performance: Optimize heap handling when mitigate dc flush

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

level_zero/core/source/cmdlist/cmdlist.cpp

@@ -79,6 +79,17 @@ void CommandList::removeHostPtrAllocations() {
     hostPtrMap.clear();
 }
 
+void CommandList::forceDcFlushForDcFlushMitigation() {
+    if (this->device && this->device->getProductHelper().isDcFlushMitigated()) {
+        for (const auto &engine : this->device->getNEODevice()->getMemoryManager()->getRegisteredEngines(this->device->getNEODevice()->getRootDeviceIndex())) {
+            if (engine.commandStreamReceiver->isDirectSubmissionEnabled()) {
+                engine.commandStreamReceiver->registerDcFlushForDcMitigation();
+                engine.commandStreamReceiver->flushTagUpdate();
+            }
+        }
+    }
+}
+
 void CommandList::removeMemoryPrefetchAllocations() {
     if (this->performMemoryPrefetch) {
         auto prefetchManager = this->device->getDriverHandle()->getMemoryManager()->getPrefetchManager();

level_zero/core/source/cmdlist/cmdlist.h

@@ -200,6 +200,8 @@ struct CommandList : _ze_command_list_handle_t {
         return commandListPerThreadScratchSize[slotId];
     }
 
+    void forceDcFlushForDcFlushMitigation();
+
     void setOrdinal(uint32_t ord) { ordinal = ord; }
     void setCommandListPerThreadScratchSize(uint32_t slotId, uint32_t size) {
         UNRECOVERABLE_IF(slotId > 1);

level_zero/core/source/cmdlist/cmdlist_hw.inl

@@ -157,6 +157,8 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::reset() {
 
     this->inOrderPatchCmds.clear();
 
+    this->forceDcFlushForDcFlushMitigation();
+
     return ZE_RESULT_SUCCESS;
 }
 

level_zero/core/source/cmdlist/cmdlist_imp.cpp

@@ -76,6 +76,8 @@ ze_result_t CommandListImp::destroy() {
         }
     }
 
+    this->forceDcFlushForDcFlushMitigation();
+
     delete this;
     return ZE_RESULT_SUCCESS;
 }