//-----------------------------------------------------------------------------
/// Return GPU-time in text format, to be parsed by the Client and displayed as its own timeline.
/// \return A line-delimited, ASCII-encoded, version of the GPU Trace data.
//-----------------------------------------------------------------------------
std::string VktTraceAnalyzerLayer::GetGPUTraceTXT()
{
gtASCIIString appendString = "";
VktFrameProfilerLayer* pProfilerLayer = VktFrameProfilerLayer::Instance();
WaitAndFetchResults(pProfilerLayer);
// During QueueSubmit we stored ProfilerResults in mEntriesWithProfilingResults. Form a response using it here.
ProfilerResultsMap& profiledCmdBufResultsMap = pProfilerLayer->GetCmdBufProfilerResultsMap();
// Gather all profiler results
if (!profiledCmdBufResultsMap.empty())
{
std::vector<ProfilerResult*> flatResults;
for (ProfilerResultsMap::iterator profIt = profiledCmdBufResultsMap.begin(); profIt != profiledCmdBufResultsMap.end(); ++profIt)
{
QueueWrapperToProfilingResultsMap& resultsPerThread = profIt->second;
for (QueueWrapperToProfilingResultsMap::iterator queuesIt = resultsPerThread.begin();
queuesIt != resultsPerThread.end();
++queuesIt)
{
// This structure holds all of the profiler results that were collected at QueueSubmit. The form is LinkId->ProfilerResult.
const SampleIdToProfilerResultMap* pResults = queuesIt->second;
for (SampleIdToProfilerResultMap::const_iterator sampleIdIt = pResults->begin();
sampleIdIt != pResults->end();
++sampleIdIt)
{
ProfilerResult* pResult = sampleIdIt->second;
pResult->measurementInfo.idInfo.pWrappedQueue = queuesIt->first;
flatResults.push_back(pResult);
}
}
}
const UINT numResults = (UINT)flatResults.size();
// We'll need to insert the GPU Trace section header before the response data, even if there aren't any results.
appendString += "//==GPU Trace==";
appendString += "\n";
appendString += "//API=";
appendString += GetAPIString();
appendString += "\n";
appendString += "//CommandBufEventCount=";
appendString += IntToString((INT)numResults);
appendString += "\n";
#ifdef WIN32
sort(flatResults.begin(), flatResults.end(), SortByStartTime);
for (UINT i = 0; i < numResults; i++)
{
ProfilerResultToStr(flatResults[i], appendString);
}
#else
ProfilerResult* pFlatResults = new ProfilerResult[numResults];
for (UINT i = 0; i < numResults; i++)
{
pFlatResults[i] = *(flatResults[i]);
}
qsort(pFlatResults, numResults, sizeof(ProfilerResult), (compfn)SortByStartTime);
for (UINT i = 0; i < numResults; i++)
{
ProfilerResultToStr(&pFlatResults[i], appendString);
}
delete[] pFlatResults;
pFlatResults = nullptr;
#endif
}
else
{
appendString += "NODATA";
}
return appendString.asCharArray();
}
//-----------------------------------------------------------------------------
/// Submit command buffers and gather results.
/// \param queue The queue issued work to.
/// \param submitCount The number of submits.
/// \param pSubmits The submit info structures.
/// \param fence The fence wrapping this submit.
//-----------------------------------------------------------------------------
VkResult VktWrappedQueue::QueueSubmit(VkQueue queue, uint32_t submitCount, const VkSubmitInfo* pSubmits, VkFence fence)
{
m_executionID++;
VkResult result = VK_INCOMPLETE;
VktTraceAnalyzerLayer* pTraceAnalyzer = VktTraceAnalyzerLayer::Instance();
VktFrameProfilerLayer* pFrameProfiler = VktFrameProfilerLayer::Instance();
// Use this calibration timestamp structure to convert GPU events to the CPU timeline.
CalibrationTimestampPair calibrationTimestamps = {};
calibrationTimestamps.mQueueCanBeTimestamped = true;
VkFence fenceToWaitOn = fence;
bool usingInternalFence = false;
std::vector<VktWrappedCmdBuf*> wrappedCmdBufs;
GatherWrappedCommandBufs(submitCount, pSubmits, wrappedCmdBufs);
for (UINT i = 0; i < wrappedCmdBufs.size(); i++)
{
wrappedCmdBufs[i]->SetProfilerExecutionId(m_executionID);
wrappedCmdBufs[i]->IncrementSubmitCount();
}
// Surround the execution of CommandBuffers with timestamps so we can determine when the GPU work occurred in the CPU timeline.
if (pTraceAnalyzer->ShouldCollectTrace() && pFrameProfiler->ShouldCollectGPUTime())
{
// Collect calibration timestamps in case we need to align GPU events against the CPU timeline.
if (calibrationTimestamps.mQueueCanBeTimestamped)
{
pFrameProfiler->CollectCalibrationTimestamps(this, &calibrationTimestamps);
}
else
{
Log(logTRACE, "Did not collect calibration timestamps for Queue '0x%p'\n", this);
}
// Inject our own fence if the app did not supply one
if (fenceToWaitOn == VK_NULL_HANDLE)
{
// Create internal fence
VkFenceCreateInfo fenceCreateInfo = {};
VkResult fenceResult = VK_INCOMPLETE;
fenceResult = device_dispatch_table(queue)->CreateFence(m_createInfo.device, &fenceCreateInfo, nullptr, &fenceToWaitOn);
VKT_ASSERT(fenceResult == VK_SUCCESS);
usingInternalFence = true;
}
}
// Invoke the real call to execute on the GPU
result = QueueSubmit_ICD(queue, submitCount, pSubmits, fenceToWaitOn);
if (pTraceAnalyzer->ShouldCollectTrace() && pFrameProfiler->ShouldCollectGPUTime())
{
// Collect the CPU and GPU frequency to convert timestamps.
QueryPerformanceFrequency(&calibrationTimestamps.cpuFrequency);
#if GATHER_PROFILER_RESULTS_WITH_WORKERS
SpawnWorker(&calibrationTimestamps, this, fenceToWaitOn, usingInternalFence, wrappedCmdBufs);
#else
VkResult waitResult = VK_TIMEOUT;
#if GPU_FENCES_FOR_PROFILER_WAIT
do
{
waitResult = device_dispatch_table(m_createInfo.device)->WaitForFences(m_createInfo.device, 1, &fenceToWaitOn, VK_TRUE, GPU_FENCE_TIMEOUT_TIME);
} while (waitResult == VK_TIMEOUT);
#else
waitResult = device_dispatch_table(queue)->QueueWaitIdle(queue);
#endif
if (calibrationTimestamps.mQueueCanBeTimestamped)
{
// Put all results into thread ID 0 bucket
const UINT32 threadID = 0;
std::vector<ProfilerResult> results;
for (UINT i = 0; i < wrappedCmdBufs.size(); i++)
{
ProfilerResultCode getResultsResult = PROFILER_FAIL;
getResultsResult = wrappedCmdBufs[i]->GetCmdBufResultsST(results);
VKT_ASSERT(getResultsResult != PROFILER_FAIL);
}
pFrameProfiler->VerifyAlignAndStoreResults(this, results, &calibrationTimestamps, threadID, VktTraceAnalyzerLayer::Instance()->GetFrameStartTime());
// Free the fence we created earlier
if (usingInternalFence)
{
device_dispatch_table(m_createInfo.device)->DestroyFence(m_createInfo.device, fenceToWaitOn, nullptr);
}
}
else
{
Log(logTRACE, "Didn't collect calibration timestamps for Queue '0x%p'.\n", this);
}
#endif
}
#if GATHER_PROFILER_RESULTS_WITH_WORKERS == 0
for (UINT i = 0; i < wrappedCmdBufs.size(); i++)
{
wrappedCmdBufs[i]->DestroyDynamicProfilers();
}
#endif
return result;
}
//-----------------------------------------------------------------------------
/// Return GPU-time in text format, to be parsed by the Client and displayed as its own timeline.
/// \return A line-delimited, ASCII-encoded, version of the GPU Trace data.
//-----------------------------------------------------------------------------
std::string VktTraceAnalyzerLayer::GetGPUTraceTXT()
{
gtASCIIString appendString = "";
VktFrameProfilerLayer* pProfilerLayer = VktFrameProfilerLayer::Instance();
WaitAndFetchResults(pProfilerLayer);
// Query the CPU clock frequency so we can accurately convert to wall clock time.
GPS_TIMESTAMP cpuClockFrequency;
QueryPerformanceFrequency(&cpuClockFrequency);
// During QueueSubmit we stored ProfilerResults in mEntriesWithProfilingResults. Form a response using it here.
ProfilerResultsMap& profiledCmdBufResultsMap = pProfilerLayer->GetCmdBufProfilerResultsMap();
// During QueueSubmit we stored ProfilerResults in mEntriesWithProfilingResults. Form a response using it here.
if (!profiledCmdBufResultsMap.empty())
{
// Keep a count of the number of lines that we'll write in the response string.
UINT numResponseLines = 0;
gtASCIIString profiledCommandsLinesStr;
ProfilerResultsMap::iterator profIt;
for (profIt = profiledCmdBufResultsMap.begin(); profIt != profiledCmdBufResultsMap.end(); ++profIt)
{
QueueWrapperToProfilingResultsMap::iterator queuesWithProfilingResults;
QueueWrapperToProfilingResultsMap& resultsPerThread = profIt->second;
for (queuesWithProfilingResults = resultsPerThread.begin(); queuesWithProfilingResults != resultsPerThread.end(); ++queuesWithProfilingResults)
{
VktWrappedQueue* pWrappedQueue = queuesWithProfilingResults->first;
const double timestampFrequency = pWrappedQueue->GetTimestampFrequency();
// This structure holds all of the profiler results that were collected at QueueSubmit. The form is LinkId->ProfilerResult.
const SampleIdToProfilerResultMap* pQueueResults = queuesWithProfilingResults->second;
SampleIdToProfilerResultMap::const_iterator queueResultsIterInner;
for (queueResultsIterInner = pQueueResults->begin(); queueResultsIterInner != pQueueResults->end(); ++queueResultsIterInner)
{
UINT64 sampleId = queueResultsIterInner->first;
const ProfilerResult* pResult = queueResultsIterInner->second;
// Convert timestamps to milliseconds by using the clock frequency.
#ifndef CODEXL_GRAPHICS
double preStartTimestamp = (pResult->timestampResult.rawClocks.preStart / timestampFrequency) * 1000.0;
#endif
double startTimestamp = (pResult->timestampResult.rawClocks.start / timestampFrequency) * 1000.0;
double endTimestamp = (pResult->timestampResult.rawClocks.end / timestampFrequency) * 1000.0;
FuncId funcId = (FuncId)pResult->measurementInfo.idInfo.funcId;
gtASCIIString funcName = GetFunctionNameFromId(funcId);
gtASCIIString retVal = "void";
gtASCIIString params = "";
if (pResult->measurementInfo.idInfo.funcId != FuncId_WholeCmdBuf)
{
VktAPIEntry* pResultEntry = pProfilerLayer->FindInvocationBySampleId(sampleId);
if (pResultEntry != nullptr)
{
// Convert the functionID and return values from integers into full strings that we can use in the response.
funcName = GetFunctionNameFromId(pResultEntry->mFunctionId);
retVal = (pResultEntry->m_returnValue != -1) ? VktUtil::WriteResultCodeEnumAsString(pResultEntry->m_returnValue) : "void";
params = pResultEntry->mParameters.asCharArray();
}
}
UINT queueIndex = pWrappedQueue->GetQueueIndex();
#ifndef CODEXL_GRAPHICS
std::string queueInfo = "";
#else
std::string queueInfo = "0";
#endif
// Vulkan Response line format:
// CommandQueuePtr CommandBufferType CommandBufferPtr APIType FuncId Vulkan_FuncName(Args) = ReturnValue StartTime EndTime SampleId
profiledCommandsLinesStr += "0x";
profiledCommandsLinesStr += IntToString(queueIndex);
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += queueInfo.c_str();
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += "0x";
profiledCommandsLinesStr += UINT64ToHexString((UINT64)pResult->measurementInfo.idInfo.pWrappedCmdBuf->AppHandle());
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += IntToString(VktTraceAnalyzerLayer::Instance()->GetAPIGroupFromAPI(funcId));
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += IntToString(funcId);
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += "Vulkan_";
profiledCommandsLinesStr += funcName;
profiledCommandsLinesStr += "(";
profiledCommandsLinesStr += params;
profiledCommandsLinesStr += ") = ";
profiledCommandsLinesStr += retVal;
#ifndef CODEXL_GRAPHICS
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += DoubleToString(preStartTimestamp);
#endif
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += DoubleToString(startTimestamp);
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += DoubleToString(endTimestamp);
profiledCommandsLinesStr += " ";
profiledCommandsLinesStr += UINT64ToString(pResult->measurementInfo.idInfo.sampleId);
profiledCommandsLinesStr += "\n";
// We just added another line to our response buffer. Increment the count that the client will read.
numResponseLines++;
}
}
}
// We'll need to insert the GPU Trace section header before the response data, even if there aren't any results.
appendString += "//==GPU Trace==";
appendString += "\n";
appendString += "//CommandBufEventCount=";
appendString += DWORDToString(numResponseLines);
appendString += "\n";
// Include the response lines after the section header.
if (numResponseLines > 0)
{
appendString += profiledCommandsLinesStr;
}
}
else
{
appendString += "NODATA";
}
return appendString.asCharArray();
}
