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

AUB CSR with a separate function to submit batch buffer

Browse Source 
Move code related to batch buffer submission from flush to dedicated function
Add parameter to pass aub file name from AubCenter to AubManager

Change-Id: I20abb3c8bd92114b3bc9caa2a6291dc1bbddad2a
This commit is contained in:
Milczarek, Slawomir
2018-11-25 14:58:12 -08:00
parent 360c107b41
commit cd5f85052e
17 changed files with 219 additions and 181 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
runtime/aub/aub_center.cpp
View File

@@ -10,10 +10,9 @@
#include "runtime/os_interface/debug_settings_manager.h"
namespace OCLRT {
AubCenter::AubCenter(const HardwareInfo *pHwInfo, bool localMemoryEnabled) {
AubCenter::AubCenter(const HardwareInfo *pHwInfo, bool localMemoryEnabled, const std::string &aubFileName) {
if (DebugManager.flags.UseAubStream.get()) {
std::string filename("aub.aub");
aubManager.reset(AubDump::AubManager::create(pHwInfo->pPlatform->eRenderCoreFamily, 1, 1, localMemoryEnabled, filename));
aubManager.reset(AubDump::AubManager::create(pHwInfo->pPlatform->eRenderCoreFamily, 1, 1, localMemoryEnabled, pHwInfo->capabilityTable.aubDeviceId, aubFileName));
}
addressMapper = std::make_unique<AddressMapper>();
streamProvider = std::make_unique<AubFileStreamProvider>();

2
runtime/aub/aub_center.h
View File

@@ -17,7 +17,7 @@ struct HardwareInfo;
class AubCenter {
public:
AubCenter(const HardwareInfo *pHwInfo, bool localMemoryEnabled);
AubCenter(const HardwareInfo *pHwInfo, bool localMemoryEnabled, const std::string &aubFileName);
virtual ~AubCenter() = default;
void initPhysicalAddressAllocator(PhysicalAddressAllocator *pPhysicalAddressAllocator) {

2
runtime/aub_mem_dump/aub_stream_stubs.cpp
View File

@@ -9,7 +9,7 @@
namespace AubDump {
AubManager *AubManager::create(uint32_t gfxFamily, uint32_t devicesCount, size_t memoryBankSizeInGB, bool localMemorySupported, std::string &aubFileName) {
AubManager *AubManager::create(uint32_t gfxFamily, uint32_t devicesCount, size_t memoryBankSizeInGB, bool localMemorySupported, uint32_t deviceId, const std::string &aubFileName) {
return nullptr;
}

1
runtime/command_stream/aub_command_stream_receiver_hw.h
View File

@@ -53,6 +53,7 @@ class AUBCommandStreamReceiverHw : public CommandStreamReceiverSimulatedHw<GfxFa
void activateAubSubCapture(const MultiDispatchInfo &dispatchInfo) override;
// Family specific version
MOCKABLE_VIRTUAL void submitBatchBuffer(size_t engineIndex, uint64_t batchBufferGpuAddress, const void *batchBuffer, size_t batchBufferSize, uint32_t memoryBank, uint64_t entryBits);
void submitLRCA(EngineInstanceT engineInstance, const MiContextDescriptorReg &contextDescriptor);
MOCKABLE_VIRTUAL void pollForCompletion(EngineInstanceT engineInstance);
void initEngineMMIO(EngineInstanceT engineInstance);

331
runtime/command_stream/aub_command_stream_receiver_hw.inl
View File

@@ -36,7 +36,7 @@ AUBCommandStreamReceiverHw<GfxFamily>::AUBCommandStreamReceiverHw(const Hardware
subCaptureManager(std::make_unique<AubSubCaptureManager>(fileName)),
standalone(standalone) {
executionEnvironment.initAubCenter(&this->peekHwInfo(), this->localMemoryEnabled);
executionEnvironment.initAubCenter(&this->peekHwInfo(), this->localMemoryEnabled, fileName);
auto aubCenter = executionEnvironment.aubCenter.get();
UNRECOVERABLE_IF(nullptr == aubCenter);
@@ -173,9 +173,13 @@ const std::string &AUBCommandStreamReceiverHw<GfxFamily>::getFileName() {
template <typename GfxFamily>
void AUBCommandStreamReceiverHw<GfxFamily>::initializeEngine(size_t engineIndex) {
auto engineInstance = allEngineInstances[engineIndex];
auto mmioBase = getCsTraits(engineInstance).mmioBase;
auto csTraits = getCsTraits(engineInstance);
auto &engineInfo = engineInfoTable[engineIndex];
if (engineInfo.pLRCA) {
return;
}
this->initGlobalMMIO();
initEngineMMIO(engineInstance);
this->initAdditionalMMIO();
@@ -211,11 +215,10 @@ void AUBCommandStreamReceiverHw<GfxFamily>::initializeEngine(size_t engineIndex)
AubGTTData data = {0};
getGTTData(reinterpret_cast<void *>(physHWSP), data);
AUB::reserveAddressGGTT(*stream, engineInfo.ggttHWSP, sizeHWSP, physHWSP, data);
stream->writeMMIO(mmioBase + 0x2080, engineInfo.ggttHWSP);
stream->writeMMIO(csTraits.mmioBase + 0x2080, engineInfo.ggttHWSP);
}
// Allocate the LRCA
auto csTraits = getCsTraits(engineInstance);
const size_t sizeLRCA = csTraits.sizeLRCA;
const size_t alignLRCA = csTraits.alignLRCA;
auto pLRCABase = alignedMalloc(sizeLRCA, alignLRCA);
@@ -280,6 +283,8 @@ void AUBCommandStreamReceiverHw<GfxFamily>::initializeEngine(size_t engineIndex)
// Create a context to facilitate AUB dumping of memory using PPGTT
addContextToken(getDumpHandle());
DEBUG_BREAK_IF(!engineInfo.pLRCA);
}
template <typename GfxFamily>
@@ -325,14 +330,8 @@ FlushStamp AUBCommandStreamReceiverHw<GfxFamily>::flush(BatchBuffer &batchBuffer
auto streamLocked = getAubStream()->lockStream();
auto engineIndex = getEngineIndex(engineType);
auto engineInstance = allEngineInstances[engineIndex];
engineType = engineInstance.type;
uint32_t mmioBase = getCsTraits(engineInstance).mmioBase;
auto &engineInfo = engineInfoTable[engineIndex];
if (!engineInfo.pLRCA) {
initializeEngine(engineIndex);
DEBUG_BREAK_IF(!engineInfo.pLRCA);
}
initializeEngine(engineIndex);
// Write our batch buffer
auto pBatchBuffer = ptrOffset(batchBuffer.commandBufferAllocation->getUnderlyingBuffer(), batchBuffer.startOffset);
@@ -352,30 +351,6 @@ FlushStamp AUBCommandStreamReceiverHw<GfxFamily>::flush(BatchBuffer &batchBuffer
}
}
{
{
std::ostringstream str;
str << "ppgtt: " << std::hex << std::showbase << pBatchBuffer;
getAubStream()->addComment(str.str().c_str());
}
auto physBatchBuffer = ppgtt->map(static_cast<uintptr_t>(batchBufferGpuAddress), sizeBatchBuffer,
getPPGTTAdditionalBits(batchBuffer.commandBufferAllocation),
this->getMemoryBank(batchBuffer.commandBufferAllocation));
AubHelperHw<GfxFamily> aubHelperHw(this->localMemoryEnabled);
AUB::reserveAddressPPGTT(*stream, static_cast<uintptr_t>(batchBufferGpuAddress), sizeBatchBuffer, physBatchBuffer,
getPPGTTAdditionalBits(batchBuffer.commandBufferAllocation),
aubHelperHw);
AUB::addMemoryWrite(
*stream,
physBatchBuffer,
pBatchBuffer,
sizeBatchBuffer,
this->getAddressSpace(AubMemDump::DataTypeHintValues::TraceBatchBufferPrimary),
AubMemDump::DataTypeHintValues::TraceBatchBufferPrimary);
}
if (this->standalone) {
if (this->dispatchMode == DispatchMode::ImmediateDispatch) {
if (!DebugManager.flags.FlattenBatchBufferForAUBDump.get()) {
@@ -387,129 +362,8 @@ FlushStamp AUBCommandStreamReceiverHw<GfxFamily>::flush(BatchBuffer &batchBuffer
}
}
processResidency(allocationsForResidency, osContext);
if (DebugManager.flags.AddPatchInfoCommentsForAUBDump.get()) {
addGUCStartMessage(static_cast<uint64_t>(reinterpret_cast<std::uintptr_t>(pBatchBuffer)), engineType);
addPatchInfoComments();
}
// Add a batch buffer start to the ring buffer
auto previousTail = engineInfo.tailRingBuffer;
{
typedef typename GfxFamily::MI_LOAD_REGISTER_IMM MI_LOAD_REGISTER_IMM;
typedef typename GfxFamily::MI_BATCH_BUFFER_START MI_BATCH_BUFFER_START;
typedef typename GfxFamily::MI_NOOP MI_NOOP;
auto pTail = ptrOffset(engineInfo.pRingBuffer, engineInfo.tailRingBuffer);
auto ggttTail = ptrOffset(engineInfo.ggttRingBuffer, engineInfo.tailRingBuffer);
auto sizeNeeded =
sizeof(MI_BATCH_BUFFER_START) +
sizeof(MI_LOAD_REGISTER_IMM);
auto tailAlignment = sizeof(uint64_t);
sizeNeeded = alignUp(sizeNeeded, tailAlignment);
if (engineInfo.tailRingBuffer + sizeNeeded >= engineInfo.sizeRingBuffer) {
// Pad the remaining ring with NOOPs
auto sizeToWrap = engineInfo.sizeRingBuffer - engineInfo.tailRingBuffer;
memset(pTail, 0, sizeToWrap);
// write remaining ring
auto physDumpStart = ggtt->map(ggttTail, sizeToWrap, this->getGTTBits(), getMemoryBankForGtt());
AUB::addMemoryWrite(
*stream,
physDumpStart,
pTail,
sizeToWrap,
this->getAddressSpace(AubMemDump::DataTypeHintValues::TraceCommandBuffer),
AubMemDump::DataTypeHintValues::TraceCommandBuffer);
previousTail = 0;
engineInfo.tailRingBuffer = 0;
pTail = engineInfo.pRingBuffer;
} else if (engineInfo.tailRingBuffer == 0) {
// Add a LRI if this is our first submission
auto lri = MI_LOAD_REGISTER_IMM::sInit();
lri.setRegisterOffset(mmioBase + 0x2244);
lri.setDataDword(0x00010000);
*(MI_LOAD_REGISTER_IMM *)pTail = lri;
pTail = ((MI_LOAD_REGISTER_IMM *)pTail) + 1;
}
// Add our BBS
auto bbs = MI_BATCH_BUFFER_START::sInit();
bbs.setBatchBufferStartAddressGraphicsaddress472(static_cast<uint64_t>(batchBufferGpuAddress));
bbs.setAddressSpaceIndicator(MI_BATCH_BUFFER_START::ADDRESS_SPACE_INDICATOR_PPGTT);
*(MI_BATCH_BUFFER_START *)pTail = bbs;
pTail = ((MI_BATCH_BUFFER_START *)pTail) + 1;
// Compute our new ring tail.
engineInfo.tailRingBuffer = (uint32_t)ptrDiff(pTail, engineInfo.pRingBuffer);
// Add NOOPs as needed as our tail needs to be aligned
while (engineInfo.tailRingBuffer % tailAlignment) {
*(MI_NOOP *)pTail = MI_NOOP::sInit();
pTail = ((MI_NOOP *)pTail) + 1;
engineInfo.tailRingBuffer = (uint32_t)ptrDiff(pTail, engineInfo.pRingBuffer);
}
UNRECOVERABLE_IF((engineInfo.tailRingBuffer % tailAlignment) != 0);
// Only dump the new commands
auto ggttDumpStart = ptrOffset(engineInfo.ggttRingBuffer, previousTail);
auto dumpStart = ptrOffset(engineInfo.pRingBuffer, previousTail);
auto dumpLength = engineInfo.tailRingBuffer - previousTail;
// write ring
{
std::ostringstream str;
str << "ggtt: " << std::hex << std::showbase << ggttDumpStart;
getAubStream()->addComment(str.str().c_str());
}
auto physDumpStart = ggtt->map(ggttDumpStart, dumpLength, this->getGTTBits(), getMemoryBankForGtt());
AUB::addMemoryWrite(
*stream,
physDumpStart,
dumpStart,
dumpLength,
this->getAddressSpace(AubMemDump::DataTypeHintValues::TraceCommandBuffer),
AubMemDump::DataTypeHintValues::TraceCommandBuffer);
// update the ring mmio tail in the LRCA
{
std::ostringstream str;
str << "ggtt: " << std::hex << std::showbase << engineInfo.ggttLRCA + 0x101c;
getAubStream()->addComment(str.str().c_str());
}
auto physLRCA = ggtt->map(engineInfo.ggttLRCA, sizeof(engineInfo.tailRingBuffer), this->getGTTBits(), getMemoryBankForGtt());
AUB::addMemoryWrite(
*stream,
physLRCA + 0x101c,
&engineInfo.tailRingBuffer,
sizeof(engineInfo.tailRingBuffer),
this->getAddressSpace(getCsTraits(engineInstance).aubHintLRCA));
DEBUG_BREAK_IF(engineInfo.tailRingBuffer >= engineInfo.sizeRingBuffer);
}
// Submit our execlist by submitting to the execlist submit ports
{
typename AUB::MiContextDescriptorReg contextDescriptor = {{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}};
contextDescriptor.sData.Valid = true;
contextDescriptor.sData.ForcePageDirRestore = false;
contextDescriptor.sData.ForceRestore = false;
contextDescriptor.sData.Legacy = true;
contextDescriptor.sData.FaultSupport = 0;
contextDescriptor.sData.PrivilegeAccessOrPPGTT = true;
contextDescriptor.sData.ADor64bitSupport = AUB::Traits::addressingBits > 32;
auto ggttLRCA = engineInfo.ggttLRCA;
contextDescriptor.sData.LogicalRingCtxAddress = ggttLRCA / 4096;
contextDescriptor.sData.ContextID = 0;
submitLRCA(engineInstance, contextDescriptor);
}
submitBatchBuffer(engineIndex, batchBufferGpuAddress, pBatchBuffer, sizeBatchBuffer, this->getMemoryBank(batchBuffer.commandBufferAllocation), getPPGTTAdditionalBits(batchBuffer.commandBufferAllocation));
pollForCompletion(engineInstance);
if (this->standalone) {
@@ -567,6 +421,158 @@ bool AUBCommandStreamReceiverHw<GfxFamily>::addPatchInfoComments() {
return true;
}
template <typename GfxFamily>
void AUBCommandStreamReceiverHw<GfxFamily>::submitBatchBuffer(size_t engineIndex, uint64_t batchBufferGpuAddress, const void *batchBuffer, size_t batchBufferSize, uint32_t memoryBank, uint64_t entryBits) {
auto engineInstance = allEngineInstances[engineIndex];
auto csTraits = getCsTraits(engineInstance);
auto &engineInfo = engineInfoTable[engineIndex];
{
{
std::ostringstream str;
str << "ppgtt: " << std::hex << std::showbase << batchBuffer;
getAubStream()->addComment(str.str().c_str());
}
auto physBatchBuffer = ppgtt->map(static_cast<uintptr_t>(batchBufferGpuAddress), batchBufferSize, entryBits, memoryBank);
AubHelperHw<GfxFamily> aubHelperHw(this->localMemoryEnabled);
AUB::reserveAddressPPGTT(*stream, static_cast<uintptr_t>(batchBufferGpuAddress), batchBufferSize, physBatchBuffer,
entryBits, aubHelperHw);
AUB::addMemoryWrite(
*stream,
physBatchBuffer,
batchBuffer,
batchBufferSize,
this->getAddressSpace(AubMemDump::DataTypeHintValues::TraceBatchBufferPrimary),
AubMemDump::DataTypeHintValues::TraceBatchBufferPrimary);
}
if (DebugManager.flags.AddPatchInfoCommentsForAUBDump.get()) {
addGUCStartMessage(static_cast<uint64_t>(reinterpret_cast<std::uintptr_t>(batchBuffer)), engineInstance.type);
addPatchInfoComments();
}
// Add a batch buffer start to the ring buffer
auto previousTail = engineInfo.tailRingBuffer;
{
typedef typename GfxFamily::MI_LOAD_REGISTER_IMM MI_LOAD_REGISTER_IMM;
typedef typename GfxFamily::MI_BATCH_BUFFER_START MI_BATCH_BUFFER_START;
typedef typename GfxFamily::MI_NOOP MI_NOOP;
auto pTail = ptrOffset(engineInfo.pRingBuffer, engineInfo.tailRingBuffer);
auto ggttTail = ptrOffset(engineInfo.ggttRingBuffer, engineInfo.tailRingBuffer);
auto sizeNeeded =
sizeof(MI_BATCH_BUFFER_START) +
sizeof(MI_LOAD_REGISTER_IMM);
auto tailAlignment = sizeof(uint64_t);
sizeNeeded = alignUp(sizeNeeded, tailAlignment);
if (engineInfo.tailRingBuffer + sizeNeeded >= engineInfo.sizeRingBuffer) {
// Pad the remaining ring with NOOPs
auto sizeToWrap = engineInfo.sizeRingBuffer - engineInfo.tailRingBuffer;
memset(pTail, 0, sizeToWrap);
// write remaining ring
auto physDumpStart = ggtt->map(ggttTail, sizeToWrap, this->getGTTBits(), getMemoryBankForGtt());
AUB::addMemoryWrite(
*stream,
physDumpStart,
pTail,
sizeToWrap,
this->getAddressSpace(AubMemDump::DataTypeHintValues::TraceCommandBuffer),
AubMemDump::DataTypeHintValues::TraceCommandBuffer);
previousTail = 0;
engineInfo.tailRingBuffer = 0;
pTail = engineInfo.pRingBuffer;
} else if (engineInfo.tailRingBuffer == 0) {
// Add a LRI if this is our first submission
auto lri = MI_LOAD_REGISTER_IMM::sInit();
lri.setRegisterOffset(csTraits.mmioBase + 0x2244);
lri.setDataDword(0x00010000);
*(MI_LOAD_REGISTER_IMM *)pTail = lri;
pTail = ((MI_LOAD_REGISTER_IMM *)pTail) + 1;
}
// Add our BBS
auto bbs = MI_BATCH_BUFFER_START::sInit();
bbs.setBatchBufferStartAddressGraphicsaddress472(static_cast<uint64_t>(batchBufferGpuAddress));
bbs.setAddressSpaceIndicator(MI_BATCH_BUFFER_START::ADDRESS_SPACE_INDICATOR_PPGTT);
*(MI_BATCH_BUFFER_START *)pTail = bbs;
pTail = ((MI_BATCH_BUFFER_START *)pTail) + 1;
// Compute our new ring tail.
engineInfo.tailRingBuffer = (uint32_t)ptrDiff(pTail, engineInfo.pRingBuffer);
// Add NOOPs as needed as our tail needs to be aligned
while (engineInfo.tailRingBuffer % tailAlignment) {
*(MI_NOOP *)pTail = MI_NOOP::sInit();
pTail = ((MI_NOOP *)pTail) + 1;
engineInfo.tailRingBuffer = (uint32_t)ptrDiff(pTail, engineInfo.pRingBuffer);
}
UNRECOVERABLE_IF((engineInfo.tailRingBuffer % tailAlignment) != 0);
// Only dump the new commands
auto ggttDumpStart = ptrOffset(engineInfo.ggttRingBuffer, previousTail);
auto dumpStart = ptrOffset(engineInfo.pRingBuffer, previousTail);
auto dumpLength = engineInfo.tailRingBuffer - previousTail;
// write ring
{
std::ostringstream str;
str << "ggtt: " << std::hex << std::showbase << ggttDumpStart;
getAubStream()->addComment(str.str().c_str());
}
auto physDumpStart = ggtt->map(ggttDumpStart, dumpLength, this->getGTTBits(), getMemoryBankForGtt());
AUB::addMemoryWrite(
*stream,
physDumpStart,
dumpStart,
dumpLength,
this->getAddressSpace(AubMemDump::DataTypeHintValues::TraceCommandBuffer),
AubMemDump::DataTypeHintValues::TraceCommandBuffer);
// update the ring mmio tail in the LRCA
{
std::ostringstream str;
str << "ggtt: " << std::hex << std::showbase << engineInfo.ggttLRCA + 0x101c;
getAubStream()->addComment(str.str().c_str());
}
auto physLRCA = ggtt->map(engineInfo.ggttLRCA, sizeof(engineInfo.tailRingBuffer), this->getGTTBits(), getMemoryBankForGtt());
AUB::addMemoryWrite(
*stream,
physLRCA + 0x101c,
&engineInfo.tailRingBuffer,
sizeof(engineInfo.tailRingBuffer),
this->getAddressSpace(csTraits.aubHintLRCA));
DEBUG_BREAK_IF(engineInfo.tailRingBuffer >= engineInfo.sizeRingBuffer);
}
// Submit our execlist by submitting to the execlist submit ports
{
typename AUB::MiContextDescriptorReg contextDescriptor = {{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}};
contextDescriptor.sData.Valid = true;
contextDescriptor.sData.ForcePageDirRestore = false;
contextDescriptor.sData.ForceRestore = false;
contextDescriptor.sData.Legacy = true;
contextDescriptor.sData.FaultSupport = 0;
contextDescriptor.sData.PrivilegeAccessOrPPGTT = true;
contextDescriptor.sData.ADor64bitSupport = AUB::Traits::addressingBits > 32;
auto ggttLRCA = engineInfo.ggttLRCA;
contextDescriptor.sData.LogicalRingCtxAddress = ggttLRCA / 4096;
contextDescriptor.sData.ContextID = 0;
submitLRCA(engineInstance, contextDescriptor);
}
}
template <typename GfxFamily>
void AUBCommandStreamReceiverHw<GfxFamily>::submitLRCA(EngineInstanceT engineInstance, const typename AUBCommandStreamReceiverHw<GfxFamily>::MiContextDescriptorReg &contextDescriptor) {
auto mmioBase = getCsTraits(engineInstance).mmioBase;
@@ -617,17 +623,16 @@ bool AUBCommandStreamReceiverHw<GfxFamily>::writeMemory(GraphicsAllocation &gfxA
if ((size == 0) || !gfxAllocation.isAubWritable())
return false;
{
std::ostringstream str;
str << "ppgtt: " << std::hex << std::showbase << gpuAddress << " end address: " << gpuAddress + size << " cpu address: " << cpuAddress << " device mask: " << gfxAllocation.devicesBitfield << " size: " << std::dec << size;
getAubStream()->addComment(str.str().c_str());
}
if (cpuAddress == nullptr) {
DEBUG_BREAK_IF(gfxAllocation.isLocked());
cpuAddress = this->getMemoryManager()->lockResource(&gfxAllocation);
gfxAllocation.setLocked(true);
}
{
std::ostringstream str;
str << "ppgtt: " << std::hex << std::showbase << gpuAddress << " end address: " << gpuAddress + size << " cpu address: " << cpuAddress << " device mask: " << gfxAllocation.devicesBitfield << " size: " << std::dec << size;
getAubStream()->addComment(str.str().c_str());
}
AubHelperHw<GfxFamily> aubHelperHw(this->localMemoryEnabled);

4
runtime/execution_environment/execution_environment.cpp
View File

@@ -23,9 +23,9 @@ ExecutionEnvironment::ExecutionEnvironment() = default;
ExecutionEnvironment::~ExecutionEnvironment() = default;
extern CommandStreamReceiver *createCommandStream(const HardwareInfo *pHwInfo, ExecutionEnvironment &executionEnvironment);
void ExecutionEnvironment::initAubCenter(const HardwareInfo *pHwInfo, bool localMemoryEnabled) {
void ExecutionEnvironment::initAubCenter(const HardwareInfo *pHwInfo, bool localMemoryEnabled, const std::string &aubFileName) {
if (!aubCenter) {
aubCenter.reset(new AubCenter(pHwInfo, localMemoryEnabled));
aubCenter.reset(new AubCenter(pHwInfo, localMemoryEnabled, aubFileName));
}
}
void ExecutionEnvironment::initGmm(const HardwareInfo *hwInfo) {

2
runtime/execution_environment/execution_environment.h
View File

@@ -37,7 +37,7 @@ class ExecutionEnvironment : public ReferenceTrackedObject<ExecutionEnvironment>
ExecutionEnvironment();
~ExecutionEnvironment() override;
MOCKABLE_VIRTUAL void initAubCenter(const HardwareInfo *hwInfo, bool localMemoryEnabled);
MOCKABLE_VIRTUAL void initAubCenter(const HardwareInfo *hwInfo, bool localMemoryEnabled, const std::string &aubFileName);
void initGmm(const HardwareInfo *hwInfo);
bool initializeCommandStreamReceiver(const HardwareInfo *pHwInfo, uint32_t deviceIndex, uint32_t deviceCsrIndex);
void initializeMemoryManager(bool enable64KBpages, bool enableLocalMemory, uint32_t deviceIndex, uint32_t deviceCsrIndex);

2
runtime/platform/platform.cpp
View File

@@ -168,7 +168,7 @@ bool Platform::initialize() {
CommandStreamReceiverType csrType = this->devices[0]->getEngine(0).commandStreamReceiver->getType();
if (csrType != CommandStreamReceiverType::CSR_HW) {
executionEnvironment->initAubCenter(&hwInfo[0], this->devices[0]->getEnableLocalMemory());
executionEnvironment->initAubCenter(&hwInfo[0], this->devices[0]->getEnableLocalMemory(), "aubfile");
}
this->fillGlobalDispatchTable();