Media
/
compute-runtime
mirror of https://github.com/intel/compute-runtime.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Enhance wait method in L0 command queue 

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk 2021-06-30 09:10:41 +00:00 committed by Compute-Runtime-Automation
parent 234b2a748e
commit fffa22b17d
3 changed files with 35 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
level_zero/core/source/cmdqueue/cmdqueue.cpp
View File

@ -66,7 +66,7 @@ void CommandQueueImp::submitBatchBuffer(size_t offset, NEO::ResidencyContainer &
commandStream->getGraphicsAllocation()->updateResidencyTaskCount(csr->peekTaskCount() + 1, csr->getOsContext().getContextId()); commandStream->getGraphicsAllocation()->updateResidencyTaskCount(csr->peekTaskCount() + 1, csr->getOsContext().getContextId());
csr->submitBatchBuffer(batchBuffer, csr->getResidencyAllocations()); csr->submitBatchBuffer(batchBuffer, csr->getResidencyAllocations());
buffers.setCurrentFlushStamp(csr->obtainCurrentFlushStamp()); buffers.setCurrentFlushStamp(csr->peekTaskCount(), csr->obtainCurrentFlushStamp());
} }
ze_result_t CommandQueueImp::synchronize(uint64_t timeout) { ze_result_t CommandQueueImp::synchronize(uint64_t timeout) {
@ -151,8 +151,8 @@ ze_result_t CommandQueueImp::CommandBufferManager::initialize(Device *device, si
memset(buffers[BUFFER_ALLOCATION::FIRST]->getUnderlyingBuffer(), 0, buffers[BUFFER_ALLOCATION::FIRST]->getUnderlyingBufferSize()); memset(buffers[BUFFER_ALLOCATION::FIRST]->getUnderlyingBuffer(), 0, buffers[BUFFER_ALLOCATION::FIRST]->getUnderlyingBufferSize());
memset(buffers[BUFFER_ALLOCATION::SECOND]->getUnderlyingBuffer(), 0, buffers[BUFFER_ALLOCATION::SECOND]->getUnderlyingBufferSize()); memset(buffers[BUFFER_ALLOCATION::SECOND]->getUnderlyingBuffer(), 0, buffers[BUFFER_ALLOCATION::SECOND]->getUnderlyingBufferSize());
flushId[BUFFER_ALLOCATION::FIRST] = 0u; flushId[BUFFER_ALLOCATION::FIRST] = std::make_pair(0u, 0u);
flushId[BUFFER_ALLOCATION::SECOND] = 0u; flushId[BUFFER_ALLOCATION::SECOND] = std::make_pair(0u, 0u);
return ZE_RESULT_SUCCESS; return ZE_RESULT_SUCCESS;
} }
@ -174,10 +174,10 @@ void CommandQueueImp::CommandBufferManager::switchBuffers(NEO::CommandStreamRece
bufferUse = BUFFER_ALLOCATION::FIRST; bufferUse = BUFFER_ALLOCATION::FIRST;
} }
NEO::FlushStamp completionId = flushId[bufferUse]; auto completionId = flushId[bufferUse];
if (completionId != 0u) { if (completionId.second != 0u) {
UNRECOVERABLE_IF(csr == nullptr); UNRECOVERABLE_IF(csr == nullptr);
csr->waitForFlushStamp(completionId); csr->waitForTaskCountWithKmdNotifyFallback(completionId.first, completionId.second, false, false);
} }
} }

6
level_zero/core/source/cmdqueue/cmdqueue_imp.h
View File

@ -43,13 +43,13 @@ struct CommandQueueImp : public CommandQueue {
return buffers[bufferUse]; return buffers[bufferUse];
} }
void setCurrentFlushStamp(NEO::FlushStamp flushStamp) { void setCurrentFlushStamp(uint32_t taskCount, NEO::FlushStamp flushStamp) {
flushId[bufferUse] = flushStamp; flushId[bufferUse] = std::make_pair(taskCount, flushStamp);
} }
private: private:
NEO::GraphicsAllocation *buffers[BUFFER_ALLOCATION::COUNT]; NEO::GraphicsAllocation *buffers[BUFFER_ALLOCATION::COUNT];
NEO::FlushStamp flushId[BUFFER_ALLOCATION::COUNT]; std::pair<uint32_t, NEO::FlushStamp> flushId[BUFFER_ALLOCATION::COUNT];
BUFFER_ALLOCATION bufferUse = BUFFER_ALLOCATION::FIRST; BUFFER_ALLOCATION bufferUse = BUFFER_ALLOCATION::FIRST;
}; };
static constexpr size_t defaultQueueCmdBufferSize = 128 * MemoryConstants::kiloByte; static constexpr size_t defaultQueueCmdBufferSize = 128 * MemoryConstants::kiloByte;

28
level_zero/core/test/unit_tests/sources/cmdqueue/test_cmdqueue.cpp
View File

@ -98,7 +98,7 @@ TEST_F(CommandQueueCreate, whenSynchronizeByPollingTaskCountThenCallsPrintOutput
commandQueue->destroy(); commandQueue->destroy();
} }
TEST_F(CommandQueueCreate, whenReserveLinearStreamThenBufferAllocationSwitched) { HWTEST_F(CommandQueueCreate, whenReserveLinearStreamThenBufferAllocationSwitched) {
const ze_command_queue_desc_t desc{}; const ze_command_queue_desc_t desc{};
ze_result_t returnValue; ze_result_t returnValue;
@ -115,14 +115,38 @@ TEST_F(CommandQueueCreate, whenReserveLinearStreamThenBufferAllocationSwitched)
auto firstAllocation = commandQueue->commandStream->getGraphicsAllocation(); auto firstAllocation = commandQueue->commandStream->getGraphicsAllocation();
EXPECT_EQ(firstAllocation, commandQueue->buffers.getCurrentBufferAllocation()); EXPECT_EQ(firstAllocation, commandQueue->buffers.getCurrentBufferAllocation());
uint32_t currentTaskCount = 33u;
auto &csr = neoDevice->getUltCommandStreamReceiver<FamilyType>();
csr.latestWaitForCompletionWithTimeoutTaskCount = currentTaskCount;
commandQueue->commandStream->getSpace(maxSize - 16u); commandQueue->commandStream->getSpace(maxSize - 16u);
commandQueue->buffers.setCurrentFlushStamp(121u, 121u);
size_t nextSize = 16u + 16u; size_t nextSize = 16u + 16u;
commandQueue->reserveLinearStreamSize(nextSize); commandQueue->reserveLinearStreamSize(nextSize);
auto secondAllocation = commandQueue->commandStream->getGraphicsAllocation(); auto secondAllocation = commandQueue->commandStream->getGraphicsAllocation();
EXPECT_EQ(secondAllocation, commandQueue->buffers.getCurrentBufferAllocation()); EXPECT_EQ(secondAllocation, commandQueue->buffers.getCurrentBufferAllocation());
EXPECT_NE(firstAllocation, secondAllocation); EXPECT_NE(firstAllocation, secondAllocation);
EXPECT_EQ(csr.latestWaitForCompletionWithTimeoutTaskCount, currentTaskCount);
commandQueue->commandStream->getSpace(maxSize - 16u);
commandQueue->buffers.setCurrentFlushStamp(244u, 244u);
commandQueue->reserveLinearStreamSize(nextSize);
auto thirdAllocation = commandQueue->commandStream->getGraphicsAllocation();
EXPECT_EQ(thirdAllocation, commandQueue->buffers.getCurrentBufferAllocation());
EXPECT_EQ(thirdAllocation, firstAllocation);
EXPECT_NE(thirdAllocation, secondAllocation);
EXPECT_EQ(csr.latestWaitForCompletionWithTimeoutTaskCount, 121u);
commandQueue->commandStream->getSpace(maxSize - 16u);
commandQueue->reserveLinearStreamSize(nextSize);
auto fourthAllocation = commandQueue->commandStream->getGraphicsAllocation();
EXPECT_EQ(fourthAllocation, commandQueue->buffers.getCurrentBufferAllocation());
EXPECT_EQ(fourthAllocation, secondAllocation);
EXPECT_NE(fourthAllocation, firstAllocation);
EXPECT_EQ(csr.latestWaitForCompletionWithTimeoutTaskCount, 244u);
commandQueue->destroy(); commandQueue->destroy();
} }