feature: enable dispatch primary batch buffer from immediate

Related-To: NEO-10356

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>

Author: zzdanowicz

level_zero/core/source/cmdlist/cmdlist_hw.inl

@@ -246,7 +246,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::initialize(Device *device, NEO
     this->l1CachePolicyData.init(productHelper);
     this->cmdListHeapAddressModel = L0GfxCoreHelper::getHeapAddressModel(rootDeviceEnvironment);
     this->dummyBlitWa.rootDeviceEnvironment = &(neoDevice->getRootDeviceEnvironmentRef());
-    this->dispatchCmdListBatchBufferAsPrimary = L0GfxCoreHelper::dispatchCmdListBatchBufferAsPrimary(rootDeviceEnvironment, !isImmediateType());
+    this->dispatchCmdListBatchBufferAsPrimary = L0GfxCoreHelper::dispatchCmdListBatchBufferAsPrimary(rootDeviceEnvironment, !(this->internalUsage && isImmediateType()));
     this->useOnlyGlobalTimestamps = gfxCoreHelper.useOnlyGlobalTimestamps();
     this->maxFillPaternSizeForCopyEngine = productHelper.getMaxFillPaternSizeForCopyEngine();
     this->heaplessModeEnabled = compilerProductHelper.isHeaplessModeEnabled();

level_zero/core/source/cmdqueue/cmdqueue.cpp

@@ -98,7 +98,7 @@ ze_result_t CommandQueueImp::initialize(bool copyOnly, bool isInternal, bool imm
         auto &productHelper = rootDeviceEnvironment.getHelper<NEO::ProductHelper>();
         this->doubleSbaWa = productHelper.isAdditionalStateBaseAddressWARequired(hwInfo);
         this->cmdListHeapAddressModel = L0GfxCoreHelper::getHeapAddressModel(rootDeviceEnvironment);
-        this->dispatchCmdListBatchBufferAsPrimary = L0GfxCoreHelper::dispatchCmdListBatchBufferAsPrimary(rootDeviceEnvironment, !immediateCmdListQueue);
+        this->dispatchCmdListBatchBufferAsPrimary = L0GfxCoreHelper::dispatchCmdListBatchBufferAsPrimary(rootDeviceEnvironment, !(internalUsage && immediateCmdListQueue));
         auto &compilerProductHelper = rootDeviceEnvironment.getHelper<NEO::CompilerProductHelper>();
         this->heaplessModeEnabled = compilerProductHelper.isHeaplessModeEnabled();
         this->heaplessStateInitEnabled = compilerProductHelper.isHeaplessStateInitEnabled(this->heaplessModeEnabled);

level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_2.cpp

@@ -1729,14 +1729,14 @@ HWTEST2_F(CommandListCreateTests, givenPlatformNotSupportsSharedHeapsWhenImmedia
 
 using PrimaryBatchBufferCmdListTest = Test<PrimaryBatchBufferCmdListFixture>;
 
-HWTEST_F(PrimaryBatchBufferCmdListTest, givenForcedPrimaryBatchBufferWhenRegularAndImmediateObjectCreatedThenRegularSetPrimaryFlagAndImmediateNot) {
+HWTEST_F(PrimaryBatchBufferCmdListTest, givenForcedPrimaryBatchBufferWhenRegularAndImmediateObjectCreatedThenRegularAndImmediateSetPrimaryFlag) {
     EXPECT_TRUE(commandList->dispatchCmdListBatchBufferAsPrimary);
     EXPECT_TRUE(commandQueue->dispatchCmdListBatchBufferAsPrimary);
 
-    EXPECT_FALSE(commandListImmediate->dispatchCmdListBatchBufferAsPrimary);
+    EXPECT_TRUE(commandListImmediate->dispatchCmdListBatchBufferAsPrimary);
     ASSERT_NE(nullptr, commandListImmediate->cmdQImmediate);
     auto immediateCmdQueue = static_cast<L0::ult::CommandQueue *>(commandListImmediate->cmdQImmediate);
-    EXPECT_FALSE(immediateCmdQueue->dispatchCmdListBatchBufferAsPrimary);
+    EXPECT_TRUE(immediateCmdQueue->dispatchCmdListBatchBufferAsPrimary);
 }
 
 HWTEST_F(PrimaryBatchBufferCmdListTest, givenPrimaryBatchBufferWhenAppendingKernelAndClosingCommandListThenExpectAlignedSpaceForBatchBufferStart) {
@@ -1851,6 +1851,46 @@ HWTEST_F(PrimaryBatchBufferCmdListTest, givenRegularCmdListWhenFlushingThenPassS
     EXPECT_TRUE(ultCsr->latestFlushedBatchBuffer.hasStallingCmds);
 }
 
+HWTEST_F(PrimaryBatchBufferCmdListTest, givenRegularCmdListWhenNoPreambleExpectedAndForceBbStartThenDispatchBbStart) {
+    using MI_BATCH_BUFFER_START = typename FamilyType::MI_BATCH_BUFFER_START;
+
+    ze_group_count_t groupCount{1, 1, 1};
+    CmdListKernelLaunchParams launchParams = {};
+    auto result = commandList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+    result = commandList->close();
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+    auto cmdListHandle = commandList->toHandle();
+
+    auto regularCmdBufferAllocation = commandList->getCmdContainer().getCommandStream()->getGraphicsAllocation();
+
+    // 1st dispatch can carry state preamble
+    result = commandQueue->executeCommandLists(1, &cmdListHandle, nullptr, true, nullptr);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+
+    auto &queueStream = commandQueue->commandStream;
+
+    auto offsetBefore = queueStream.getUsed();
+    commandQueue->triggerBbStartJump();
+    result = commandQueue->executeCommandLists(1, &cmdListHandle, nullptr, true, nullptr);
+    EXPECT_EQ(ZE_RESULT_SUCCESS, result);
+    auto offsetAfter = queueStream.getUsed();
+
+    GenCmdList cmdList;
+    ASSERT_TRUE(FamilyType::Parse::parseCommandBuffer(
+        cmdList,
+        ptrOffset(queueStream.getCpuBase(), offsetBefore),
+        offsetAfter - offsetBefore));
+    ASSERT_NE(0u, cmdList.size());
+
+    auto firstCmdIt = cmdList.begin();
+    auto bbStartCmd = genCmdCast<MI_BATCH_BUFFER_START *>(*firstCmdIt);
+    ASSERT_NE(nullptr, bbStartCmd);
+
+    EXPECT_EQ(regularCmdBufferAllocation->getGpuAddress(), bbStartCmd->getBatchBufferStartAddress());
+    EXPECT_EQ(MI_BATCH_BUFFER_START::SECOND_LEVEL_BATCH_BUFFER::SECOND_LEVEL_BATCH_BUFFER_FIRST_LEVEL_BATCH, bbStartCmd->getSecondLevelBatchBuffer());
+}
+
 HWTEST2_F(PrimaryBatchBufferCmdListTest, givenRelaxedOrderingAndRegularCmdListAndSubmittedToImmediateWhenFlushingThenPassStallingCmdsInfo, IsAtLeastXeHpcCore) {
     DebugManagerStateRestore restore;
     debugManager.flags.DirectSubmissionRelaxedOrdering.set(1);