intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-25 13:33:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: clear deferred TS nodes after in-order CL sync

Browse Source 
Related-To: NEO-7966

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2023-05-29 11:59:13 +00:00
committed by Compute-Runtime-Automation
parent c544004b8e
commit 2dd7a9cdc3
2 changed files with 34 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
View File

@@ -1077,6 +1077,9 @@ template <GFXCORE_FAMILY gfxCoreFamily>
ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::synchronizeInOrderExecution(uint64_t timeout) const {
using TSPacketType = typename GfxFamily::TimestampPacketType;
NEO::TimestampPacketContainer nodesToRelease;
nodesToRelease.swapNodes(*this->deferredTimestampPackets);
std::chrono::high_resolution_clock::time_point waitStartTime, lastHangCheckTime, now;
uint64_t timeDiff = 0;
@@ -1110,6 +1113,10 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::synchronizeInOrderExe
timeDiff = std::chrono::duration_cast<std::chrono::nanoseconds>(now - waitStartTime).count();
} while (timeDiff < timeout);
if (status == ZE_RESULT_NOT_READY) {
nodesToRelease.moveNodesToNewContainer(*this->deferredTimestampPackets);
}
return status;
}

27
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp
View File

@@ -1258,6 +1258,33 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenCallingSyncThenHandleCompleti
}
}
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenCallingSyncThenHandleDeferredNodesLifeCycle, IsAtLeastXeHpCore) {
auto immCmdList = createImmCmdList<gfxCoreFamily>();
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
immCmdList->appendLaunchKernel(kernel->toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
auto node = getLatestTsNode(immCmdList.get());
setTimestampPacketContextEndValue<FamilyType>(node, 1);
EXPECT_EQ(1u, immCmdList->timestampPacketContainer->peekNodes().size());
EXPECT_EQ(2u, immCmdList->deferredTimestampPackets->peekNodes().size());
EXPECT_EQ(ZE_RESULT_NOT_READY, immCmdList->hostSynchronize(1));
EXPECT_EQ(1u, immCmdList->timestampPacketContainer->peekNodes().size());
EXPECT_EQ(2u, immCmdList->deferredTimestampPackets->peekNodes().size());
setTimestampPacketContextEndValue<FamilyType>(node, 0x1234);
EXPECT_EQ(ZE_RESULT_SUCCESS, immCmdList->hostSynchronize(std::numeric_limits<uint64_t>::max()));
EXPECT_EQ(1u, immCmdList->timestampPacketContainer->peekNodes().size());
EXPECT_EQ(0u, immCmdList->deferredTimestampPackets->peekNodes().size());
}
HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenDoingCpuCopyThenSynchronize, IsAtLeastXeHpCore) {
auto immCmdList = createImmCmdList<gfxCoreFamily>();
immCmdList->copyThroughLockedPtrEnabled = true;