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

Do not bind command buffer separately

Browse Source 
Improve performance by binding the command buffer together with other
allocations if VM_BIND feature is available. Remove the legacy
flag PassBoundBOToExec from DebugManager to simplify the logic.
Adapt unit tests and reuse handy macros to generate proxy mock-methods.

Related-To: NEO-7348
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
This commit is contained in:
Maciej Bielski
2022-09-29 18:31:22 +00:00
committed by Compute-Runtime-Automation
parent e6daa207ad
commit 11eb0aa769
9 changed files with 73 additions and 217 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
opencl/test/unit_test/os_interface/linux/cl_drm_memory_manager_tests.cpp
View File

@ -449,7 +449,7 @@ HWTEST_F(ClDrmMemoryManagerTest, givenDrmMemoryManagerWhenTiledImageIsBeingCreat
device->setPreemptionMode(PreemptionMode::Disabled);
auto csr = static_cast<TestedDrmCommandStreamReceiver<DEFAULT_TEST_FAMILY_NAME> *>(device->getDefaultEngine().commandStreamReceiver);
csr->callHwFlush = false;
csr->flushInternalCallBase = false;
mock->ioctl_expected.gemCreate = 1;
mock->ioctl_expected.gemSetTiling = 1;

64
opencl/test/unit_test/os_interface/linux/drm_command_stream_tests_2.cpp
View File

@ -1400,70 +1400,6 @@ struct MockMergeResidencyContainerMemoryOperationsHandler : public DrmMemoryOper
(OsContext * osContext, ArrayRef<GraphicsAllocation *> gfxAllocations, bool evictable));
};
HWTEST_TEMPLATED_F(DrmCommandStreamEnhancedTest, givenMakeResidentWithinOsContextFailsThenFlushReturnsError) {
struct MockDrmCsr : public DrmCommandStreamReceiver<FamilyType> {
using DrmCommandStreamReceiver<FamilyType>::DrmCommandStreamReceiver;
SubmissionStatus flushInternal(const BatchBuffer &batchBuffer, const ResidencyContainer &allocationsForResidency) override {
return SubmissionStatus::SUCCESS;
}
};
mock->bindAvailable = true;
executionEnvironment->rootDeviceEnvironments[0]->memoryOperationsInterface.reset(new MockMergeResidencyContainerMemoryOperationsHandler());
auto operationHandler = static_cast<MockMergeResidencyContainerMemoryOperationsHandler *>(executionEnvironment->rootDeviceEnvironments[0]->memoryOperationsInterface.get());
operationHandler->makeResidentWithinOsContextResult = NEO::MemoryOperationsStatus::FAILED;
auto osContext = std::make_unique<OsContextLinux>(*mock, 0u,
EngineDescriptorHelper::getDefaultDescriptor(HwHelper::get(defaultHwInfo->platform.eRenderCoreFamily).getGpgpuEngineInstances(*defaultHwInfo)[0],
PreemptionHelper::getDefaultPreemptionMode(*defaultHwInfo)));
auto commandBuffer = mm->allocateGraphicsMemoryWithProperties(MockAllocationProperties{csr->getRootDeviceIndex(), MemoryConstants::pageSize});
LinearStream cs(commandBuffer);
CommandStreamReceiverHw<FamilyType>::addBatchBufferEnd(cs, nullptr);
EncodeNoop<FamilyType>::alignToCacheLine(cs);
BatchBuffer batchBuffer{cs.getGraphicsAllocation(), 0, 0, nullptr, false, false, QueueThrottle::MEDIUM, QueueSliceCount::defaultSliceCount, cs.getUsed(), &cs, nullptr, false};
MockDrmCsr mockCsr(*executionEnvironment, rootDeviceIndex, 1, gemCloseWorkerMode::gemCloseWorkerInactive);
mockCsr.setupContext(*osContext.get());
auto res = mockCsr.flush(batchBuffer, mockCsr.getResidencyAllocations());
EXPECT_GT(operationHandler->makeResidentWithinOsContextCalled, 0u);
EXPECT_EQ(NEO::SubmissionStatus::FAILED, res);
mm->freeGraphicsMemory(commandBuffer);
}
HWTEST_TEMPLATED_F(DrmCommandStreamEnhancedTest, givenMakeResidentWithinOsContextOutOfMemoryThenFlushReturnsError) {
struct MockDrmCsr : public DrmCommandStreamReceiver<FamilyType> {
using DrmCommandStreamReceiver<FamilyType>::DrmCommandStreamReceiver;
SubmissionStatus flushInternal(const BatchBuffer &batchBuffer, const ResidencyContainer &allocationsForResidency) override {
return SubmissionStatus::SUCCESS;
}
};
mock->bindAvailable = true;
executionEnvironment->rootDeviceEnvironments[0]->memoryOperationsInterface.reset(new MockMergeResidencyContainerMemoryOperationsHandler());
auto operationHandler = static_cast<MockMergeResidencyContainerMemoryOperationsHandler *>(executionEnvironment->rootDeviceEnvironments[0]->memoryOperationsInterface.get());
operationHandler->makeResidentWithinOsContextResult = NEO::MemoryOperationsStatus::OUT_OF_MEMORY;
auto osContext = std::make_unique<OsContextLinux>(*mock, 0u,
EngineDescriptorHelper::getDefaultDescriptor(HwHelper::get(defaultHwInfo->platform.eRenderCoreFamily).getGpgpuEngineInstances(*defaultHwInfo)[0],
PreemptionHelper::getDefaultPreemptionMode(*defaultHwInfo)));
auto commandBuffer = mm->allocateGraphicsMemoryWithProperties(MockAllocationProperties{csr->getRootDeviceIndex(), MemoryConstants::pageSize});
LinearStream cs(commandBuffer);
CommandStreamReceiverHw<FamilyType>::addBatchBufferEnd(cs, nullptr);
EncodeNoop<FamilyType>::alignToCacheLine(cs);
BatchBuffer batchBuffer{cs.getGraphicsAllocation(), 0, 0, nullptr, false, false, QueueThrottle::MEDIUM, QueueSliceCount::defaultSliceCount, cs.getUsed(), &cs, nullptr, false};
MockDrmCsr mockCsr(*executionEnvironment, rootDeviceIndex, 1, gemCloseWorkerMode::gemCloseWorkerInactive);
mockCsr.setupContext(*osContext.get());
auto res = mockCsr.flush(batchBuffer, mockCsr.getResidencyAllocations());
EXPECT_GT(operationHandler->makeResidentWithinOsContextCalled, 0u);
EXPECT_EQ(NEO::SubmissionStatus::OUT_OF_MEMORY, res);
mm->freeGraphicsMemory(commandBuffer);
}
HWTEST_TEMPLATED_F(DrmCommandStreamEnhancedTest, givenMergeWithResidencyContainerFailsThenFlushReturnsError) {
struct MockDrmCsr : public DrmCommandStreamReceiver<FamilyType> {
using DrmCommandStreamReceiver<FamilyType>::DrmCommandStreamReceiver;

6
opencl/test/unit_test/os_interface/linux/linux_create_command_queue_with_properties_tests.cpp
View File

@ -116,7 +116,7 @@ HWTEST_F(clCreateCommandQueueWithPropertiesLinux, givenPropertiesWithClQueueSlic
cl_queue_properties properties[] = {CL_QUEUE_SLICE_COUNT_INTEL, newSliceCount, 0};
auto mockCsr = new TestedDrmCommandStreamReceiver<FamilyType>(*mdevice->executionEnvironment, rootDeviceIndex, 1);
mockCsr->callHwFlush = false;
mockCsr->flushInternalCallBase = false;
mdevice->resetCommandStreamReceiver(mockCsr);
cl_command_queue cmdQ = clCreateCommandQueueWithProperties(context.get(), clDevice, properties, &retVal);
@ -162,7 +162,7 @@ HWTEST_F(clCreateCommandQueueWithPropertiesLinux, givenSameSliceCountAsRecentlyS
cl_queue_properties properties[] = {CL_QUEUE_SLICE_COUNT_INTEL, newSliceCount, 0};
auto mockCsr = new TestedDrmCommandStreamReceiver<FamilyType>(*mdevice->executionEnvironment, rootDeviceIndex, 1);
mockCsr->callHwFlush = false;
mockCsr->flushInternalCallBase = false;
mdevice->resetCommandStreamReceiver(mockCsr);
cl_command_queue cmdQ = clCreateCommandQueueWithProperties(context.get(), clDevice, properties, &retVal);
@ -207,7 +207,7 @@ HWTEST_F(clCreateCommandQueueWithPropertiesLinux, givenPropertiesWithClQueueSlic
cl_queue_properties properties[] = {CL_QUEUE_SLICE_COUNT_INTEL, newSliceCount, 0};
auto mockCsr = new TestedDrmCommandStreamReceiver<FamilyType>(*mdevice->executionEnvironment, rootDeviceIndex, 1);
mockCsr->callHwFlush = false;
mockCsr->flushInternalCallBase = false;
mdevice->resetCommandStreamReceiver(mockCsr);
cl_command_queue cmdQ = clCreateCommandQueueWithProperties(context.get(), clDevice, properties, &retVal);