Make sure to use steady clock, high resolution was responsible for time going backwards and deadlocks on GLsync waits.

GLES backend has a bug where the tests timeout!?

Vulkan has a bug because new Buffer API is not ready yet.

Author: devshgraphicsprogramming

examples_tests/48.ArithmeticUnitTest/main.cpp

@@ -201,6 +201,7 @@ bool validateResults(ILogicalDevice* device, const uint32_t* inputData, const ui
 	auto dataFromBuffer = reinterpret_cast<uint32_t*>(mem->getMappedPointer());
 	const uint32_t subgroupSize = (*dataFromBuffer++);
 
+	// TODO: parallel for
 	// now check if the data obtained has valid values
 	uint32_t* tmp = new uint32_t[workgroupSize];
 	uint32_t* ballotInput = new uint32_t[workgroupSize];
@@ -281,7 +282,7 @@ bool runTest(
 	passed = validateResults<Arithmetic,max_op>(device, inputData, workgroupSize, workgroupCount, buffers[6].get(),logger)&&passed;
 	if(is_workgroup_test)
 	{
-		passed = validateResults<Arithmetic,ballot>(device, inputData, workgroupSize, workgroupCount, buffers[7].get(),logger) && passed;
+		passed = validateResults<Arithmetic,ballot>(device, inputData, workgroupSize, workgroupCount, buffers[7].get(),logger)&&passed;
 	}
 
 	return passed;
@@ -300,7 +301,7 @@ class ArythmeticUnitTestApp : public NonGraphicalApplicationBase
 	void onAppInitialized_impl() override
 	{
 		CommonAPI::InitOutput initOutput;
-		CommonAPI::InitWithNoExt(initOutput, nbl::video::EAT_OPENGL, "Subgroup Arithmetic Test");
+		CommonAPI::InitWithNoExt(initOutput, video::EAT_OPENGL, "Subgroup Arithmetic Test");
 		gl = std::move(initOutput.apiConnection);
 		gpuPhysicalDevice = std::move(initOutput.physicalDevice);
 		logicalDevice = std::move(initOutput.logicalDevice);
@@ -335,7 +336,7 @@ class ArythmeticUnitTestApp : public NonGraphicalApplicationBase
 			params.queueFamilyIndexCount = 0;
 			params.queueFamilyIndices = nullptr;
 			params.sharingMode = ESM_CONCURRENT;
-			params.usage = IGPUBuffer::EUF_STORAGE_BUFFER_BIT;
+			params.usage = core::bitflag(IGPUBuffer::EUF_STORAGE_BUFFER_BIT)|IGPUBuffer::EUF_TRANSFER_SRC_BIT;
 			IDriverMemoryBacked::SDriverMemoryRequirements reqs;
 			reqs.vulkanReqs.memoryTypeBits = ~0u;
 			reqs.vulkanReqs.alignment = 256u;
@@ -424,7 +425,7 @@ class ArythmeticUnitTestApp : public NonGraphicalApplicationBase
 		core::smart_refctd_ptr<IGPUCommandBuffer> cmdbuf;
 		logicalDevice->createCommandBuffers(cmdPool.get(), IGPUCommandBuffer::EL_PRIMARY, 1u, &cmdbuf);
 		computeQueue->startCapture();
-		for (uint32_t workgroupSize = 1u; workgroupSize <= 1024u; workgroupSize++)
+		for (uint32_t workgroupSize=45u; workgroupSize<=1024u; workgroupSize++)
 		{
 			core::smart_refctd_ptr<IGPUComputePipeline> pipelines[kTestTypeCount];
 			for (uint32_t i = 0u; i < kTestTypeCount; i++)

examples_tests/Readme.md

@@ -79,7 +79,7 @@ N = No support
 | 45.BRDFEvalTest                 | S            | S                | S            | S            | S               | S            | N                 | S              |                                                   |
 | 46.SamplingValidation           | S            | S                | S            | S            | S               | S            | N                 | S              |                                                   |
 | 47.DerivMapTest                 | Y            | Y                | B            | S            | N               | S            | N                 | N              |                                                   |
-| 48.ArithmeticUnitTest           | Y            | Y                | Y            | S            | N               | S            | N                 | S              |                                                   |
+| 48.ArithmeticUnitTest           | Y            | B                | B            | S            | S               | S            | N                 | S              |                                                   |
 | 49.ComputeFFT                   | S            | N                | S            | S            | N               | S            | N                 | N              |                                                   |
 | 50.NewAPITest                   | W            | W                | W            | W            | W               | W            | W                 | W              |                                                   |
 | 51.RadixSort                    | W            | N                | W            | W            | N               | W            | N                 | W              |                                                   |

include/nbl/core/EventDeferredHandler.h

@@ -105,7 +105,7 @@ class DeferredEventHandlerST
                 auto prev = mEvents.before_begin();
                 for (auto it = mEvents.begin(); it!=mEvents.end(); )
                 {
-                    if (it->m_event.wait_until(std::chrono::high_resolution_clock::now()+std::chrono::microseconds(250ull)))
+                    if (it->m_event.wait_until(std::chrono::steady_clock::now()+std::chrono::microseconds(250ull)))
                     {
                         it->m_function();
                         it = mEvents.erase_after(prev);

include/nbl/video/IGPUFence.h

@@ -65,10 +65,11 @@ class GPUEventWrapper : public core::Uncopyable
         return *this;
     }
 
-    template<class Clock=std::chrono::high_resolution_clock, class Duration=typename Clock::duration>
+    template<class Clock=std::chrono::steady_clock, class Duration=typename Clock::duration>
     inline static std::chrono::time_point<Clock,Duration> default_wait()
     {
-        return std::chrono::high_resolution_clock::now()+std::chrono::nanoseconds(50000ull); // 50 us
+        //return typename Clock::now()+std::chrono::nanoseconds(50000ull); // 50 us
+        return Clock::now()+std::chrono::nanoseconds(50000ull); // 50 us
     }
 
     IGPUFence::E_STATUS waitFenceWrapper(IGPUFence* fence, uint64_t timeout);

include/nbl/video/alloc/SubAllocatedDataBuffer.h

@@ -227,7 +227,7 @@ class SubAllocatedDataBuffer : protected core::impl::FriendOfHeterogenousMemoryA
             return multi_alloc(GPUEventWrapper::default_wait(),count,std::forward<Args>(args)...);
         }
         //!
-        template<class Clock=typename std::chrono::high_resolution_clock, typename... Args>
+        template<class Clock=typename std::chrono::steady_clock, typename... Args>
         inline size_type    multi_alloc(const std::chrono::time_point<Clock>& maxWaitPoint, const Args&... args) noexcept
         {
             #ifdef _NBL_DEBUG

include/nbl/video/utilities/CPropertyPoolHandler.h

@@ -187,7 +187,7 @@ class CPropertyPoolHandler final : public core::IReferenceCounted, public core::
 			StreamingTransientDataBufferMT<>* const upBuff, IGPUCommandBuffer* const cmdbuf, IGPUFence* const fence, IGPUQueue* const queue,
 			const asset::SBufferBinding<video::IGPUBuffer>& scratch, UpStreamingRequest*& requests, const uint32_t requestCount,
 			uint32_t& waitSemaphoreCount, IGPUSemaphore* const*& semaphoresToWaitBeforeOverwrite, const asset::E_PIPELINE_STAGE_FLAGS*& stagesToWaitForPerSemaphore,
-			system::logger_opt_ptr logger, const std::chrono::high_resolution_clock::time_point& maxWaitPoint=std::chrono::high_resolution_clock::now()+std::chrono::microseconds(500u)
+			system::logger_opt_ptr logger, const std::chrono::steady_clock::time_point& maxWaitPoint=std::chrono::steady_clock::now()+std::chrono::microseconds(500u)
 		);
 
 		// utility to help you fill out the tail move scatter request after the free, properly, returns if you actually need to transfer anything

include/nbl/video/utilities/IUtilities.h

@@ -317,7 +317,7 @@ namespace nbl::video
                 const uint32_t subSize = core::min(alllocationSize,size);
                 // cannot use `multi_place` because of the extra padding size we could have added
                 uint32_t localOffset = video::StreamingTransientDataBufferMT<>::invalid_address;
-                m_defaultUploadBuffer.get()->multi_alloc(std::chrono::high_resolution_clock::now()+std::chrono::microseconds(500u),1u,&localOffset,&alllocationSize,&alignment);
+                m_defaultUploadBuffer.get()->multi_alloc(std::chrono::steady_clock::now()+std::chrono::microseconds(500u),1u,&localOffset,&alllocationSize,&alignment);
                 // copy only the unpadded part
                 if (localOffset != video::StreamingTransientDataBufferMT<>::invalid_address)
                 {

src/nbl/video/COpenGLSync.cpp

@@ -32,7 +32,7 @@ uint64_t COpenGLSync::prewait() const
     if (sync)
         return 0ull;
 
-    using clock_t = std::chrono::high_resolution_clock;
+    using clock_t = std::chrono::steady_clock;
     auto start = clock_t::now();
     while (!sync)
     { 
@@ -53,7 +53,7 @@ COpenGLSync::E_STATUS COpenGLSync::waitCPU(IOpenGL_FunctionTable* _gl, uint64_t
         return (cachedRetval = ES_TIMEOUT_EXPIRED);
     timeout -= spintime;
 
-    GLenum status = _gl->glSync.pglClientWaitSync(sync, lockedTable?GL_SYNC_FLUSH_COMMANDS_BIT:0, timeout); // GL_SYNC_FLUSH_COMMANDS_BIT to flags?
+    GLenum status = _gl->glSync.pglClientWaitSync(sync, lockedTable ? GL_SYNC_FLUSH_COMMANDS_BIT:0, timeout);
     switch (status)
     {
     case GL_ALREADY_SIGNALED:

src/nbl/video/IOpenGL_LogicalDevice.h

@@ -852,7 +852,12 @@ class IOpenGL_LogicalDevice : public ILogicalDevice, protected impl::IOpenGL_Log
                         if(notFirstRun && now>=timeoutPoint)
                             return IGPUFence::ES_TIMEOUT;
                         else if (_waitAll) // all fences have to get signalled anyway so no use round robining
-                            timeout = std::chrono::duration_cast<std::chrono::nanoseconds>(timeoutPoint-now).count();
+                        {
+                            if (timeoutPoint>now)
+                                timeout = std::chrono::duration_cast<std::chrono::nanoseconds>(timeoutPoint-now).count();
+                            else
+                                timeout = 0ull;
+                        }
                         else if (i==0u) // if we're only looking for one to succeed then poll with increasing timeouts until deadline
                             timeout <<= 1u;
                     }

src/nbl/video/utilities/CPropertyPoolHandler.cpp

@@ -174,7 +174,7 @@ uint32_t CPropertyPoolHandler::transferProperties(
 	StreamingTransientDataBufferMT<>* const upBuff, IGPUCommandBuffer* const cmdbuf, IGPUFence* const fence, IGPUQueue* const queue,
 	const asset::SBufferBinding<video::IGPUBuffer>& scratch, UpStreamingRequest* &requests, const uint32_t requestCount,
 	uint32_t& waitSemaphoreCount, IGPUSemaphore* const*& semaphoresToWaitBeforeOverwrite, const asset::E_PIPELINE_STAGE_FLAGS*& stagesToWaitForPerSemaphore,
-	system::logger_opt_ptr logger, const std::chrono::high_resolution_clock::time_point& maxWaitPoint
+	system::logger_opt_ptr logger, const std::chrono::steady_clock::time_point& maxWaitPoint
 )
 {
 	if (!requestCount)