/// Hook the load library functions
void HookLoadLibrary()
{
LogTrace(traceENTER, "");
AMDT::BeginHook();
LONG error = AMDT::HookAPICall(&(PVOID&)Real_LoadLibraryA, Mine_LoadLibraryA);
PsAssert(error == NO_ERROR);
error = AMDT::HookAPICall(&(PVOID&)Real_LoadLibraryExA, Mine_LoadLibraryExA);
PsAssert(error == NO_ERROR);
error = AMDT::HookAPICall(&(PVOID&)Real_LoadLibraryW, Mine_LoadLibraryW);
PsAssert(error == NO_ERROR);
error = AMDT::HookAPICall(&(PVOID&)Real_LoadLibraryExW, Mine_LoadLibraryExW);
PsAssert(error == NO_ERROR);
error = AMDT::HookAPICall(&(PVOID&)Real_FreeLibrary, Mine_FreeLibrary);
PsAssert(error == NO_ERROR);
if (AMDT::EndHook() != NO_ERROR)
{
Log(logERROR, "HookLoadLibrary() failed\n");
}
// Load the "dxgi.dll" at the beginning as we missed the dxgi loading sometimes.
// TODO: Fix this: This function is called from DllMain(), and according to the
// Microsoft documentation, LoadLibrary() should not be called from DllMain()
LoadLibraryA("dxgi.dll");
LogTrace(traceEXIT, "");
return;
}
//-----------------------------------------------------------------------------
/// Release global shared memory region so it can be accessed by other processes.
//-----------------------------------------------------------------------------
void SharedGlobal::Unlock(void)
{
PsAssert(this != NULL);
PsAssert(m_Mutex != NULL);
m_Mutex->unlock();
}
//-----------------------------------------------------------
// Stops capturing D3DPerf markers
//-----------------------------------------------------------
LONG CAPTURE_D3DPerfMarkers_Unhook()
{
LogTrace(traceENTER, "");
ScopeLock t(&s_mtx);
if (dwAttached <= 0)
{
Log(logERROR, "Trying to Detach ID3D10Device twice!\n");
return -1;
}
s_pCaptureLayer = NULL;
LONG error;
AMDT::BeginHook();
if (Real_D3DPERF_BeginEvent != NULL)
{
error = AMDT::UnhookAPICall(&(PVOID&)Real_D3DPERF_BeginEvent, Mine_D3DPERF_BeginEvent);
PsAssert(error == NO_ERROR);
}
if (Real_D3DPERF_EndEvent != NULL)
{
error = AMDT::UnhookAPICall(&(PVOID&)Real_D3DPERF_EndEvent, Mine_D3DPERF_EndEvent);
PsAssert(error == NO_ERROR);
}
if (Real_D3DPERF_SetMarker != NULL)
{
error = AMDT::UnhookAPICall(&(PVOID&)Real_D3DPERF_SetMarker, Mine_D3DPERF_SetMarker);
PsAssert(error == NO_ERROR);
}
if (Real_D3DPERF_SetRegion != NULL)
{
error = AMDT::UnhookAPICall(&(PVOID&)Real_D3DPERF_SetRegion, Mine_D3DPERF_SetRegion);
PsAssert(error == NO_ERROR);
}
error = AMDT::EndHook();
if (error != NO_ERROR)
{
Log(logERROR, "failed\n");
}
else
{
dwAttached--;
}
LogTrace(traceEXIT, "");
return error;
}
//-----------------------------------------------------------------------------
/// Lock global shared memory region for exclusive access.
/// \return true if the shared memory could be locked; false otherwise
//-----------------------------------------------------------------------------
bool SharedGlobal::Lock(void)
{
PsAssert(this != NULL);
PsAssert(m_Mutex != NULL);
if (m_Mutex->lock() == false)
{
Log(logERROR, "Error occurred while waiting for Mutex :%d\n", osGetLastSystemError());
return false;
}
return true;
}
//--------------------------------------------------------------------------
/// Returns the values of editable commands
/// \return XML containing the values of editable commands
//--------------------------------------------------------------------------
string CommandProcessor::GetEditableCommandValues()
{
stringstream strOut;
// give derived classes a chance to update / add settings.
strOut << GetDerivedSettings();
// now add this processors editable commands (Settings).
CommandList::const_iterator objIter;
for (objIter = m_Commands.begin(); objIter < m_Commands.end(); ++objIter)
{
CommandResponse* pObj = *objIter;
PsAssert(pObj != NULL);
if (pObj->GetEditableContent() != NOT_EDITABLE)
{
strOut << "<";
strOut << pObj->GetTagName();
strOut << " name='";
strOut << pObj->GetDisplayName();
strOut << "' url='";
strOut << GetFullPathString().asCharArray();
strOut << "'>";
strOut << pObj->GetEditableContentValue();
strOut << "</";
strOut << pObj->GetTagName();
strOut << ">";
}
}
return strOut.str();
}
////////////////////////////////////////////////////////////////////////
/// Writes a message to the console
/// \param msg The message to display.
////////////////////////////////////////////////////////////////////////
void ClientRequestThread::OutputScreenMessage(const char* msg)
{
PsAssert(s_output_criticalSection != NULL);
s_output_criticalSection->enter();
LogConsole(logMESSAGE, "%s\n", msg);
s_output_criticalSection->leave();
}
////////////////////////////////////////////////////////////////////////
/// Writes an error to the console. It clears the error before exiting.
/// \param errmsg The message to display.
////////////////////////////////////////////////////////////////////////
void ClientRequestThread::OutputScreenError(const char* errmsg)
{
PsAssert(s_output_criticalSection != NULL);
s_output_criticalSection->enter();
LogConsole(logERROR, "%s - %i\n", errmsg, NetSocket::LastError());
s_output_criticalSection->leave();
}
void DestroyResponse(CommunicationID& rRequestID, Response** ppResponse)
{
PsAssert(ppResponse != NULL);
if (*ppResponse == NULL)
{
return;
}
// protect the maps from being changed by other threads using the mutex
ScopeLock lock(s_mutex);
// remove from streaming map
if ((*ppResponse)->m_bStreamingEnabled)
{
ResponseMap::iterator iter = g_streamingResponseMap.find(rRequestID);
if (iter != g_streamingResponseMap.end())
{
g_streamingResponseMap.erase(iter);
}
}
SAFE_DELETE(*ppResponse);
rRequestID = 0;
}
//-----------------------------------------------------------------------------
/// Set a path in the global data block. All paths are assumed to be PS_MAX_PATH size.
///
/// \param offset The offset from the start of the structure to the path of interest.
/// It is calculated automatically at compile time using the offsetof functionality
/// through the SG_SET_PATH macro
/// \param path The path to set at the specified offset
/// \return true if the path could be set; false otherwise
//-----------------------------------------------------------------------------
bool SharedGlobal::SetPath(size_t offset, const char* path)
{
PsAssert(this != NULL);
PsAssert(path != NULL);
PsAssert(m_MapFile != NULL);
PsAssert(m_MapFile->Get() != NULL);
if (Lock())
{
strcpy_s(&((char*)m_MapFile->Get())[offset], PS_MAX_PATH, path);
Unlock();
return (true);
}
return (false);
}
//--------------------------------------------------------------------------
/// Registers the specified CommandProcessor
/// \param pTagName The XML tag name of the processor.
/// \param pDisplayName The name to display in the client for this command
/// \param pID The ID that should target a command to the CommandProcessor
/// \param pTitlePrefix A customizable prefix that is prepended to the display name
/// \param eDisplayMode Indicates whether or not to display the command in the client
/// \param rComProc The commandProcessor that should be targeted when a
/// command with the supplied ID is received
//--------------------------------------------------------------------------
void CommandProcessor::AddProcessor(const char* pTagName,
const char* pDisplayName,
const char* pID,
const char* pTitlePrefix,
UIDisplayMode eDisplayMode,
CommandProcessor& rComProc)
{
PsAssert(pTagName != NULL);
PsAssert(pDisplayName != NULL);
PsAssert(pID != NULL);
rComProc.SetTagName(pTagName);
rComProc.SetID(pID);
rComProc.SetDisplayName(pDisplayName);
rComProc.SetTitlePrefix(pTitlePrefix);
rComProc.SetUIDisplayMode(eDisplayMode);
rComProc.SetParent(this);
m_Processors.push_back(&rComProc);
}
//-----------------------------------------------------------------------------
/// Verify, align, and store the new profiler results.
/// \param pQueue The Queue used to collect the results to verify.
/// \param results The vector of profiler results to verify.
/// \param pTimestampPair A pair of calibration timestamps used to align CPU and GPU timestamps.
/// \param threadID The ThreadId that the results are collected with.
/// \param frameStartTime The start time of the frame as collected on the CPU.
//-----------------------------------------------------------------------------
void VktFrameProfilerLayer::VerifyAlignAndStoreResults(
VktWrappedQueue* pQueue,
std::vector<ProfilerResult>& results,
CalibrationTimestampPair* pTimestampPair,
UINT32 threadID,
GPS_TIMESTAMP frameStartTime)
{
#if MANUAL_TIMESTAMP_CALIBRATION == 0
UNREFERENCED_PARAMETER(pTimestampPair);
UNREFERENCED_PARAMETER(frameStartTime);
#endif
SampleIdToProfilerResultMap* pResultMap = FindOrCreateProfilerResultsMap(pQueue, threadID);
PsAssert(pResultMap != nullptr);
if (pResultMap != nullptr)
{
for (size_t resultIndex = 0; resultIndex < results.size(); ++resultIndex)
{
ProfilerResult& currentResult = results[resultIndex];
const UINT64 sampleId = currentResult.measurementInfo.idInfo.sampleId;
// Verify that the timestamps retrieved from the profiler appear to be valid.
if (ValidateProfilerResult(currentResult) == true)
{
#if MANUAL_TIMESTAMP_CALIBRATION
// Now attempt to align the profiled GPU timestamps with the traced API calls on the CPU.
bool bAlignedSuccessfully = AlignProfilerResultWithCPUTimeline(currentResult, pTimestampPair, frameStartTime);
#else
bool bAlignedSuccessfully = true;
#endif
// @TODO - determine if this is what we want to do
// Make zero-duration case equal to 1 clock cycle
if (currentResult.timestampResult.rawClocks.start == currentResult.timestampResult.rawClocks.end)
{
currentResult.timestampResult.rawClocks.end++;
}
if (bAlignedSuccessfully)
{
// Store the final adjusted profiler results if they're valid.
ProfilerResult* pNewResult = new ProfilerResult;
CopyProfilerResult(pNewResult, ¤tResult);
(*pResultMap)[sampleId] = pNewResult;
}
else
{
Log(logERROR, "Command with SampleId %d failed to align with CPU timeline.\n", sampleId);
}
}
}
}
}
//--------------------------------------------------------------------------
/// Registers the specified CommandObject
/// \param eType The type of content that this command should be responded with
/// \param pTagName The name to assign to nodes in XML in the CommandTree
/// \param pDisplayName The name to give the CommandResponse object
/// \param pURL The URL that should activate the CommandResponse object
/// \param eDisplayMode Indicates whether or not to display the command in the client
/// \param eIncludeFlag Indicates whether or not to include the command in the CommandTree
/// \param rComObj The CommandResponse object that will be activated when
/// the supplied URL is received as a command
//--------------------------------------------------------------------------
void CommandProcessor::AddCommand(ContentType eType,
const char* pTagName,
const char* pDisplayName,
const char* pURL,
UIDisplayMode eDisplayMode,
TreeInclude eIncludeFlag,
CommandResponse& rComObj)
{
PsAssert(pTagName != NULL);
PsAssert(pDisplayName != NULL);
PsAssert(pURL != NULL);
rComObj.SetTagName(pTagName);
rComObj.SetDisplayName(pDisplayName);
rComObj.SetURL(pURL);
rComObj.SetContentType(eType);
rComObj.SetUIDisplayMode(eDisplayMode);
rComObj.SetTreeInclude(eIncludeFlag);
m_Commands.push_back(&rComObj);
}
//-----------------------------------------------------------------------------
/// Get a path from the global data block. All paths are assumed to be PS_MAX_PATH size.
///
/// \param offset The offset from the start of the structure to the path of interest.
/// It is calculated automatically at compile time using the offsetof functionality
/// through the SG_GET_PATH macro
///
/// \return A shadow copy of the path if maintained in the class. This allows us to directly pass back
/// a char * without concerns about locking/unlocking the shared memory region.
//-----------------------------------------------------------------------------
const char* SharedGlobal::GetPath(size_t offset)
{
PsAssert(this != NULL);
PsAssert(m_MapFile != NULL);
PsAssert(m_MapFile->Get() != NULL);
char* src = & ((char*)m_MapFile->Get())[offset];
char* dst = & ((char*) & m_Shadow)[offset];
size_t nSize = sizeof((&m_Shadow)[offset]);
if (Lock())
{
// copy shared memory copy of string into local shadow copy
memcpy_s(dst, nSize, src, PS_MAX_PATH);
Unlock();
return (dst);
}
return (NULL);
}
//-----------------------------------------------------------------------------
/// Provides access to the single instance of this class in the process. If the
/// class does not exist, it will initialize it and log any errors that occur.
/// \return pointer to the instance of this class.
//-----------------------------------------------------------------------------
SharedGlobal* SharedGlobal::Instance(void)
{
static SharedGlobal* sg = new SharedGlobal;
PsAssert(sg != NULL);
if (!sg->m_bInitialized && !sg->Initialize())
{
LogConsole(logERROR, "Unable to create SharedGlobal data\n");
delete sg;
sg = NULL;
}
return (sg);
}
/// Unhook the load library functions
void UnhookLoadLibrary()
{
LogTrace(traceENTER, "");
AMDT::BeginHook();
LONG error = AMDT::UnhookAPICall(&(PVOID&)Real_LoadLibraryA, Mine_LoadLibraryA);
PsAssert(error == NO_ERROR);
error = AMDT::UnhookAPICall(&(PVOID&)Real_LoadLibraryExA, Mine_LoadLibraryExA);
PsAssert(error == NO_ERROR);
error = AMDT::UnhookAPICall(&(PVOID&)Real_LoadLibraryW, Mine_LoadLibraryW);
PsAssert(error == NO_ERROR);
error = AMDT::UnhookAPICall(&(PVOID&)Real_LoadLibraryExW, Mine_LoadLibraryExW);
PsAssert(error == NO_ERROR);
error = AMDT::UnhookAPICall(&(PVOID&)Real_FreeLibrary, Mine_FreeLibrary);
PsAssert(error == NO_ERROR);
if (AMDT::EndHook() != NO_ERROR)
{
Log(logERROR, "UnhookLoadLibrary() failed\n");
}
// Restore Real functions to original values in case they aren't restored correctly by the unhook call
Real_LoadLibraryA = LoadLibraryA;
Real_LoadLibraryExA = LoadLibraryExA;
Real_LoadLibraryW = LoadLibraryW;
Real_LoadLibraryExW = LoadLibraryExW;
Real_FreeLibrary = FreeLibrary;
LogTrace(traceEXIT, "");
return;
}
//--------------------------------------------------------------------------
/// Allows the application to send an error response for this command.
/// These errors also get added to the log
/// \param pError Error message to send
//--------------------------------------------------------------------------
void CommandResponse::SendError(const char* pError, ...)
{
PsAssert(pError != NULL);
char errString[ COMMAND_MAX_LENGTH ];
va_list arg_ptr;
// we prepend "Error: to the string for the HTTP response - but not for the logfile
// pLogString tracks the start of the formatted string to pass to the Logfile
int nLen = sprintf_s(errString, COMMAND_MAX_LENGTH, "Error: ");
if (nLen < 0)
{
Log(logERROR, "String length is less than 0\n");
return;
}
char* pLogString = &errString[nLen];
va_start(arg_ptr, pError);
vsprintf_s(pLogString, COMMAND_MAX_LENGTH - nLen, pError, arg_ptr);
va_end(arg_ptr);
Log(logERROR, "%s\n", pLogString);
// send the error back to all received requests
for (std::list< CommunicationID >::const_iterator iRequestID = m_requestIDs.begin(); iRequestID != m_requestIDs.end(); ++iRequestID)
{
bool bResult = SendResponse(*iRequestID, "text/plain", errString, (unsigned int) strlen(errString), m_bStreamingEnabled);
if (bResult == false)
{
Log(logERROR, "Failed to send error to request %u\n", *iRequestID);
m_bStreamingEnabled = false;
m_eResponseState = ERROR_SENDING_RESPONSE;
}
}
if (m_bStreamingEnabled == false)
{
// we sent a response and the request was not for streaming data, so unset the request
m_requestIDs.clear();
}
}
//--------------------------------------------------------------------------
/// Processes the specified CommandObject to see if it is targeting any of
/// the added Processors
/// \param rIncomingCommand the incoming command that should be handled
/// \return true if the command could be processed; false otherwise
//--------------------------------------------------------------------------
bool CommandProcessor::ProcessProcessors(CommandObject& rIncomingCommand)
{
ProcessorList::const_iterator it;
for (it = m_Processors.begin() ; it < m_Processors.end(); ++it)
{
CommandProcessor* pProc = *it;
PsAssert(pProc != NULL);
if (rIncomingCommand.IsCommand(pProc->GetID()))
{
pProc->Process(rIncomingCommand);
return true;
}
}
return false;
}
//--------------------------------------------------------------------------
/// Returns the number of commands that are editable.
/// \return Number of editable commands (0-n).
//--------------------------------------------------------------------------
unsigned int CommandProcessor::GetEditableCount()
{
unsigned int nCount = 0;
CommandList::const_iterator objIter;
for (objIter = m_Commands.begin(); objIter < m_Commands.end(); ++objIter)
{
CommandResponse* pObj = *objIter;
PsAssert(pObj != NULL);
if (pObj->GetEditableContent() != NOT_EDITABLE)
{
nCount++;
}
}
return nCount;
}
//-----------------------------------------------------------------------------
/// ShouldResponseBeSent
///
/// indicates whether a response should be sent to the specified request based
/// on whether or not the request is streaming and is rate limited. If the
/// request is rate limited, but a response can be sent, the "last sent time"
/// will be updated if bUpdateTime is true
///
/// \param requestID id of the request that may or may not be rate limited
/// \param bUpdateTime indicates whether or not the "last sent time" will be
/// updated if the response is rate limited, but allowed to be sent
///
/// \return true if a response should NOT be sent; false otherwise
//-----------------------------------------------------------------------------
bool ShouldResponseBeSent(CommunicationID requestID, bool bUpdateTime)
{
// protect the maps from being changed by other threads using the mutex
ScopeLock lock(s_mutex);
ResponseMap::iterator iterResponse = g_streamingResponseMap.find(requestID);
if (iterResponse == g_streamingResponseMap.end())
{
// don't limit the send because we don't even know it is streaming
return false;
}
Response* pResponse = iterResponse->second;
PsAssert(pResponse != NULL);
// if this is a streaming request, only send if rate allows
if (pResponse->m_bStreamingEnabled == true)
{
if (pResponse->m_dwMaxStreamsPerSecond == COMM_MAX_STREAM_RATE ||
pResponse->m_dwMaxStreamsPerSecond == 0)
{
return false;
}
DWORD dwCurrTime = g_streamTimer.GetAbsolute();
if (dwCurrTime - pResponse->m_dwLastSent >= 1000 / pResponse->m_dwMaxStreamsPerSecond)
{
if (bUpdateTime)
{
pResponse->m_dwLastSent = dwCurrTime;
}
return false;
}
else
{
return true;
}
}
return false;
}
//-----------------------------------------------------------------------------
/// Get the map used to associate queue with its command lists.
/// \param inQueue Input queue.
/// \returns The CommandQueueToCommandListMap used to associate a Queue with the CommandLists it utilizes.
//-----------------------------------------------------------------------------
D3D12_COMMAND_LIST_TYPE DX12FrameProfilerLayer::GetCommandListTypeFromCommandQueue(Wrapped_ID3D12CommandQueue* inQueue)
{
// Use the metadata object to retrieve the cached CreateInfo, and then pull the type out. Start with "Direct".
D3D12_COMMAND_LIST_TYPE commandListType = D3D12_COMMAND_LIST_TYPE_DIRECT;
PsAssert(inQueue != nullptr);
if (inQueue != nullptr)
{
DX12ObjectDatabaseProcessor* databaseProcessor = DX12ObjectDatabaseProcessor::Instance();
DX12WrappedObjectDatabase* objectDatabase = static_cast<DX12WrappedObjectDatabase*>(databaseProcessor->GetObjectDatabase());
IDX12InstanceBase* queueMetadata = objectDatabase->GetMetadataObject((IUnknown*)inQueue);
Wrapped_ID3D12CommandQueueCreateInfo* queueCreateInfo = static_cast<Wrapped_ID3D12CommandQueueCreateInfo*>(queueMetadata->GetCreateInfoStruct());
D3D12_COMMAND_QUEUE_DESC* queueDesc = queueCreateInfo->GetDescription();
commandListType = queueDesc->Type;
}
return commandListType;
}
//-----------------------------------------------------------------------------
/// Set the internal flag that determines if GPU command profiling is enabled.
/// \param inbProfilingEnabled The flag used to enable or disable profiling.
//-----------------------------------------------------------------------------
void DX12FrameProfilerLayer::SetProfilingEnabled(bool inbProfilingEnabled)
{
ModernAPIFrameProfilerLayer::SetProfilingEnabled(inbProfilingEnabled);
for (CommandQueueToCommandListMap::iterator it = mCommandQueueTracker.begin();
it != mCommandQueueTracker.end();
++it)
{
Wrapped_ID3D12Device* pDevice = static_cast<Wrapped_ID3D12Device*>(DX12Util::SafeGetDevice(it->first));
if (pDevice != nullptr)
{
HRESULT stateSet = E_FAIL;
stateSet = pDevice->mRealDevice->SetStablePowerState(inbProfilingEnabled);
PsAssert(stateSet == S_OK);
pDevice->Release();
}
}
}
bool IsToken(char** sIn, const char* sTok)
{
size_t dwTokLen = strlen(sTok);
size_t nStringLength = strlen(*sIn);
if (_strnicmp(*sIn, sTok, dwTokLen) == 0)
{
// Check to see if we are going to increment the pointer beond the end of the array.
PsAssert(dwTokLen <= nStringLength);
if (dwTokLen > nStringLength)
{
Log(logERROR, "IsToken: buffer overrun. Str = %s, Tok = %s\n", *sIn, sTok);
return false;
}
*sIn += dwTokLen;
return true;
}
return false;
}
//-----------------------------------------------------------------------------
/// Verify, align, and store the new profiler results.
/// \param pQueue The Queue used to collect the results to verify.
/// \param results The vector of profiler results to verify.
/// \param pTimestampPair A pair of calibration timestamps used to align CPU and GPU timestamps.
/// \param threadID The ThreadId that the results are collected with.
/// \param frameStartTime The start time of the frame as collected on the CPU.
//-----------------------------------------------------------------------------
void DX12FrameProfilerLayer::VerifyAlignAndStoreResults(
Wrapped_ID3D12CommandQueue* pQueue,
std::vector<ProfilerResult>& results,
CalibrationTimestampPair* pTimestampPair,
UINT32 threadID,
GPS_TIMESTAMP frameStartTime)
{
SampleIdToProfilerResultMap* pResultMap = FindOrCreateProfilerResultsMap(pQueue, threadID);
PsAssert(pResultMap != nullptr);
if (pResultMap != nullptr)
{
for (size_t resultIndex = 0; resultIndex < results.size(); ++resultIndex)
{
ProfilerResult& currentResult = results[resultIndex];
const UINT64 sampleId = currentResult.measurementInfo.idInfo.mSampleId;
// Verify that the timestamps retrieved from the profiler appear to be valid.
if (ValidateProfilerResult(currentResult) == true)
{
// Now attempt to align the profiled GPU timestamps with the traced API calls on the CPU.
bool bAlignedSuccessfully = AlignProfilerResultWithCPUTimeline(currentResult, pTimestampPair, frameStartTime);
if (bAlignedSuccessfully)
{
// Store the final adjusted profiler results if they're valid.
ProfilerResult* pNewResult = new ProfilerResult;
CopyProfilerResult(pNewResult, ¤tResult);
(*pResultMap)[sampleId] = pNewResult;
}
else
{
Log(logERROR, "Command with SampleId %d failed to align with CPU timeline.\n", sampleId);
}
}
}
}
}
//--------------------------------------------------------------------------
/// Retrieve an existing SampleInfo instance, or create a new one, for the given ThreadId.
/// \param inThreadId The current thread requesting a SampleInfo structure.
/// \returns A SampleInfo instance to be used with the given ThreadId.
//--------------------------------------------------------------------------
ModernAPIFrameProfilerLayer::SampleInfo* ModernAPIFrameProfilerLayer::GetSampleInfoForThread(DWORD inThreadId)
{
SampleInfo* pResult = nullptr;
if (mSampleIdMap.find(inThreadId) != mSampleIdMap.end())
{
// mSampleIdMap already has a key for the incoming thread. Update the current sampleInfo.
pResult = mSampleIdMap[inThreadId];
}
else
{
// We need to insert a new key into the mSampleIdMap. Need to lock it first.
ScopeLock sampleLock(&mUniqueSampleIdMutex);
pResult = new SampleInfo;
mSampleIdMap[inThreadId] = pResult;
}
PsAssert(pResult != nullptr);
return pResult;
}
//--------------------------------------------------------------------------
/// \param pName the full name of this object
//--------------------------------------------------------------------------
void CommandResponse::SetDisplayName(const char* pName)
{
PsAssert(pName != NULL);
m_pDisplayName = pName;
}
//--------------------------------------------------------------------------
/// Return a string with all of the logged API call data in line-delimited
/// text format. This is used within the Timeline view in the client.
/// \return A string of all of the logged API calls captured during frame render.
//--------------------------------------------------------------------------
std::string MultithreadedTraceAnalyzerLayer::GetAPITraceTXT()
{
// A switch to determine at the last moment whether or not we should send our generated response back to the client.
bool bWriteResponseString = false;
// Concatenate all of the logged call lines into a single string that we can send to the client.
std::stringstream traceString;
std::map<DWORD, ThreadTraceData*>::iterator traceIter;
for (traceIter = mThreadTraces.begin(); traceIter != mThreadTraces.end(); ++traceIter)
{
ThreadTraceData* currentTrace = traceIter->second;
const TimingLog& currentTimer = currentTrace->mAPICallTimer;
GPS_TIMESTAMP timeFrequency = currentTimer.GetTimeFrequency();
const GPS_TIMESTAMP frameStartTime = mFramestartTime;
size_t numEntries = currentTrace->mLoggedCallVector.size();
// When using the updated trace format, include a preamble section for each traced thread.
#if defined(CODEXL_GRAPHICS)
// Write the trace type, API, ThreadID, and count of APIs traced.
traceString << "//==API Trace==" << std::endl;
traceString << "//API=" << GetAPIString() << std::endl;
traceString << "//ThreadID=" << traceIter->first << std::endl;
traceString << "//ThreadAPICount=" << numEntries << std::endl;
#endif
for (size_t entryIndex = 0; entryIndex < numEntries; ++entryIndex)
{
// Get each logged call by index.
const CallsTiming& callTiming = currentTimer.GetTimingByIndex(entryIndex);
double deltaStartTime, deltaEndTime;
bool conversionResults = currentTimer.ConvertTimestampToDoubles(callTiming.m_startTime,
callTiming.m_endTime,
deltaStartTime,
deltaEndTime,
frameStartTime,
&timeFrequency);
// We should always be able to convert from GPS_TIMESTAMPs to doubles.
PsAssert(conversionResults == true);
(void)conversionResults;
const APIEntry* callEntry = currentTrace->mLoggedCallVector[entryIndex];
// This exists as a sanity check. If a duration stretches past this point, we can be pretty sure something is messed up.
// This signal value is basically random, with the goal of it being large enough to catch any obvious duration errors.
if (deltaEndTime > 8000000000.0f)
{
const char* functionName = callEntry->GetAPIName();
Log(logWARNING, "The duration for APIEntry '%s' with index '%d' is suspicious. Tracing the application may have hung, producing inflated results.\n", functionName, entryIndex);
}
callEntry->AppendAPITraceLine(traceString, deltaStartTime, deltaEndTime);
}
bWriteResponseString = true;
}
// If for some reason we failed to write a valid response string, reply with a known failure signal so the client handles it properly.
if (!bWriteResponseString)
{
traceString << "NODATA";
}
return traceString.str();
}
//-----------------------------------------------------------------------------
bool MakeResponse(CommunicationID requestID, Response** ppResponse)
{
// protect the maps from being changed by other threads using the mutex
ScopeLock lock(s_mutex);
PsAssert(ppResponse != NULL);
// first see if we already have this ID as a streaming response
ResponseMap::iterator iterResponse = g_streamingResponseMap.find(requestID);
if (iterResponse != g_streamingResponseMap.end())
{
*ppResponse = iterResponse->second;
return true;
}
// otherwise we need to create a new response based on the original request
// so get the request
RequestMap::iterator iterRequest = g_requestMap.find(requestID);
if (iterRequest == g_requestMap.end())
{
// the original request couldn't be found, so return failure
return false;
}
// need to create a new response
if (PsNew(*ppResponse) == false)
{
return false;
}
HTTPRequestHeader* pRequest = iterRequest->second;
PsAssert(pRequest != NULL);
if (pRequest->GetReceivedOverSocket() == true)
{
(*ppResponse)->client_socket = pRequest->GetClientSocket();
}
else
{
#if defined (_WIN32)
(*ppResponse)->client_socket = NetSocket::CreateFromDuplicate(pRequest->GetProtoInfo());
#else
// create a new socket and connect to the streamSocket on the server
(*ppResponse)->client_socket = NetSocket::Create();
if ((*ppResponse)->client_socket != NULL)
{
osPortAddress portAddress((unsigned short)pRequest->GetPort());
(*ppResponse)->client_socket->Connect(portAddress);
}
#endif
}
if ((*ppResponse)->client_socket == NULL)
{
int Err = NetSocket::LastError();
Log(logERROR, "Could not create socket: NetSocket failed with error: %ld\n", Err);
return false;
}
// see if this should be added as a streaming response
gtASCIIString strUrl(pRequest->GetUrl());
int32 iStream = strUrl.find(STR_STREAM_TOKEN);
if (iStream >= 0)
{
const char* pBuf = strUrl.asCharArray();
const char* pRate = &pBuf[ iStream + strlen(STR_STREAM_TOKEN)];
unsigned int uRate = 0;
// try to get the rate from the command;
if (sscanf_s(pRate, "%u", &uRate) < 1)
{
// default to max rate
uRate = COMM_MAX_STREAM_RATE;
}
// set the response as streaming with the specified rate
(*ppResponse)->m_bStreamingEnabled = true;
(*ppResponse)->m_dwMaxStreamsPerSecond = uRate;
g_streamingResponseMap[ requestID ] = *ppResponse;
}
else
{
// streaming requests need to be kept around so that
// additional responses can be directed to the right place,
// HOWEVER, non-streaming requests only get a single response
// and we just created the response for it, so it is safe
// to remove the request from the requestMap. This will
// help keep communication quick as the number of incoming
// requests grows.
RemoveRequest(requestID);
}
return true;
}
//--------------------------------------------------------------------------
/// Processes the specified CommandObject to see if it should activate any
/// of the added Commands
/// \param rIncomingCommand the incoming command that should be handled
/// \return true if the command could be processed; false otherwise
//--------------------------------------------------------------------------
bool CommandProcessor::ProcessCommands(CommandObject& rIncomingCommand)
{
CommandList::const_iterator it;
for (it = m_Commands.begin() ; it < m_Commands.end(); ++it)
{
CommandResponse* pComm = *it;
PsAssert(pComm != NULL);
if (rIncomingCommand.IsCommand(pComm->GetURL()))
{
float dummy;
pComm->m_bStreamingEnabled = rIncomingCommand.GetParam("Stream", dummy);
if (pComm->GetParams(rIncomingCommand))
{
// skip over the parsed parameters
pComm->SkipParsedParams(rIncomingCommand);
// only set the command to be active if
// - the command is not editable (the client has requested data) or
// - AutoReply is false (another part of the server wants to react / respond) or
// - this is the last command and no response has been set yet
if (pComm->GetEditableContent() == NOT_EDITABLE ||
pComm->GetEditableContentAutoReply() == false ||
(rIncomingCommand.HasAnotherCommand() == false &&
rIncomingCommand.GetResponseState() == NO_RESPONSE))
{
pComm->SetActiveRequest(rIncomingCommand);
}
if (rIncomingCommand.HasAnotherCommand() == false)
{
rIncomingCommand.SetState(DELAYED_RESPONSE);
if (pComm->GetEditableContent() != NOT_EDITABLE &&
rIncomingCommand.GetResponseState() != SENT_RESPONSE)
{
if (pComm->GetEditableContentAutoReply())
{
// if this is an editable value,
// send back a response
pComm->Send("OK");
rIncomingCommand.SetState(SENT_RESPONSE);
}
}
}
return true;
}
else
{
// return variable's value
if (pComm->GetEditableContent() != NOT_EDITABLE &&
rIncomingCommand.GetResponseState() != SENT_RESPONSE &&
pComm->GetEditableContentAutoReply()
)
{
pComm->SetActiveRequest(rIncomingCommand);
rIncomingCommand.SetState(DELAYED_RESPONSE);
pComm->Send(pComm->GetEditableContentValue().c_str());
rIncomingCommand.SetState(SENT_RESPONSE);
return true;
}
return false;
}
}
}
return false;
}
//--------------------------------------------------------------------------
/// Should be called by an API-specific FrameDebugger at each drawcall
/// \param rDrawCall the API-specific DrawCall that is being executed
//--------------------------------------------------------------------------
void FrameDebugger::OnDrawCall(IDrawCall& rDrawCall)
{
m_ulDrawCallCounter++;
if (m_drawCallList.IsActive())
{
m_LastDrawCall = &rDrawCall;
gtASCIIString xmlString = rDrawCall.GetXML();
xmlString += XML("hash", rDrawCall.GetHash().asCharArray());
m_drawCallList += GetDrawCallXML(m_ulDrawCallCounter, xmlString.asCharArray()) ;
}
// Could collect drawcall XML here if desired
if (IsDrawCallEnabled(m_ulDrawCallCounter))
{
// Make sure to not render after the frame debugger's break point
if (IsTargetDrawCall(m_ulDrawCallCounter) == false)
{
// execute the drawcall for the app
rDrawCall.Execute();
}
else
{
// have the pipeline process non-hud commands before drawing anything on HUD
OnDrawCallAtBreakPointPreHUD(rDrawCall);
// allow the API-specific FrameDebugger to decide whether or not to do the drawcall at the breakpoint
DoDrawCallAtBreakPoint(rDrawCall);
// Send the current draw call back to the client.
if (m_CurrentDrawCall.IsActive())
{
m_CurrentDrawCall.Send(rDrawCall.GetXML().asCharArray());
}
bool bRes = BeginHUD();
PsAssert(bRes != false);
if (bRes == false)
{
Log(logERROR, "BeginHUD() failed");
return;
}
// show the wireframe overlay
if (m_bWireframeOverlay)
{
DoWireframeOverlay(rDrawCall, m_fWireframeOverlayColor[0], m_fWireframeOverlayColor[1], m_fWireframeOverlayColor[2], m_fWireframeOverlayColor[3]);
}
if (m_bAutoRenderTarget)
{
DoAutoRenderTarget();
}
OnDrawCallAtBreakPoint(rDrawCall);
EndHUD();
// skip the rest of the draws if HUD is being configured
if (false == m_bConfigHUD)
{
// When the frame debugger is enabled we process the messages from inside the FD and no longer re-render the Frame Capture. This speeds up processing of messages
// in apps that are running very slowly
for (int ProcessedCommands = 0;;)
{
if (m_drawCallList.IsActive() == true || m_Stats.IsActive() == true)
{
break;
}
// have the pipeline process non-hud commands before drawing anything on HUD
OnDrawCallAtBreakPointPreHUD(rDrawCall);
OnDrawCallAtBreakPoint(rDrawCall);
gtASCIIString SuCmd = PeekPendingRequests();
// Check for non FD command, if this changes then we have to re-render the HUD etc. So we break out
if (SuCmd.length() != 0)
{
char* pCmd = (char*)SuCmd.asCharArray();
ProcessedCommands++;
if (strstr(pCmd, "FD/Pipeline") == NULL)
{
LogConsole(logMESSAGE, "Processed in FD %3u\n", ProcessedCommands);
break;
}
GetSinglePendingRequest();
}
}
}
}
}
}
//--------------------------------------------------------------------------
/// Sets the current tag name.
/// \param pTagName The tag name.
//--------------------------------------------------------------------------
void CommandResponse::SetTagName(const char* pTagName)
{
PsAssert(pTagName != NULL);
m_pTagName = pTagName;
}
