr;
protectedSubmitInfo.protectedSubmit = true;
submitInfo.pNext = &protectedSubmitInfo;
}
if (commandBufferHandle != VK_NULL_HANDLE)
{
submitInfo.commandBufferCount = 1;
submitInfo.pCommandBuffers = &commandBufferHandle;
}
if (waitSemaphore != nullptr)
{
submitInfo.waitSemaphoreCount = 1;
submitInfo.pWaitSemaphores = waitSemaphore->ptr();
submitInfo.pWaitDstStageMask = &waitSemaphoreStageMask;
}
return queueSubmit(context, contextPriority, submitInfo, fence, submitQueueSerial);
}
angle::Result CommandQueue::queueSubmit(Context *context,
egl::ContextPriority contextPriority,
const VkSubmitInfo &submitInfo,
const Fence *fence,
Serial submitQueueSerial)
{
ANGLE_TRACE_EVENT0("gpu.angle", "CommandQueue::queueSubmit");
RendererVk *renderer = context->getRenderer();
if (kOutputVmaStatsString)
{
renderer->outputVmaStatString();
}
VkFence fenceHandle = fence ? fence->getHandle() : VK_NULL_HANDLE;
VkQueue queue = getQueue(contextPriority);
ANGLE_VK_TRY(context, vkQueueSubmit(queue, 1, &submitInfo, fenceHandle));
mLastSubmittedQueueSerial = submitQueueSerial;
++mPerfCounters.vkQueueSubmitCallsTotal;
++mPerfCounters.vkQueueSubmitCallsPerFrame;
return angle::Result::Continue;
}
void CommandQueue::resetPerFramePerfCounters()
{
mPerfCounters.commandQueueSubmitCallsPerFrame = 0;
mPerfCounters.vkQueueSubmitCallsPerFrame = 0;
}
VkResult CommandQueue::queuePresent(egl::ContextPriority contextPriority,
const VkPresentInfoKHR &presentInfo)
{
VkQueue queue = getQueue(contextPriority);
return vkQueuePresentKHR(queue, &presentInfo);
}
bool CommandQueue::isBusy() const
{
return mLastSubmittedQueueSerial > getLastCompletedQueueSerial();
}
// QueuePriorities:
constexpr float kVulkanQueuePriorityLow = 0.0;
constexpr float kVulkanQueuePriorityMedium = 0.4;
constexpr float kVulkanQueuePriorityHigh = 1.0;
const float QueueFamily::kQueuePriorities[static_cast<uint32_t>(egl::ContextPriority::EnumCount)] =
{kVulkanQueuePriorityMedium, kVulkanQueuePriorityHigh, kVulkanQueuePriorityLow};
egl::ContextPriority DeviceQueueMap::getDevicePriority(egl::ContextPriority priority) const
{
return mPriorities[priority];
}
DeviceQueueMap::~DeviceQueueMap() {}
DeviceQueueMap &DeviceQueueMap::operator=(const DeviceQueueMap &other)
{
ASSERT(this != &other);
if ((this != &other) && other.valid())
{
mIndex = other.mIndex;
mIsProtected = other.mIsProtected;
mPriorities[egl::ContextPriority::Low] = other.mPriorities[egl::ContextPriority::Low];
mPriorities[egl::ContextPriority::Medium] = other.mPriorities[egl::ContextPriority::Medium];
mPriorities[egl::ContextPriority::High] = other.mPriorities[egl::ContextPriority::High];
*static_cast<angle::PackedEnumMap<egl::ContextPriority, VkQueue> *>(this) = other;
}
return *this;
}
void QueueFamily::getDeviceQueue(VkDevice device,
bool makeProtected,
uint32_t queueIndex,
VkQueue *queue)
{
if (makeProtected)
{
VkDeviceQueueInfo2 queueInfo2 = {};
queueInfo2.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2;
queueInfo2.flags = VK_DEVICE_QUEUE_CREATE_PROTECTED_BIT;
queueInfo2.queueFamilyIndex = mIndex;
queueInfo2.queueIndex = queueIndex;
vkGetDeviceQueue2(device, &queueInfo2, queue);
}
else
{
vkGetDeviceQueue(device, mIndex, queueIndex, queue);
}
}
DeviceQueueMap QueueFamily::initializeQueueMap(VkDevice device,
bool makeProtected,
uint32_t queueIndex,
uint32_t queueCount)
{
// QueueIndexing:
constexpr uint32_t kQueueIndexMedium = 0;
constexpr uint32_t kQueueIndexHigh = 1;
constexpr uint32_t kQueueIndexLow = 2;
ASSERT(queueCount);
ASSERT((queueIndex + queueCount) <= mProperties.queueCount);
DeviceQueueMap queueMap(mIndex, makeProtected);
getDeviceQueue(device, makeProtected, queueIndex + kQueueIndexMedium,
&queueMap[egl::ContextPriority::Medium]);
queueMap.mPriorities[egl::ContextPriority::Medium] = egl::ContextPriority::Medium;
// If at least 2 queues, High has its own queue
if (queueCount > 1)
{
getDeviceQueue(device, makeProtected, queueIndex + kQueueIndexHigh,
&queueMap[egl::ContextPriority::High]);
queueMap.mPriorities[egl::ContextPriority::High] = egl::ContextPriority::High;
}
else
{
queueMap[egl::ContextPriority::High] = queueMap[egl::ContextPriority::Medium];
queueMap.mPriorities[egl::ContextPriority::High] = egl::ContextPriority::Medium;
}
// If at least 3 queues, Low has its own queue. Adjust Low priority.
if (queueCount > 2)
{
getDeviceQueue(device, makeProtected, queueIndex + kQueueIndexLow,
&queueMap[egl::ContextPriority::Low]);
queueMap.mPriorities[egl::ContextPriority::Low] = egl::ContextPriority::Low;
}
else
{
queueMap[egl::ContextPriority::Low] = queueMap[egl::ContextPriority::Medium];
queueMap.mPriorities[egl::ContextPriority::Low] = egl::ContextPriority::Medium;
}
return queueMap;
}
void QueueFamily::initialize(const VkQueueFamilyProperties &queueFamilyProperties, uint32_t index)
{
mProperties = queueFamilyProperties;
mIndex = index;
}
uint32_t QueueFamily::FindIndex(const std::vector<VkQueueFamilyProperties> &queueFamilyProperties,
VkQueueFlags flags,
int32_t matchNumber,
uint32_t *matchCount)
{
uint32_t index = QueueFamily::kInvalidIndex;
uint32_t count = 0;
for (uint32_t familyIndex = 0; familyIndex < queueFamilyProperties.size(); ++familyIndex)
{
const auto &queueInfo = queueFamilyProperties[familyIndex];
if ((queueInfo.queueFlags & flags) == flags)
{
ASSERT(queueInfo.queueCount > 0);
count++;
if ((index == QueueFamily::kInvalidIndex) && (matchNumber-- == 0))
{
index = familyIndex;
}
}
}
if (matchCount)
{
*matchCount = count;
}
return index;
}
// ScopedCommandQueueLock implementation
ScopedCommandQueueLock::~ScopedCommandQueueLock()
{
// Before unlocking the mutex, see if device loss has occured, and if so handle it.
if (mRenderer->isDeviceLost())
{
mRenderer->handleDeviceLostNoLock();
}
mLock.unlock();
}
} // namespace vk
} // namespace rx