intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-24 12:23:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

Allocate buffers in local memory for PVC multi root device platforms (1/n)

Browse Source 
PVC platform with no support for atomic operations on system memory
must always allocate buffers in local memory to avoid atomic access violation.
Note: the feature is being implemented under the new registry key
AllocateBuffersInLocalMemoryForMultiRootDeviceContexts (disabled by default)

Related-To: NEO-7092

Signed-off-by: Milczarek, Slawomir <slawomir.milczarek@intel.com>
This commit is contained in:
Milczarek, Slawomir
2022-10-24 23:25:04 +00:00
committed by Compute-Runtime-Automation
parent b0c97e49ea
commit 25a5ed0dca
15 changed files with 137 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
opencl/source/memory_manager/migration_controller.cpp
View File

@@ -27,12 +27,17 @@ void MigrationController::handleMigration(Context &context, CommandStreamReceive
if (migrationSyncData->getCurrentLocation() != targetRootDeviceIndex) {
migrateMemory(context, *memoryManager, memObj, targetRootDeviceIndex);
}
migrationSyncData->signalUsage(targetCsr.getTagAddress(), targetCsr.peekTaskCount() + 1);
if (!context.getSpecialQueue(targetRootDeviceIndex)->isWaitForTimestampsEnabled()) {
migrationSyncData->signalUsage(targetCsr.getTagAddress(), targetCsr.peekTaskCount() + 1);
}
}
void MigrationController::migrateMemory(Context &context, MemoryManager &memoryManager, MemObj *memObj, uint32_t targetRootDeviceIndex) {
auto &multiGraphicsAllocation = memObj->getMultiGraphicsAllocation();
auto migrationSyncData = multiGraphicsAllocation.getMigrationSyncData();
if (migrationSyncData->isMigrationInProgress()) {
return;
}
auto sourceRootDeviceIndex = migrationSyncData->getCurrentLocation();
if (sourceRootDeviceIndex == std::numeric_limits<uint32_t>::max()) {