performance: Optimize make resident for ULLS light

Do not check if ULLS light is active during every Csr::makeResident
call. Store that information once during ULLS init.

Related-To: NEO-13922

Signed-off-by: Lukasz Jobczyk <[email protected]>

Author: LukaszJobczyk

shared/source/command_stream/command_stream_receiver.cpp

@@ -156,7 +156,7 @@ void CommandStreamReceiver::makeResident(GraphicsAllocation &gfxAllocation) {
     gfxAllocation.updateTaskCount(submissionTaskCount, osContext->getContextId());
 
     if (gfxAllocation.isResidencyTaskCountBelow(submissionTaskCount, osContext->getContextId())) {
-        auto pushAllocations = !this->osContext->isDirectSubmissionLightActive();
+        auto pushAllocations = this->pushAllocationsForMakeResident;
 
         if (debugManager.flags.MakeEachAllocationResident.get() != -1) {
             pushAllocations = !debugManager.flags.MakeEachAllocationResident.get();

shared/source/command_stream/command_stream_receiver.h

@@ -688,6 +688,7 @@ class CommandStreamReceiver : NEO::NonCopyableAndNonMovableClass {
     bool lastSystolicPipelineSelectMode = false;
     bool requiresInstructionCacheFlush = false;
     bool requiresDcFlush = false;
+    bool pushAllocationsForMakeResident = true;
 
     bool localMemoryEnabled = false;
     bool pageTableManagerInitialized = false;

shared/source/command_stream/command_stream_receiver_hw_base.inl

@@ -1411,8 +1411,11 @@ inline bool CommandStreamReceiverHw<GfxFamily>::initDirectSubmission() {
                     this->overrideDispatchPolicy(DispatchMode::immediateDispatch);
                 }
             }
+            this->osContext->setDirectSubmissionActive();
+            if (this->osContext->isDirectSubmissionLightActive()) {
+                this->pushAllocationsForMakeResident = false;
+            }
         }
-        this->osContext->setDirectSubmissionActive();
     }
     return ret;
 }

shared/test/common/mocks/linux/mock_drm_command_stream_receiver.h

@@ -36,6 +36,7 @@ class TestedDrmCommandStreamReceiver : public DrmCommandStreamReceiver<GfxFamily
     using CommandStreamReceiver::immWritePostSyncWriteOffset;
     using CommandStreamReceiver::latestSentTaskCount;
     using CommandStreamReceiver::makeResident;
+    using CommandStreamReceiver::pushAllocationsForMakeResident;
     using CommandStreamReceiver::tagAddress;
     using CommandStreamReceiver::taskCount;
     using CommandStreamReceiver::timeStampPostSyncWriteOffset;

shared/test/unit_test/os_interface/linux/drm_command_stream_tests_1.cpp

@@ -963,6 +963,19 @@ HWTEST_TEMPLATED_F(DrmCommandStreamDirectSubmissionTest, givenDirectSubmissionLi
     executionEnvironment->rootDeviceEnvironments[rootDeviceIndex]->memoryOperationsInterface.reset(oldMemoryOperationsInterface);
 }
 
+HWTEST_TEMPLATED_F(DrmCommandStreamDirectSubmissionTest, givenDirectSubmissionLightWhenMakeResidentThenDoNotAddToCsrResidencyContainer) {
+    auto testedCsr = static_cast<TestedDrmCommandStreamReceiver<FamilyType> *>(csr);
+    testedCsr->directSubmission.reset();
+    csr->initDirectSubmission();
+    EXPECT_FALSE(testedCsr->pushAllocationsForMakeResident);
+    EXPECT_EQ(testedCsr->getResidencyAllocations().size(), 0u);
+
+    DrmAllocation graphicsAllocation(0, 1u /*num gmms*/, AllocationType::unknown, nullptr, nullptr, 1024, static_cast<osHandle>(1u), MemoryPool::memoryNull);
+    csr->makeResident(graphicsAllocation);
+
+    EXPECT_EQ(testedCsr->getResidencyAllocations().size(), 0u);
+}
+
 template <typename GfxFamily>
 struct MockDrmDirectSubmission : public DrmDirectSubmission<GfxFamily, RenderDispatcher<GfxFamily>> {
     using DrmDirectSubmission<GfxFamily, RenderDispatcher<GfxFamily>>::currentTagData;