compute-runtime/unit_tests/event/event_tests.cpp

1514 lines
56 KiB
C++
Raw Normal View History

/*
* Copyright (c) 2017 - 2018, Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#include "event_fixture.h"
#include "runtime/event/perf_counter.h"
#include "unit_tests/os_interface/mock_performance_counters.h"
#include "runtime/command_queue/command_queue_hw.h"
#include "runtime/command_stream/command_stream_receiver.h"
#include "runtime/memory_manager/surface.h"
#include "runtime/os_interface/os_interface.h"
#include "runtime/helpers/hw_info.h"
#include "runtime/helpers/task_information.h"
#include "unit_tests/mocks/mock_context.h"
#include "unit_tests/mocks/mock_command_queue.h"
#include "unit_tests/mocks/mock_csr.h"
#include "unit_tests/mocks/mock_event.h"
#include "unit_tests/mocks/mock_program.h"
#include "unit_tests/mocks/mock_kernel.h"
#include "unit_tests/mocks/mock_mdi.h"
#include "unit_tests/helpers/debug_manager_state_restore.h"
#include <memory>
#include <type_traits>
#include "test.h"
TEST(Event, NonCopyable) {
EXPECT_FALSE(std::is_move_constructible<Event>::value);
EXPECT_FALSE(std::is_copy_constructible<Event>::value);
}
TEST(Event, NonAssignable) {
EXPECT_FALSE(std::is_move_assignable<Event>::value);
EXPECT_FALSE(std::is_copy_assignable<Event>::value);
}
TEST(Event, dontUpdateExecutionStatusOnNotReadyEvent) {
MockContext ctx;
MockCommandQueue cmdQ(&ctx, nullptr, 0);
Event event(&cmdQ, CL_COMMAND_NDRANGE_KERNEL, Event::eventNotReady, 0);
EXPECT_FALSE(event.peekIsBlocked());
EXPECT_EQ(CL_QUEUED, event.peekExecutionStatus());
event.updateExecutionStatus();
EXPECT_EQ(CL_QUEUED, event.peekExecutionStatus());
}
TEST(Event, givenEventThatStatusChangeWhenPeekIsCalledThenEventIsNotUpdated) {
MockContext ctx;
MockCommandQueue cmdQ(&ctx, nullptr, 0);
struct mockEvent : public Event {
using Event::Event;
void updateExecutionStatus() override {
callCount++;
}
uint32_t callCount = 0u;
};
mockEvent event(&cmdQ, CL_COMMAND_NDRANGE_KERNEL, Event::eventNotReady, 0);
EXPECT_EQ(0u, event.callCount);
event.peekExecutionStatus();
EXPECT_EQ(0u, event.callCount);
event.updateEventAndReturnCurrentStatus();
EXPECT_EQ(1u, event.callCount);
event.updateEventAndReturnCurrentStatus();
EXPECT_EQ(2u, event.callCount);
}
TEST(Event, givenEventWithHigherTaskCountWhenLowerTaskCountIsBeingSetThenTaskCountRemainsUnmodifed) {
Event *event = new Event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 4, 10);
EXPECT_EQ(10u, event->peekTaskCount());
event->updateTaskCount(8);
EXPECT_EQ(10u, event->peekTaskCount());
delete event;
}
TEST(Event_, testGetTaskLevel) {
class TempEvent : public Event {
public:
TempEvent() : Event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 5, 7){};
uint32_t getTaskLevel() override {
return Event::getTaskLevel();
}
};
TempEvent event;
// taskLevel and getTaskLevel() should give the same result
EXPECT_EQ(5u, event.taskLevel);
EXPECT_EQ(5u, event.getTaskLevel());
}
TEST(Event_, testGetTaskCount) {
Event event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 5, 7);
EXPECT_EQ(7u, event.getCompletionStamp());
}
TEST(Event_, testGetEventInfoReturnsTheCQ) {
auto ctx = std::unique_ptr<Context>(new MockContext());
auto cmdQ = std::unique_ptr<CommandQueue>(new MockCommandQueue(ctx.get(), nullptr, 0));
Event *event = new Event(cmdQ.get(), CL_COMMAND_NDRANGE_KERNEL, 1, 5);
cl_event clEvent = event;
cl_command_queue cmdQResult = nullptr;
size_t sizeReturned = 0;
auto result = clGetEventInfo(clEvent, CL_EVENT_COMMAND_QUEUE, 0, nullptr, &sizeReturned);
EXPECT_EQ(CL_SUCCESS, result);
EXPECT_EQ(sizeof(cl_command_queue), sizeReturned);
result = clGetEventInfo(clEvent, CL_EVENT_COMMAND_QUEUE, sizeof(cmdQResult), &cmdQResult, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
EXPECT_EQ(cmdQ.get(), cmdQResult);
EXPECT_EQ(sizeReturned, sizeof(cmdQResult));
delete event;
}
TEST(Event, givenCommandQueueWhenEventIsCreatedWithCommandQueueThenCommandQueueInternalRefCountIsIncremented) {
MockContext ctx;
MockCommandQueue cmdQ(&ctx, nullptr, 0);
auto intitialRefCount = cmdQ.getRefInternalCount();
Event *event = new Event(&cmdQ, CL_COMMAND_NDRANGE_KERNEL, 4, 10);
auto newRefCount = cmdQ.getRefInternalCount();
EXPECT_EQ(intitialRefCount + 1, newRefCount);
delete event;
auto finalRefCount = cmdQ.getRefInternalCount();
EXPECT_EQ(intitialRefCount, finalRefCount);
}
TEST(Event, givenCommandQueueWhenEventIsCreatedWithoutCommandQueueThenCommandQueueInternalRefCountIsNotModified) {
MockContext ctx;
MockCommandQueue cmdQ(&ctx, nullptr, 0);
auto intitialRefCount = cmdQ.getRefInternalCount();
Event *event = new Event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 4, 10);
auto newRefCount = cmdQ.getRefInternalCount();
EXPECT_EQ(intitialRefCount, newRefCount);
delete event;
auto finalRefCount = cmdQ.getRefInternalCount();
EXPECT_EQ(intitialRefCount, finalRefCount);
}
TEST(Event, currentCmdQVirtualEventSetToFalseInCtor) {
Event *event = new Event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 4, 10);
EXPECT_FALSE(event->isCurrentCmdQVirtualEvent());
delete event;
}
TEST(Event, setCurrentCmdQVirtualEven) {
Event *event = new Event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 4, 10);
event->setCurrentCmdQVirtualEvent(true);
EXPECT_TRUE(event->isCurrentCmdQVirtualEvent());
delete event;
}
TEST(Event, waitForEventsFlushesAllQueues) {
class MockCommandQueueWithFlushCheck : public MockCommandQueue {
public:
MockCommandQueueWithFlushCheck() = delete;
MockCommandQueueWithFlushCheck(MockCommandQueueWithFlushCheck &) = delete;
MockCommandQueueWithFlushCheck(Context &context, Device *device) : MockCommandQueue(&context, device, nullptr) {
}
cl_int flush() override {
flushCounter++;
return CL_SUCCESS;
}
uint32_t flushCounter = 0;
};
std::unique_ptr<Device> device(DeviceHelper<>::create());
MockContext context;
std::unique_ptr<MockCommandQueueWithFlushCheck> cmdQ1(new MockCommandQueueWithFlushCheck(context, device.get()));
std::unique_ptr<Event> event1(new Event(cmdQ1.get(), CL_COMMAND_NDRANGE_KERNEL, 4, 10));
std::unique_ptr<MockCommandQueueWithFlushCheck> cmdQ2(new MockCommandQueueWithFlushCheck(context, device.get()));
std::unique_ptr<Event> event2(new Event(cmdQ2.get(), CL_COMMAND_NDRANGE_KERNEL, 5, 20));
cl_event eventWaitlist[] = {event1.get(), event2.get()};
Event::waitForEvents(2, eventWaitlist);
EXPECT_EQ(1u, cmdQ1->flushCounter);
EXPECT_EQ(1u, cmdQ2->flushCounter);
}
TEST(Event, waitForEventsWithNotReadyEventDoesNotFlushQueue) {
class MockCommandQueueWithFlushCheck : public MockCommandQueue {
public:
MockCommandQueueWithFlushCheck() = delete;
MockCommandQueueWithFlushCheck(MockCommandQueueWithFlushCheck &) = delete;
MockCommandQueueWithFlushCheck(Context &context, Device *device) : MockCommandQueue(&context, device, nullptr) {
}
cl_int flush() override {
flushCounter++;
return CL_SUCCESS;
}
uint32_t flushCounter = 0;
};
std::unique_ptr<Device> device(DeviceHelper<>::create());
MockContext context;
std::unique_ptr<MockCommandQueueWithFlushCheck> cmdQ1(new MockCommandQueueWithFlushCheck(context, device.get()));
std::unique_ptr<Event> event1(new Event(cmdQ1.get(), CL_COMMAND_NDRANGE_KERNEL, Event::eventNotReady, 0));
cl_event eventWaitlist[] = {event1.get()};
Event::waitForEvents(1, eventWaitlist);
EXPECT_EQ(0u, cmdQ1->flushCounter);
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_COMMAND_EXECUTION_STATUS_sizeReturned) {
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 1, 5);
cl_int eventStatus = -1;
size_t sizeReturned = 0;
auto result = clGetEventInfo(&event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(eventStatus), &eventStatus, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
EXPECT_EQ(sizeReturned, sizeof(eventStatus));
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_COMMAND_EXECUTION_STATUS_returns_CL_SUBMITTED_HW_LT_Event) {
uint32_t tagHW = 4;
uint32_t taskCount = 5;
*pTagMemory = tagHW;
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, taskCount);
cl_int eventStatus = -1;
size_t sizeReturned = 0;
auto result = clGetEventInfo(&event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(eventStatus), &eventStatus, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
// If tagCS < taskCount, we always return submitted (ie. no buffering!)
EXPECT_EQ(CL_SUBMITTED, eventStatus);
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_COMMAND_EXECUTION_STATUS_returns_CL_COMPLETE_HW_EQ_Event) {
uint32_t tagHW = 5;
uint32_t taskCount = 5;
*pTagMemory = tagHW;
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, taskCount);
cl_int eventStatus = -1;
size_t sizeReturned = 0;
auto result = clGetEventInfo(&event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(eventStatus), &eventStatus, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
// If tagCS == event.taskCount, the event is completed.
EXPECT_EQ(CL_COMPLETE, eventStatus);
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_COMMAND_EXECUTION_STATUS_returns_CL_SUBMITTED_HW_GT_Event) {
uint32_t tagHW = 6;
uint32_t taskCount = 5;
*pTagMemory = tagHW;
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, taskCount);
cl_int eventStatus = -1;
size_t sizeReturned = 0;
auto result = clGetEventInfo(&event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(eventStatus), &eventStatus, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
// If tagCS > taskCount, the event is not completed.
EXPECT_EQ(CL_COMPLETE, eventStatus);
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_COMMAND_EXECUTION_STATUS_returnsSetStatus) {
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, Event::eventNotReady, Event::eventNotReady);
cl_int eventStatus = -1;
event.setStatus(-1);
auto result = clGetEventInfo(&event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(eventStatus), &eventStatus, 0);
EXPECT_EQ(CL_SUCCESS, result);
EXPECT_EQ(-1, eventStatus);
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_REFERENCE_COUNT_new_Event) {
uint32_t tagEvent = 5;
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, tagEvent);
cl_uint refCount = 0;
size_t sizeReturned = 0;
auto result = clGetEventInfo(&event, CL_EVENT_REFERENCE_COUNT, sizeof(refCount), &refCount, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
EXPECT_EQ(sizeof(refCount), sizeReturned);
EXPECT_EQ(1u, refCount);
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_REFERENCE_COUNT_Retain_Event) {
uint32_t tagEvent = 5;
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, tagEvent);
event.retain();
cl_uint refCount = 0;
size_t sizeReturned = 0;
auto result = clGetEventInfo(&event, CL_EVENT_REFERENCE_COUNT, sizeof(refCount), &refCount, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
EXPECT_EQ(sizeof(refCount), sizeReturned);
EXPECT_EQ(2u, refCount);
event.release();
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_REFERENCE_COUNT_Retain_Release_Event) {
uint32_t tagEvent = 5;
Event *pEvent = new Event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, tagEvent);
ASSERT_NE(nullptr, pEvent);
pEvent->retain();
auto retVal = pEvent->getReference();
EXPECT_EQ(2, retVal);
cl_uint refCount = 0;
size_t sizeReturned = 0;
auto result = clGetEventInfo(pEvent, CL_EVENT_REFERENCE_COUNT, sizeof(refCount), &refCount, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
EXPECT_EQ(sizeof(refCount), sizeReturned);
EXPECT_EQ(2u, refCount);
pEvent->release();
retVal = pEvent->getReference();
EXPECT_EQ(1, retVal);
delete pEvent;
}
TEST_F(EventTest, GetEventInfo_CL_EVENT_CONTEXT) {
uint32_t tagEvent = 5;
Event *pEvent = new Event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, tagEvent);
ASSERT_NE(nullptr, pEvent);
cl_context context;
size_t sizeReturned = 0;
auto result = clGetEventInfo(pEvent, CL_EVENT_CONTEXT, sizeof(context), &context, &sizeReturned);
ASSERT_EQ(CL_SUCCESS, result);
EXPECT_EQ(sizeof(context), sizeReturned);
cl_context qCtx = (cl_context)&mockContext;
EXPECT_EQ(qCtx, context);
delete pEvent;
}
TEST_F(EventTest, GetEventInfo_InvalidParam) {
uint32_t tagEvent = 5;
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, tagEvent);
cl_int eventStatus = -1;
auto result = clGetEventInfo(&event, -1, sizeof(eventStatus), &eventStatus, nullptr);
EXPECT_EQ(CL_INVALID_VALUE, result);
}
TEST_F(EventTest, Event_Wait_NonBlocking) {
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, Event::eventNotReady);
auto result = event.wait(false);
EXPECT_FALSE(result);
}
TEST_F(EventTest, givenEventContainingCommandQueueWhenItsStatusIsUpdatedToCompletedThenTemporaryAllocationsAreDeleted) {
auto memoryManager = pCmdQ->getDevice().getMemoryManager();
void *ptr = (void *)0x1000;
size_t size = 4096;
auto temporary = memoryManager->allocateGraphicsMemory(size, ptr);
temporary->taskCount = 3;
memoryManager->storeAllocation(std::unique_ptr<GraphicsAllocation>(temporary), TEMPORARY_ALLOCATION);
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 3, 3);
EXPECT_EQ(1u, memoryManager->hostPtrManager.getFragmentCount());
event.updateExecutionStatus();
EXPECT_EQ(0u, memoryManager->hostPtrManager.getFragmentCount());
}
class SurfaceMock : public Surface {
public:
SurfaceMock() {
resident = nonResident = completionStamp = 0;
};
~SurfaceMock() override{};
void makeResident(CommandStreamReceiver &csr) override {
if (parent) {
parent->resident++;
} else {
resident++;
}
if (this->graphicsAllocation) {
csr.makeResident(*graphicsAllocation);
}
};
void setCompletionStamp(CompletionStamp &cs, Device *pDevice, CommandQueue *pCmdQ) override {
if (parent) {
parent->completionStamp++;
} else {
completionStamp++;
}
};
Surface *duplicate() override {
return new SurfaceMock(this);
};
SurfaceMock *parent = nullptr;
std::atomic<uint32_t> resident;
std::atomic<uint32_t> nonResident;
std::atomic<uint32_t> completionStamp;
GraphicsAllocation *graphicsAllocation = nullptr;
protected:
SurfaceMock(SurfaceMock *parent) : parent(parent){};
};
TEST_F(InternalsEventTest, resizeCmdQueueHeapsWhenKernelOparationHeapsAreBigger) {
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, 0);
IndirectHeap &cmdQueueDsh = pCmdQ->getIndirectHeap(IndirectHeap::DYNAMIC_STATE, 4096);
IndirectHeap &cmdQueueIsh = pCmdQ->getIndirectHeap(IndirectHeap::INSTRUCTION, 4096);
IndirectHeap &cmdQueueIoh = pCmdQ->getIndirectHeap(IndirectHeap::INDIRECT_OBJECT, 4096);
IndirectHeap &cmdQueueSsh = pCmdQ->getIndirectHeap(IndirectHeap::SURFACE_STATE, 4096);
auto requestedSize = cmdQueueDsh.getMaxAvailableSpace() * 2;
auto cmdStream = new LinearStream(alignedMalloc(requestedSize, requestedSize), requestedSize);
auto createFullHeap = [](size_t size) {
auto heap = new IndirectHeap(alignedMalloc(size, size), size);
heap->getSpace(heap->getAvailableSpace());
return heap;
};
auto dsh = createFullHeap(requestedSize);
auto ish = createFullHeap(requestedSize);
auto ioh = createFullHeap(requestedSize);
auto ssh = createFullHeap(maxSshSize);
using UniqueIH = std::unique_ptr<IndirectHeap>;
auto kernelOperation = new KernelOperation(std::unique_ptr<LinearStream>(cmdStream), UniqueIH(dsh),
UniqueIH(ish), UniqueIH(ioh), UniqueIH(ssh));
std::vector<Surface *> v;
SurfaceMock *surface = new SurfaceMock;
v.push_back(surface);
PreemptionMode preemptionMode = pDevice->getPreemptionMode();
auto cmdComputeKernel = new CommandComputeKernel(*pCmdQ, pDevice->getCommandStreamReceiver(),
std::unique_ptr<KernelOperation>(kernelOperation), v, false, false, false, nullptr, preemptionMode);
EXPECT_LT(cmdQueueDsh.getMaxAvailableSpace(), dsh->getMaxAvailableSpace());
EXPECT_EQ(requestedSize, ish->getMaxAvailableSpace());
EXPECT_LT(cmdQueueIoh.getMaxAvailableSpace(), ioh->getMaxAvailableSpace());
EXPECT_EQ(maxSshSize, ssh->getMaxAvailableSpace());
cmdComputeKernel->submit(0, false);
EXPECT_GE(cmdQueueDsh.getMaxAvailableSpace(), dsh->getMaxAvailableSpace());
EXPECT_GE(cmdQueueIsh.getMaxAvailableSpace(), ish->getMaxAvailableSpace());
EXPECT_GE(cmdQueueIoh.getMaxAvailableSpace(), ioh->getMaxAvailableSpace());
EXPECT_GE(cmdQueueSsh.getMaxAvailableSpace(), ssh->getMaxAvailableSpace());
delete pCmdQ;
delete cmdComputeKernel;
}
TEST_F(InternalsEventTest, processBlockedCommandsKernelOperation) {
MockEvent<Event> event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, 0);
auto cmdStream = new LinearStream(alignedMalloc(4096, 4096), 4096);
auto dsh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ish = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ioh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ssh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
using UniqueIH = std::unique_ptr<IndirectHeap>;
auto blockedCommandsData = new KernelOperation(std::unique_ptr<LinearStream>(cmdStream), UniqueIH(dsh),
UniqueIH(ish), UniqueIH(ioh), UniqueIH(ssh));
auto &csr = pDevice->getCommandStreamReceiver();
std::vector<Surface *> v;
SurfaceMock *surface = new SurfaceMock;
surface->graphicsAllocation = new GraphicsAllocation((void *)0x1234, 100u);
PreemptionMode preemptionMode = pDevice->getPreemptionMode();
v.push_back(surface);
auto cmd = new CommandComputeKernel(*pCmdQ, csr, std::unique_ptr<KernelOperation>(blockedCommandsData), v, false, false, false, nullptr, preemptionMode);
event.setCommand(std::unique_ptr<Command>(cmd));
auto taskLevelBefore = csr.peekTaskLevel();
event.submitCommand(false);
auto taskLevelAfter = csr.peekTaskLevel();
EXPECT_EQ(taskLevelBefore + 1, taskLevelAfter);
delete pCmdQ;
EXPECT_EQ(surface->resident, 1u);
EXPECT_FALSE(surface->graphicsAllocation->isResident());
delete surface->graphicsAllocation;
EXPECT_EQ(surface->completionStamp, 1u);
}
TEST_F(InternalsEventTest, processBlockedCommandsAbortKernelOperation) {
MockEvent<Event> event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, 0);
auto cmdStream = new LinearStream(alignedMalloc(4096, 4096), 4096);
auto dsh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ish = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ioh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ssh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
using UniqueIH = std::unique_ptr<IndirectHeap>;
auto blockedCommandsData = new KernelOperation(std::unique_ptr<LinearStream>(cmdStream), UniqueIH(dsh),
UniqueIH(ish), UniqueIH(ioh), UniqueIH(ssh));
auto &csr = pDevice->getCommandStreamReceiver();
std::vector<Surface *> v;
NullSurface *surface = new NullSurface;
v.push_back(surface);
PreemptionMode preemptionMode = pDevice->getPreemptionMode();
auto cmd = new CommandComputeKernel(*pCmdQ, csr, std::unique_ptr<KernelOperation>(blockedCommandsData), v, false, false, false, nullptr, preemptionMode);
event.setCommand(std::unique_ptr<Command>(cmd));
auto taskLevelBefore = csr.peekTaskLevel();
event.submitCommand(true);
auto taskLevelAfter = csr.peekTaskLevel();
EXPECT_EQ(taskLevelBefore, taskLevelAfter);
delete pCmdQ;
}
TEST_F(InternalsEventTest, givenBlockedKernelWithPrintfWhenSubmittedThenPrintOutput) {
testing::internal::CaptureStdout();
MockEvent<Event> event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, 0);
auto cmdStream = new LinearStream(alignedMalloc(4096, 4096), 4096);
auto dsh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ish = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ioh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ssh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
using UniqueIH = std::unique_ptr<IndirectHeap>;
auto blockedCommandsData = new KernelOperation(std::unique_ptr<LinearStream>(cmdStream), UniqueIH(dsh),
UniqueIH(ish), UniqueIH(ioh), UniqueIH(ssh));
SPatchAllocateStatelessPrintfSurface *pPrintfSurface = new SPatchAllocateStatelessPrintfSurface();
pPrintfSurface->DataParamOffset = 0;
pPrintfSurface->DataParamSize = 8;
char *testString = new char[sizeof("test")];
strcpy_s(testString, sizeof("test"), "test");
PrintfStringInfo printfStringInfo;
printfStringInfo.SizeInBytes = sizeof("test");
printfStringInfo.pStringData = testString;
KernelInfo *pKernelInfo = new KernelInfo();
pKernelInfo->patchInfo.pAllocateStatelessPrintfSurface = pPrintfSurface;
pKernelInfo->patchInfo.stringDataMap.insert(std::make_pair(0, printfStringInfo));
MockProgram *pProgram = new MockProgram(mockContext, false);
uint64_t crossThread[10];
MockKernel *pKernel = new MockKernel(pProgram, *pKernelInfo, *pDevice);
pKernel->setCrossThreadData(&crossThread, sizeof(uint64_t) * 8);
MockMultiDispatchInfo multiDispatchInfo(pKernel);
std::unique_ptr<PrintfHandler> printfHandler(PrintfHandler::create(multiDispatchInfo, *pDevice));
printfHandler.get()->prepareDispatch(multiDispatchInfo);
auto surface = printfHandler.get()->getSurface();
auto printfSurface = reinterpret_cast<uint32_t *>(surface->getUnderlyingBuffer());
printfSurface[0] = 8;
printfSurface[1] = 0;
auto &csr = pDevice->getCommandStreamReceiver();
std::vector<Surface *> v;
PreemptionMode preemptionMode = pDevice->getPreemptionMode();
auto cmd = new CommandComputeKernel(*pCmdQ, csr, std::unique_ptr<KernelOperation>(blockedCommandsData), v, false, false, false, std::move(printfHandler), preemptionMode, pKernel);
event.setCommand(std::unique_ptr<Command>(cmd));
event.submitCommand(false);
std::string output = testing::internal::GetCapturedStdout();
EXPECT_STREQ("test", output.c_str());
EXPECT_FALSE(surface->isResident());
delete pPrintfSurface;
delete pKernelInfo;
pKernel->decRefInternal();
pProgram->decRefInternal();
delete pCmdQ;
}
TEST_F(InternalsEventTest, processBlockedCommandsMapOperation) {
MockEvent<Event> event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, 0);
auto &csr = pDevice->getCommandStreamReceiver();
auto buffer = new MockBuffer;
MemObjSizeArray size = {{1, 1, 1}};
MemObjOffsetArray offset = {{0, 0, 0}};
event.setCommand(std::unique_ptr<Command>(new CommandMapUnmap(MAP, *buffer, size, offset, false, csr, *pCmdQ)));
auto taskLevelBefore = csr.peekTaskLevel();
event.submitCommand(false);
auto taskLevelAfter = csr.peekTaskLevel();
EXPECT_EQ(taskLevelBefore + 1, taskLevelAfter);
buffer->decRefInternal();
delete pCmdQ;
}
TEST_F(InternalsEventTest, processBlockedCommandsMapOperationNonZeroCopyBuffer) {
MockEvent<Event> event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, 0);
auto &csr = pDevice->getCommandStreamReceiver();
auto buffer = new UnalignedBuffer;
MemObjSizeArray size = {{1, 1, 1}};
MemObjOffsetArray offset = {{0, 0, 0}};
event.setCommand(std::unique_ptr<Command>(new CommandMapUnmap(MAP, *buffer, size, offset, false, csr, *pCmdQ)));
auto taskLevelBefore = csr.peekTaskLevel();
event.submitCommand(false);
auto taskLevelAfter = csr.peekTaskLevel();
EXPECT_EQ(taskLevelBefore + 1, taskLevelAfter);
buffer->decRefInternal();
delete pCmdQ;
}
uint32_t commands[] = {
CL_COMMAND_NDRANGE_KERNEL,
CL_COMMAND_TASK,
CL_COMMAND_NATIVE_KERNEL,
CL_COMMAND_READ_BUFFER,
CL_COMMAND_WRITE_BUFFER,
CL_COMMAND_COPY_BUFFER,
CL_COMMAND_READ_IMAGE,
CL_COMMAND_WRITE_IMAGE,
CL_COMMAND_COPY_IMAGE,
CL_COMMAND_COPY_IMAGE_TO_BUFFER,
CL_COMMAND_COPY_BUFFER_TO_IMAGE,
CL_COMMAND_MAP_BUFFER,
CL_COMMAND_MAP_IMAGE,
CL_COMMAND_UNMAP_MEM_OBJECT,
CL_COMMAND_MARKER,
CL_COMMAND_ACQUIRE_GL_OBJECTS,
CL_COMMAND_RELEASE_GL_OBJECTS,
CL_COMMAND_READ_BUFFER_RECT,
CL_COMMAND_WRITE_BUFFER_RECT,
CL_COMMAND_COPY_BUFFER_RECT,
CL_COMMAND_BARRIER,
CL_COMMAND_MIGRATE_MEM_OBJECTS,
CL_COMMAND_FILL_BUFFER,
CL_COMMAND_FILL_IMAGE,
CL_COMMAND_SVM_FREE,
CL_COMMAND_SVM_MEMCPY,
CL_COMMAND_SVM_MEMFILL,
CL_COMMAND_SVM_MAP,
CL_COMMAND_SVM_UNMAP,
};
class InternalsEventProfilingTest : public InternalsEventTest,
public ::testing::WithParamInterface<uint32_t> {
void SetUp() override {
InternalsEventTest::SetUp();
}
void TearDown() override {
InternalsEventTest::TearDown();
}
};
TEST_P(InternalsEventProfilingTest, GivenProfilingWhenEventCreatedThenProfilingSet) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
std::unique_ptr<CommandQueue> pCmdQ(new CommandQueue(mockContext, pDevice, props));
std::unique_ptr<MockEvent<Event>> event(new MockEvent<Event>(pCmdQ.get(), GetParam(), 0, 0));
EXPECT_TRUE(event.get()->isProfilingEnabled());
}
INSTANTIATE_TEST_CASE_P(InternalsEventProfilingTest,
InternalsEventProfilingTest,
::testing::ValuesIn(commands));
TEST_F(InternalsEventTest, GivenProfilingWhenUserEventCreatedThenProfilingNotSet) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
std::unique_ptr<CommandQueue> pCmdQ(new CommandQueue(mockContext, pDevice, props));
std::unique_ptr<MockEvent<Event>> event(new MockEvent<Event>(pCmdQ.get(), CL_COMMAND_USER, 0, 0));
EXPECT_FALSE(event.get()->isProfilingEnabled());
}
TEST_F(InternalsEventTest, GIVENProfilingWHENMapOperationTHENTimesSet) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
MockEvent<Event> *event = new MockEvent<Event>(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
auto &csr = pDevice->getCommandStreamReceiver();
UnalignedBuffer buffer;
MemObjSizeArray size = {{1, 1, 1}};
MemObjOffsetArray offset = {{0, 0, 0}};
event->setCommand(std::unique_ptr<Command>(new CommandMapUnmap(MAP, buffer, size, offset, false, csr, *pCmdQ)));
auto taskLevelBefore = csr.peekTaskLevel();
event->submitCommand(false);
uint64_t submitTime = 0ULL;
event->getEventProfilingInfo(CL_PROFILING_COMMAND_SUBMIT, sizeof(uint64_t), &submitTime, 0);
EXPECT_NE(0ULL, submitTime);
auto taskLevelAfter = csr.peekTaskLevel();
delete event;
EXPECT_EQ(taskLevelBefore + 1, taskLevelAfter);
delete pCmdQ;
}
TEST_F(InternalsEventTest, processBlockedCommandsUnMapOperation) {
MockEvent<Event> event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, 0, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
auto &csr = pDevice->getCommandStreamReceiver();
auto buffer = new UnalignedBuffer;
MemObjSizeArray size = {{1, 1, 1}};
MemObjOffsetArray offset = {{0, 0, 0}};
event.setCommand(std::unique_ptr<Command>(new CommandMapUnmap(UNMAP, *buffer, size, offset, false, csr, *pCmdQ)));
auto taskLevelBefore = csr.peekTaskLevel();
event.submitCommand(false);
auto taskLevelAfter = csr.peekTaskLevel();
EXPECT_EQ(taskLevelBefore + 1, taskLevelAfter);
buffer->decRefInternal();
delete pCmdQ;
}
TEST_F(InternalsEventTest, processBlockedCommandsUnMapOperationNonZeroCopyBuffer) {
MockEvent<Event> event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, 0, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
auto &csr = pDevice->getCommandStreamReceiver();
auto buffer = new UnalignedBuffer;
MemObjSizeArray size = {{1, 1, 1}};
MemObjOffsetArray offset = {{0, 0, 0}};
event.setCommand(std::unique_ptr<Command>(new CommandMapUnmap(UNMAP, *buffer, size, offset, false, csr, *pCmdQ)));
auto taskLevelBefore = csr.peekTaskLevel();
event.submitCommand(false);
auto taskLevelAfter = csr.peekTaskLevel();
EXPECT_EQ(taskLevelBefore + 1, taskLevelAfter);
buffer->decRefInternal();
delete pCmdQ;
}
HWTEST_F(InternalsEventTest, givenCpuProfilingPathWhenEnqueuedMarkerThenDontUseTimeStampNode) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
MockEvent<Event> *event = new MockEvent<Event>(pCmdQ, CL_COMMAND_MARKER, 0, 0);
event->setCPUProfilingPath(true);
auto &csr = pDevice->getCommandStreamReceiver();
event->setCommand(std::unique_ptr<Command>(new CommandMarker(*pCmdQ, csr, CL_COMMAND_MARKER, 4096u)));
event->submitCommand(false);
uint64_t submit, start, end;
event->getEventProfilingInfo(CL_PROFILING_COMMAND_SUBMIT, sizeof(uint64_t), &submit, 0);
event->getEventProfilingInfo(CL_PROFILING_COMMAND_START, sizeof(uint64_t), &start, 0);
event->getEventProfilingInfo(CL_PROFILING_COMMAND_END, sizeof(uint64_t), &end, 0);
EXPECT_LT(0u, submit);
EXPECT_LT(submit, start);
EXPECT_LT(start, end);
delete event;
delete pCmdQ;
}
struct InternalsEventWithPerfCountersTest
: public InternalsEventTest,
public PerformanceCountersFixture {
void SetUp() override {
PerformanceCountersFixture::SetUp();
InternalsEventTest::SetUp();
createPerfCounters();
performanceCountersBase->initialize(platformDevices[0]);
pDevice->setPerfCounters(performanceCountersBase.get());
}
void TearDown() override {
performanceCountersBase.release();
InternalsEventTest::TearDown();
PerformanceCountersFixture::TearDown();
}
};
HWTEST_F(InternalsEventWithPerfCountersTest, givenCpuProfilingPerfCountersPathWhenEnqueuedMarkerThenDontUseTimeStampNodePerfCounterNode) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
bool ret = false;
ret = pCmdQ->setPerfCountersEnabled(true, 1);
EXPECT_TRUE(ret);
ret = pCmdQ->setPerfCountersEnabled(true, 1);
EXPECT_TRUE(ret);
MockEvent<Event> *event = new MockEvent<Event>(pCmdQ, CL_COMMAND_MARKER, 0, 0);
event->setCPUProfilingPath(true);
auto &csr = pDevice->getCommandStreamReceiver();
event->setCommand(std::unique_ptr<Command>(new CommandMarker(*pCmdQ, csr, CL_COMMAND_MARKER, 4096u)));
event->submitCommand(false);
uint64_t submit, start, end;
event->getEventProfilingInfo(CL_PROFILING_COMMAND_SUBMIT, sizeof(uint64_t), &submit, 0);
event->getEventProfilingInfo(CL_PROFILING_COMMAND_START, sizeof(uint64_t), &start, 0);
event->getEventProfilingInfo(CL_PROFILING_COMMAND_END, sizeof(uint64_t), &end, 0);
EXPECT_LT(0u, submit);
EXPECT_LT(submit, start);
EXPECT_LT(start, end);
delete event;
delete pCmdQ;
}
HWTEST_F(InternalsEventWithPerfCountersTest, givenCpuProfilingPerfCountersPathWhenEnqueuedMarkerThenUseTimeStampNodePerfCounterNode) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
pCmdQ->setPerfCountersEnabled(true, 1);
MockEvent<Event> *event = new MockEvent<Event>(pCmdQ, CL_COMMAND_MARKER, 0, 0);
event->setCPUProfilingPath(true);
HwPerfCounter *perfCounter = event->getHwPerfCounter();
ASSERT_NE(nullptr, perfCounter);
HwTimeStamps *timeStamps = event->getHwTimeStamp();
ASSERT_NE(nullptr, timeStamps);
auto &csr = pDevice->getCommandStreamReceiver();
event->setCommand(std::unique_ptr<Command>(new CommandMarker(*pCmdQ, csr, CL_COMMAND_MARKER, 4096u)));
event->submitCommand(false);
uint64_t submit, start, end;
event->getEventProfilingInfo(CL_PROFILING_COMMAND_SUBMIT, sizeof(uint64_t), &submit, 0);
event->getEventProfilingInfo(CL_PROFILING_COMMAND_START, sizeof(uint64_t), &start, 0);
event->getEventProfilingInfo(CL_PROFILING_COMMAND_END, sizeof(uint64_t), &end, 0);
EXPECT_LT(0u, submit);
EXPECT_LT(submit, start);
EXPECT_LT(start, end);
delete event;
delete pCmdQ;
}
TEST_F(InternalsEventWithPerfCountersTest, IsPerfCounter_Enabled) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
pCmdQ->setPerfCountersEnabled(true, 2);
Event *ev = new Event(pCmdQ, CL_COMMAND_COPY_BUFFER, 3, 0);
EXPECT_TRUE(ev->isProfilingEnabled());
EXPECT_TRUE(ev->isPerfCountersEnabled());
delete ev;
delete pCmdQ;
}
TEST(Event, GivenNoContextOnDeletionDeletesSelf) {
UserEvent *ue = new UserEvent();
auto autoptr = ue->release();
ASSERT_TRUE(autoptr.isUnused());
}
HWTEST_F(InternalsEventTest, GivenBufferWithoutZeroCopyOnCommandMapOrUnmapFlushesPreviousTasksBeforeMappingOrUnmapping) {
struct MockNonZeroCopyBuff : UnalignedBuffer {
MockNonZeroCopyBuff(int32_t &executionStamp)
: executionStamp(executionStamp), dataTransferedStamp(-1) {
hostPtr = &dataTransferedStamp;
memoryStorage = &executionStamp;
size = sizeof(executionStamp);
hostPtrMinSize = size;
}
void setIsZeroCopy() {
isZeroCopy = false;
}
void swapCopyDirection() {
std::swap(hostPtr, memoryStorage);
}
int32_t &executionStamp;
int32_t dataTransferedStamp;
};
int32_t executionStamp = 0;
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, 0, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
MockNonZeroCopyBuff buffer(executionStamp);
MockCsr<FamilyType> csr(executionStamp);
csr.setMemoryManager(pDevice->getMemoryManager());
MemObjSizeArray size = {{4, 1, 1}};
MemObjOffsetArray offset = {{0, 0, 0}};
auto commandMap = std::unique_ptr<Command>(new CommandMapUnmap(MAP, buffer, size, offset, false, csr, *pCmdQ));
EXPECT_EQ(0, executionStamp);
EXPECT_EQ(-1, csr.flushTaskStamp);
EXPECT_EQ(-1, buffer.dataTransferedStamp);
auto latestSentFlushTaskCount = csr.peekLatestSentTaskCount();
commandMap->submit(0, false);
EXPECT_EQ(1, executionStamp);
EXPECT_EQ(0, csr.flushTaskStamp);
EXPECT_EQ(1, buffer.dataTransferedStamp);
auto latestSentFlushTaskCountAfterSubmit = csr.peekLatestSentTaskCount();
EXPECT_GT(latestSentFlushTaskCountAfterSubmit, latestSentFlushTaskCount);
executionStamp = 0;
csr.flushTaskStamp = -1;
buffer.dataTransferedStamp = -1;
buffer.swapCopyDirection();
auto commandUnMap = std::unique_ptr<Command>(new CommandMapUnmap(UNMAP, buffer, size, offset, false, csr, *pCmdQ));
EXPECT_EQ(0, executionStamp);
EXPECT_EQ(-1, csr.flushTaskStamp);
EXPECT_EQ(-1, buffer.dataTransferedStamp);
commandUnMap->submit(0, false);
EXPECT_EQ(1, executionStamp);
EXPECT_EQ(0, csr.flushTaskStamp);
EXPECT_EQ(1, buffer.dataTransferedStamp);
EXPECT_EQ(nullptr, commandUnMap->getCommandStream());
delete pCmdQ;
}
TEST(EventCallback, CallbackAfterStatusOverrideUsesNewStatus) {
struct ClbFuncTempStruct {
static void CL_CALLBACK ClbFuncT(cl_event e, cl_int status, void *retStatus) {
*((cl_int *)retStatus) = status;
}
};
cl_int retStatus = 7;
Event::Callback clb(nullptr, ClbFuncTempStruct::ClbFuncT, CL_COMPLETE, &retStatus);
EXPECT_EQ(CL_COMPLETE, clb.getCallbackExecutionStatusTarget());
clb.execute();
EXPECT_EQ(CL_COMPLETE, retStatus);
retStatus = 7;
clb.overrideCallbackExecutionStatusTarget(-1);
EXPECT_EQ(-1, clb.getCallbackExecutionStatusTarget());
clb.execute();
EXPECT_EQ(-1, retStatus);
}
TEST_F(EventTest, WhensetTimeStampThenCorrectValues) {
MyEvent ev(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 3, 0);
TimeStampData inTimeStamp = {1ULL, 2ULL};
ev.setSubmitTimeStamp(&inTimeStamp);
TimeStampData outtimeStamp = {0, 0};
outtimeStamp = ev.getSubmitTimeStamp();
EXPECT_EQ(1ULL, outtimeStamp.GPUTimeStamp);
EXPECT_EQ(2ULL, outtimeStamp.CPUTimeinNS);
inTimeStamp.GPUTimeStamp = 3;
inTimeStamp.CPUTimeinNS = 4;
ev.setQueueTimeStamp(&inTimeStamp);
outtimeStamp = ev.getQueueTimeStamp();
EXPECT_EQ(3ULL, outtimeStamp.GPUTimeStamp);
EXPECT_EQ(4ULL, outtimeStamp.CPUTimeinNS);
}
TEST_F(EventTest, WhensetCPUTimeStampThenCorrectTimes) {
MyEvent ev(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 3, 0);
ev.setProfilingEnabled(true);
ev.setQueueTimeStamp();
TimeStampData outtimeStamp = {0, 0};
outtimeStamp = ev.getQueueTimeStamp();
EXPECT_NE(0ULL, outtimeStamp.CPUTimeinNS);
EXPECT_EQ(0ULL, outtimeStamp.GPUTimeStamp);
ev.setSubmitTimeStamp();
outtimeStamp = ev.getSubmitTimeStamp();
EXPECT_NE(0ULL, outtimeStamp.CPUTimeinNS);
EXPECT_EQ(0ULL, outtimeStamp.GPUTimeStamp);
ev.setStartTimeStamp();
uint64_t outCPUtimeStamp = ev.getStartTimeStamp();
EXPECT_NE(0ULL, outCPUtimeStamp);
ev.setEndTimeStamp();
outCPUtimeStamp = ev.getEndTimeStamp();
EXPECT_NE(0ULL, outCPUtimeStamp);
outCPUtimeStamp = ev.getCompleteTimeStamp();
EXPECT_NE(0ULL, outCPUtimeStamp);
}
TEST_F(EventTest, GIVENNoQueueWhensetCPUTimeStampThenTimesNotSet) {
MyEvent ev(nullptr, CL_COMMAND_COPY_BUFFER, 3, 0);
ev.setQueueTimeStamp();
TimeStampData outtimeStamp = {0, 0};
outtimeStamp = ev.getQueueTimeStamp();
EXPECT_EQ(0ULL, outtimeStamp.CPUTimeinNS);
EXPECT_EQ(0ULL, outtimeStamp.GPUTimeStamp);
ev.setSubmitTimeStamp();
outtimeStamp = ev.getSubmitTimeStamp();
EXPECT_EQ(0ULL, outtimeStamp.CPUTimeinNS);
EXPECT_EQ(0ULL, outtimeStamp.GPUTimeStamp);
ev.setStartTimeStamp();
uint64_t outCPUtimeStamp = ev.getStartTimeStamp();
EXPECT_EQ(0ULL, outCPUtimeStamp);
ev.setEndTimeStamp();
outCPUtimeStamp = ev.getEndTimeStamp();
EXPECT_EQ(0ULL, outCPUtimeStamp);
outCPUtimeStamp = ev.getCompleteTimeStamp();
EXPECT_EQ(0ULL, outCPUtimeStamp);
}
TEST_F(EventTest, getHwTimeStampsReturnsValidPointer) {
std::unique_ptr<Event> event(new Event(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 0, 0));
ASSERT_NE(nullptr, event);
HwTimeStamps *timeStamps = event->getHwTimeStamp();
ASSERT_NE(nullptr, timeStamps);
//this should not cause any heap corruptions
ASSERT_EQ(0ULL, timeStamps->GlobalStartTS);
ASSERT_EQ(0ULL, timeStamps->ContextStartTS);
ASSERT_EQ(0ULL, timeStamps->GlobalEndTS);
ASSERT_EQ(0ULL, timeStamps->ContextEndTS);
ASSERT_EQ(0ULL, timeStamps->GlobalCompleteTS);
ASSERT_EQ(0ULL, timeStamps->ContextCompleteTS);
HwTimeStamps *timeStamps2 = event->getHwTimeStamp();
ASSERT_EQ(timeStamps, timeStamps2);
}
TEST_F(EventTest, getHwTimeStampsAllocationReturnsValidPointer) {
std::unique_ptr<Event> event(new Event(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 0, 0));
ASSERT_NE(nullptr, event);
GraphicsAllocation *allocation = event->getHwTimeStampAllocation();
ASSERT_NE(nullptr, allocation);
void *memoryStorage = allocation->getUnderlyingBuffer();
size_t memoryStorageSize = allocation->getUnderlyingBufferSize();
EXPECT_NE(nullptr, memoryStorage);
EXPECT_GT(memoryStorageSize, 0u);
}
TEST_F(EventTest, hwTimeStampsMemoryIsPlacedInGraphicsAllocation) {
std::unique_ptr<Event> event(new Event(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 0, 0));
ASSERT_NE(nullptr, event);
HwTimeStamps *timeStamps = event->getHwTimeStamp();
ASSERT_NE(nullptr, timeStamps);
GraphicsAllocation *allocation = event->getHwTimeStampAllocation();
ASSERT_NE(nullptr, allocation);
void *memoryStorage = allocation->getUnderlyingBuffer();
size_t graphicsAllocationSize = allocation->getUnderlyingBufferSize();
uintptr_t timeStampAddress = reinterpret_cast<uintptr_t>(timeStamps);
uintptr_t graphicsAllocationStart = reinterpret_cast<uintptr_t>(memoryStorage);
if (!((timeStampAddress >= graphicsAllocationStart) &&
((timeStampAddress + sizeof(HwTimeStamps)) <= (graphicsAllocationStart + graphicsAllocationSize)))) {
EXPECT_TRUE(false);
}
}
TEST_F(EventTest, getHwPerfCounterReturnsValidPointer) {
std::unique_ptr<Event> event(new Event(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 0, 0));
ASSERT_NE(nullptr, event);
HwPerfCounter *perfCounter = event->getHwPerfCounter();
ASSERT_NE(nullptr, perfCounter);
ASSERT_EQ(0ULL, perfCounter->HWTimeStamp.GlobalStartTS);
ASSERT_EQ(0ULL, perfCounter->HWTimeStamp.ContextStartTS);
ASSERT_EQ(0ULL, perfCounter->HWTimeStamp.GlobalEndTS);
ASSERT_EQ(0ULL, perfCounter->HWTimeStamp.ContextEndTS);
ASSERT_EQ(0ULL, perfCounter->HWTimeStamp.GlobalCompleteTS);
ASSERT_EQ(0ULL, perfCounter->HWTimeStamp.ContextCompleteTS);
HwPerfCounter *perfCounter2 = event->getHwPerfCounter();
ASSERT_EQ(perfCounter, perfCounter2);
}
TEST_F(EventTest, getHwPerfCounterAllocationReturnsValidPointer) {
std::unique_ptr<Event> event(new Event(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 0, 0));
ASSERT_NE(nullptr, event);
GraphicsAllocation *allocation = event->getHwPerfCounterAllocation();
ASSERT_NE(nullptr, allocation);
void *memoryStorage = allocation->getUnderlyingBuffer();
size_t memoryStorageSize = allocation->getUnderlyingBufferSize();
EXPECT_NE(nullptr, memoryStorage);
EXPECT_GT(memoryStorageSize, 0u);
}
TEST_F(EventTest, hwPerfCounterMemoryIsPlacedInGraphicsAllocation) {
std::unique_ptr<Event> event(new Event(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 0, 0));
ASSERT_NE(nullptr, event);
HwPerfCounter *perfCounter = event->getHwPerfCounter();
ASSERT_NE(nullptr, perfCounter);
GraphicsAllocation *allocation = event->getHwPerfCounterAllocation();
ASSERT_NE(nullptr, allocation);
void *memoryStorage = allocation->getUnderlyingBuffer();
size_t graphicsAllocationSize = allocation->getUnderlyingBufferSize();
uintptr_t perfCounterAddress = reinterpret_cast<uintptr_t>(perfCounter);
uintptr_t graphicsAllocationStart = reinterpret_cast<uintptr_t>(memoryStorage);
if (!((perfCounterAddress >= graphicsAllocationStart) &&
((perfCounterAddress + sizeof(HwPerfCounter)) <= (graphicsAllocationStart + graphicsAllocationSize)))) {
EXPECT_TRUE(false);
}
}
TEST_F(EventTest, IsPerfCounter_DisabledByNullQueue) {
Event ev(nullptr, CL_COMMAND_COPY_BUFFER, 3, 0);
EXPECT_FALSE(ev.isProfilingEnabled());
EXPECT_FALSE(ev.isPerfCountersEnabled());
}
TEST_F(EventTest, IsPerfCounter_DisabledByNoProfiling) {
Event ev(pCmdQ, CL_COMMAND_COPY_BUFFER, 3, 0);
EXPECT_FALSE(ev.isProfilingEnabled());
EXPECT_FALSE(ev.isPerfCountersEnabled());
}
TEST_F(InternalsEventTest, IsPerfCounter_DisabledByNoPerfCounter) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
Event *ev = new Event(pCmdQ, CL_COMMAND_COPY_BUFFER, 3, 0);
EXPECT_TRUE(ev->isProfilingEnabled());
EXPECT_FALSE(ev->isPerfCountersEnabled());
delete ev;
delete pCmdQ;
}
TEST_F(InternalsEventWithPerfCountersTest, SetPerfCounter_negativeInvalidASInterface) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
performanceCountersBase->setAutoSamplingStartFunc(autoSamplingStartFailing);
bool ret = false;
ret = pCmdQ->setPerfCountersEnabled(true, 1);
EXPECT_FALSE(ret);
delete pCmdQ;
}
TEST_F(InternalsEventWithPerfCountersTest, SetPerfCounter_AvailFalse) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
CommandQueue *pCmdQ = new CommandQueue(mockContext, pDevice, props);
bool ret = false;
ret = pCmdQ->setPerfCountersEnabled(true, 1);
EXPECT_TRUE(ret);
performanceCountersBase->setAvailableFlag(false);
ret = pCmdQ->setPerfCountersEnabled(false, 0);
EXPECT_TRUE(ret);
performanceCountersBase->shutdown();
delete pCmdQ;
}
TEST_F(EventTest, GivenNullptrWhenpeekIsSubmittedThenFalse) {
Event ev(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 3, 0);
bool executionStatus = ev.peekIsSubmitted(nullptr);
EXPECT_NE(true, executionStatus);
}
TEST_F(EventTest, GivenCL_SUBMITTEDWhenpeekIsSubmittedThenTrue) {
Event ev(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 3, 0);
int32_t executionStatusSnapshot = CL_SUBMITTED;
bool executionStatus = ev.peekIsSubmitted(&executionStatusSnapshot);
EXPECT_EQ(true, executionStatus);
}
TEST_F(EventTest, GivenCompletedEventWhenQueryingExecutionStatusAfterFlushThenCsrIsNotFlushed) {
cl_int ret;
*pDevice->getTagAddress() = 3;
Event ev(this->pCmdQ, CL_COMMAND_COPY_BUFFER, 3, 3);
auto &csr = this->pCmdQ->getDevice().getCommandStreamReceiver();
auto previousTaskLevel = csr.peekTaskLevel();
EXPECT_GT(3u, previousTaskLevel);
ret = clFlush(this->pCmdQ);
ASSERT_EQ(CL_SUCCESS, ret);
cl_int execState;
ret = clGetEventInfo(&ev, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(execState), &execState, nullptr);
ASSERT_EQ(CL_SUCCESS, ret);
EXPECT_EQ(previousTaskLevel, csr.peekTaskLevel());
}
HWTEST_F(EventTest, submitCommandOnEventCreatedOnMapBufferWithoutCommandUpdatesTaskCount) {
MockEvent<Event> ev(this->pCmdQ, CL_COMMAND_MAP_BUFFER, Event::eventNotReady, Event::eventNotReady);
EXPECT_EQ(Event::eventNotReady, ev.peekTaskCount());
ev.submitCommand(false);
EXPECT_EQ(0u, ev.peekTaskCount());
}
HWTEST_F(EventTest, submitCommandOnEventCreatedOnMapImageWithoutCommandUpdatesTaskCount) {
MockEvent<Event> ev(this->pCmdQ, CL_COMMAND_MAP_IMAGE, Event::eventNotReady, Event::eventNotReady);
EXPECT_EQ(Event::eventNotReady, ev.peekTaskCount());
ev.submitCommand(false);
EXPECT_EQ(0u, ev.peekTaskCount());
}
TEST_F(EventTest, givenCmdQueueWithoutProfilingWhenIsCpuProfilingIsCalledThenFalseIsReturned) {
MockEvent<Event> ev(this->pCmdQ, CL_COMMAND_MAP_IMAGE, Event::eventNotReady, Event::eventNotReady);
bool cpuProfiling = ev.isCPUProfilingPath() != 0;
EXPECT_FALSE(cpuProfiling);
}
TEST_F(EventTest, givenCmdQueueWithProfilingWhenIsCpuProfilingIsCalledThenTrueIsReturned) {
const cl_queue_properties props[3] = {CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0};
std::unique_ptr<CommandQueue> pCmdQ(new CommandQueue(&mockContext, pDevice, props));
MockEvent<Event> ev(pCmdQ.get(), CL_COMMAND_MAP_IMAGE, Event::eventNotReady, Event::eventNotReady);
bool cpuProfiling = ev.isCPUProfilingPath() != 0;
EXPECT_TRUE(cpuProfiling);
}
TEST(EventCallback, GivenEventWithCallbacksOnPeekHasCallbacksReturnsTrue) {
DebugManagerStateRestore dbgRestore;
DebugManager.flags.EnableAsyncEventsHandler.set(false);
struct ClbFuncTempStruct {
static void CL_CALLBACK ClbFuncT(cl_event, cl_int, void *) {
}
};
struct SmallMockEvent : Event {
SmallMockEvent()
: Event(nullptr, CL_COMMAND_COPY_BUFFER, 0, 0) {
this->parentCount = 1; // block event
}
};
{
SmallMockEvent ev;
EXPECT_FALSE(ev.peekHasCallbacks());
}
{
SmallMockEvent ev;
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_SUBMITTED, nullptr);
EXPECT_TRUE(ev.peekHasCallbacks());
ev.decRefInternal();
}
{
SmallMockEvent ev;
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_RUNNING, nullptr);
EXPECT_TRUE(ev.peekHasCallbacks());
ev.decRefInternal();
}
{
SmallMockEvent ev;
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_COMPLETE, nullptr);
EXPECT_TRUE(ev.peekHasCallbacks());
ev.decRefInternal();
}
{
SmallMockEvent ev;
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_SUBMITTED, nullptr);
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_COMPLETE, nullptr);
EXPECT_TRUE(ev.peekHasCallbacks());
ev.decRefInternal();
ev.decRefInternal();
}
{
SmallMockEvent ev;
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_RUNNING, nullptr);
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_COMPLETE, nullptr);
EXPECT_TRUE(ev.peekHasCallbacks());
ev.decRefInternal();
ev.decRefInternal();
}
{
SmallMockEvent ev;
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_SUBMITTED, nullptr);
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_RUNNING, nullptr);
ev.addCallback(ClbFuncTempStruct::ClbFuncT, CL_COMPLETE, nullptr);
EXPECT_TRUE(ev.peekHasCallbacks());
ev.decRefInternal();
ev.decRefInternal();
ev.decRefInternal();
}
}
TEST_F(EventTest, addChildForEventUncompleted) {
VirtualEvent virtualEvent;
{
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
event.addChild(virtualEvent);
EXPECT_NE(0U, virtualEvent.peekNumEventsBlockingThis());
}
}
TEST(Event, whenCreatingRegularEventsThenExternalSynchronizationIsNotRequired) {
Event *event = new Event(nullptr, 0, 0, 0);
EXPECT_FALSE(event->isExternallySynchronized());
event->release();
UserEvent *userEvent = new UserEvent();
EXPECT_FALSE(userEvent->isExternallySynchronized());
userEvent->release();
VirtualEvent *virtualEvent = new VirtualEvent();
EXPECT_FALSE(virtualEvent->isExternallySynchronized());
virtualEvent->release();
}
TEST_F(EventTest, addChildForEventCompleted) {
VirtualEvent virtualEvent;
{
Event event(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
event.setStatus(CL_COMPLETE);
event.addChild(virtualEvent);
EXPECT_EQ(0U, virtualEvent.peekNumEventsBlockingThis());
}
}
HWTEST_F(InternalsEventTest, givenCommandWhenSubmitCalledThenUpdateFlushStamp) {
auto pCmdQ = std::unique_ptr<CommandQueue>(new CommandQueue(mockContext, pDevice, 0));
MockEvent<Event> *event = new MockEvent<Event>(pCmdQ.get(), CL_COMMAND_MARKER, 0, 0);
auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();
csr.flushStamp->setStamp(5);
FlushStamp expectedFlushStamp = 0;
EXPECT_EQ(expectedFlushStamp, event->flushStamp->peekStamp());
event->setCommand(std::unique_ptr<Command>(new CommandMarker(*pCmdQ.get(), csr, CL_COMMAND_MARKER, 4096u)));
event->submitCommand(false);
EXPECT_EQ(csr.flushStamp->peekStamp(), event->flushStamp->peekStamp());
delete event;
}
HWTEST_F(InternalsEventTest, givenAbortedCommandWhenSubmitCalledThenDontUpdateFlushStamp) {
auto pCmdQ = std::unique_ptr<CommandQueue>(new CommandQueue(mockContext, pDevice, 0));
MockEvent<Event> *event = new MockEvent<Event>(pCmdQ.get(), CL_COMMAND_MARKER, 0, 0);
auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();
csr.flushStamp->setStamp(5);
auto cmdStream = new LinearStream(alignedMalloc(4096, 4096), 4096);
auto dsh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ish = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ioh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
auto ssh = new IndirectHeap(alignedMalloc(4096, 4096), 4096);
using UniqueIH = std::unique_ptr<IndirectHeap>;
auto blockedCommandsData = new KernelOperation(std::unique_ptr<LinearStream>(cmdStream), UniqueIH(dsh),
UniqueIH(ish), UniqueIH(ioh), UniqueIH(ssh));
PreemptionMode preemptionMode = pDevice->getPreemptionMode();
std::vector<Surface *> v;
auto cmd = new CommandComputeKernel(*pCmdQ, csr, std::unique_ptr<KernelOperation>(blockedCommandsData), v, false, false, false, nullptr, preemptionMode);
event->setCommand(std::unique_ptr<Command>(cmd));
FlushStamp expectedFlushStamp = 0;
EXPECT_EQ(expectedFlushStamp, event->flushStamp->peekStamp());
event->submitCommand(true);
EXPECT_EQ(expectedFlushStamp, event->flushStamp->peekStamp());
delete event;
}
TEST(EventLockerTests, givenEventWhenEventLockerIsUsedThenOwnershipIsAutomaticallyReleased) {
Event ev(nullptr, CL_COMMAND_COPY_BUFFER, 3, 0);
{
TakeOwnershipWrapper<Event> locker(ev);
EXPECT_TRUE(ev.hasOwnership());
}
EXPECT_FALSE(ev.hasOwnership());
}
TEST(EventLockerTests, givenEventWhenEventLockerIsUsedAndUnlockedThenOwnershipIsReleased) {
Event ev(nullptr, CL_COMMAND_COPY_BUFFER, 3, 0);
{
TakeOwnershipWrapper<Event> locker(ev);
locker.unlock();
EXPECT_FALSE(ev.hasOwnership());
}
EXPECT_FALSE(ev.hasOwnership());
}
TEST(EventLockerTests, givenEventWhenEventLockerIsUsedAndlockedThenOwnershipIsAcquiredAgain) {
Event ev(nullptr, CL_COMMAND_COPY_BUFFER, 3, 0);
{
TakeOwnershipWrapper<Event> locker(ev);
locker.unlock();
locker.lock();
EXPECT_TRUE(ev.hasOwnership());
}
EXPECT_FALSE(ev.hasOwnership());
}
TEST(EventLockerTests, givenEventWhenEventLockerIsLockedTwiceThenOwnershipIsReleaseAfterLeavingTheScope) {
Event ev(nullptr, CL_COMMAND_COPY_BUFFER, 3, 0);
{
TakeOwnershipWrapper<Event> locker(ev);
locker.lock();
EXPECT_TRUE(ev.hasOwnership());
}
EXPECT_FALSE(ev.hasOwnership());
}
TEST(EventsDebug, givenEventWhenTrackingOfParentsIsOnThenTrackParents) {
DebugManagerStateRestore stateRestore;
DebugManager.flags.TrackParentEvents.set(true);
Event event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
Event event2(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
auto &parentEvents = event.getParentEvents();
auto &parentEvents2 = event2.getParentEvents();
EXPECT_EQ(0u, parentEvents.size());
EXPECT_EQ(0u, parentEvents2.size());
event.addChild(event2);
EXPECT_EQ(0u, parentEvents.size());
EXPECT_EQ(1u, parentEvents2.size());
EXPECT_EQ(&event, parentEvents2.at(0));
event.setStatus(CL_COMPLETE);
}
TEST(EventsDebug, givenEventWhenTrackingOfParentsIsOffThenDoNotTrackParents) {
DebugManagerStateRestore stateRestore;
DebugManager.flags.TrackParentEvents.set(false);
Event event(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
Event event2(nullptr, CL_COMMAND_NDRANGE_KERNEL, 0, 0);
auto &parentEvents = event.getParentEvents();
auto &parentEvents2 = event2.getParentEvents();
EXPECT_EQ(0u, parentEvents.size());
EXPECT_EQ(0u, parentEvents2.size());
event.addChild(event2);
EXPECT_EQ(0u, parentEvents.size());
EXPECT_EQ(0u, parentEvents2.size());
event.setStatus(CL_COMPLETE);
}