intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-04 07:14:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: waiting for in-order dependencies with TimestampPacket

Browse Source 
Related-To: NEO-7966

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2023-05-24 10:16:21 +00:00
committed by Compute-Runtime-Automation
parent 223eb3e3fc
commit f98ac7098b
6 changed files with 47 additions and 111 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_1.cpp
View File

@@ -1315,6 +1315,10 @@ HWTEST2_F(CommandListCreate, givenInOrderExecutionWhenDispatchingRelaxedOrdering
size_t offset = cmdStream->getUsed();
ASSERT_EQ(1u, cmdList->timestampPacketContainer->peekNodes().size());
auto previousNode = cmdList->timestampPacketContainer->peekNodes()[0];
cmdList->appendLaunchKernel(kernel.toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
GenCmdList genCmdList;
@@ -1329,7 +1333,9 @@ HWTEST2_F(CommandListCreate, givenInOrderExecutionWhenDispatchingRelaxedOrdering
lrrCmd++;
lrrCmd++;
EXPECT_TRUE(RelaxedOrderingCommandsHelper::verifyConditionalDataMemBbStart<FamilyType>(lrrCmd, 0, cmdList->inOrderDependencyCounterAllocation->getGpuAddress(), 2, NEO::CompareOperation::Less, true));
auto compareAddress = NEO::TimestampPacketHelper::getContextEndGpuAddress(*previousNode);
EXPECT_TRUE(RelaxedOrderingCommandsHelper::verifyConditionalDataMemBbStart<FamilyType>(lrrCmd, 0, compareAddress, 1, NEO::CompareOperation::Equal, true));
}
TEST_F(CommandListCreate, GivenGpuHangWhenCreatingImmCmdListWithSyncModeAndAppendBarrierThenAppendBarrierReturnsDeviceLost) {

110
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp
View File

@@ -734,7 +734,7 @@ struct InOrderCmdListTests : public CommandListAppendLaunchKernel {
std::vector<std::unique_ptr<MockEvent>> events;
};
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenResetEventCalledThenResetEventState, IsAtLeastSkl) {
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenResetEventCalledThenResetEventState, IsAtLeastXeHpCore) {
auto immCmdList = createImmCmdList<gfxCoreFamily>();
auto eventPool = createEvents(3, false);
@@ -763,6 +763,8 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenSubmittingThenProgramSemaphor
auto offset = cmdStream->getUsed();
auto previousNode = immCmdList->timestampPacketContainer->peekNodes()[0];
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
GenCmdList cmdList;
@@ -777,9 +779,11 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenSubmittingThenProgramSemaphor
auto semaphoreCmd = genCmdCast<MI_SEMAPHORE_WAIT *>(*itor);
auto compareAddress = NEO::TimestampPacketHelper::getContextEndGpuAddress(*previousNode);
EXPECT_EQ(1u, semaphoreCmd->getSemaphoreDataDword());
EXPECT_EQ(immCmdList->inOrderDependencyCounterAllocation->getGpuAddress(), semaphoreCmd->getSemaphoreGraphicsAddress());
EXPECT_EQ(MI_SEMAPHORE_WAIT::COMPARE_OPERATION::COMPARE_OPERATION_SAD_GREATER_THAN_OR_EQUAL_SDD, semaphoreCmd->getCompareOperation());
EXPECT_EQ(compareAddress, semaphoreCmd->getSemaphoreGraphicsAddress());
EXPECT_EQ(MI_SEMAPHORE_WAIT::COMPARE_OPERATION::COMPARE_OPERATION_SAD_NOT_EQUAL_SDD, semaphoreCmd->getCompareOperation());
}
HWTEST2_F(InOrderCmdListTests, givenInOrderEventModeWhenSubmittingThenProgramSemaphoreForEvent, IsAtLeastXeHpCore) {
@@ -817,86 +821,18 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderEventModeWhenSubmittingThenProgramSem
auto semaphoreCmd = genCmdCast<MI_SEMAPHORE_WAIT *>(*itor);
EXPECT_EQ(2u, semaphoreCmd->getSemaphoreDataDword());
EXPECT_EQ(immCmdList->inOrderDependencyCounterAllocation->getGpuAddress(), semaphoreCmd->getSemaphoreGraphicsAddress());
EXPECT_EQ(MI_SEMAPHORE_WAIT::COMPARE_OPERATION::COMPARE_OPERATION_SAD_GREATER_THAN_OR_EQUAL_SDD, semaphoreCmd->getCompareOperation());
auto gpuAddr = events[0]->getCompletionFieldGpuAddress(this->device);
EXPECT_EQ(static_cast<uint32_t>(Event::State::STATE_CLEARED), semaphoreCmd->getSemaphoreDataDword());
EXPECT_EQ(gpuAddr, semaphoreCmd->getSemaphoreGraphicsAddress());
EXPECT_EQ(MI_SEMAPHORE_WAIT::COMPARE_OPERATION::COMPARE_OPERATION_SAD_NOT_EQUAL_SDD, semaphoreCmd->getCompareOperation());
}
HWTEST2_F(InOrderCmdListTests, givenInOrderEventModeWhenWaitingForEventFromPreviousAppendThenSkip, IsAtLeastXeHpCore) {
using MI_SEMAPHORE_WAIT = typename FamilyType::MI_SEMAPHORE_WAIT;
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenDispatchingThenHandleTimestampPacketResidency, IsAtLeastXeHpCore) {
DebugManager.flags.DisableTimestampPacketOptimizations.set(1); // Create new allocation for node each time, to test residency handling
auto immCmdList = createImmCmdList<gfxCoreFamily>();
auto eventPool = createEvents(1, false);
auto cmdStream = immCmdList->getCmdContainer().getCommandStream();
auto event0Handle = events[0]->toHandle();
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, event0Handle, 0, nullptr, launchParams, false);
auto offset = cmdStream->getUsed();
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 1, &event0Handle, launchParams, false);
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
cmdList,
ptrOffset(cmdStream->getCpuBase(), offset),
cmdStream->getUsed() - offset));
auto itor = find<typename FamilyType::MI_SEMAPHORE_WAIT *>(cmdList.begin(), cmdList.end());
ASSERT_NE(cmdList.end(), itor);
itor = find<typename FamilyType::MI_SEMAPHORE_WAIT *>(++itor, cmdList.end());
EXPECT_EQ(cmdList.end(), itor);
}
HWTEST2_F(InOrderCmdListTests, givenInOrderEventModeWhenSubmittingFromDifferentCmdListThenProgramSemaphoreForEvent, IsAtLeastSkl) {
using MI_SEMAPHORE_WAIT = typename FamilyType::MI_SEMAPHORE_WAIT;
auto immCmdList1 = createImmCmdList<gfxCoreFamily>();
auto immCmdList2 = createImmCmdList<gfxCoreFamily>();
auto eventPool = createEvents(1, false);
auto cmdStream = immCmdList2->getCmdContainer().getCommandStream();
auto event0Handle = events[0]->toHandle();
auto ultCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(device->getNEODevice()->getDefaultEngine().commandStreamReceiver);
ultCsr->storeMakeResidentAllocations = true;
immCmdList1->appendLaunchKernel(kernel->toHandle(), &groupCount, event0Handle, 0, nullptr, launchParams, false);
EXPECT_EQ(1u, ultCsr->makeResidentAllocations[immCmdList1->inOrderDependencyCounterAllocation]);
immCmdList2->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 1, &event0Handle, launchParams, false);
EXPECT_EQ(2u, ultCsr->makeResidentAllocations[immCmdList1->inOrderDependencyCounterAllocation]);
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(cmdList, cmdStream->getCpuBase(), cmdStream->getUsed()));
auto itor = find<typename FamilyType::MI_SEMAPHORE_WAIT *>(cmdList.begin(), cmdList.end());
ASSERT_NE(cmdList.end(), itor);
auto semaphoreCmd = genCmdCast<MI_SEMAPHORE_WAIT *>(*itor);
EXPECT_EQ(1u, semaphoreCmd->getSemaphoreDataDword());
EXPECT_NE(immCmdList1->inOrderDependencyCounterAllocation->getGpuAddress(), immCmdList2->inOrderDependencyCounterAllocation->getGpuAddress());
EXPECT_EQ(immCmdList1->inOrderDependencyCounterAllocation->getGpuAddress(), semaphoreCmd->getSemaphoreGraphicsAddress());
EXPECT_EQ(MI_SEMAPHORE_WAIT::COMPARE_OPERATION::COMPARE_OPERATION_SAD_GREATER_THAN_OR_EQUAL_SDD, semaphoreCmd->getCompareOperation());
}
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenDispatchingThenHandleDependencyCounter, IsAtLeastSkl) {
auto immCmdList = createImmCmdList<gfxCoreFamily>();
EXPECT_NE(nullptr, immCmdList->inOrderDependencyCounterAllocation);
EXPECT_EQ(AllocationType::TIMESTAMP_PACKET_TAG_BUFFER, immCmdList->inOrderDependencyCounterAllocation->getAllocationType());
@@ -906,12 +842,22 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenDispatchingThenHandleDependen
ultCsr->storeMakeResidentAllocations = true;
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
EXPECT_EQ(1u, immCmdList->inOrderDependencyCounter);
EXPECT_EQ(1u, ultCsr->makeResidentAllocations[immCmdList->inOrderDependencyCounterAllocation]);
auto node0 = immCmdList->timestampPacketContainer->peekNodes()[0];
ultCsr->getTimestampPacketAllocator()->getTag();
EXPECT_EQ(1u, ultCsr->makeResidentAllocations[node0->getBaseGraphicsAllocation()->getGraphicsAllocation(0)]);
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
EXPECT_EQ(2u, immCmdList->inOrderDependencyCounter);
EXPECT_EQ(2u, ultCsr->makeResidentAllocations[immCmdList->inOrderDependencyCounterAllocation]);
auto node1 = immCmdList->timestampPacketContainer->peekNodes()[0];
ultCsr->getTimestampPacketAllocator()->getTag();
EXPECT_EQ(2u, ultCsr->makeResidentAllocations[node0->getBaseGraphicsAllocation()->getGraphicsAllocation(0)]);
EXPECT_EQ(1u, ultCsr->makeResidentAllocations[node1->getBaseGraphicsAllocation()->getGraphicsAllocation(0)]);
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
auto node2 = immCmdList->timestampPacketContainer->peekNodes()[0];
ultCsr->getTimestampPacketAllocator()->getTag();
EXPECT_EQ(2u, ultCsr->makeResidentAllocations[node0->getBaseGraphicsAllocation()->getGraphicsAllocation(0)]); // not used anymore
EXPECT_EQ(2u, ultCsr->makeResidentAllocations[node1->getBaseGraphicsAllocation()->getGraphicsAllocation(0)]);
EXPECT_EQ(1u, ultCsr->makeResidentAllocations[node2->getBaseGraphicsAllocation()->getGraphicsAllocation(0)]);
}
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenAddingRelaxedOrderingEventsThenConfigureRegistersFirst, IsAtLeastXeHpCore) {