intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[21/n] Remove Instruction Heap from enqueue path.

Browse Source 
- This removes Instruction Heap allocation from enqueue path
- Blocked path is handled as well
- Heap is no longer allocated on demand it is bind to kernelInfo.

Change-Id: I54545beceed3404ee0330a8bac2b0934944cac30
This commit is contained in:
Mrozek, Michal
2018-03-28 19:21:07 +02:00
parent f2b96fa508
commit 2be5934096
21 changed files with 49 additions and 277 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
unit_tests/execution_model/scheduler_dispatch_tests.cpp
View File

@@ -67,11 +67,9 @@ HWTEST_F(ExecutionModelSchedulerFixture, dispatchScheduler) {
EXPECT_NE(nullptr, executionModelDsh);
size_t minRequiredSizeForSchedulerIH = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minRequiredSizeForSchedulerSSH = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
// Setup heaps in pCmdQ
LinearStream &commandStream = getCommandStream<FamilyType, CL_COMMAND_NDRANGE_KERNEL>(*pCmdQ, false, false, &scheduler);
pCmdQ->getIndirectHeap(IndirectHeap::INSTRUCTION, minRequiredSizeForSchedulerIH);
pCmdQ->getIndirectHeap(IndirectHeap::SURFACE_STATE, minRequiredSizeForSchedulerSSH);
dispatchScheduler<FamilyType>(
@@ -185,11 +183,9 @@ HWTEST_F(ExecutionModelSchedulerFixture, dispatchSchedulerDoesNotUseStandardCmdQ
DeviceQueueHw<FamilyType> *pDevQueueHw = castToObject<DeviceQueueHw<FamilyType>>(pDevQueue);
SchedulerKernel &scheduler = BuiltIns::getInstance().getSchedulerKernel(*context);
size_t minRequiredSizeForSchedulerIH = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minRequiredSizeForSchedulerSSH = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
// Setup heaps in pCmdQ
getCommandStream<FamilyType, CL_COMMAND_NDRANGE_KERNEL>(*pCmdQ, false, false, &scheduler);
pCmdQ->getIndirectHeap(IndirectHeap::INSTRUCTION, minRequiredSizeForSchedulerIH);
pCmdQ->getIndirectHeap(IndirectHeap::SURFACE_STATE, minRequiredSizeForSchedulerSSH);
dispatchScheduler<FamilyType>(
@@ -218,11 +214,9 @@ HWTEST_F(ParentKernelCommandQueueFixture, dispatchSchedulerWithEarlyReturnSetToF
SchedulerKernel &scheduler = BuiltIns::getInstance().getSchedulerKernel(*context);
size_t minRequiredSizeForSchedulerIH = KernelCommandsHelper<FamilyType>::getSizeRequiredIH(scheduler);
size_t minRequiredSizeForSchedulerSSH = KernelCommandsHelper<FamilyType>::getSizeRequiredSSH(scheduler);
// Setup heaps in pCmdQ
LinearStream &commandStream = getCommandStream<FamilyType, CL_COMMAND_NDRANGE_KERNEL>(*pCmdQ, false, false, &scheduler);
pCmdQ->getIndirectHeap(IndirectHeap::INSTRUCTION, minRequiredSizeForSchedulerIH);
pCmdQ->getIndirectHeap(IndirectHeap::SURFACE_STATE, minRequiredSizeForSchedulerSSH);
dispatchScheduler<FamilyType>(

25
unit_tests/execution_model/submit_blocked_parent_kernel_tests.cpp
View File

@@ -65,9 +65,9 @@ class MockDeviceQueueHwWithCriticalSectionRelease : public DeviceQueueHw<GfxFami
return igilCmdQueue->m_controls.m_CriticalSection == DeviceQueueHw<GfxFamily>::ExecutionModelCriticalSection::Free;
}
void setupIndirectState(IndirectHeap &instructionHeap, IndirectHeap &surfaceStateHeap, Kernel *parentKernel, uint32_t parentIDCount) override {
void setupIndirectState(IndirectHeap &surfaceStateHeap, Kernel *parentKernel, uint32_t parentIDCount) override {
indirectStateSetup = true;
return BaseClass::setupIndirectState(instructionHeap, surfaceStateHeap, parentKernel, parentIDCount);
return BaseClass::setupIndirectState(surfaceStateHeap, parentKernel, parentIDCount);
}
void addExecutionModelCleanUpSection(Kernel *parentKernel, HwTimeStamps *hwTimeStamp, uint32_t taskCount) override {
cleanupSectionAdded = true;
@@ -103,16 +103,13 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenLockedEMcritcalSectionWhenParentK
IndirectHeap *dsh = new IndirectHeap(alignedMalloc(dshSize, alignement), dshSize);
dsh->getSpace(mockDevQueue.getDshOffset());
size_t minSizeISHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minSizeSSHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
KernelOperation *blockedCommandData = new KernelOperation(std::unique_ptr<LinearStream>(new LinearStream()),
std::unique_ptr<IndirectHeap>(dsh),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)));
blockedCommandData->instructionHeapSizeEM = minSizeISHForEM;
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
std::vector<Surface *> surfaces;
@@ -166,13 +163,10 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenParentKernelWhenCommandIsSubmitte
KernelOperation *blockedCommandData = new KernelOperation(std::unique_ptr<LinearStream>(new LinearStream()),
std::unique_ptr<IndirectHeap>(dsh),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)));
size_t minSizeISHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minSizeSSHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
blockedCommandData->instructionHeapSizeEM = minSizeISHForEM;
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
std::vector<Surface *> surfaces;
@@ -211,13 +205,10 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenParentKernelWhenCommandIsSubmitte
KernelOperation *blockedCommandData = new KernelOperation(std::unique_ptr<LinearStream>(new LinearStream()),
std::unique_ptr<IndirectHeap>(dsh),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)));
size_t minSizeISHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minSizeSSHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
blockedCommandData->instructionHeapSizeEM = minSizeISHForEM;
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
std::vector<Surface *> surfaces;
@@ -251,13 +242,10 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenBlockedParentKernelWithProfilingW
KernelOperation *blockedCommandData = new KernelOperation(std::unique_ptr<LinearStream>(new LinearStream()),
std::unique_ptr<IndirectHeap>(dsh),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)));
size_t minSizeISHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minSizeSSHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
blockedCommandData->instructionHeapSizeEM = minSizeISHForEM;
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
std::vector<Surface *> surfaces;
@@ -294,13 +282,10 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenParentKernelWhenCommandIsSubmitte
KernelOperation *blockedCommandData = new KernelOperation(std::unique_ptr<LinearStream>(new LinearStream()),
std::unique_ptr<IndirectHeap>(dsh),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)));
size_t minSizeISHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minSizeSSHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
blockedCommandData->instructionHeapSizeEM = minSizeISHForEM;
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
std::vector<Surface *> surfaces;
@@ -326,7 +311,6 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenUsedSSHWhenParentKernelIsSubmitte
MockCommandQueue cmdQ(context, device, properties);
size_t minSizeISHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minSizeSSHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
size_t heapSize = 20;
@@ -343,10 +327,8 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenUsedSSHWhenParentKernelIsSubmitte
KernelOperation *blockedCommandData = new KernelOperation(std::unique_ptr<LinearStream>(new LinearStream()),
std::unique_ptr<IndirectHeap>(dsh),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)));
blockedCommandData->instructionHeapSizeEM = minSizeISHForEM;
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
std::vector<Surface *> surfaces;
@@ -370,7 +352,6 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenNotUsedSSHWhenParentKernelIsSubmi
parentKernel->createReflectionSurface();
context->setDefaultDeviceQueue(&mockDevQueue);
size_t minSizeISHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::INSTRUCTION>(*parentKernel);
size_t minSizeSSHForEM = KernelCommandsHelper<FamilyType>::template getSizeRequiredForExecutionModel<IndirectHeap::SURFACE_STATE>(*parentKernel);
size_t heapSize = 20;
@@ -392,10 +373,8 @@ HWTEST_F(ParentKernelCommandQueueFixture, givenNotUsedSSHWhenParentKernelIsSubmi
KernelOperation *blockedCommandData = new KernelOperation(std::unique_ptr<LinearStream>(new LinearStream()),
std::unique_ptr<IndirectHeap>(dsh),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(new IndirectHeap(alignedMalloc(heapSize, alignement), heapSize)),
std::unique_ptr<IndirectHeap>(ssh));
blockedCommandData->instructionHeapSizeEM = minSizeISHForEM;
blockedCommandData->surfaceStateHeapSizeEM = minSizeSSHForEM;
PreemptionMode preemptionMode = device->getPreemptionMode();
std::vector<Surface *> surfaces;