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

fix: capability to write memory chunk in aub/tbx mode

Browse Source 
Related-To: GSD-6604

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2023-10-18 13:54:46 +00:00
committed by Compute-Runtime-Automation
parent faaceaaff8
commit 25195ebc96
22 changed files with 223 additions and 85 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
shared/test/common/libult/ult_command_stream_receiver.h
View File

@@ -33,6 +33,14 @@ struct WaitUserFenceParams {
bool forceRetStatusValue = true;
};
struct WriteMemoryParams {
GraphicsAllocation *latestGfxAllocation = nullptr;
uint64_t latestGpuVaChunkOffset = 0;
size_t latestChunkSize = 0;
uint32_t callCount = 0;
bool latestChunkedMode = false;
};
template <typename GfxFamily>
class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, public NonCopyableOrMovableClass {
using BaseClass = CommandStreamReceiverHw<GfxFamily>;
@@ -207,6 +215,17 @@ class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, publ
return BaseClass::flushImmediateTask(immediateCommandStream, immediateCommandStreamStart, dispatchFlags, device);
}
bool writeMemory(GraphicsAllocation &gfxAllocation, bool isChunkCopy, uint64_t gpuVaChunkOffset, size_t chunkSize) override {
writeMemoryParams.callCount++;
writeMemoryParams.latestGfxAllocation = &gfxAllocation;
writeMemoryParams.latestChunkedMode = isChunkCopy;
writeMemoryParams.latestGpuVaChunkOffset = gpuVaChunkOffset;
writeMemoryParams.latestChunkSize = chunkSize;
return BaseClass::writeMemory(gfxAllocation, isChunkCopy, gpuVaChunkOffset, chunkSize);
}
size_t getPreferredTagPoolSize() const override {
return BaseClass::getPreferredTagPoolSize() + 1;
}
@@ -447,6 +466,7 @@ class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, publ
TaskCountType latestSentTaskCountValueDuringFlush = 0;
WaitParams latestWaitForCompletionWithTimeoutWaitParams{0};
WaitUserFenceParams waitUserFenecParams;
WriteMemoryParams writeMemoryParams;
TaskCountType flushBcsTaskReturnValue{};
LinearStream *lastFlushedCommandStream = nullptr;

4
shared/test/common/mocks/mock_aub_csr.h
View File

@@ -103,8 +103,8 @@ struct MockAubCsr : public AUBCommandStreamReceiverHw<GfxFamily> {
batchBufferGpuAddressPassed = batchBufferGpuAddress;
}
void writeMemoryWithAubManager(GraphicsAllocation &graphicsAllocation) override {
CommandStreamReceiverSimulatedHw<GfxFamily>::writeMemoryWithAubManager(graphicsAllocation);
void writeMemoryWithAubManager(GraphicsAllocation &graphicsAllocation, bool isChunkCopy, uint64_t gpuVaChunkOffset, size_t chunkSize) override {
CommandStreamReceiverSimulatedHw<GfxFamily>::writeMemoryWithAubManager(graphicsAllocation, isChunkCopy, gpuVaChunkOffset, chunkSize);
writeMemoryWithAubManagerCalled = true;
}

4
shared/test/common/mocks/mock_tbx_csr.h
View File

@@ -33,8 +33,8 @@ class MockTbxCsr : public TbxCommandStreamReceiverHw<GfxFamily> {
initializeEngineCalled = true;
}
void writeMemoryWithAubManager(GraphicsAllocation &graphicsAllocation) override {
CommandStreamReceiverSimulatedHw<GfxFamily>::writeMemoryWithAubManager(graphicsAllocation);
void writeMemoryWithAubManager(GraphicsAllocation &graphicsAllocation, bool isChunkCopy, uint64_t gpuVaChunkOffset, size_t chunkSize) override {
CommandStreamReceiverSimulatedHw<GfxFamily>::writeMemoryWithAubManager(graphicsAllocation, isChunkCopy, gpuVaChunkOffset, chunkSize);
writeMemoryWithAubManagerCalled = true;
}

2
shared/test/unit_test/command_stream/aub_command_stream_receiver_1_tests.cpp
View File

@@ -35,7 +35,7 @@ struct MockAubCsrToTestDumpAubNonWritable : public AUBCommandStreamReceiverHw<Gf
using AUBCommandStreamReceiverHw<GfxFamily>::AUBCommandStreamReceiverHw;
using AUBCommandStreamReceiverHw<GfxFamily>::dumpAubNonWritable;
bool writeMemory(GraphicsAllocation &gfxAllocation) override {
bool writeMemory(GraphicsAllocation &gfxAllocation, bool isChunkCopy, uint64_t gpuVaChunkOffset, size_t chunkSize) override {
return true;
}
};

4
shared/test/unit_test/command_stream/aub_command_stream_receiver_3_tests.cpp
View File

@@ -114,7 +114,7 @@ HWTEST_F(AubCsrTest, WhenWriteWithAubManagerIsCalledThenAubManagerIsInvokedWithC
PreemptionHelper::getDefaultPreemptionMode(hwInfo)));
aubCsr->setupContext(*osContext);
aubCsr->writeMemoryWithAubManager(*allocation);
aubCsr->writeMemoryWithAubManager(*allocation, false, 0, 0);
EXPECT_TRUE(aubManager.writeMemory2Called);
EXPECT_EQ(AubMemDump::DataTypeHintValues::TraceBatchBuffer, aubManager.hintToWriteMemory);
@@ -122,7 +122,7 @@ HWTEST_F(AubCsrTest, WhenWriteWithAubManagerIsCalledThenAubManagerIsInvokedWithC
auto allocation2 = executionEnvironment->memoryManager->allocateGraphicsMemoryWithProperties(MockAllocationProperties{rootDeviceIndex, true, MemoryConstants::pageSize, AllocationType::LINEAR_STREAM});
aubCsr->writeMemoryWithAubManager(*allocation2);
aubCsr->writeMemoryWithAubManager(*allocation2, true, 1, 1);
EXPECT_TRUE(aubManager.writeMemory2Called);
EXPECT_EQ(AubMemDump::DataTypeHintValues::TraceNotype, aubManager.hintToWriteMemory);

24
shared/test/unit_test/command_stream/command_stream_receiver_simulated_tests.cpp
View File

@@ -313,7 +313,7 @@ HWTEST_F(CommandStreamSimulatedTests, givenSimulatedCommandStreamReceiverWhenClo
GraphicsAllocation graphicsAllocation{0, AllocationType::UNKNOWN,
&dummy, 0, 0, sizeof(dummy), MemoryPool::MemoryNull, MemoryManager::maxOsContextCount};
graphicsAllocation.storageInfo.cloningOfPageTables = true;
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_FALSE(mockHardwareContext->writeMemory2Called);
EXPECT_TRUE(mockManager->writeMemory2Called);
@@ -339,7 +339,7 @@ HWTEST_F(CommandStreamSimulatedTests, givenCompressedAllocationWhenCloningPageTa
graphicsAllocation.setDefaultGmm(&gmm);
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_FALSE(mockHardwareContext->writeMemory2Called);
EXPECT_TRUE(mockManager->writeMemory2Called);
@@ -369,7 +369,7 @@ HWTEST_F(CommandStreamSimulatedTests, givenUncachedAllocationWhenCloningPageTabl
graphicsAllocation.setDefaultGmm(&gmm);
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_FALSE(mockHardwareContext->writeMemory2Called);
EXPECT_TRUE(mockManager->writeMemory2Called);
@@ -396,7 +396,7 @@ HWTEST_F(CommandStreamSimulatedTests, givenTileInstancedAllocationWhenWriteMemor
graphicsAllocation.storageInfo.cloningOfPageTables = false;
graphicsAllocation.storageInfo.tileInstanced = true;
graphicsAllocation.storageInfo.memoryBanks = 0b11u;
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_TRUE(firstMockHardwareContext->writeMemory2Called);
EXPECT_EQ(0b01u, firstMockHardwareContext->memoryBanksPassed);
@@ -431,7 +431,7 @@ HWTEST_F(CommandStreamSimulatedTests, givenCompressedTileInstancedAllocationWhen
graphicsAllocation.setDefaultGmm(&gmm);
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_TRUE(firstMockHardwareContext->writeMemory2Called);
EXPECT_EQ(1u, firstMockHardwareContext->storedAllocationParams.size());
@@ -470,7 +470,7 @@ HWTEST_F(CommandStreamSimulatedTests, givenUncachedTileInstancedAllocationWhenWr
graphicsAllocation.setDefaultGmm(&gmm);
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_TRUE(firstMockHardwareContext->writeMemory2Called);
EXPECT_EQ(1u, firstMockHardwareContext->storedAllocationParams.size());
@@ -500,7 +500,7 @@ HWTEST_F(CommandStreamSimulatedTests, givenTileInstancedAllocationWithMissingMem
graphicsAllocation.storageInfo.cloningOfPageTables = false;
graphicsAllocation.storageInfo.tileInstanced = true;
graphicsAllocation.storageInfo.memoryBanks = 2u;
EXPECT_THROW(csr->writeMemoryWithAubManager(graphicsAllocation), std::exception);
EXPECT_THROW(csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0), std::exception);
EXPECT_FALSE(firstMockHardwareContext->writeMemory2Called);
EXPECT_FALSE(secondMockHardwareContext->writeMemory2Called);
}
@@ -514,7 +514,7 @@ HWTEST_F(CommandStreamSimulatedTests, givenCommandBufferAllocationWhenWriteMemor
GraphicsAllocation graphicsAllocation{0, AllocationType::COMMAND_BUFFER,
&dummy, 0, 0, sizeof(dummy), MemoryPool::MemoryNull, MemoryManager::maxOsContextCount};
graphicsAllocation.storageInfo.cloningOfPageTables = true;
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_EQ(AubMemDump::DataTypeHintValues::TraceBatchBuffer, mockManager->hintToWriteMemory);
EXPECT_TRUE(mockManager->writeMemory2Called);
@@ -548,13 +548,13 @@ HWTEST_F(CommandStreamSimulatedTests, givenSpecificMemoryPoolAllocationWhenWrite
GraphicsAllocation graphicsAllocation{0, AllocationType::COMMAND_BUFFER,
&dummy, 0, 0, sizeof(dummy), poolsWith4kPages[i], MemoryManager::maxOsContextCount};
graphicsAllocation.storageInfo.cloningOfPageTables = true;
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_TRUE(mockManager->writeMemory2Called);
EXPECT_EQ(MemoryConstants::pageSize, mockManager->writeMemoryPageSizePassed);
graphicsAllocation.storageInfo.cloningOfPageTables = false;
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
if (graphicsAllocation.isAllocatedInLocalMemoryPool()) {
EXPECT_TRUE(mockHardwareContext->writeMemory2Called);
@@ -581,13 +581,13 @@ HWTEST_F(CommandStreamSimulatedTests, givenSpecificMemoryPoolAllocationWhenWrite
GraphicsAllocation graphicsAllocation{0, AllocationType::COMMAND_BUFFER,
&dummy, 0, 0, sizeof(dummy), poolsWith64kPages[i], MemoryManager::maxOsContextCount};
graphicsAllocation.storageInfo.cloningOfPageTables = true;
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
EXPECT_TRUE(mockManager->writeMemory2Called);
EXPECT_EQ(MemoryConstants::pageSize64k, mockManager->writeMemoryPageSizePassed);
graphicsAllocation.storageInfo.cloningOfPageTables = false;
csr->writeMemoryWithAubManager(graphicsAllocation);
csr->writeMemoryWithAubManager(graphicsAllocation, false, 0, 0);
if (graphicsAllocation.isAllocatedInLocalMemoryPool()) {
EXPECT_TRUE(mockHardwareContext->writeMemory2Called);

13
shared/test/unit_test/command_stream/command_stream_receiver_tests.cpp
View File

@@ -1454,6 +1454,19 @@ TEST(CommandStreamReceiverSimpleTest, givenVariousDataSetsWhenVerifyingMemoryThe
EXPECT_TRUE(csr.expectMemory(setA2, setB2, setSize, compareNotEqual));
}
TEST(CommandStreamReceiverSimpleTest, givenBaseCsrWhenWritingMemoryThenReturnFalse) {
MockExecutionEnvironment executionEnvironment;
executionEnvironment.prepareRootDeviceEnvironments(1);
executionEnvironment.initializeMemoryManager();
DeviceBitfield deviceBitfield(1);
MockCommandStreamReceiver csr(executionEnvironment, 0, deviceBitfield);
MockGraphicsAllocation mockAllocation;
EXPECT_FALSE(csr.writeMemory(mockAllocation));
}
TEST(CommandStreamReceiverSimpleTest, givenNewResourceFlushDisabledWhenProvidingNeverUsedAllocationTaskCountThenDoNotMarkNewResourceTrue) {
MockExecutionEnvironment executionEnvironment;
executionEnvironment.prepareRootDeviceEnvironments(1);

37
shared/test/unit_test/command_stream/command_stream_receiver_with_aub_dump_tests.cpp
View File

@@ -305,6 +305,43 @@ HWTEST_P(CommandStreamReceiverWithAubDumpTest, givenCommandStreamReceiverWithAub
memoryManager->freeGraphicsMemoryImpl(gfxAllocation);
}
HWTEST_P(CommandStreamReceiverWithAubDumpTest, givenCommandStreamReceiverWithAubDumpWhenWriteMemoryIsCalledThenBothCommandStreamReceiversAreCalled) {
MockGraphicsAllocation mockAllocation;
EXPECT_EQ(0u, csrWithAubDump->writeMemoryParams.callCount);
if (createAubCSR) {
EXPECT_EQ(0u, csrWithAubDump->getAubMockCsr().writeMemoryParams.callCount);
}
csrWithAubDump->writeMemory(mockAllocation, false, 0, 0);
EXPECT_EQ(1u, csrWithAubDump->writeMemoryParams.callCount);
EXPECT_EQ(&mockAllocation, csrWithAubDump->writeMemoryParams.latestGfxAllocation);
EXPECT_FALSE(csrWithAubDump->writeMemoryParams.latestChunkedMode);
if (createAubCSR) {
EXPECT_EQ(1u, csrWithAubDump->getAubMockCsr().writeMemoryParams.callCount);
EXPECT_EQ(&mockAllocation, csrWithAubDump->getAubMockCsr().writeMemoryParams.latestGfxAllocation);
EXPECT_FALSE(csrWithAubDump->getAubMockCsr().writeMemoryParams.latestChunkedMode);
}
csrWithAubDump->writeMemory(mockAllocation, true, 1, 2);
EXPECT_EQ(2u, csrWithAubDump->writeMemoryParams.callCount);
EXPECT_TRUE(csrWithAubDump->writeMemoryParams.latestChunkedMode);
EXPECT_EQ(&mockAllocation, csrWithAubDump->writeMemoryParams.latestGfxAllocation);
EXPECT_EQ(1u, csrWithAubDump->writeMemoryParams.latestGpuVaChunkOffset);
EXPECT_EQ(2u, csrWithAubDump->writeMemoryParams.latestChunkSize);
if (createAubCSR) {
EXPECT_EQ(2u, csrWithAubDump->getAubMockCsr().writeMemoryParams.callCount);
EXPECT_TRUE(csrWithAubDump->getAubMockCsr().writeMemoryParams.latestChunkedMode);
EXPECT_EQ(&mockAllocation, csrWithAubDump->getAubMockCsr().writeMemoryParams.latestGfxAllocation);
EXPECT_EQ(1u, csrWithAubDump->getAubMockCsr().writeMemoryParams.latestGpuVaChunkOffset);
EXPECT_EQ(2u, csrWithAubDump->getAubMockCsr().writeMemoryParams.latestChunkSize);
}
}
HWTEST_F(CommandStreamReceiverWithAubDumpSimpleTest, givenCsrWithAubDumpWhenCreatingAubCsrThenInitializeTagAllocation) {
auto executionEnvironment = pDevice->getExecutionEnvironment();
executionEnvironment->initializeMemoryManager();

2
shared/test/unit_test/command_stream/tbx_command_stream_tests.cpp
View File

@@ -72,7 +72,7 @@ struct MockTbxCsrToTestDumpTbxNonWritable : public TbxCommandStreamReceiverHw<Gf
using TbxCommandStreamReceiverHw<GfxFamily>::TbxCommandStreamReceiverHw;
using TbxCommandStreamReceiverHw<GfxFamily>::dumpTbxNonWritable;
bool writeMemory(GraphicsAllocation &gfxAllocation) override {
bool writeMemory(GraphicsAllocation &gfxAllocation, bool isChunkCopy, uint64_t gpuVaChunkOffset, size_t chunkSize) override {
return true;
}
};

6
shared/test/unit_test/mocks/mock_csr_simulated_common_hw.h
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2021-2022 Intel Corporation
* Copyright (C) 2021-2023 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -28,8 +28,8 @@ class MockSimulatedCsrHw : public CommandStreamReceiverSimulatedHw<GfxFamily> {
bool writeMemory(GraphicsAllocation &gfxAllocation) override {
return true;
}
void writeMemoryWithAubManager(GraphicsAllocation &graphicsAllocation) override {
CommandStreamReceiverSimulatedHw<GfxFamily>::writeMemoryWithAubManager(graphicsAllocation);
void writeMemoryWithAubManager(GraphicsAllocation &graphicsAllocation, bool isChunkCopy, uint64_t gpuVaChunkOffset, size_t chunkSize) override {
CommandStreamReceiverSimulatedHw<GfxFamily>::writeMemoryWithAubManager(graphicsAllocation, isChunkCopy, gpuVaChunkOffset, chunkSize);
}
void writeMMIO(uint32_t offset, uint32_t value) override {}
bool isMultiOsContextCapable() const override {