SEA_EXPORT void ITTAPI __itt_api_init(__itt_global* pGlob, __itt_group_id id)
{
if (!g_bInitialized)
{
g_bInitialized = true;
__itt_global* pGlobal = GetITTGlobal();
__itt_mutex_init(&pGlobal->mutex);
pGlobal->mutex_initialized = 1;
sea::CIttLocker locker;
__itt_api_init(pGlobal, id);
pGlobal->api_initialized = 1;
}
const char* procname = sea::GetProcessName(true);
sea::SModuleInfo mdlinfo = sea::Fn2Mdl(pGlob);
VerbosePrint("IntelSEAPI init is called from process '%s' at module '%s'\n", procname, mdlinfo.path.c_str());
if (GetITTGlobal() != pGlob)
ChainGlobal(pGlob);
sea::FillApiList(pGlob->api_list_ptr);
for (___itt_domain* pDomain = pGlob->domain_list; pDomain; pDomain = pDomain->next)
{
FIX_DOMAIN(pDomain);
sea::InitDomain(pDomain);
}
for (__itt_string_handle* pStr = pGlob->string_list; pStr; pStr = pStr->next)
{
FIX_STRING(pStr);
sea::ReportString(const_cast<__itt_string_handle *>(pStr));
}
if (pGlob->version_build > 20120000) //counter_list was not yet invented that time
{
for (__itt_counter_info_t* pCounter = pGlob->counter_list; pCounter; pCounter = pCounter->next)
{
FIX_COUNTER(pCounter);
VerbosePrint("Fixed counter: %s | %s\n", pCounter->domainA, pCounter->nameA);
}
}
sea::ReportModule(pGlob);
static bool bInitialized = false;
if (!bInitialized)
{
bInitialized = true;
sea::InitSEA();
#ifdef _WIN32
EventRegisterIntelSEAPI();
#endif
atexit(AtExit);
}
}
int GlobalInit()
{
static const char var_name[] = INTEL_LIBITTNOTIFY BIT_SUFFIX;
static const char jit_var_name[] = INTEL_JIT_PROFILER BIT_SUFFIX;
sea::SModuleInfo mdlinfo = sea::Fn2Mdl((void*)GlobalInit);
VerbosePrint("IntelSEAPI: %s=%s | Loaded from: %s\n", var_name, get_environ_value(var_name).c_str(), mdlinfo.path.c_str());
std::string value = var_name;
value += "=";
value += mdlinfo.path;
std::string jit_val = jit_var_name;
jit_val += "=" + mdlinfo.path;
setenv(_strdup(value.c_str()));
VerbosePrint("IntelSEAPI: setting %s\n", value.c_str());
setenv(_strdup(jit_val.c_str()));
VerbosePrint("IntelSEAPI: setting %s\n", jit_val.c_str());
return 1;
}
inline CRecorder* GetFile(const SRecord& record)
{
DomainExtra* pDomainExtra = reinterpret_cast<DomainExtra*>(record.domain.extra2);
if (!pDomainExtra || !pDomainExtra->bHasDomainPath)
return nullptr;
static thread_local SThreadRecord* pThreadRecord = nullptr;
if (pThreadRecord) {} else pThreadRecord = GetThreadRecord();
if (pThreadRecord->bRemoveFiles)
{
pThreadRecord->pLastRecorder = nullptr;
pThreadRecord->pLastDomain = nullptr;
pThreadRecord->bRemoveFiles = false;
pThreadRecord->files.clear();
}
//with very high probability the same thread will write into the same domain
if (pThreadRecord->pLastRecorder && pThreadRecord->pLastDomain == record.domain.nameA && (100 > pThreadRecord->nSpeedupCounter++))
return reinterpret_cast<CRecorder*>(pThreadRecord->pLastRecorder);
pThreadRecord->nSpeedupCounter = 0; //we can't avoid checking ring size
pThreadRecord->pLastDomain = record.domain.nameA;
auto it = pThreadRecord->files.find(record.domain.nameA);
CRecorder* pRecorder = nullptr;
if (it != pThreadRecord->files.end())
{
pRecorder = &it->second;
uint64_t diff = record.rf.nanoseconds - pRecorder->GetCreationTime();
//just checking pointer of g_spCutName.get() is thread safe without any locks: we don't access internals. And if it's the same we work with the old path.
//but if it's changed we will lock and access the value below
if (pRecorder->SameCut(g_spCutName.get()) && (!g_nRingBuffer || (diff < g_nRingBuffer)))
{
pThreadRecord->pLastRecorder = pRecorder;
return pRecorder; //normal flow
}
pRecorder->Close(); //time to create new file
}
if (!pRecorder)
{
pRecorder = &pThreadRecord->files[record.domain.nameA];
}
CIttLocker lock; //locking only on file creation
if (pDomainExtra->strDomainPath.empty())//this is theoretically possible because we check pDomainExtra->bHasDomainPath without lock above
{
pThreadRecord->pLastRecorder = nullptr;
return nullptr;
}
std::shared_ptr<std::string> spCutName = g_spCutName;
CTraceEventFormat::SRegularFields rf = CTraceEventFormat::GetRegularFields();
char path[1024] = {};
_sprintf(path, "%s%llu%s%s.sea",
pDomainExtra->strDomainPath.c_str(),
(unsigned long long)rf.tid,
spCutName ? (std::string("!") + *spCutName).c_str() : "",
(g_nRingBuffer ? ((pRecorder->GetCount() % 2) ? "-1" : "-0") : "")
);
try {
VerbosePrint("Opening: %s\n", path);
pRecorder->Init(path, rf.nanoseconds, spCutName.get());
}
catch (const std::exception& exc)
{
VerbosePrint("Exception: %s\n", exc.what());
pThreadRecord->files.erase(record.domain.nameA);
pRecorder = nullptr;
}
pThreadRecord->pLastRecorder = pRecorder;
return pRecorder;
}
bool EmberRender(EmberOptions& opt)
{
#ifdef USECL EmberCLns::OpenCLInfo& info(EmberCLns::OpenCLInfo::Instance());#endif
std::cout.imbue(std::locale(""));
if (opt.DumpArgs())
cout << opt.GetValues(OPT_USE_RENDER) << endl;
if (opt.OpenCLInfo())
{
cout << "\nOpenCL Info: " << endl;
cout << info.DumpInfo();
return true;
}
Timing t;
bool writeSuccess = false;
byte* finalImagep;
uint padding;
size_t i, channels;
size_t strips;
size_t iterCount;
string filename;
string inputPath = GetPath(opt.Input());
ostringstream os;
pair<size_t, size_t> p;
vector<Ember<T>> embers;
vector<byte> finalImage;
EmberStats stats;
EmberReport emberReport;
EmberImageComments comments;
XmlToEmber<T> parser;
EmberToXml<T> emberToXml;
vector<QTIsaac<ISAAC_SIZE, ISAAC_INT>> randVec;
const vector<pair<size_t, size_t>> devices = Devices(opt.Devices());
unique_ptr<RenderProgress<T>> progress(new RenderProgress<T>());
unique_ptr<Renderer<T, float>> renderer(CreateRenderer<T>(opt.EmberCL() ? OPENCL_RENDERER : CPU_RENDERER, devices, false, 0, emberReport));
vector<string> errorReport = emberReport.ErrorReport();
if (!errorReport.empty())
emberReport.DumpErrorReport();
if (!renderer.get())
{
cout << "Renderer creation failed, exiting." << endl;
return false;
}
if (opt.EmberCL() && renderer->RendererType() != OPENCL_RENDERER)//OpenCL init failed, so fall back to CPU.
opt.EmberCL(false);
if (!InitPaletteList<T>(opt.PalettePath()))
return false;
if (!ParseEmberFile(parser, opt.Input(), embers))
return false;
if (!opt.EmberCL())
{
if (opt.ThreadCount() == 0)
{
cout << "Using " << Timing::ProcessorCount() << " automatically detected threads." << endl;
opt.ThreadCount(Timing::ProcessorCount());
}
else
{
cout << "Using " << opt.ThreadCount() << " manually specified threads." << endl;
}
renderer->ThreadCount(opt.ThreadCount(), opt.IsaacSeed() != "" ? opt.IsaacSeed().c_str() : nullptr);
}
else
{
cout << "Using OpenCL to render." << endl;
if (opt.Verbose())
{
for (auto& device : devices)
{
cout << "Platform: " << info.PlatformName(device.first) << endl;
cout << "Device: " << info.DeviceName(device.first, device.second) << endl;
}
}
if (opt.ThreadCount() > 1)
cout << "Cannot specify threads with OpenCL, using 1 thread." << endl;
opt.ThreadCount(1);
renderer->ThreadCount(opt.ThreadCount(), opt.IsaacSeed() != "" ? opt.IsaacSeed().c_str() : nullptr);
if (opt.BitsPerChannel() != 8)
{
cout << "Bits per channel cannot be anything other than 8 with OpenCL, setting to 8." << endl;
opt.BitsPerChannel(8);
}
}
if (opt.Format() != "jpg" &&
opt.Format() != "png" &&
opt.Format() != "ppm" &&
opt.Format() != "bmp")
{
cout << "Format must be jpg, png, ppm, or bmp not " << opt.Format() << ". Setting to jpg." << endl;
}
channels = opt.Format() == "png" ? 4 : 3;
if (opt.BitsPerChannel() == 16 && opt.Format() != "png")
{
cout << "Support for 16 bits per channel images is only present for the png format. Setting to 8." << endl;
opt.BitsPerChannel(8);
}
else if (opt.BitsPerChannel() != 8 && opt.BitsPerChannel() != 16)
{
cout << "Unexpected bits per channel specified " << opt.BitsPerChannel() << ". Setting to 8." << endl;
opt.BitsPerChannel(8);
}
if (opt.InsertPalette() && opt.BitsPerChannel() != 8)
{
cout << "Inserting palette only supported with 8 bits per channel, insertion will not take place." << endl;
opt.InsertPalette(false);
}
if (opt.AspectRatio() < 0)
{
cout << "Invalid pixel aspect ratio " << opt.AspectRatio() << endl << ". Must be positive, setting to 1." << endl;
opt.AspectRatio(1);
}
if (!opt.Out().empty() && (embers.size() > 1))
{
cout << "Single output file " << opt.Out() << " specified for multiple images. Changing to use prefix of badname-changethis instead. Always specify prefixes when reading a file with multiple embers." << endl;
opt.Out("");
opt.Prefix("badname-changethis");
}
//Final setup steps before running.
os.imbue(std::locale(""));
padding = uint(std::log10(double(embers.size()))) + 1;
renderer->EarlyClip(opt.EarlyClip());
renderer->YAxisUp(opt.YAxisUp());
renderer->LockAccum(opt.LockAccum());
renderer->InsertPalette(opt.InsertPalette());
renderer->PixelAspectRatio(T(opt.AspectRatio()));
renderer->Transparency(opt.Transparency());
renderer->NumChannels(channels);
renderer->BytesPerChannel(opt.BitsPerChannel() / 8);
renderer->Priority(eThreadPriority(Clamp<intmax_t>(intmax_t(opt.Priority()), intmax_t(eThreadPriority::LOWEST), intmax_t(eThreadPriority::HIGHEST))));
renderer->Callback(opt.DoProgress() ? progress.get() : nullptr);
for (i = 0; i < embers.size(); i++)
{
if (opt.Verbose() && embers.size() > 1)
cout << "\nFlame = " << i + 1 << "/" << embers.size() << endl;
else if (embers.size() > 1)
VerbosePrint(endl);
if (opt.Supersample() > 0)
embers[i].m_Supersample = opt.Supersample();
if (opt.SubBatchSize() != DEFAULT_SBS)
embers[i].m_SubBatchSize = opt.SubBatchSize();
embers[i].m_TemporalSamples = 1;//Force temporal samples to 1 for render.
embers[i].m_Quality *= T(opt.QualityScale());
embers[i].m_FinalRasW = size_t(T(embers[i].m_FinalRasW) * opt.SizeScale());
embers[i].m_FinalRasH = size_t(T(embers[i].m_FinalRasH) * opt.SizeScale());
embers[i].m_PixelsPerUnit *= T(opt.SizeScale());
if (embers[i].m_FinalRasW == 0 || embers[i].m_FinalRasH == 0)
{
cout << "Output image " << i << " has dimension 0: " << embers[i].m_FinalRasW << ", " << embers[i].m_FinalRasH << ". Setting to 1920 x 1080." << endl;
embers[i].m_FinalRasW = 1920;
embers[i].m_FinalRasH = 1080;
}
//Cast to double in case the value exceeds 2^32.
double imageMem = double(renderer->NumChannels()) * double(embers[i].m_FinalRasW)
* double(embers[i].m_FinalRasH) * double(renderer->BytesPerChannel());
double maxMem = pow(2.0, double((sizeof(void*) * 8) - 1));
if (imageMem > maxMem)//Ensure the max amount of memory for a process is not exceeded.
{
cout << "Image " << i << " size > " << maxMem << ". Setting to 1920 x 1080." << endl;
embers[i].m_FinalRasW = 1920;
embers[i].m_FinalRasH = 1080;
}
stats.Clear();
renderer->SetEmber(embers[i]);
renderer->PrepFinalAccumVector(finalImage);//Must manually call this first because it could be erroneously made smaller due to strips if called inside Renderer::Run().
if (opt.Strips() > 1)
{
strips = opt.Strips();
}
else
{
p = renderer->MemoryRequired(1, true, false);//No threaded write for render, only for animate.
strips = CalcStrips(double(p.second), double(renderer->MemoryAvailable()), opt.UseMem());
if (strips > 1)
VerbosePrint("Setting strips to " << strips << " with specified memory usage of " << opt.UseMem());
}
strips = VerifyStrips(embers[i].m_FinalRasH, strips,
[&](const string& s) { cout << s << endl; },//Greater than height.
[&](const string& s) { cout << s << endl; },//Mod height != 0.
[&](const string& s) { cout << s << endl; });//Final strips value to be set.
//For testing incremental renderer.
//int sb = 1;
//bool resume = false, success = false;
//do
//{
// success = renderer->Run(finalImage, 0, sb, false/*resume == false*/) == RENDER_OK;
// sb++;
// resume = true;
//}
//while (success && renderer->ProcessState() != ACCUM_DONE);
StripsRender<T>(renderer.get(), embers[i], finalImage, 0, strips, opt.YAxisUp(),
[&](size_t strip)//Pre strip.
{
if (opt.Verbose() && (strips > 1) && strip > 0)
cout << endl;
if (strips > 1)
VerbosePrint("Strip = " << (strip + 1) << "/" << strips);
},
[&](size_t strip)//Post strip.
{
progress->Clear();
stats += renderer->Stats();
},
[&](size_t strip)//Error.
{
cout << "Error: image rendering failed, skipping to next image." << endl;
renderer->DumpErrorReport();//Something went wrong, print errors.
},
//Final strip.
//Original wrote every strip as a full image which could be very slow with many large images.
//Only write once all strips for this image are finished.
[&](Ember<T>& finalEmber)
{
if (!opt.Out().empty())
{
filename = opt.Out();
}
else if (opt.NameEnable() && !finalEmber.m_Name.empty())
{
filename = inputPath + opt.Prefix() + finalEmber.m_Name + opt.Suffix() + "." + opt.Format();
}
else
{
ostringstream fnstream;
fnstream << inputPath << opt.Prefix() << setfill('0') << setw(padding) << i << opt.Suffix() << "." << opt.Format();
filename = fnstream.str();
}
//TotalIterCount() is actually using ScaledQuality() which does not get reset upon ember assignment,
//so it ends up using the correct value for quality * strips.
iterCount = renderer->TotalIterCount(1);
comments = renderer->ImageComments(stats, opt.PrintEditDepth(), opt.IntPalette(), opt.HexPalette());
os.str("");
os << comments.m_NumIters << " / " << iterCount << " (" << std::fixed << std::setprecision(2) << ((double(stats.m_Iters) / double(iterCount)) * 100) << "%)";
VerbosePrint("\nIters ran/requested: " + os.str());
if (!opt.EmberCL()) VerbosePrint("Bad values: " << stats.m_Badvals);
VerbosePrint("Render time: " + t.Format(stats.m_RenderMs));
VerbosePrint("Pure iter time: " + t.Format(stats.m_IterMs));
VerbosePrint("Iters/sec: " << size_t(stats.m_Iters / (stats.m_IterMs / 1000.0)) << endl);
VerbosePrint("Writing " + filename);
if ((opt.Format() == "jpg" || opt.Format() == "bmp") && renderer->NumChannels() == 4)
RgbaToRgb(finalImage, finalImage, renderer->FinalRasW(), renderer->FinalRasH());
finalImagep = finalImage.data();
writeSuccess = false;
if (opt.Format() == "png")
writeSuccess = WritePng(filename.c_str(), finalImagep, finalEmber.m_FinalRasW, finalEmber.m_FinalRasH, opt.BitsPerChannel() / 8, opt.PngComments(), comments, opt.Id(), opt.Url(), opt.Nick());
else if (opt.Format() == "jpg")
writeSuccess = WriteJpeg(filename.c_str(), finalImagep, finalEmber.m_FinalRasW, finalEmber.m_FinalRasH, int(opt.JpegQuality()), opt.JpegComments(), comments, opt.Id(), opt.Url(), opt.Nick());
else if (opt.Format() == "ppm")
writeSuccess = WritePpm(filename.c_str(), finalImagep, finalEmber.m_FinalRasW, finalEmber.m_FinalRasH);
else if (opt.Format() == "bmp")
writeSuccess = WriteBmp(filename.c_str(), finalImagep, finalEmber.m_FinalRasW, finalEmber.m_FinalRasH);
if (!writeSuccess)
cout << "Error writing " << filename << endl;
});
if (opt.EmberCL() && opt.DumpKernel())
{
if (auto rendererCL = dynamic_cast<RendererCL<T, float>*>(renderer.get()))
{
cout << "Iteration kernel:\n" <<
rendererCL->IterKernel() << "\n\n" <<
"Density filter kernel:\n" <<
rendererCL->DEKernel() << "\n\n" <<
"Final accumulation kernel:\n" <<
rendererCL->FinalAccumKernel() << endl;
}
}
VerbosePrint("Done.");
}
t.Toc("\nFinished in: ", true);
return true;
}
