fix: Release temporary allocations from bcs split

Related-To: NEO-7933

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>

level_zero/core/source/device/bcs_split.h

@@ -12,7 +12,7 @@
 #include "shared/source/sku_info/sku_info_base.h"
 
 #include "level_zero/core/source/cmdlist/cmdlist_hw_immediate.h"
-#include "level_zero/core/source/cmdqueue/cmdqueue.h"
+#include "level_zero/core/source/cmdqueue/cmdqueue_imp.h"
 #include "level_zero/core/source/context/context.h"
 #include "level_zero/core/source/event/event.h"
 
@@ -118,6 +118,10 @@ struct BcsSplit {
 
             totalSize -= localSize;
             engineCount--;
+
+            if (hSignalEvent) {
+                Event::fromHandle(hSignalEvent)->appendAdditionalCsr(static_cast<CommandQueueImp *>(cmdQsForSplit[i])->getCsr());
+            }
         }
 
         cmdList->addEventsToCmdList(static_cast<uint32_t>(cmdQsForSplit.size()), eventHandles.data(), hasRelaxedOrderingDependencies, false);

level_zero/core/source/event/event.cpp

@@ -384,7 +384,8 @@ void Event::resetPackets(bool resetAllPackets) {
     cpuStartTimestamp = 0;
     gpuStartTimestamp = 0;
     gpuEndTimestamp = 0;
-    this->csr = this->device->getNEODevice()->getDefaultEngine().commandStreamReceiver;
+    this->csrs.clear();
+    this->csrs.push_back(this->device->getNEODevice()->getDefaultEngine().commandStreamReceiver);
 }
 
 void Event::setIsCompleted() {

level_zero/core/source/event/event.h

@@ -135,7 +135,15 @@ struct Event : _ze_event_handle_t {
         return isTimestampEvent || usingContextEndOffset;
     }
     void setCsr(NEO::CommandStreamReceiver *csr) {
-        this->csr = csr;
+        this->csrs[0] = csr;
+    }
+    void appendAdditionalCsr(NEO::CommandStreamReceiver *additonalCsr) {
+        for (const auto &csr : csrs) {
+            if (csr == additonalCsr) {
+                return;
+            }
+        }
+        csrs.push_back(additonalCsr);
     }
 
     void increaseKernelCount();
@@ -226,7 +234,7 @@ struct Event : _ze_event_handle_t {
 
     // Metric streamer instance associated with the event.
     MetricStreamer *metricStreamer = nullptr;
-    NEO::CommandStreamReceiver *csr = nullptr;
+    StackVec<NEO::CommandStreamReceiver *, 1> csrs;
     void *hostAddress = nullptr;
     Device *device = nullptr;
     EventPool *eventPool = nullptr;

level_zero/core/source/event/event_impl.inl

@@ -42,7 +42,7 @@ Event *Event::create(EventPool *eventPool, const ze_event_desc_t *desc, Device *
     event->hostAddress = reinterpret_cast<void *>(baseHostAddr + event->eventPoolOffset);
     event->signalScope = desc->signal;
     event->waitScope = desc->wait;
-    event->csr = csr;
+    event->csrs.push_back(csr);
     event->maxKernelCount = eventPool->getMaxKernelCount();
     event->maxPacketCount = eventPool->getEventMaxPackets();
     event->isFromIpcPool = eventPool->getImportedIpcPool();
@@ -167,10 +167,14 @@ ze_result_t EventImp<TagSizeT>::queryStatusEventPackets() {
         }
     }
     if (this->downloadAllocationRequired) {
-        this->csr->downloadAllocations();
+        for (auto &csr : csrs) {
+            csr->downloadAllocations();
+        }
     }
     this->setIsCompleted();
-    this->csr->getInternalAllocationStorage()->cleanAllocationList(this->csr->peekTaskCount(), NEO::AllocationUsage::TEMPORARY_ALLOCATION);
+    for (auto &csr : csrs) {
+        csr->getInternalAllocationStorage()->cleanAllocationList(csr->peekTaskCount(), NEO::AllocationUsage::TEMPORARY_ALLOCATION);
+    }
     return ZE_RESULT_SUCCESS;
 }
 
@@ -180,7 +184,9 @@ ze_result_t EventImp<TagSizeT>::queryStatus() {
         hostEventSetValue(metricStreamer->getNotificationState());
     }
     if (this->downloadAllocationRequired) {
-        this->csr->downloadAllocation(this->getAllocation(this->device));
+        for (auto &csr : csrs) {
+            csr->downloadAllocation(this->getAllocation(this->device));
+        }
     }
 
     if (!this->isFromIpcPool && isAlreadyCompleted()) {
@@ -292,7 +298,7 @@ ze_result_t EventImp<TagSizeT>::hostSynchronize(uint64_t timeout) {
 
     ze_result_t ret = ZE_RESULT_NOT_READY;
 
-    if (this->csr->getType() == NEO::CommandStreamReceiverType::CSR_AUB) {
+    if (this->csrs[0]->getType() == NEO::CommandStreamReceiverType::CSR_AUB) {
         return ZE_RESULT_SUCCESS;
     }
 
@@ -320,7 +326,7 @@ ze_result_t EventImp<TagSizeT>::hostSynchronize(uint64_t timeout) {
 
         if (elapsedTimeSinceGpuHangCheck.count() >= this->gpuHangCheckPeriod.count()) {
             lastHangCheckTime = currentTime;
-            if (this->csr->isGpuHangDetected()) {
+            if (this->csrs[0]->isGpuHangDetected()) {
                 if (device->getNEODevice()->getRootDeviceEnvironment().assertHandler.get()) {
                     device->getNEODevice()->getRootDeviceEnvironment().assertHandler->printAssertAndAbort();
                 }