intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-29 09:03:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Revert "performance: Signal inOrder counter with pipe control, part 4"

Browse Source 
This reverts commit 332002d9ea.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
This commit is contained in:
Compute-Runtime-Validation
2024-12-28 01:32:41 +01:00
committed by Compute-Runtime-Automation
parent 8cd7ce1b66
commit c27921767b
3 changed files with 37 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
level_zero/core/source/cmdlist/cmdlist_hw_xehp_and_later.inl
View File

@@ -304,9 +304,9 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
if (!eventForInOrderExec->getAllocation(this->device) && Event::standaloneInOrderTimestampAllocationEnabled()) {
eventForInOrderExec->resetInOrderTimestampNode(device->getInOrderTimestampAllocator()->getTag());
}
if ((!compactEvent && this->heaplessModeEnabled) || this->asMutable() || !eventForInOrderExec->isCounterBased()) {
if (!compactEvent || this->asMutable() || !compactEvent->isCounterBased()) {
dispatchEventPostSyncOperation(eventForInOrderExec, nullptr, launchParams.outListCommands, Event::STATE_CLEARED, false, false, false, false, false);
} else if (compactEvent) {
} else {
eventAddress = eventForInOrderExec->getPacketAddress(this->device);
isTimestampEvent = true;
if (!launchParams.omitAddingEventResidency) {

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

@@ -2852,8 +2852,9 @@ HWTEST2_F(CommandListAppendLaunchKernel,
auto eventCompletionAddress = event->getCompletionFieldGpuAddress(device);
ASSERT_EQ(heapless ? 0u : 1u, outCbEventCmds.size());
size_t expectedSdi = heapless ? 0 : 1;
ASSERT_EQ(heapless ? 0u : 2u, outCbEventCmds.size());
size_t expectedSdi = heapless ? 0 : commandList->inOrderAtomicSignalingEnabled ? 1
: 2;
auto storeDataImmList = findAll<MI_STORE_DATA_IMM *>(cmdList.begin(), cmdList.end());
ASSERT_EQ(expectedSdi, storeDataImmList.size());
@@ -2861,6 +2862,14 @@ HWTEST2_F(CommandListAppendLaunchKernel,
ASSERT_EQ(1u, computeWalkerList.size());
auto semaphoreWaitList = findAll<MI_SEMAPHORE_WAIT *>(cmdList.begin(), cmdList.end());
ASSERT_EQ(heapless ? 0u : 1u, semaphoreWaitList.size());
if (!heapless) {
EXPECT_EQ(CommandToPatch::CbEventTimestampClearStoreDataImm, outCbEventCmds[0].type);
EXPECT_EQ(*storeDataImmList[0], outCbEventCmds[0].pDestination);
auto storeDataImmCmd = genCmdCast<MI_STORE_DATA_IMM *>(outCbEventCmds[0].pDestination);
ASSERT_NE(nullptr, storeDataImmCmd);
EXPECT_EQ(eventCompletionAddress, storeDataImmCmd->getAddress());
}
EXPECT_EQ(launchParams.outWalker, *computeWalkerList[0]);
ASSERT_NE(nullptr, launchParams.outWalker);
@@ -2879,9 +2888,9 @@ HWTEST2_F(CommandListAppendLaunchKernel,
if (!heapless) {
EXPECT_EQ(CommandToPatch::CbEventTimestampPostSyncSemaphoreWait, outCbEventCmds[0].type);
EXPECT_EQ(*semaphoreWaitList[0], outCbEventCmds[0].pDestination);
auto semaphoreWaitCmd = genCmdCast<MI_SEMAPHORE_WAIT *>(outCbEventCmds[0].pDestination);
EXPECT_EQ(CommandToPatch::CbEventTimestampPostSyncSemaphoreWait, outCbEventCmds[1].type);
EXPECT_EQ(*semaphoreWaitList[0], outCbEventCmds[1].pDestination);
auto semaphoreWaitCmd = genCmdCast<MI_SEMAPHORE_WAIT *>(outCbEventCmds[1].pDestination);
ASSERT_NE(nullptr, semaphoreWaitCmd);
EXPECT_EQ(eventCompletionAddress, semaphoreWaitCmd->getSemaphoreGraphicsAddress());
}

46
level_zero/core/test/unit_tests/sources/cmdlist/test_in_order_cmdlist_1.cpp
View File

@@ -2574,26 +2574,24 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenProgrammingTimestampEventThen
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::Parse::parseCommandBuffer(cmdList, cmdStream->getCpuBase(), cmdStream->getUsed()));
if (immCmdList->isHeaplessModeEnabled()) {
auto sdiItor = find<MI_STORE_DATA_IMM *>(cmdList.begin(), cmdList.end());
ASSERT_NE(cmdList.end(), sdiItor);
auto sdiItor = find<MI_STORE_DATA_IMM *>(cmdList.begin(), cmdList.end());
ASSERT_NE(cmdList.end(), sdiItor);
auto sdiCmd = genCmdCast<MI_STORE_DATA_IMM *>(*sdiItor);
ASSERT_NE(nullptr, sdiCmd);
auto sdiCmd = genCmdCast<MI_STORE_DATA_IMM *>(*sdiItor);
ASSERT_NE(nullptr, sdiCmd);
EXPECT_EQ(events[0]->getCompletionFieldGpuAddress(device), sdiCmd->getAddress());
EXPECT_EQ(0u, sdiCmd->getStoreQword());
EXPECT_EQ(Event::STATE_CLEARED, sdiCmd->getDataDword0());
}
EXPECT_EQ(events[0]->getCompletionFieldGpuAddress(device), sdiCmd->getAddress());
EXPECT_EQ(0u, sdiCmd->getStoreQword());
EXPECT_EQ(Event::STATE_CLEARED, sdiCmd->getDataDword0());
auto eventBaseGpuVa = events[0]->getPacketAddress(device);
auto eventEndGpuVa = events[0]->getCompletionFieldGpuAddress(device);
auto walkerItor = NEO::UnitTestHelper<FamilyType>::findWalkerTypeCmd(cmdList.begin(), cmdList.end());
auto walkerItor = NEO::UnitTestHelper<FamilyType>::findWalkerTypeCmd(sdiItor, cmdList.end());
ASSERT_NE(cmdList.end(), walkerItor);
WalkerVariant walkerVariant = NEO::UnitTestHelper<FamilyType>::getWalkerVariant(*walkerItor);
std::visit([eventBaseGpuVa, eventEndGpuVa, &immCmdList](auto &&walker) {
std::visit([eventBaseGpuVa, eventEndGpuVa, &immCmdList, &sdiCmd](auto &&walker) {
auto &postSync = walker->getPostSync();
using PostSyncType = std::decay_t<decltype(postSync)>;
@@ -2606,7 +2604,7 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenProgrammingTimestampEventThen
EXPECT_EQ(eventEndGpuVa, semaphoreCmd->getSemaphoreGraphicsAddress());
EXPECT_EQ(MI_SEMAPHORE_WAIT::COMPARE_OPERATION::COMPARE_OPERATION_SAD_NOT_EQUAL_SDD, semaphoreCmd->getCompareOperation());
auto sdiCmd = genCmdCast<MI_STORE_DATA_IMM *>(++semaphoreCmd);
sdiCmd = genCmdCast<MI_STORE_DATA_IMM *>(++semaphoreCmd);
ASSERT_NE(nullptr, sdiCmd);
EXPECT_EQ(immCmdList->inOrderExecInfo->getBaseDeviceAddress(), sdiCmd->getAddress());
@@ -2720,23 +2718,21 @@ HWTEST2_F(InOrderCmdListTests, givenRelaxedOrderingWhenProgrammingTimestampEvent
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::Parse::parseCommandBuffer(cmdList, cmdStream->getCpuBase(), immCmdList->flushData[1]));
if (immCmdList->isHeaplessModeEnabled()) {
auto sdiItor = find<MI_STORE_DATA_IMM *>(cmdList.begin(), cmdList.end());
ASSERT_NE(cmdList.end(), sdiItor);
auto sdiItor = find<MI_STORE_DATA_IMM *>(cmdList.begin(), cmdList.end());
ASSERT_NE(cmdList.end(), sdiItor);
auto sdiCmd = genCmdCast<MI_STORE_DATA_IMM *>(*sdiItor);
ASSERT_NE(nullptr, sdiCmd);
auto sdiCmd = genCmdCast<MI_STORE_DATA_IMM *>(*sdiItor);
ASSERT_NE(nullptr, sdiCmd);
EXPECT_EQ(events[0]->getCompletionFieldGpuAddress(device), sdiCmd->getAddress());
EXPECT_EQ(0u, sdiCmd->getStoreQword());
EXPECT_EQ(Event::STATE_CLEARED, sdiCmd->getDataDword0());
EXPECT_EQ(events[0]->getCompletionFieldGpuAddress(device), sdiCmd->getAddress());
EXPECT_EQ(0u, sdiCmd->getStoreQword());
EXPECT_EQ(Event::STATE_CLEARED, sdiCmd->getDataDword0());
auto sdiOffset = ptrDiff(sdiCmd, cmdStream->getCpuBase());
EXPECT_TRUE(sdiOffset >= immCmdList->flushData[0]);
EXPECT_TRUE(sdiOffset < immCmdList->flushData[1]);
}
auto sdiOffset = ptrDiff(sdiCmd, cmdStream->getCpuBase());
EXPECT_TRUE(sdiOffset >= immCmdList->flushData[0]);
EXPECT_TRUE(sdiOffset < immCmdList->flushData[1]);
auto walkerItor = NEO::UnitTestHelper<FamilyType>::findWalkerTypeCmd(cmdList.begin(), cmdList.end());
auto walkerItor = NEO::UnitTestHelper<FamilyType>::findWalkerTypeCmd(sdiItor, cmdList.end());
ASSERT_NE(cmdList.end(), walkerItor);
auto eventBaseGpuVa = events[0]->getPacketAddress(device);