intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: in-order flush is host visible if dc flush not required

Browse Source 
Related-To: NEO-7966

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2023-10-31 14:00:02 +00:00
committed by Compute-Runtime-Automation
parent 330fb40107
commit 130fbced5b
3 changed files with 41 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
level_zero/core/test/unit_tests/mocks/mock_cmdlist.h
View File

@@ -61,6 +61,7 @@ struct WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>
using BaseClass::currentDynamicStateBaseAddress;
using BaseClass::currentIndirectObjectBaseAddress;
using BaseClass::currentSurfaceStateBaseAddress;
using BaseClass::dcFlushSupport;
using BaseClass::device;
using BaseClass::dispatchCmdListBatchBufferAsPrimary;
using BaseClass::doubleSbaWa;
@@ -168,6 +169,7 @@ struct WhiteBox<L0::CommandListCoreFamilyImmediate<gfxCoreFamily>>
using BaseClass::compactL3FlushEvent;
using BaseClass::compactL3FlushEventPacket;
using BaseClass::csr;
using BaseClass::dcFlushSupport;
using BaseClass::device;
using BaseClass::doubleSbaWa;
using BaseClass::engineGroupType;

44
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp
View File

@@ -878,9 +878,11 @@ HWTEST2_F(InOrderCmdListTests, givenCmdListsWhenDispatchingThenUseInternalTaskCo
{
immCmdList0->hostSynchronize(0);
EXPECT_EQ(1u, ultCsr->latestWaitForCompletionWithTimeoutTaskCount.load());
EXPECT_EQ(1u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled.load());
immCmdList1->hostSynchronize(0);
EXPECT_EQ(2u, ultCsr->latestWaitForCompletionWithTimeoutTaskCount.load());
EXPECT_EQ(2u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled.load());
}
// implicit wait
@@ -894,12 +896,20 @@ HWTEST2_F(InOrderCmdListTests, givenCmdListsWhenDispatchingThenUseInternalTaskCo
ASSERT_EQ(result, ZE_RESULT_SUCCESS);
uint32_t hostCopyData = 0;
auto hostAddress0 = static_cast<uint64_t *>(immCmdList0->inOrderExecInfo->inOrderDependencyCounterAllocation.getUnderlyingBuffer());
auto hostAddress1 = static_cast<uint64_t *>(immCmdList1->inOrderExecInfo->inOrderDependencyCounterAllocation.getUnderlyingBuffer());
*hostAddress0 = 1;
*hostAddress1 = 1;
immCmdList0->appendMemoryCopy(deviceAlloc, &hostCopyData, 1, nullptr, 0, nullptr, false, false);
EXPECT_EQ(1u, ultCsr->latestWaitForCompletionWithTimeoutTaskCount.load());
EXPECT_EQ(immCmdList0->dcFlushSupport ? 1u : 2u, ultCsr->latestWaitForCompletionWithTimeoutTaskCount.load());
EXPECT_EQ(immCmdList0->dcFlushSupport ? 3u : 2u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled.load());
immCmdList1->appendMemoryCopy(deviceAlloc, &hostCopyData, 1, nullptr, 0, nullptr, false, false);
EXPECT_EQ(2u, ultCsr->latestWaitForCompletionWithTimeoutTaskCount.load());
EXPECT_EQ(immCmdList0->dcFlushSupport ? 4u : 2u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled.load());
context->freeMem(deviceAlloc);
}
@@ -2080,27 +2090,45 @@ HWTEST2_F(InOrderCmdListTests, givenHostVisibleEventOnLatestFlushWhenCallingSync
EXPECT_FALSE(immCmdList->latestFlushIsHostVisible);
immCmdList->appendLaunchKernel(kernel->toHandle(), groupCount, events[0]->toHandle(), 0, nullptr, launchParams, false);
EXPECT_FALSE(immCmdList->latestFlushIsHostVisible);
EXPECT_EQ(immCmdList->dcFlushSupport ? false : true, immCmdList->latestFlushIsHostVisible);
EXPECT_EQ(0u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(0u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
immCmdList->hostSynchronize(0, 1, false);
EXPECT_EQ(0u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(1u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
if (immCmdList->dcFlushSupport) {
EXPECT_EQ(0u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(1u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
} else {
EXPECT_EQ(1u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(0u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
}
events[0]->signalScope = ZE_EVENT_SCOPE_FLAG_HOST;
immCmdList->appendLaunchKernel(kernel->toHandle(), groupCount, events[0]->toHandle(), 0, nullptr, launchParams, false);
EXPECT_TRUE(immCmdList->latestFlushIsHostVisible);
immCmdList->hostSynchronize(0, 1, false);
EXPECT_EQ(1u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(1u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
if (immCmdList->dcFlushSupport) {
EXPECT_EQ(1u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(1u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
} else {
EXPECT_EQ(2u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(0u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
}
// handle post sync operations
immCmdList->hostSynchronize(0, 1, true);
EXPECT_EQ(1u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(2u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
if (immCmdList->dcFlushSupport) {
EXPECT_EQ(1u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(2u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
} else {
EXPECT_EQ(2u, immCmdList->synchronizeInOrderExecutionCalled);
EXPECT_EQ(1u, ultCsr->waitForCompletionWithTimeoutTaskCountCalled);
}
}
using NonPostSyncWalkerMatcher = IsWithinGfxCore<IGFX_GEN9_CORE, IGFX_GEN12LP_CORE>;