intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-12 09:30:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: use proper pointer inside appendWriteGlobalTimestamp

Browse Source 
Resolves: NEO-8105
Signed-off-by: Katarzyna Cencelewska <katarzyna.cencelewska@intel.com>
This commit is contained in:
Katarzyna Cencelewska
2023-09-27 00:10:51 +00:00
committed by Compute-Runtime-Automation
parent 96a7f63371
commit e71db368db
8 changed files with 106 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -2539,12 +2539,18 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWriteGlobalTimestamp(
appendEventForProfiling(signalEvent, true, false);
auto allocationStruct = getAlignedAllocationData(this->device, dstptr, sizeof(uint64_t), false);
if (allocationStruct.alloc == nullptr) {
return ZE_RESULT_ERROR_OUT_OF_DEVICE_MEMORY;
}
commandContainer.addToResidencyContainer(allocationStruct.alloc);
if (isCopyOnly()) {
NEO::MiFlushArgs args{this->dummyBlitWa};
args.timeStampOperation = true;
args.commandWithPostSync = true;
NEO::EncodeMiFlushDW<GfxFamily>::programWithWa(*commandContainer.getCommandStream(),
reinterpret_cast<uint64_t>(dstptr),
allocationStruct.alignedAllocationPtr,
0,
args);
makeResidentDummyAllocation();
@@ -2555,7 +2561,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWriteGlobalTimestamp(
NEO::MemorySynchronizationCommands<GfxFamily>::addBarrierWithPostSyncOperation(
*commandContainer.getCommandStream(),
NEO::PostSyncMode::Timestamp,
reinterpret_cast<uint64_t>(dstptr),
allocationStruct.alignedAllocationPtr,
0,
this->device->getNEODevice()->getRootDeviceEnvironment(),
args);
@@ -2568,12 +2574,6 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWriteGlobalTimestamp(
handleInOrderDependencyCounter();
}
auto allocationStruct = getAlignedAllocationData(this->device, dstptr, sizeof(uint64_t), false);
if (allocationStruct.alloc == nullptr) {
return ZE_RESULT_ERROR_OUT_OF_DEVICE_MEMORY;
}
commandContainer.addToResidencyContainer(allocationStruct.alloc);
addToMappedEventList(signalEvent);
return ZE_RESULT_SUCCESS;

50
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_4.cpp
View File

@@ -39,13 +39,21 @@ HWTEST2_F(CommandListCreate, givenCopyOnlyCommandListWhenAppendWriteGlobalTimest
ze_result_t returnValue;
std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::Copy, 0u, returnValue));
auto &commandContainer = commandList->getCmdContainer();
auto memoryManager = static_cast<MockMemoryManager *>(neoDevice->getMemoryManager());
memoryManager->returnFakeAllocation = true;
uint64_t timestampAddress = 0xfffffffffff0L;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(timestampAddress);
uint64_t dstAddress = 0xfffffffffff0L;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(dstAddress);
commandContainer.getResidencyContainer().clear();
const auto commandStreamOffset = commandContainer.getCommandStream()->getUsed();
commandList->appendWriteGlobalTimestamp(dstptr, nullptr, 0, nullptr);
auto residencyContainer = commandContainer.getResidencyContainer();
auto timestampAlloc = residencyContainer[0];
EXPECT_EQ(dstAddress, reinterpret_cast<uint64_t>(timestampAlloc->getUnderlyingBuffer()));
auto timestampAddress = timestampAlloc->getGpuAddress();
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
cmdList,
@@ -57,9 +65,9 @@ HWTEST2_F(CommandListCreate, givenCopyOnlyCommandListWhenAppendWriteGlobalTimest
ASSERT_NE(0u, iterator.size());
for (auto it : iterator) {
auto cmd = genCmdCast<MI_FLUSH_DW *>(*it);
auto postSyncDestAddress = cmd->getDestinationAddress();
if ((cmd->getPostSyncOperation() == MI_FLUSH_DW::POST_SYNC_OPERATION_WRITE_TIMESTAMP_REGISTER) &&
(cmd->getDestinationAddress() == timestampAddress)) {
(postSyncDestAddress == timestampAddress)) {
postSyncFound = true;
}
}
@@ -100,6 +108,40 @@ HWTEST2_F(CommandListCreate, givenCommandListWhenAppendWriteGlobalTimestampCalle
EXPECT_TRUE(foundTimestampPipeControl);
}
HWTEST2_F(CommandListCreate, givenMovedDstptrWhenAppendWriteGlobalTimestampCalledThenPipeControlWithProperTimestampEncoded, IsAtLeastSkl) {
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
using POST_SYNC_OPERATION = typename PIPE_CONTROL::POST_SYNC_OPERATION;
ze_result_t returnValue;
std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, returnValue));
auto &commandContainer = commandList->getCmdContainer();
uint64_t timestampAddress = 0x123456785500;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(timestampAddress) + 1;
uint64_t expectedTimestampAddress = timestampAddress + sizeof(uint64_t);
const auto commandStreamOffset = commandContainer.getCommandStream()->getUsed();
commandList->appendWriteGlobalTimestamp(dstptr, nullptr, 0, nullptr);
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
cmdList,
ptrOffset(commandContainer.getCommandStream()->getCpuBase(), commandStreamOffset),
commandContainer.getCommandStream()->getUsed() - commandStreamOffset));
auto pcList = findAll<PIPE_CONTROL *>(cmdList.begin(), cmdList.end());
ASSERT_NE(0u, pcList.size());
bool foundTimestampPipeControl = false;
for (auto it : pcList) {
auto cmd = genCmdCast<PIPE_CONTROL *>(*it);
if (cmd->getPostSyncOperation() == POST_SYNC_OPERATION::POST_SYNC_OPERATION_WRITE_TIMESTAMP) {
EXPECT_TRUE(cmd->getCommandStreamerStallEnable());
EXPECT_FALSE(cmd->getDcFlushEnable());
EXPECT_EQ(expectedTimestampAddress, NEO::UnitTestHelper<FamilyType>::getPipeControlPostSyncAddress(*cmd));
foundTimestampPipeControl = true;
}
}
EXPECT_TRUE(foundTimestampPipeControl);
}
HWTEST2_F(CommandListCreate, givenCommandListWhenAppendWriteGlobalTimestampCalledThenTimestampAllocationIsInsideResidencyContainer, IsAtLeastSkl) {
ze_result_t returnValue;
std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, NEO::EngineGroupType::RenderCompute, 0u, returnValue));

24
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_signal_event.cpp
View File

@@ -109,8 +109,10 @@ HWTEST2_F(CommandListAppendSignalEvent, givenCommandListWhenAppendWriteGlobalTim
using POST_SYNC_OPERATION = typename PIPE_CONTROL::POST_SYNC_OPERATION;
auto &commandContainer = commandList->getCmdContainer();
uint64_t timestampAddress = 0x12345678555500;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(timestampAddress);
uint64_t dstAddress = 0x12345678555500;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(dstAddress);
commandContainer.getResidencyContainer().clear();
commandList->appendWriteGlobalTimestamp(dstptr, event->toHandle(), 0, nullptr);
@@ -118,6 +120,9 @@ HWTEST2_F(CommandListAppendSignalEvent, givenCommandListWhenAppendWriteGlobalTim
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
cmdList, ptrOffset(commandContainer.getCommandStream()->getCpuBase(), 0), commandContainer.getCommandStream()->getUsed()));
auto residencyContainer = commandContainer.getResidencyContainer();
auto timestampAlloc = residencyContainer[0];
auto itorPC = find<PIPE_CONTROL *>(cmdList.begin(), cmdList.end());
EXPECT_NE(cmdList.end(), itorPC);
auto cmd = genCmdCast<PIPE_CONTROL *>(*itorPC);
@@ -129,6 +134,8 @@ HWTEST2_F(CommandListAppendSignalEvent, givenCommandListWhenAppendWriteGlobalTim
}
EXPECT_TRUE(cmd->getCommandStreamerStallEnable());
EXPECT_FALSE(cmd->getDcFlushEnable());
EXPECT_EQ(dstAddress, reinterpret_cast<uint64_t>(timestampAlloc->getUnderlyingBuffer()));
auto timestampAddress = timestampAlloc->getGpuAddress();
EXPECT_EQ(timestampAddress, NEO::UnitTestHelper<FamilyType>::getPipeControlPostSyncAddress(*cmd));
itorPC++;
@@ -482,8 +489,12 @@ HWTEST2_F(CommandListAppendUsedPacketSignalEvent,
using POST_SYNC_OPERATION = typename PIPE_CONTROL::POST_SYNC_OPERATION;
auto &commandContainer = commandList->getCmdContainer();
uint64_t timestampAddress = 0x12345678555500;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(timestampAddress);
auto memoryManager = static_cast<MockMemoryManager *>(neoDevice->getMemoryManager());
memoryManager->returnFakeAllocation = true;
uint64_t dstAddress = 0x123456785500;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(dstAddress);
commandContainer.getResidencyContainer().clear();
constexpr uint32_t packets = 2u;
@@ -493,6 +504,11 @@ HWTEST2_F(CommandListAppendUsedPacketSignalEvent,
commandList->appendWriteGlobalTimestamp(dstptr, event->toHandle(), 0, nullptr);
EXPECT_EQ(packets, event->getPacketsInUse());
auto residencyContainer = commandContainer.getResidencyContainer();
auto timestampAlloc = residencyContainer[1];
EXPECT_EQ(dstAddress, reinterpret_cast<uint64_t>(timestampAlloc->getUnderlyingBuffer()));
auto timestampAddress = timestampAlloc->getGpuAddress();
auto eventGpuAddress = event->getGpuAddress(device);
uint64_t contextStartAddress = eventGpuAddress + event->getContextStartOffset();
uint64_t globalStartAddress = eventGpuAddress + event->getGlobalStartOffset();

19
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_wait_on_events.cpp
View File

@@ -479,16 +479,27 @@ HWTEST2_F(CommandListAppendWaitOnEvent, givenCommandListWhenAppendWriteGlobalTim
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
using POST_SYNC_OPERATION = typename PIPE_CONTROL::POST_SYNC_OPERATION;
uint64_t timestampAddress = 0x12345678555500;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(timestampAddress);
auto memoryManager = static_cast<MockMemoryManager *>(neoDevice->getMemoryManager());
memoryManager->returnFakeAllocation = true;
uint64_t dstAddress = 0x123456785500;
uint64_t *dstptr = reinterpret_cast<uint64_t *>(dstAddress);
auto &commandContainer = commandList->getCmdContainer();
commandContainer.getResidencyContainer().clear();
ze_event_handle_t hEventHandle = event->toHandle();
commandList->appendWriteGlobalTimestamp(dstptr, nullptr, 1, &hEventHandle);
auto residencyContainer = commandContainer.getResidencyContainer();
auto timestampAlloc = residencyContainer[1];
EXPECT_EQ(dstAddress, reinterpret_cast<uint64_t>(timestampAlloc->getUnderlyingBuffer()));
auto timestampAddress = timestampAlloc->getGpuAddress();
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
cmdList, ptrOffset(commandList->getCmdContainer().getCommandStream()->getCpuBase(), 0),
commandList->getCmdContainer().getCommandStream()->getUsed()));
cmdList, ptrOffset(commandContainer.getCommandStream()->getCpuBase(), 0),
commandContainer.getCommandStream()->getUsed()));
auto itor = find<MI_SEMAPHORE_WAIT *>(cmdList.begin(), cmdList.end());
ASSERT_NE(cmdList.end(), itor);