Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Prefetch kmd-migrated shared allocation to multiple subdevices

Browse Source 
Related-To: NEO-6740

Signed-off-by: Slawomir Milczarek <slawomir.milczarek@intel.com>
This commit is contained in:
Slawomir Milczarek
2022-11-16 23:43:35 +00:00
committed by Compute-Runtime-Automation
parent 145f249f0c
commit ba003dd7dc
10 changed files with 100 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
shared/source/debug_settings/debug_variables_base.inl
View File

@ -254,6 +254,7 @@ DECLARE_DEBUG_VARIABLE(bool, PrintTimestampPacketContents, false, "prints all ti
DECLARE_DEBUG_VARIABLE(bool, WddmResidencyLogger, false, "gather Wddm residency statistics to file")
DECLARE_DEBUG_VARIABLE(bool, PrintBOCreateDestroyResult, false, "tracks the result of creation and destruction of BOs")
DECLARE_DEBUG_VARIABLE(bool, PrintBOBindingResult, false, "tracks the result of binding and unbinding of BOs")
DECLARE_DEBUG_VARIABLE(bool, PrintBOPrefetchingResult, false, "tracks the result of prefetching BOs")
DECLARE_DEBUG_VARIABLE(bool, PrintTagAllocationAddress, false, "Print tag allocation address for each engine")
DECLARE_DEBUG_VARIABLE(bool, ProvideVerboseImplicitFlush, false, "provides verbose messages about implicit flush mechanism")
DECLARE_DEBUG_VARIABLE(bool, PrintBlitDispatchDetails, false, "Print blit dispatch details")

4
shared/source/memory_manager/graphics_allocation.cpp
View File

@ -116,6 +116,10 @@ void GraphicsAllocation::prepareHostPtrForResidency(CommandStreamReceiver *csr)
}
}
uint32_t GraphicsAllocation::getNumHandlesForKmdSharedAllocation(uint32_t numBanks) {
return (numBanks > 1) && (DebugManager.flags.CreateKmdMigratedSharedAllocationWithMultipleBOs.get() != 0) ? numBanks : 1u;
}
constexpr uint32_t GraphicsAllocation::objectNotUsed;
constexpr uint32_t GraphicsAllocation::objectNotResident;
constexpr uint32_t GraphicsAllocation::objectAlwaysResident;

2
shared/source/memory_manager/graphics_allocation.h
View File

@ -217,6 +217,8 @@ class GraphicsAllocation : public IDNode<GraphicsAllocation> {
type == AllocationType::DEBUG_SBA_TRACKING_BUFFER;
}
static uint32_t getNumHandlesForKmdSharedAllocation(uint32_t numBanks);
void *getReservedAddressPtr() const {
return this->reservedAddressRangeInfo.addressPtr;
}

32
shared/source/os_interface/linux/drm_allocation.cpp
View File

@ -179,6 +179,21 @@ int DrmAllocation::bindBOs(OsContext *osContext, uint32_t vmHandleId, std::vecto
return 0;
}
bool DrmAllocation::prefetchBO(BufferObject *bo, uint32_t subDeviceId) {
auto drm = bo->peekDrm();
auto ioctlHelper = drm->getIoctlHelper();
auto memoryClassDevice = ioctlHelper->getDrmParamValue(DrmParam::MemoryClassDevice);
auto region = static_cast<uint32_t>((memoryClassDevice << 16u) | subDeviceId);
auto vmId = drm->getVirtualMemoryAddressSpace(subDeviceId);
auto result = ioctlHelper->setVmPrefetch(bo->peekAddress(), bo->peekSize(), region, vmId);
PRINT_DEBUG_STRING(DebugManager.flags.PrintBOPrefetchingResult.get(), stdout,
"prefetch BO=%d to VM %u, range: %llx - %llx, size: %lld, region: %x, result: %d\n",
bo->peekHandle(), vmId, bo->peekAddress(), ptrOffset(bo->peekAddress(), bo->peekSize()), bo->peekSize(), region, result);
return result;
}
void DrmAllocation::registerBOBindExtHandle(Drm *drm) {
if (!drm->resourceRegistrationEnabled()) {
return;
@ -318,15 +333,18 @@ bool DrmAllocation::setMemAdvise(Drm *drm, MemAdviseFlags flags) {
bool DrmAllocation::setMemPrefetch(Drm *drm, uint32_t subDeviceId) {
bool success = true;
auto ioctlHelper = drm->getIoctlHelper();
auto memoryClassDevice = ioctlHelper->getDrmParamValue(DrmParam::MemoryClassDevice);
auto vmId = drm->getVirtualMemoryAddressSpace(subDeviceId);
auto numHandles = GraphicsAllocation::getNumHandlesForKmdSharedAllocation(storageInfo.getNumBanks());
for (auto bo : bufferObjects) {
if (bo != nullptr) {
auto region = static_cast<uint32_t>((memoryClassDevice << 16u) | subDeviceId);
success &= ioctlHelper->setVmPrefetch(bo->peekAddress(), bo->peekSize(), region, vmId);
if (numHandles > 1) {
for (uint8_t subDeviceId = 0u; subDeviceId < EngineLimits::maxHandleCount; subDeviceId++) {
if (storageInfo.memoryBanks.test(subDeviceId)) {
auto bo = this->getBOs()[subDeviceId];
success &= prefetchBO(bo, subDeviceId);
}
}
} else {
auto bo = this->getBO();
success = prefetchBO(bo, subDeviceId);
}
return success;

1
shared/source/os_interface/linux/drm_allocation.h
View File

@ -109,6 +109,7 @@ class DrmAllocation : public GraphicsAllocation {
MOCKABLE_VIRTUAL int makeBOsResident(OsContext *osContext, uint32_t vmHandleId, std::vector<BufferObject *> *bufferObjects, bool bind);
MOCKABLE_VIRTUAL int bindBO(BufferObject *bo, OsContext *osContext, uint32_t vmHandleId, std::vector<BufferObject *> *bufferObjects, bool bind);
MOCKABLE_VIRTUAL int bindBOs(OsContext *osContext, uint32_t vmHandleId, std::vector<BufferObject *> *bufferObjects, bool bind);
MOCKABLE_VIRTUAL bool prefetchBO(BufferObject *bo, uint32_t subDeviceId);
MOCKABLE_VIRTUAL void registerBOBindExtHandle(Drm *drm);
void freeRegisteredBOBindExtHandles(Drm *drm);
void linkWithRegisteredHandle(uint32_t handle);

11
shared/source/os_interface/linux/drm_memory_manager.cpp
View File

@ -1859,15 +1859,12 @@ GraphicsAllocation *DrmMemoryManager::createSharedUnifiedMemoryAllocation(const
BufferObjects bos{};
auto currentAddress = cpuPointer;
auto remainingSize = size;
auto getNumHandles = [](uint32_t numBanks) -> uint32_t {
return (numBanks > 1) && (DebugManager.flags.CreateKmdMigratedSharedAllocationWithMultipleBOs.get() != 0) ? numBanks : 1u;
};
auto numHandles = GraphicsAllocation::getNumHandlesForKmdSharedAllocation(allocationData.storageInfo.getNumBanks());
auto handles = getNumHandles(allocationData.storageInfo.getNumBanks());
for (auto handleId = 0u; handleId < handles; handleId++) {
for (auto handleId = 0u; handleId < numHandles; handleId++) {
uint32_t handle = 0;
auto currentSize = alignUp(remainingSize / (handles - handleId), MemoryConstants::pageSize64k);
auto currentSize = alignUp(remainingSize / (numHandles - handleId), MemoryConstants::pageSize64k);
if (currentSize == 0) {
break;
}
@ -1917,7 +1914,7 @@ GraphicsAllocation *DrmMemoryManager::createSharedUnifiedMemoryAllocation(const
}
return nullptr;
}
if (handles > 1) {
if (numHandles > 1) {
allocation->storageInfo = allocationData.storageInfo;
}

15
shared/test/common/mocks/linux/mock_drm_allocation.h
View File

@ -42,7 +42,12 @@ class MockDrmAllocation : public DrmAllocation {
using DrmAllocation::memoryPool;
using DrmAllocation::registeredBoBindHandles;
MockDrmAllocation(AllocationType allocationType, MemoryPool pool) : DrmAllocation(0, allocationType, nullptr, nullptr, 0, static_cast<size_t>(0), pool) {
MockDrmAllocation(AllocationType allocationType, MemoryPool pool)
: DrmAllocation(0, allocationType, nullptr, nullptr, 0, static_cast<size_t>(0), pool) {
}
MockDrmAllocation(AllocationType allocationType, MemoryPool pool, BufferObjects &bos)
: DrmAllocation(0, 0, allocationType, bos, nullptr, 0, static_cast<size_t>(0), pool) {
}
void registerBOBindExtHandle(Drm *drm) override {
@ -61,12 +66,20 @@ class MockDrmAllocation : public DrmAllocation {
return bindBOsRetValue;
}
bool prefetchBO(BufferObject *bo, uint32_t subDeviceId) override {
prefetchBOCalled = true;
subDeviceIdsReceived.push_back(subDeviceId);
return DrmAllocation::prefetchBO(bo, subDeviceId);
}
ADDMETHOD_NOBASE(makeBOsResident, int, 0, (OsContext * osContext, uint32_t vmHandleId, std::vector<BufferObject *> *bufferObjects, bool bind));
bool registerBOBindExtHandleCalled = false;
bool markedForCapture = false;
bool bindBOsCalled = false;
int bindBOsRetValue = 0;
bool prefetchBOCalled = false;
std::vector<uint32_t> subDeviceIdsReceived;
};
} // namespace NEO

1
shared/test/common/test_files/igdrcl.config
View File

@ -91,6 +91,7 @@ PrintTimestampPacketContents = 0
WddmResidencyLogger = 0
PrintBOCreateDestroyResult = 0
PrintBOBindingResult = 0
PrintBOPrefetchingResult = 0
PrintDriverDiagnostics = -1
PrintDeviceAndEngineIdOnSubmission = 0
EnableDirectSubmission = -1

14
shared/test/unit_test/memory_manager/graphics_allocation_tests.cpp
View File

@ -5,6 +5,7 @@
*
*/
#include "shared/test/common/helpers/debug_manager_state_restore.h"
#include "shared/test/common/mocks/mock_aub_csr.h"
#include "shared/test/common/mocks/mock_command_stream_receiver.h"
#include "shared/test/common/mocks/mock_execution_environment.h"
@ -197,6 +198,19 @@ TEST(GraphicsAllocationTest, whenAllocationTypeIsImageThenAllocationIsNotLockabl
EXPECT_FALSE(GraphicsAllocation::isLockable(AllocationType::IMAGE));
}
TEST(GraphicsAllocationTest, givenNumMemoryBanksWhenGettingNumHandlesForKmdSharedAllocationThenReturnCorrectValue) {
DebugManagerStateRestore restore;
EXPECT_EQ(1u, GraphicsAllocation::getNumHandlesForKmdSharedAllocation(1));
EXPECT_EQ(2u, GraphicsAllocation::getNumHandlesForKmdSharedAllocation(2));
DebugManager.flags.CreateKmdMigratedSharedAllocationWithMultipleBOs.set(0);
EXPECT_EQ(1u, GraphicsAllocation::getNumHandlesForKmdSharedAllocation(2));
DebugManager.flags.CreateKmdMigratedSharedAllocationWithMultipleBOs.set(1);
EXPECT_EQ(2u, GraphicsAllocation::getNumHandlesForKmdSharedAllocation(2));
}
TEST(GraphicsAllocationTest, givenDefaultAllocationWhenGettingNumHandlesThenOneIsReturned) {
MockGraphicsAllocation graphicsAllocation;
EXPECT_EQ(1u, graphicsAllocation.getNumGmms());

35
shared/test/unit_test/os_interface/linux/drm_memory_manager_tests.cpp
View File

@ -4162,7 +4162,35 @@ TEST_F(DrmMemoryManagerTest, givenDrmMemoryManagerWhenSetMemAdviseIsCalledThenUp
}
}
TEST_F(DrmMemoryManagerTest, givenDrmMemoryManagerWhenSetMemPrefetchIsCalledThenReturnTrue) {
TEST_F(DrmMemoryManagerTest, givenDrmMemoryManagerWhenSetMemPrefetchIsCalledThenPrefetchBOsToMultipleSubDevices) {
TestedDrmMemoryManager memoryManager(false, false, false, *executionEnvironment);
BufferObject bo0(mock, 3, 1, 1024, 0);
BufferObject bo1(mock, 3, 2, 1024, 0);
BufferObjects bos{&bo0, &bo1};
MockDrmAllocation drmAllocation(AllocationType::UNIFIED_SHARED_MEMORY, MemoryPool::LocalMemory, bos);
drmAllocation.storageInfo.memoryBanks = 0x3;
memoryManager.registeredEngines = EngineControlContainer{this->device->allEngines};
for (auto engine : memoryManager.registeredEngines) {
engine.osContext->incRefInternal();
}
EXPECT_TRUE(memoryManager.setMemPrefetch(&drmAllocation, 0, rootDeviceIndex));
EXPECT_TRUE(drmAllocation.bindBOsCalled);
EXPECT_TRUE(drmAllocation.prefetchBOCalled);
ASSERT_EQ(2u, drmAllocation.subDeviceIdsReceived.size());
for (uint32_t subDeviceId = 0; subDeviceId < drmAllocation.subDeviceIdsReceived.size(); subDeviceId++) {
EXPECT_EQ(subDeviceId, drmAllocation.subDeviceIdsReceived[subDeviceId]);
}
}
TEST_F(DrmMemoryManagerTest, givenCreateKmdMigratedSharedAllocationWithMultipleBOsSetToFalseWhenSetMemPrefetchIsCalledThenPrefetchBOToSubdevice) {
DebugManagerStateRestore restorer;
DebugManager.flags.CreateKmdMigratedSharedAllocationWithMultipleBOs.set(false);
TestedDrmMemoryManager memoryManager(false, false, false, *executionEnvironment);
BufferObject bo(mock, 3, 1, 1024, 0);
@ -4177,6 +4205,10 @@ TEST_F(DrmMemoryManagerTest, givenDrmMemoryManagerWhenSetMemPrefetchIsCalledThen
EXPECT_TRUE(memoryManager.setMemPrefetch(&drmAllocation, 0, rootDeviceIndex));
EXPECT_TRUE(drmAllocation.bindBOsCalled);
EXPECT_TRUE(drmAllocation.prefetchBOCalled);
ASSERT_EQ(1u, drmAllocation.subDeviceIdsReceived.size());
EXPECT_EQ(0u, drmAllocation.subDeviceIdsReceived[0]);
}
TEST_F(DrmMemoryManagerTest, givenDrmMemoryManagerWhenSetMemPrefetchFailsToBindBufferObjectThenReturnFalse) {
@ -4196,6 +4228,7 @@ TEST_F(DrmMemoryManagerTest, givenDrmMemoryManagerWhenSetMemPrefetchFailsToBindB
EXPECT_FALSE(memoryManager.setMemPrefetch(&drmAllocation, 0, rootDeviceIndex));
EXPECT_TRUE(drmAllocation.bindBOsCalled);
EXPECT_FALSE(drmAllocation.prefetchBOCalled);
}
TEST_F(DrmMemoryManagerTest, givenPageFaultIsUnSupportedWhenCallingBindBoOnBufferAllocationThenAllocationShouldNotPageFaultAndExplicitResidencyIsNotRequired) {