Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-10 12:53:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

Queue stall mode for RelaxedOrdering

Browse Source 
Related-To: NEO-7458

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2022-11-19 18:25:04 +00:00
committed by Compute-Runtime-Automation
parent bb308c04ed
commit bc619fcbec
28 changed files with 549 additions and 200 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
opencl/test/unit_test/command_queue/blit_enqueue_1_tests.cpp
View File

@ -227,6 +227,16 @@ HWTEST_TEMPLATED_F(BlitAuxTranslationTests, whenFlushTagUpdateThenMiFlushDwIsFlu
EXPECT_NE(cmdFound, cmdListBcs.end());
}
HWTEST_TEMPLATED_F(BlitAuxTranslationTests, whenFlushTagUpdateThenSetStallingCmdsFlag) {
auto ultCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(bcsCsr);
ultCsr->recordFlusheBatchBuffer = true;
EXPECT_EQ(SubmissionStatus::SUCCESS, bcsCsr->flushTagUpdate());
EXPECT_TRUE(ultCsr->latestFlushedBatchBuffer.hasStallingCmds);
}
HWTEST_TEMPLATED_F(BlitAuxTranslationTests, givenBlitTranslationWhenConstructingCommandBufferThenSynchronizeBcsOutput) {
using XY_COPY_BLT = typename FamilyType::XY_COPY_BLT;
using MI_FLUSH_DW = typename FamilyType::MI_FLUSH_DW;
@ -1412,7 +1422,7 @@ HWTEST_TEMPLATED_F(BlitEnqueueTaskCountTests, givenMarkerThatFollowsCopyOperatio
auto offset = mockCmdQueue->getCS(0).getUsed();
//marker needs to program semaphore
// marker needs to program semaphore
commandQueue->enqueueMarkerWithWaitList(0, nullptr, &outEvent1);
auto cmdListQueue = getCmdList<FamilyType>(mockCmdQueue->getCS(0), offset);
@ -1475,7 +1485,7 @@ HWTEST_TEMPLATED_F(BlitEnqueueTaskCountTests, givenMarkerThatFollowsCopyOperatio
auto ultGpgpuCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(gpgpuCsr);
auto ultBcsCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(bcsCsr);
//make sure we wait for both
// make sure we wait for both
clWaitForEvents(1, &outEvent1);
EXPECT_EQ(ultBcsCsr->latestWaitForCompletionWithTimeoutTaskCount, ultBcsCsr->taskCount);
EXPECT_EQ(ultGpgpuCsr->latestWaitForCompletionWithTimeoutTaskCount, ultGpgpuCsr->taskCount);
@ -1503,7 +1513,7 @@ HWTEST_TEMPLATED_F(BlitEnqueueTaskCountTests, givenMarkerThatFollowsCopyOperatio
auto ultGpgpuCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(gpgpuCsr);
auto ultBcsCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(bcsCsr);
//make sure we wait for both
// make sure we wait for both
clWaitForEvents(1, &outEvent2);
EXPECT_EQ(ultBcsCsr->latestWaitForCompletionWithTimeoutTaskCount, ultBcsCsr->taskCount);
EXPECT_EQ(ultGpgpuCsr->latestWaitForCompletionWithTimeoutTaskCount, ultGpgpuCsr->taskCount);

15
opencl/test/unit_test/os_interface/linux/drm_command_stream_tests_2.cpp
View File

@ -558,7 +558,8 @@ HWTEST_TEMPLATED_F(DrmCommandStreamEnhancedTest, GivenFlushMultipleTimesThenSucc
cs.replaceGraphicsAllocation(commandBuffer);
CommandStreamReceiverHw<FamilyType>::addBatchBufferEnd(cs, nullptr);
EncodeNoop<FamilyType>::alignToCacheLine(cs);
BatchBuffer batchBuffer2{cs.getGraphicsAllocation(), 8, 0, 0, nullptr, false, false, QueueThrottle::MEDIUM, QueueSliceCount::defaultSliceCount, cs.getUsed(), &cs, nullptr, false};
BatchBuffer batchBuffer2 = BatchBufferHelper::createDefaultBatchBuffer(cs.getGraphicsAllocation(), &cs, cs.getUsed());
batchBuffer.startOffset = 8;
csr->flush(batchBuffer2, csr->getResidencyAllocations());
auto allocation = mm->allocateGraphicsMemoryWithProperties(MockAllocationProperties{csr->getRootDeviceIndex(), MemoryConstants::pageSize});
@ -577,7 +578,8 @@ HWTEST_TEMPLATED_F(DrmCommandStreamEnhancedTest, GivenFlushMultipleTimesThenSucc
cs.replaceGraphicsAllocation(commandBuffer2);
CommandStreamReceiverHw<FamilyType>::addBatchBufferEnd(cs, nullptr);
EncodeNoop<FamilyType>::alignToCacheLine(cs);
BatchBuffer batchBuffer3{cs.getGraphicsAllocation(), 16, 0, 0, nullptr, false, false, QueueThrottle::MEDIUM, QueueSliceCount::defaultSliceCount, cs.getUsed(), &cs, nullptr, false};
BatchBuffer batchBuffer3 = BatchBufferHelper::createDefaultBatchBuffer(cs.getGraphicsAllocation(), &cs, cs.getUsed());
batchBuffer.startOffset = 16;
csr->flush(batchBuffer3, csr->getResidencyAllocations());
csr->makeSurfacePackNonResident(csr->getResidencyAllocations(), true);
mm->freeGraphicsMemory(allocation);
@ -590,7 +592,8 @@ HWTEST_TEMPLATED_F(DrmCommandStreamEnhancedTest, GivenFlushMultipleTimesThenSucc
cs.replaceGraphicsAllocation(commandBuffer2);
CommandStreamReceiverHw<FamilyType>::addBatchBufferEnd(cs, nullptr);
EncodeNoop<FamilyType>::alignToCacheLine(cs);
BatchBuffer batchBuffer4{cs.getGraphicsAllocation(), 24, 0, 0, nullptr, false, false, QueueThrottle::MEDIUM, QueueSliceCount::defaultSliceCount, cs.getUsed(), &cs, nullptr, false};
BatchBuffer batchBuffer4 = BatchBufferHelper::createDefaultBatchBuffer(cs.getGraphicsAllocation(), &cs, cs.getUsed());
batchBuffer.startOffset = 24;
csr->flush(batchBuffer4, csr->getResidencyAllocations());
}
@ -644,7 +647,8 @@ HWTEST_TEMPLATED_F(DrmCommandStreamEnhancedTest, GivenNotAlignedWhenFlushingThen
CommandStreamReceiverHw<FamilyType>::addBatchBufferEnd(cs, nullptr);
EncodeNoop<FamilyType>::alignToCacheLine(cs);
BatchBuffer batchBuffer{cs.getGraphicsAllocation(), 4, 0, 0, nullptr, false, false, QueueThrottle::MEDIUM, QueueSliceCount::defaultSliceCount, cs.getUsed(), &cs, nullptr, false};
BatchBuffer batchBuffer = BatchBufferHelper::createDefaultBatchBuffer(cs.getGraphicsAllocation(), &cs, cs.getUsed());
batchBuffer.startOffset = 4;
csr->flush(batchBuffer, csr->getResidencyAllocations());
}
@ -661,7 +665,8 @@ HWTEST_TEMPLATED_F(DrmCommandStreamEnhancedTest, GivenCheckDrmFreeWhenFlushingTh
csr->makeResident(*allocation);
CommandStreamReceiverHw<FamilyType>::addBatchBufferEnd(cs, nullptr);
EncodeNoop<FamilyType>::alignToCacheLine(cs);
BatchBuffer batchBuffer{cs.getGraphicsAllocation(), 4, 0, 0, nullptr, false, false, QueueThrottle::MEDIUM, QueueSliceCount::defaultSliceCount, cs.getUsed(), &cs, nullptr, false};
BatchBuffer batchBuffer = BatchBufferHelper::createDefaultBatchBuffer(cs.getGraphicsAllocation(), &cs, cs.getUsed());
batchBuffer.startOffset = 4;
csr->flush(batchBuffer, csr->getResidencyAllocations());
csr->makeNonResident(*allocation);
mm->freeGraphicsMemory(allocation);