intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-11 16:45:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

Store SSH per root device in Kernel

Browse Source 
Related-To: NEO-5001
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2020-11-23 18:01:38 +00:00
committed by Compute-Runtime-Automation
parent 52d96af5f0
commit 7ec69c33f9
40 changed files with 231 additions and 220 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
opencl/test/unit_test/execution_model/enqueue_execution_model_kernel_tests.cpp
View File

@@ -297,7 +297,7 @@ HWCMDTEST_P(IGFX_GEN8_CORE, ParentKernelEnqueueTest, givenParentKernelWhenEnqueu
BlockKernelManager *blockManager = pProgram->getBlockKernelManager();
uint32_t blockCount = static_cast<uint32_t>(blockManager->getCount());
size_t parentKernelSSHSize = pKernel->getSurfaceStateHeapSize();
size_t parentKernelSSHSize = pKernel->getSurfaceStateHeapSize(rootDeviceIndex);
MockMultiDispatchInfo multiDispatchInfo(pClDevice, pKernel);
@@ -340,7 +340,7 @@ HWCMDTEST_P(IGFX_GEN8_CORE, ParentKernelEnqueueTest, givenParentKernelWhenEnqueu
EXPECT_EQ(0, memcmp(srcSurfaceState, dstSurfaceState, sizeof(RENDER_SURFACE_STATE)));
}
blockSSH = ptrOffset(blockSSH, blockKernel->getSurfaceStateHeapSize());
blockSSH = ptrOffset(blockSSH, blockKernel->getSurfaceStateHeapSize(rootDeviceIndex));
}
delete blockKernel;

4
opencl/test/unit_test/execution_model/parent_kernel_dispatch_tests.cpp
View File

@@ -132,7 +132,7 @@ HWCMDTEST_P(IGFX_GEN8_CORE, ParentKernelDispatchTest, givenParentKernelWhenQueue
EXPECT_LE(pKernel->getKernelInfo().heapInfo.SurfaceStateHeapSize, ssh.getMaxAvailableSpace());
size_t minRequiredSize = HardwareCommandsHelper<FamilyType>::getTotalSizeRequiredSSH(multiDispatchInfo);
size_t minRequiredSizeForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*pKernel);
size_t minRequiredSizeForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*pKernel, rootDeviceIndex);
EXPECT_LE(minRequiredSize + minRequiredSizeForEM, ssh.getMaxAvailableSpace());
}
@@ -162,7 +162,7 @@ HWCMDTEST_P(IGFX_GEN8_CORE, ParentKernelDispatchTest, givenParentKernelWhenQueue
ASSERT_NE(nullptr, blockedCommandsData);
size_t minRequiredSize = HardwareCommandsHelper<FamilyType>::getTotalSizeRequiredSSH(multiDispatchInfo) + UnitTestHelper<FamilyType>::getDefaultSshUsage();
size_t minRequiredSizeForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*pKernel);
size_t minRequiredSizeForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*pKernel, rootDeviceIndex);
size_t sshUsed = blockedCommandsData->ssh->getUsed();

6
opencl/test/unit_test/execution_model/scheduler_dispatch_tests.cpp
View File

@@ -53,7 +53,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ExecutionModelSchedulerFixture, WhenDispatchingSched
EXPECT_NE(nullptr, executionModelDsh);
size_t minRequiredSizeForSchedulerSSH = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minRequiredSizeForSchedulerSSH = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
// Setup heaps in pCmdQ
MultiDispatchInfo multiDispatchinfo(&scheduler);
LinearStream &commandStream = getCommandStream<FamilyType, CL_COMMAND_NDRANGE_KERNEL>(*pCmdQ, CsrDependencies(),
@@ -174,7 +174,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ExecutionModelSchedulerFixture, WhenDispatchingSched
DeviceQueueHw<FamilyType> *pDevQueueHw = castToObject<DeviceQueueHw<FamilyType>>(pDevQueue);
SchedulerKernel &scheduler = context->getSchedulerKernel();
size_t minRequiredSizeForSchedulerSSH = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minRequiredSizeForSchedulerSSH = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
// Setup heaps in pCmdQ
MultiDispatchInfo multiDispatchinfo(&scheduler);
@@ -209,7 +209,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ParentKernelCommandQueueFixture, GivenEarlyReturnSet
SchedulerKernel &scheduler = context->getSchedulerKernel();
size_t minRequiredSizeForSchedulerSSH = HardwareCommandsHelper<FamilyType>::getSizeRequiredSSH(scheduler);
size_t minRequiredSizeForSchedulerSSH = HardwareCommandsHelper<FamilyType>::getSizeRequiredSSH(scheduler, rootDeviceIndex);
// Setup heaps in pCmdQ
MultiDispatchInfo multiDispatchinfo(&scheduler);
LinearStream &commandStream = getCommandStream<FamilyType, CL_COMMAND_NDRANGE_KERNEL>(*pCmdQ, CsrDependencies(),

14
opencl/test/unit_test/execution_model/submit_blocked_parent_kernel_tests.cpp
View File

@@ -97,7 +97,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ParentKernelCommandQueueFixture, givenLockedEMcritca
dsh->getSpace(mockDevQueue.getDshOffset());
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
auto cmdStreamAllocation = device->getMemoryManager()->allocateGraphicsMemoryWithProperties({device->getRootDeviceIndex(), 4096, GraphicsAllocation::AllocationType::COMMAND_BUFFER, device->getDeviceBitfield()});
auto blockedCommandData = std::make_unique<KernelOperation>(new LinearStream(cmdStreamAllocation),
@@ -162,7 +162,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ParentKernelCommandQueueFixture, givenParentKernelWh
*pCmdQ->getGpgpuCommandStreamReceiver().getInternalAllocationStorage());
blockedCommandData->setHeaps(dsh, ioh, ssh);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
@@ -203,7 +203,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ParentKernelCommandQueueFixture, givenParentKernelWh
*pCmdQ->getGpgpuCommandStreamReceiver().getInternalAllocationStorage());
blockedCommandData->setHeaps(dsh, ioh, ssh);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
@@ -241,7 +241,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ParentKernelCommandQueueFixture, givenBlockedParentK
*pCmdQ->getGpgpuCommandStreamReceiver().getInternalAllocationStorage());
blockedCommandData->setHeaps(dsh, ioh, ssh);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
@@ -282,7 +282,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ParentKernelCommandQueueFixture, givenParentKernelWh
*pCmdQ->getGpgpuCommandStreamReceiver().getInternalAllocationStorage());
blockedCommandData->setHeaps(dsh, ioh, ssh);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
@@ -308,7 +308,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ParentKernelCommandQueueFixture, givenUsedCommandQue
MockCommandQueue cmdQ(context, device, properties);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
size_t heapSize = 20;
@@ -362,7 +362,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, ParentKernelCommandQueueFixture, givenNotUsedSSHWhen
parentKernel->createReflectionSurface();
context->setDefaultDeviceQueue(&mockDevQueue);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel);
size_t minSizeSSHForEM = HardwareCommandsHelper<FamilyType>::getSshSizeForExecutionModel(*parentKernel, rootDeviceIndex);
size_t heapSize = 20;