Michal Mrozek

Hello @liyuming1978 , I am from OpenCL driver team. Our code works as you think, this is busy wait for some time, then it switches to non busy wait. It...

Windows is a bit different. There non busy wait is only when you are on a battery. When driver is on AC then it never goes to non busy wait...

cl_cache works the same way on Linux & Windows. For Windows and non busy waits, I suggest to use throttle hints to create a command queue with CL_QUEUE_THROTTLE_LOW. This would...

Those tests measure CPU driver overhead for various API calls. They show how long it takes to execute given API (i.e. clEnqueueNDRangeKernel) in various conditions ( multiple threads, different parameter...

@stef does this tests starts to fail after some of your local changes OR it fails randomly on driver without any modifications ?

The test that fails is actually quite simple TEST_F(PageFaultManagerLinuxTest, whenPageFaultIsRaisedThenHandlerIsInvoked) { auto pageFaultManager = std::make_unique(); EXPECT_FALSE(pageFaultManager->handlerInvoked); std::raise(SIGSEGV); EXPECT_TRUE(pageFaultManager->handlerInvoked); } If it fails it means that handler was not invoked. Can...

Is this a sporadic problem ? Can you check if handler is properly intialized here: PageFaultManagerLinux::PageFaultManagerLinux() { pageFaultHandler = [&](int signal, siginfo_t *info, void *context) { if (!this->verifyPageFault(info->si_addr)) { callPreviousHandler(signal,...

This macro is not used in this test so that is intresting. Does it ever fail if you only run this one test ? (gtest filter may be used to...

I think you may be running into special case of WriteBuffer where driver optimizes the whole call and there is no GPU execution. In such case we have issue as...

Integrated GPU's goes over KMD and GuC for submissions, hence the times you observe are within expected range.