intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-24 21:18:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: iterate over indirect allocations once

Browse Source 
Related-To: NEO-11921

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
This commit is contained in:
Szymon Morek
2024-07-09 08:05:59 +00:00
committed by Compute-Runtime-Automation
parent 43c840cb29
commit 457cb005de
7 changed files with 20 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
shared/source/memory_manager/unified_memory_manager.cpp
View File

@@ -744,6 +744,7 @@ void SVMAllocsManager::makeIndirectAllocationsResident(CommandStreamReceiver &co
std::unique_lock<std::shared_mutex> lock(mtx);
bool parseAllAllocations = false;
auto entry = indirectAllocationsResidency.find(&commandStreamReceiver);
TaskCountType previousCounter = 0;
if (entry == indirectAllocationsResidency.end()) {
parseAllAllocations = true;
@@ -755,13 +756,14 @@ void SVMAllocsManager::makeIndirectAllocationsResident(CommandStreamReceiver &co
} else {
if (this->allocationsCounter > entry->second.latestResidentObjectId) {
parseAllAllocations = true;
previousCounter = entry->second.latestResidentObjectId;
entry->second.latestResidentObjectId = this->allocationsCounter;
}
entry->second.latestSentTaskCount = taskCount;
}
if (parseAllAllocations) {
auto currentCounter = this->allocationsCounter.load();
for (auto allocationId = 1u; allocationId <= currentCounter; allocationId++) {
for (auto allocationId = static_cast<uint32_t>(previousCounter + 1); allocationId <= currentCounter; allocationId++) {
makeResidentForAllocationsWithId(allocationId, commandStreamReceiver);
}
}
@@ -887,6 +889,7 @@ std::unique_lock<std::mutex> SVMAllocsManager::obtainOwnership() {
void SVMAllocsManager::insertSVMAlloc(void *svmPtr, const SvmAllocationData &allocData) {
std::unique_lock<std::shared_mutex> lock(mtx);
this->svmAllocs.insert(svmPtr, allocData);
UNRECOVERABLE_IF(internalAllocationsMap.count(allocData.getAllocId()) > 0);
for (auto alloc : allocData.gpuAllocations.getGraphicsAllocations()) {
if (alloc != nullptr) {
internalAllocationsMap.insert({allocData.getAllocId(), alloc});