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

Add missing download allocation calls

Browse Source 
Signed-off-by: Jobczyk, Lukasz <lukasz.jobczyk@intel.com>
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Jobczyk, Lukasz
2022-03-29 15:31:51 +00:00
committed by Compute-Runtime-Automation
parent 01e76998d4
commit a285712cc4
7 changed files with 59 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
shared/source/command_stream/command_stream_receiver.cpp
View File

@ -169,6 +169,7 @@ void CommandStreamReceiver::makeResidentHostPtrAllocation(GraphicsAllocation *gf
WaitStatus CommandStreamReceiver::waitForTaskCount(uint32_t requiredTaskCount) {
auto address = getTagAddress();
if (address) {
this->downloadTagAllocation();
return baseWaitFunction(address, WaitParams{false, false, 0}, requiredTaskCount);
}
@ -277,7 +278,7 @@ void CommandStreamReceiver::cleanupResources() {
}
if (tagsMultiAllocation) {
//Null tag address to prevent waiting for tag update when freeing it
// Null tag address to prevent waiting for tag update when freeing it
tagAllocation = nullptr;
tagAddress = nullptr;
DEBUG_BREAK_IF(tagAllocation != nullptr);
@ -461,6 +462,12 @@ AubSubCaptureStatus CommandStreamReceiver::checkAndActivateAubSubCapture(const s
void CommandStreamReceiver::addAubComment(const char *comment) {}
void CommandStreamReceiver::downloadAllocation(GraphicsAllocation &gfxAllocation) {
if (this->downloadAllocationImpl) {
this->downloadAllocationImpl(gfxAllocation);
}
}
void CommandStreamReceiver::startControllingDirectSubmissions() {
auto controller = this->executionEnvironment.directSubmissionController.get();
if (controller) {
@ -805,7 +812,14 @@ bool CommandStreamReceiver::checkImplicitFlushForGpuIdle() {
return false;
}
void CommandStreamReceiver::downloadTagAllocation() {
if (this->getTagAllocation()) {
this->downloadAllocation(*this->getTagAllocation());
}
}
bool CommandStreamReceiver::testTaskCountReady(volatile uint32_t *pollAddress, uint32_t taskCountToWait) {
this->downloadTagAllocation();
for (uint32_t i = 0; i < activePartitions; i++) {
if (!WaitUtils::waitFunction(pollAddress, taskCountToWait)) {
return false;

4
shared/source/command_stream/command_stream_receiver.h
View File

@ -235,7 +235,7 @@ class CommandStreamReceiver {
return scratchSpaceController.get();
}
virtual void downloadAllocation(GraphicsAllocation &gfxAllocation){};
void downloadAllocation(GraphicsAllocation &gfxAllocation);
void registerInstructionCacheFlush() {
auto mutex = obtainUniqueOwnership();
@ -324,6 +324,7 @@ class CommandStreamReceiver {
void printDeviceIndex();
void checkForNewResources(uint32_t submittedTaskCount, uint32_t allocationTaskCount, GraphicsAllocation &gfxAllocation);
bool checkImplicitFlushForGpuIdle();
void downloadTagAllocation();
MOCKABLE_VIRTUAL std::unique_lock<MutexType> obtainHostPtrSurfaceCreationLock();
std::unique_ptr<FlushStampTracker> flushStamp;
@ -356,6 +357,7 @@ class CommandStreamReceiver {
SpinLock debugPauseStateLock;
static void *asyncDebugBreakConfirmation(void *arg);
std::function<void()> debugConfirmationFunction = []() { std::cin.get(); };
std::function<void(GraphicsAllocation &)> downloadAllocationImpl;
GraphicsAllocation *tagAllocation = nullptr;
GraphicsAllocation *globalFenceAllocation = nullptr;

2
shared/source/command_stream/tbx_command_stream_receiver_hw.h
View File

@ -46,7 +46,7 @@ class TbxCommandStreamReceiverHw : public CommandStreamReceiverSimulatedHw<GfxFa
WaitStatus waitForTaskCountWithKmdNotifyFallback(uint32_t taskCountToWait, FlushStamp flushStampToWait, bool useQuickKmdSleep, QueueThrottle throttle) override;
WaitStatus waitForCompletionWithTimeout(const WaitParams &params, uint32_t taskCountToWait) override;
void downloadAllocations() override;
void downloadAllocation(GraphicsAllocation &gfxAllocation) override;
void downloadAllocationTbx(GraphicsAllocation &gfxAllocation);
void processEviction() override;
void processResidency(const ResidencyContainer &allocationsForResidency, uint32_t handleId) override;

15
shared/source/command_stream/tbx_command_stream_receiver_hw.inl
View File

@ -55,10 +55,15 @@ TbxCommandStreamReceiverHw<GfxFamily>::TbxCommandStreamReceiverHw(ExecutionEnvir
? this->peekHwInfo().capabilityTable.aubDeviceId
: static_cast<uint32_t>(debugDeviceId);
this->stream = &tbxStream;
this->downloadAllocationImpl = [this](GraphicsAllocation &graphicsAllocation) {
this->downloadAllocationTbx(graphicsAllocation);
};
}
template <typename GfxFamily>
TbxCommandStreamReceiverHw<GfxFamily>::~TbxCommandStreamReceiverHw() {
this->downloadAllocationImpl = nullptr;
if (streamInitialized) {
tbxStream.close();
}
@ -480,14 +485,14 @@ void TbxCommandStreamReceiverHw<GfxFamily>::flushSubmissionsAndDownloadAllocatio
volatile uint32_t *pollAddress = this->getTagAddress();
for (uint32_t i = 0; i < this->activePartitions; i++) {
while (*pollAddress < this->latestFlushedTaskCount) {
downloadAllocation(*this->getTagAllocation());
this->downloadAllocation(*this->getTagAllocation());
}
pollAddress = ptrOffset(pollAddress, this->postSyncWriteOffset);
}
auto lockCSR = this->obtainUniqueOwnership();
for (GraphicsAllocation *graphicsAllocation : this->allocationsForDownload) {
downloadAllocation(*graphicsAllocation);
this->downloadAllocation(*graphicsAllocation);
}
this->allocationsForDownload.clear();
}
@ -527,7 +532,7 @@ void TbxCommandStreamReceiverHw<GfxFamily>::processResidency(const ResidencyCont
}
template <typename GfxFamily>
void TbxCommandStreamReceiverHw<GfxFamily>::downloadAllocation(GraphicsAllocation &gfxAllocation) {
void TbxCommandStreamReceiverHw<GfxFamily>::downloadAllocationTbx(GraphicsAllocation &gfxAllocation) {
if (hardwareContextController) {
hardwareContextController->readMemory(gfxAllocation.getGpuAddress(), gfxAllocation.getUnderlyingBuffer(), gfxAllocation.getUnderlyingBufferSize(),
this->getMemoryBank(&gfxAllocation), MemoryConstants::pageSize64k);
@ -552,13 +557,13 @@ void TbxCommandStreamReceiverHw<GfxFamily>::downloadAllocations() {
volatile uint32_t *pollAddress = this->getTagAddress();
for (uint32_t i = 0; i < this->activePartitions; i++) {
while (*pollAddress < this->latestFlushedTaskCount) {
downloadAllocation(*this->getTagAllocation());
this->downloadAllocation(*this->getTagAllocation());
}
pollAddress = ptrOffset(pollAddress, this->postSyncWriteOffset);
}
auto lockCSR = this->obtainUniqueOwnership();
for (GraphicsAllocation *graphicsAllocation : this->allocationsForDownload) {
downloadAllocation(*graphicsAllocation);
this->downloadAllocation(*graphicsAllocation);
}
this->allocationsForDownload.clear();
}

10
shared/test/common/libult/ult_command_stream_receiver.h
View File

@ -128,6 +128,12 @@ class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, publ
const DeviceBitfield deviceBitfield)
: BaseClass(executionEnvironment, rootDeviceIndex, deviceBitfield), recursiveLockCounter(0),
recordedDispatchFlags(DispatchFlagsHelper::createDefaultDispatchFlags()) {
this->downloadAllocationImpl = [this](GraphicsAllocation &graphicsAllocation) {
this->downloadAllocationUlt(graphicsAllocation);
};
}
~UltCommandStreamReceiver() {
this->downloadAllocationImpl = nullptr;
}
static CommandStreamReceiver *create(bool withAubDump,
ExecutionEnvironment &executionEnvironment,
@ -171,7 +177,7 @@ class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, publ
downloadAllocationCalled = true;
}
void downloadAllocation(GraphicsAllocation &gfxAllocation) override {
void downloadAllocationUlt(GraphicsAllocation &gfxAllocation) {
downloadAllocationCalled = true;
}
@ -339,7 +345,7 @@ class UltCommandStreamReceiver : public CommandStreamReceiverHw<GfxFamily>, publ
bool recordFlusheBatchBuffer = false;
bool checkAndActivateAubSubCaptureCalled = false;
bool addAubCommentCalled = false;
bool downloadAllocationCalled = false;
std::atomic_bool downloadAllocationCalled = false;
bool flushBatchedSubmissionsCalled = false;
bool initProgrammingFlagsCalled = false;
bool multiOsContextCapable = false;

26
shared/test/common/mocks/mock_tbx_csr.h
View File

@ -25,7 +25,14 @@ class MockTbxCsr : public TbxCommandStreamReceiverHw<GfxFamily> {
using TbxCommandStreamReceiverHw<GfxFamily>::writeMemory;
using TbxCommandStreamReceiverHw<GfxFamily>::allocationsForDownload;
MockTbxCsr(ExecutionEnvironment &executionEnvironment, const DeviceBitfield deviceBitfield)
: TbxCommandStreamReceiverHw<GfxFamily>(executionEnvironment, 0, deviceBitfield) {}
: TbxCommandStreamReceiverHw<GfxFamily>(executionEnvironment, 0, deviceBitfield) {
this->downloadAllocationImpl = [this](GraphicsAllocation &gfxAllocation) {
this->downloadAllocationTbxMock(gfxAllocation);
};
}
~MockTbxCsr() {
this->downloadAllocationImpl = nullptr;
}
void initializeEngine() override {
TbxCommandStreamReceiverHw<GfxFamily>::initializeEngine();
@ -50,8 +57,8 @@ class MockTbxCsr : public TbxCommandStreamReceiverHw<GfxFamily> {
TbxCommandStreamReceiverHw<GfxFamily>::pollForCompletion();
pollForCompletionCalled = true;
}
void downloadAllocation(GraphicsAllocation &gfxAllocation) override {
TbxCommandStreamReceiverHw<GfxFamily>::downloadAllocation(gfxAllocation);
void downloadAllocationTbxMock(GraphicsAllocation &gfxAllocation) {
TbxCommandStreamReceiverHw<GfxFamily>::downloadAllocationTbx(gfxAllocation);
makeCoherentCalled = true;
}
void dumpAllocation(GraphicsAllocation &gfxAllocation) override {
@ -74,9 +81,18 @@ template <typename GfxFamily>
struct MockTbxCsrRegisterDownloadedAllocations : TbxCommandStreamReceiverHw<GfxFamily> {
using CommandStreamReceiver::latestFlushedTaskCount;
using CommandStreamReceiver::tagsMultiAllocation;
using TbxCommandStreamReceiverHw<GfxFamily>::TbxCommandStreamReceiverHw;
using TbxCommandStreamReceiverHw<GfxFamily>::flushSubmissionsAndDownloadAllocations;
void downloadAllocation(GraphicsAllocation &gfxAllocation) override {
MockTbxCsrRegisterDownloadedAllocations(ExecutionEnvironment &executionEnvironment, uint32_t rootDeviceIndex, const DeviceBitfield deviceBitfield)
: TbxCommandStreamReceiverHw<GfxFamily>(executionEnvironment, rootDeviceIndex, deviceBitfield) {
this->downloadAllocationImpl = [this](GraphicsAllocation &gfxAllocation) {
this->downloadAllocationTbxMock(gfxAllocation);
};
}
~MockTbxCsrRegisterDownloadedAllocations() {
this->downloadAllocationImpl = nullptr;
}
void downloadAllocationTbxMock(GraphicsAllocation &gfxAllocation) {
*reinterpret_cast<uint32_t *>(CommandStreamReceiver::getTagAllocation()->getUnderlyingBuffer()) = this->latestFlushedTaskCount;
downloadedAllocations.insert(&gfxAllocation);
}

1
shared/test/unit_test/command_stream/command_stream_receiver_tests.cpp
View File

@ -301,6 +301,7 @@ HWTEST_F(CommandStreamReceiverTest, givenGpuHangWhenWaititingForTaskCountThenGpu
constexpr auto taskCountToWait = 1;
const auto waitStatus = csr.waitForTaskCount(taskCountToWait);
EXPECT_EQ(WaitStatus::GpuHang, waitStatus);
EXPECT_TRUE(csr.downloadAllocationCalled);
}
HWTEST_F(CommandStreamReceiverTest, givenGpuHangAndNonEmptyAllocationsListWhenCallingWaitForTaskCountAndCleanAllocationListThenWaitIsCalledAndGpuHangIsReturned) {