intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 14:02:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add per-DSS back buffer programming

Browse Source 
Related-To: NEO-3220

Change-Id: Ide341205a283d8973b5c11f3a953eabbda14262f
Signed-off-by: Pawel Wilma <pawel.wilma@intel.com>
This commit is contained in:
Pawel Wilma
2019-09-04 16:44:27 +02:00
committed by sys_ocldev
parent 0cd93d6d7d
commit 849ff8c6d1
16 changed files with 142 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
unit_tests/command_stream/command_stream_receiver_flush_task_1_tests.cpp
View File

@@ -737,6 +737,7 @@ HWTEST_F(CommandStreamReceiverFlushTaskTests, flushTaskWithOnlyEnoughMemoryForPr
commandStream.getSpace(sizeof(PIPE_CONTROL));
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
commandStreamReceiver.timestampPacketWriteEnabled = false;
// Force a PIPE_CONTROL through a taskLevel transition
taskLevel = commandStreamReceiver.peekTaskLevel() + 1;

49
unit_tests/command_stream/command_stream_receiver_flush_task_3_tests.cpp
View File

@@ -1532,3 +1532,52 @@ HWTEST_F(CommandStreamReceiverFlushTaskTests, givenDcFlushArgumentIsFalseWhenCal
EXPECT_EQ(expectedDcFlush, pipeControl->getDcFlushEnable());
EXPECT_TRUE(pipeControl->getCommandStreamerStallEnable());
}
HWTEST_F(CommandStreamReceiverFlushTaskTests, whenPerDssBackBufferIsAllocatedItIsClearedInCleanupResources) {
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
ASSERT_NE(nullptr, pDevice);
commandStreamReceiver.createPerDssBackedBuffer(*pDevice);
EXPECT_NE(nullptr, commandStreamReceiver.perDssBackedBuffer);
commandStreamReceiver.cleanupResources();
EXPECT_EQ(nullptr, commandStreamReceiver.perDssBackedBuffer);
}
HWTEST_F(CommandStreamReceiverFlushTaskTests, whenPerDssBackBufferProgrammingEnabledThenAllocationIsCreated) {
DebugManagerStateRestore restore;
DebugManager.flags.ForcePerDssBackedBufferProgramming.set(true);
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
DispatchFlags dispatchFlags;
commandStreamReceiver.flushTask(commandStream,
0,
dsh,
ioh,
ssh,
taskLevel,
dispatchFlags,
*pDevice);
EXPECT_EQ(1u, commandStreamReceiver.createPerDssBackedBufferCalled);
EXPECT_NE(nullptr, commandStreamReceiver.perDssBackedBuffer);
}
HWTEST_F(CommandStreamReceiverFlushTaskTests, whenPerDssBackBufferProgrammingEnabledAndPerDssBackedBufferAlreadyPresentThenNewAllocationIsNotCreated) {
DebugManagerStateRestore restore;
DebugManager.flags.ForcePerDssBackedBufferProgramming.set(true);
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
auto memoryManager = pDevice->getMemoryManager();
commandStreamReceiver.perDssBackedBuffer = memoryManager->allocateGraphicsMemoryWithProperties(MockAllocationProperties{MemoryConstants::pageSize});
DispatchFlags dispatchFlags;
commandStreamReceiver.flushTask(commandStream,
0,
dsh,
ioh,
ssh,
taskLevel,
dispatchFlags,
*pDevice);
EXPECT_EQ(0u, commandStreamReceiver.createPerDssBackedBufferCalled);
}

18
unit_tests/command_stream/command_stream_receiver_hw_tests.cpp
View File

@@ -168,6 +168,24 @@ HWTEST_F(UltCommandStreamReceiverTest, givenPreambleSentWhenEstimatingPreambleCm
EXPECT_EQ(expectedDifference, actualDifference);
}
HWTEST_F(UltCommandStreamReceiverTest, givenPerDssBackBufferProgrammingEnabledWhenEstimatingPreambleCmdSizeThenResultIncludesPerDssBackBufferProgramingCommandsSize) {
DebugManagerStateRestore restore;
DebugManager.flags.ForcePerDssBackedBufferProgramming.set(true);
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
commandStreamReceiver.requiredThreadArbitrationPolicy = commandStreamReceiver.lastSentThreadArbitrationPolicy;
commandStreamReceiver.isPreambleSent = false;
auto preambleNotSent = commandStreamReceiver.getRequiredCmdSizeForPreamble(*pDevice);
commandStreamReceiver.isPreambleSent = true;
auto preambleSent = commandStreamReceiver.getRequiredCmdSizeForPreamble(*pDevice);
auto actualDifference = preambleNotSent - preambleSent;
auto expectedDifference = PreambleHelper<FamilyType>::getThreadArbitrationCommandsSize() + PreambleHelper<FamilyType>::getAdditionalCommandsSize(*pDevice) + PreambleHelper<FamilyType>::getPerDssBackedBufferCommandsSize(pDevice->getHardwareInfo());
EXPECT_EQ(expectedDifference, actualDifference);
}
HWCMDTEST_F(IGFX_GEN8_CORE, UltCommandStreamReceiverTest, givenMediaVfeStateDirtyEstimatingPreambleCmdSizeThenResultDependsVfeStateProgrammingCmdSize) {
typedef typename FamilyType::MEDIA_VFE_STATE MEDIA_VFE_STATE;
typedef typename FamilyType::PIPE_CONTROL PIPE_CONTROL;

2
unit_tests/gen11/test_preamble_gen11.cpp
View File

@@ -115,7 +115,7 @@ GEN11TEST_F(ThreadArbitrationGen11, givenPreambleWhenItIsProgrammedThenThreadArb
MockDevice mockDevice;
PreambleHelper<FamilyType>::programPreamble(&linearStream, mockDevice, l3Config,
ThreadArbitrationPolicy::RoundRobin,
nullptr);
nullptr, nullptr);
parseCommands<FamilyType>(cs);

2
unit_tests/gen9/skl/test_preamble_skl.cpp
View File

@@ -80,7 +80,7 @@ SKLTEST_F(ThreadArbitration, givenPreambleWhenItIsProgrammedThenThreadArbitratio
MockDevice mockDevice;
PreambleHelper<SKLFamily>::programPreamble(&linearStream, mockDevice, l3Config,
ThreadArbitrationPolicy::RoundRobin,
nullptr);
nullptr, nullptr);
parseCommands<SKLFamily>(cs);

12
unit_tests/libult/ult_command_stream_receiver.h
View File

@@ -30,6 +30,7 @@ class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, publ
using BaseClass::getScratchSpaceController;
using BaseClass::indirectHeap;
using BaseClass::iohState;
using BaseClass::perDssBackedBuffer;
using BaseClass::programPreamble;
using BaseClass::programStateSip;
using BaseClass::requiresInstructionCacheFlush;
@@ -177,6 +178,16 @@ class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, publ
CommandStreamReceiverHw<GfxFamily>::blitBuffer(blitProperites);
}
bool createPerDssBackedBuffer(Device &device) override {
createPerDssBackedBufferCalled++;
bool result = BaseClass::createPerDssBackedBuffer(device);
if (!perDssBackedBuffer) {
AllocationProperties properties{MemoryConstants::pageSize, GraphicsAllocation::AllocationType::INTERNAL_HEAP};
perDssBackedBuffer = executionEnvironment.memoryManager->allocateGraphicsMemoryWithProperties(properties);
}
return result;
}
std::atomic<uint32_t> recursiveLockCounter;
bool createPageTableManagerCalled = false;
bool recordFlusheBatchBuffer = false;
@@ -190,6 +201,7 @@ class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, publ
BatchBuffer latestFlushedBatchBuffer = {};
uint32_t latestSentTaskCountValueDuringFlush = 0;
uint32_t blitBufferCalled = 0;
uint32_t createPerDssBackedBufferCalled = 0;
std::atomic<uint32_t> latestWaitForCompletionWithTimeoutTaskCount{0};
DispatchFlags recordedDispatchFlags;
};

6
unit_tests/preamble/preamble_tests.cpp
View File

@@ -71,7 +71,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, PreambleTest, givenMidThreadPreemptionWhenPreambleIs
MockGraphicsAllocation csrSurface(reinterpret_cast<void *>(minCsrAlignment), 1024);
PreambleHelper<FamilyType>::programPreamble(&preambleStream, *mockDevice, 0U,
ThreadArbitrationPolicy::RoundRobin, &csrSurface);
ThreadArbitrationPolicy::RoundRobin, &csrSurface, nullptr);
PreemptionHelper::programStateSip<FamilyType>(preemptionStream, *mockDevice);
@@ -142,7 +142,7 @@ HWTEST_F(PreambleTest, givenKernelDebuggingActiveWhenPreambleIsProgrammedThenPro
LinearStream preambleStream(&*preambleBuffer.begin(), preambleBuffer.size());
PreambleHelper<FamilyType>::programPreamble(&preambleStream, *mockDevice, 0U,
ThreadArbitrationPolicy::RoundRobin, nullptr);
ThreadArbitrationPolicy::RoundRobin, nullptr, nullptr);
HardwareParse hwParser;
hwParser.parseCommands<FamilyType>(preambleStream);
@@ -156,7 +156,7 @@ HWTEST_F(PreambleTest, givenKernelDebuggingActiveWhenPreambleIsProgrammedThenPro
StackVec<char, 8192> preambleBuffer2(8192);
preambleStream.replaceBuffer(&*preambleBuffer2.begin(), preambleBuffer2.size());
PreambleHelper<FamilyType>::programPreamble(&preambleStream, *mockDevice, 0U,
ThreadArbitrationPolicy::RoundRobin, preemptionAllocation);
ThreadArbitrationPolicy::RoundRobin, preemptionAllocation, nullptr);
HardwareParse hwParser2;
hwParser2.parseCommands<FamilyType>(preambleStream);
cmdList = hwParser2.getCommandsList<MI_LOAD_REGISTER_IMM>();

3
unit_tests/test_files/igdrcl.config
View File

@@ -116,4 +116,5 @@ DisableAuxTranslation = 0
EnableFreeMemory = 0
OverrideStatelessMocsIndex = -1
AllocateSharedAllocationsWithCpuAndGpuStorage = -1
EnableSharedSystemUsmSupport = -1
EnableSharedSystemUsmSupport = -1
ForcePerDssBackedBufferProgramming = 0