intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-29 09:03:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add mechanism to migrate multi root device memory

Browse Source 
invalidate TLB cache if kernel requires migration


Related-To: NEO-3691

Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2021-07-01 16:00:22 +00:00
committed by Compute-Runtime-Automation
parent 714a1ebf53
commit 6f3c89decb
37 changed files with 1544 additions and 173 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
opencl/source/helpers/task_information.cpp
View File

@@ -77,7 +77,8 @@ CompletionStamp &CommandMapUnmap::submit(uint32_t taskLevel, bool terminated) {
false, //usePerDssBackedBuffer
false, //useSingleSubdevice
false, //useGlobalAtomics
1u); //numDevicesInContext
false, //areMultipleSubDevicesInContext
false); //memoryMigrationRequired
DEBUG_BREAK_IF(taskLevel >= CompletionStamp::notReady);
@@ -243,7 +244,8 @@ CompletionStamp &CommandComputeKernel::submit(uint32_t taskLevel, bool terminate
kernel->requiresPerDssBackedBuffer(), //usePerDssBackedBuffer
kernel->isSingleSubdevicePreferred(), //useSingleSubdevice
kernel->getKernelInfo().kernelDescriptor.kernelAttributes.flags.useGlobalAtomics, //useGlobalAtomics
kernel->areMultipleSubDevicesInContext()); //areMultipleSubDevicesInContext
kernel->areMultipleSubDevicesInContext(), //areMultipleSubDevicesInContext
kernel->requiresMemoryMigration()); //memoryMigrationRequired
if (commandQueue.getContext().getRootDeviceIndices().size() > 1) {
eventsRequest.fillCsrDependenciesForTaskCountContainer(dispatchFlags.csrDependencies, commandStreamReceiver);
@@ -269,6 +271,12 @@ CompletionStamp &CommandComputeKernel::submit(uint32_t taskLevel, bool terminate
gtpinNotifyPreFlushTask(&commandQueue);
if (kernel->requiresMemoryMigration()) {
for (auto &arg : kernel->getMemObjectsToMigrate()) {
MigrationController::handleMigration(commandQueue.getContext(), commandStreamReceiver, arg.second);
}
}
completionStamp = commandStreamReceiver.flushTask(*kernelOperation->commandStream,
0,
*dsh,
@@ -351,34 +359,36 @@ CompletionStamp &CommandWithoutKernel::submit(uint32_t taskLevel, bool terminate
}
}
auto rootDeviceIndex = commandStreamReceiver.getRootDeviceIndex();
DispatchFlags dispatchFlags(
{}, //csrDependencies
barrierNodes, //barrierTimestampPacketNodes
{}, //pipelineSelectArgs
commandQueue.flushStamp->getStampReference(), //flushStampReference
commandQueue.getThrottle(), //throttle
commandQueue.getDevice().getPreemptionMode(), //preemptionMode
GrfConfig::NotApplicable, //numGrfRequired
L3CachingSettings::NotApplicable, //l3CacheSettings
ThreadArbitrationPolicy::NotPresent, //threadArbitrationPolicy
AdditionalKernelExecInfo::NotApplicable, //additionalKernelExecInfo
KernelExecutionType::NotApplicable, //kernelExecutionType
MemoryCompressionState::NotApplicable, //memoryCompressionState
commandQueue.getSliceCount(), //sliceCount
true, //blocking
false, //dcFlush
false, //useSLM
true, //guardCommandBufferWithPipeControl
false, //GSBA32BitRequired
false, //requiresCoherency
commandQueue.getPriority() == QueuePriority::LOW, //lowPriority
false, //implicitFlush
commandStreamReceiver.isNTo1SubmissionModelEnabled(), //outOfOrderExecutionAllowed
false, //epilogueRequired
false, //usePerDssBackedBuffer
false, //useSingleSubdevice
false, //useGlobalAtomics
1u); //numDevicesInContext
{}, //csrDependencies
barrierNodes, //barrierTimestampPacketNodes
{}, //pipelineSelectArgs
commandQueue.flushStamp->getStampReference(), //flushStampReference
commandQueue.getThrottle(), //throttle
commandQueue.getDevice().getPreemptionMode(), //preemptionMode
GrfConfig::NotApplicable, //numGrfRequired
L3CachingSettings::NotApplicable, //l3CacheSettings
ThreadArbitrationPolicy::NotPresent, //threadArbitrationPolicy
AdditionalKernelExecInfo::NotApplicable, //additionalKernelExecInfo
KernelExecutionType::NotApplicable, //kernelExecutionType
MemoryCompressionState::NotApplicable, //memoryCompressionState
commandQueue.getSliceCount(), //sliceCount
true, //blocking
false, //dcFlush
false, //useSLM
true, //guardCommandBufferWithPipeControl
false, //GSBA32BitRequired
false, //requiresCoherency
commandQueue.getPriority() == QueuePriority::LOW, //lowPriority
false, //implicitFlush
commandStreamReceiver.isNTo1SubmissionModelEnabled(), //outOfOrderExecutionAllowed
false, //epilogueRequired
false, //usePerDssBackedBuffer
false, //useSingleSubdevice
false, //useGlobalAtomics
commandQueue.getContext().containsMultipleSubDevices(rootDeviceIndex), //areMultipleSubDevicesInContext
false); //memoryMigrationRequired
UNRECOVERABLE_IF(!kernelOperation->blitEnqueue && !commandStreamReceiver.peekTimestampPacketWriteEnabled() && commandQueue.getContext().getRootDeviceIndices().size() == 1);