From be946ae56c693fed9c240b1fde7b353241ea17a7 Mon Sep 17 00:00:00 2001
From: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
Date: Mon, 24 Feb 2025 10:50:20 +0000
Subject: [PATCH] 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 <lukasz.jobczyk@intel.com>
---
 .../command_stream/command_stream_receiver.cpp      |  2 +-
 .../source/command_stream/command_stream_receiver.h |  1 +
 .../command_stream_receiver_hw_base.inl             |  5 ++++-
 .../mocks/linux/mock_drm_command_stream_receiver.h  |  1 +
 .../linux/drm_command_stream_tests_1.cpp            | 13 +++++++++++++
 5 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/shared/source/command_stream/command_stream_receiver.cpp b/shared/source/command_stream/command_stream_receiver.cpp
index f716fd4e3b..3a8f31b0c1 100644
--- a/shared/source/command_stream/command_stream_receiver.cpp
+++ b/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();
diff --git a/shared/source/command_stream/command_stream_receiver.h b/shared/source/command_stream/command_stream_receiver.h
index d04c3b5be0..6f7b4a52d6 100644
--- a/shared/source/command_stream/command_stream_receiver.h
+++ b/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;
diff --git a/shared/source/command_stream/command_stream_receiver_hw_base.inl b/shared/source/command_stream/command_stream_receiver_hw_base.inl
index 294ce0f354..4dfe38c9ac 100644
--- a/shared/source/command_stream/command_stream_receiver_hw_base.inl
+++ b/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;
 }
diff --git a/shared/test/common/mocks/linux/mock_drm_command_stream_receiver.h b/shared/test/common/mocks/linux/mock_drm_command_stream_receiver.h
index 8b8f2f5696..0c6d3fe293 100644
--- a/shared/test/common/mocks/linux/mock_drm_command_stream_receiver.h
+++ b/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;
diff --git a/shared/test/unit_test/os_interface/linux/drm_command_stream_tests_1.cpp b/shared/test/unit_test/os_interface/linux/drm_command_stream_tests_1.cpp
index 7e40a20494..8b0b2a2d36 100644
--- a/shared/test/unit_test/os_interface/linux/drm_command_stream_tests_1.cpp
+++ b/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;