intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-10 15:12:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: flush split task count

Browse Source 
Related-To: NEO-14641

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
This commit is contained in:
Bartosz Dunajski
2025-05-29 09:02:58 +00:00
committed by Compute-Runtime-Automation
parent 0c605f2da1
commit cdc725e0fe
2 changed files with 9 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
View File

@@ -693,7 +693,7 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendMemoryCopy(
numWaitEvents, phWaitEvents, memoryCopyParams);
}
return flushImmediate(ret, true, hasStallindCmds, memoryCopyParams.relaxedOrderingDispatch, NEO::AppendOperations::kernel, memoryCopyParams.copyOffloadAllowed, hSignalEvent, false, nullptr, nullptr);
return flushImmediate(ret, true, hasStallindCmds, memoryCopyParams.relaxedOrderingDispatch, NEO::AppendOperations::kernel, memoryCopyParams.copyOffloadAllowed, hSignalEvent, isSplitNeeded, nullptr, nullptr);
}
template <GFXCORE_FAMILY gfxCoreFamily>
@@ -751,7 +751,7 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendMemoryCopyRegio
hSignalEvent, numWaitEvents, phWaitEvents, memoryCopyParams);
}
return flushImmediate(ret, true, hasStallindCmds, memoryCopyParams.relaxedOrderingDispatch, NEO::AppendOperations::kernel, memoryCopyParams.copyOffloadAllowed, hSignalEvent, false, nullptr, nullptr);
return flushImmediate(ret, true, hasStallindCmds, memoryCopyParams.relaxedOrderingDispatch, NEO::AppendOperations::kernel, memoryCopyParams.copyOffloadAllowed, hSignalEvent, isSplitNeeded, nullptr, nullptr);
}
template <GFXCORE_FAMILY gfxCoreFamily>
@@ -818,7 +818,7 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendPageFaultCopy(N
} else {
ret = CommandListCoreFamily<gfxCoreFamily>::appendPageFaultCopy(dstAllocation, srcAllocation, size, flushHost);
}
return flushImmediate(ret, false, false, relaxedOrdering, NEO::AppendOperations::kernel, false, nullptr, false, nullptr, nullptr);
return flushImmediate(ret, false, false, relaxedOrdering, NEO::AppendOperations::kernel, false, nullptr, isSplitNeeded, nullptr, nullptr);
}
template <GFXCORE_FAMILY gfxCoreFamily>

7
level_zero/core/test/unit_tests/xe_hpc_core/test_cmdqueue_xe_hpc_core.cpp
View File

@@ -963,6 +963,7 @@ HWTEST2_F(CommandQueueCommandsXeHpc, givenFlushTaskSubmissionEnabledAndSplitBcsC
auto result = commandList0->appendMemoryCopy(dstPtr, srcPtr, size, nullptr, 0, nullptr, copyParams);
ASSERT_EQ(ZE_RESULT_SUCCESS, result);
auto csr0 = commandList0->getCsr(false);
auto csr2 = static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[2])->getCsr();
auto csr3 = static_cast<CommandQueueImp *>(static_cast<DeviceImp *>(testL0Device.get())->bcsSplit.cmdQs[3])->getCsr();
@@ -974,6 +975,7 @@ HWTEST2_F(CommandQueueCommandsXeHpc, givenFlushTaskSubmissionEnabledAndSplitBcsC
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::Parse::parseCommandBuffer(cmdList, ptrOffset(cmdStream->getCpuBase(), offset), (cmdStream->getUsed() - offset)));
bool csr0TaskCountFound = false;
bool csr2TaskCountFound = false;
bool csr3TaskCountFound = false;
@@ -983,10 +985,13 @@ HWTEST2_F(CommandQueueCommandsXeHpc, givenFlushTaskSubmissionEnabledAndSplitBcsC
csr2TaskCountFound = true;
} else if (miFlushCmd->getDestinationAddress() == csr3->getTagAllocation()->getGpuAddress()) {
csr3TaskCountFound = true;
} else if (miFlushCmd->getDestinationAddress() == csr0->getTagAllocation()->getGpuAddress()) {
csr0TaskCountFound = true;
}
}
}
EXPECT_TRUE(csr0TaskCountFound);
EXPECT_TRUE(csr2TaskCountFound);
EXPECT_TRUE(csr3TaskCountFound);
@@ -1121,7 +1126,7 @@ HWTEST2_F(CommandQueueCommandsXeHpc, givenFlushTaskSubmissionEnabledAndSplitBcsC
commandList0->appendMemoryCopy(dstPtr, srcPtr, size, nullptr, 0, nullptr, copyParams);
EXPECT_TRUE(ultCsr->latestFlushedBatchBuffer.hasRelaxedOrderingDependencies);
EXPECT_FALSE(ultCsr->latestFlushedBatchBuffer.hasStallingCmds);
EXPECT_TRUE(ultCsr->latestFlushedBatchBuffer.hasStallingCmds);
directSubmission->relaxedOrderingEnabled = false;