// TestFunction_Speed
//------------------------------------------------------------------------------
void TestProjectGeneration::TestFunction_Speed() const
{
VSProjectGenerator pg;
AStackString<> baseDir( "C:\\Windows\\System32" );
Array< AString > baseDirs;
baseDirs.Append( baseDir );
// project name
pg.SetProjectName( AStackString<>( "Big" ) );
pg.SetBasePaths( baseDirs );
// platforms
Array< VSProjectConfig > configs;
VSProjectConfig cfg;
cfg.m_Platform = "Win32";
cfg.m_Config = "Debug";
configs.Append( cfg );
// files (about 5,000)
Array< AString > files;
FileIO::GetFiles( baseDir, AStackString<>( "*.mui" ), true, &files );
FileIO::GetFiles( baseDir, AStackString<>( "*.exe" ), true, &files );
FileIO::GetFiles( baseDir, AStackString<>( "*.dll" ), true, &files );
pg.AddFiles( files );
Array< VSProjectFileType > fileTypes;
{
VSProjectFileType ft;
ft.m_FileType = "CppForm";
ft.m_Pattern = "Code\\Forms\\*.h";
fileTypes.Append( ft );
ft.m_FileType = "CppControl";
ft.m_Pattern = "Controls\\*.h";
fileTypes.Append( ft );
}
AStackString<> projectFileName( "C:\\Windows\\System\\dummy.vcxproj" );
{
Timer t;
pg.GenerateVCXProj( projectFileName, configs, fileTypes );
float time = t.GetElapsed();
OUTPUT( "Gen vcxproj : %2.3fs\n", time );
}
{
Timer t;
pg.GenerateVCXProjFilters( projectFileName );
float time = t.GetElapsed();
OUTPUT( "Gen vcxproj.filters: %2.3fs\n", time );
}
}
REGISTER_TESTS_END
// TestMSVCPreprocessedOutput
//------------------------------------------------------------------------------
void TestIncludeParser::TestMSVCPreprocessedOutput() const
{
FileStream f;
TEST_ASSERT( f.Open( "Data/TestIncludeParser/fbuildcore.msvc.ii", FileStream::READ_ONLY) )
const size_t fileSize = (size_t)f.GetFileSize();
AutoPtr< char > mem( (char *)ALLOC( fileSize + 1 ) );
TEST_ASSERT( f.Read( mem.Get(), fileSize ) == fileSize );
mem.Get()[ fileSize ] = 0;
Timer t;
const size_t repeatCount( 100 );
for ( size_t i=0; i<repeatCount; ++i )
{
CIncludeParser parser;
TEST_ASSERT( parser.ParseMSCL_Preprocessed( mem.Get(), fileSize ) );
// check number of includes found to prevent future regressions
const Array< AString > & includes = parser.GetIncludes();
TEST_ASSERT( includes.GetSize() == 284 );
#ifdef DEBUG
TEST_ASSERT( parser.GetNonUniqueCount() == 381 );
#endif
}
float time = t.GetElapsed();
OUTPUT( "MSVC : %2.3fs (%2.1f MiB/sec)\n", time, ( (float)( fileSize * repeatCount / ( 1024.0f * 1024.0f ) ) / time ) );
}
// TestClangMSExtensionsPreprocessedOutput
//------------------------------------------------------------------------------
void TestIncludeParser::TestClangMSExtensionsPreprocessedOutput() const
{
FBuild fBuild; // needed fer CleanPath for relative dirs
FileStream f;
TEST_ASSERT( f.Open( "Data/TestIncludeParser/fbuildcore.clang.ms-extensions.ii", FileStream::READ_ONLY) )
const size_t fileSize = (size_t)f.GetFileSize();
AutoPtr< char > mem( (char *)ALLOC( fileSize + 1 ) );
TEST_ASSERT( f.Read( mem.Get(), fileSize ) == fileSize );
mem.Get()[ fileSize ] = 0;
Timer t;
const size_t repeatCount( 100 );
for ( size_t i=0; i<repeatCount; ++i )
{
CIncludeParser parser;
TEST_ASSERT( parser.ParseGCC_Preprocessed( mem.Get(), fileSize ) );
// check number of includes found to prevent future regressions
const Array< AString > & includes = parser.GetIncludes();
TEST_ASSERT( includes.GetSize() == 285 );
#ifdef DEBUG
TEST_ASSERT( parser.GetNonUniqueCount() == 4758 );
#endif
}
float time = t.GetElapsed();
OUTPUT( "Clang (ms-extensions): %2.3fs (%2.1f MiB/sec)\n", time, ( (float)( fileSize * repeatCount / ( 1024.0f * 1024.0f ) ) / time ) );
}
int WINAPI WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance, PSTR pScmdline, int iCmdshow)
{
if (!init_log(NULL))
return -1;
if(!Init(NULL, hInstance, hPrevInstance, pScmdline, iCmdshow))
{
return 1;
}
MSG msg = {0};
while(WM_QUIT != msg.message)
{
if(PeekMessage(&msg, nullptr, 0, 0, PM_REMOVE))
{
TranslateMessage(&msg);
DispatchMessage(&msg);
}
else
{
//no messages waiting, step the simulation forward a frame
if(time.Tick())
{
Update(time.GetElapsed());
Render();
}
}
}
// return this part of the WM_QUIT message to Windows
return msg.wParam;
}
/*static*/ void FileIO::WorkAroundForWindowsFilePermissionProblem( const AString & fileName )
{
// Sometimes after closing a file, subsequent operations on that file will
// fail. For example, trying to set the file time, or even another process
// opening the file.
//
// This seems to be a known issue in windows, with multiple potential causes
// like Virus scanners and possibly the behaviour of the kernel itself.
//
// A work-around for this problem is to attempt to open a file we just closed.
// This will sometimes fail, but if we retry until it succeeds, we avoid the
// problem on the subsequent operation.
FileStream f;
Timer timer;
while ( f.Open( fileName.Get() ) == false )
{
Thread::Sleep( 1 );
// timeout so we don't get stuck in here forever
if ( timer.GetElapsed() > 1.0f )
{
ASSERT( false && "WorkAroundForWindowsFilePermissionProblem Failed!" );
return;
}
}
f.Close();
}
// AllocateFromSystemAllocator
//------------------------------------------------------------------------------
/*static*/ float TestSmallBlockAllocator::AllocateFromSystemAllocator( const Array< uint32_t > & allocSizes, const uint32_t repeatCount )
{
const size_t numAllocs = allocSizes.GetSize();
Array< void * > allocs( numAllocs, false );
Timer timer;
for ( size_t r=0; r<repeatCount; ++r )
{
// use malloc
for ( uint32_t i=0; i<numAllocs; ++i )
{
uint32_t * mem = (uint32_t *)malloc( allocSizes[i] );
allocs.Append( mem );
}
// use free
for ( uint32_t i=0; i<numAllocs; ++i )
{
void * mem = allocs[ i ];
free( mem );
}
allocs.Clear();
}
return timer.GetElapsed();
}
int LaunchSubProcess( const AString & args )
{
// try to make a copy of our exe
AStackString<> exeName;
Env::GetExePath( exeName );
AStackString<> exeNameCopy( exeName );
exeNameCopy += ".copy";
Timer t;
while ( FileIO::FileCopy( exeName.Get(), exeNameCopy.Get() ) == false )
{
if ( t.GetElapsed() > 5.0f )
{
AStackString<> msg;
msg.Format( "Failed to make sub-process copy - error: %u (0x%x)\n\nSrc: %s\nDst: %s\n", Env::GetLastErr(), Env::GetLastErr(), exeName.Get(), exeNameCopy.Get() );
ShowMsgBox( msg.Get() );
return -2;
}
Thread::Sleep( 100 );
}
AStackString<> argsCopy( args );
argsCopy += " -subprocess";
// allow subprocess to access the mutex
g_OneProcessMutex.Unlock();
Process p;
#if defined( __WINDOWS__ )
p.DisableHandleRedirection(); // TODO:MAC TODO:LINUX is this needed?
#endif
p.Spawn( exeNameCopy.Get(), argsCopy.Get(), nullptr, nullptr );
p.Detach();
return 0;
}
REGISTER_TESTS_END
// Validate
//------------------------------------------------------------------------------
void TestTimer::Validate() const
{
Timer t;
t.Start();
const uint64_t before = t.GetNow();
Thread::Sleep( 100 ); // sleep for 100ms
const float elapsed = t.GetElapsed();
const float elapsedMS = t.GetElapsedMS();
const uint64_t after = t.GetNow();
// some time must have elapsed
TEST_ASSERT( after > before );
// sanity check
TEST_ASSERT( elapsed >= 0.001f ); // at least 1ms
TEST_ASSERT( elapsed < 1.000f ); // some sensible value
// sanity check
TEST_ASSERT( elapsedMS >= 1.0f ); // at least 1ms
TEST_ASSERT( elapsedMS < 1000.0f ); // some sensible value
}
//------------------------------------------------------------------------------
void Solver::Advance (float timeStep)
{
Timer t;
t.Start();
updateBlendValues();
mFluidHashTable[LOW]->Fill(mFluidParticles[LOW]->ActiveIDs);
mFluidHashTable[HIGH]->Fill(mFluidParticles[HIGH]->ActiveIDs);
computeDensity(LOW);
computeDensity(HIGH);
computeAcceleration(LOW);
computeAcceleration(HIGH);
integrate(HIGH, timeStep/2.0f);
mFluidHashTable[HIGH]->Fill(mFluidParticles[HIGH]->ActiveIDs);
computeDensity(HIGH);
computeAcceleration(HIGH);
integrate(HIGH, timeStep/2.0f);
integrate(LOW, timeStep);
inject();
t.Stop();
std::cout << "#LOW " << mFluidParticles[LOW]->ActiveIDs.size() << " #HIGH "
<< mFluidParticles[HIGH]->ActiveIDs.size() << " #TOTAL: "
<< mFluidParticles[LOW]->ActiveIDs.size() +
mFluidParticles[HIGH]->ActiveIDs.size() << "TIME: "
<< t.GetElapsed() << std::endl;
}
// Generate
//------------------------------------------------------------------------------
void Report::Generate( const FBuildStats & stats )
{
Timer t;
// pre-allocate a large string for output
m_Output.SetReserved( MEGABYTE );
m_Output.SetLength( 0 );
// generate some common data used in reporting
GetLibraryStats( stats );
// build the report
CreateHeader();
CreateTitle();
CreateOverview( stats );
DoCPUTimeByType( stats );
DoCacheStats( stats );
DoCPUTimeByLibrary();
DoCPUTimeByItem( stats );
DoIncludes();
CreateFooter();
// patch in time take
const float time = t.GetElapsed();
AStackString<> timeTakenBuffer;
stats.FormatTime( time, timeTakenBuffer );
char * placeholder = m_Output.Find( "^^^^ " );
memcpy( placeholder, timeTakenBuffer.Get(), timeTakenBuffer.GetLength() );
}
bool Timer::Period(Timer& T, double* t, double period) {
if (T.GetElapsed() - *t > 0) {
*t += period;
return true;
}
return false;
}
// WaitTimeout
//------------------------------------------------------------------------------
void TestSemaphore::WaitTimeout() const
{
Timer t;
Semaphore s;
s.Wait( 50 ); // wait 50ms
// ensure some sensible time has elapsed
ASSERT( t.GetElapsed() > 0.025f ); // 25ms (allow wide margin of error)
}
// AllocateFromSmallBlockAllocator
//------------------------------------------------------------------------------
/*static*/ float TestSmallBlockAllocator::AllocateFromSmallBlockAllocator( const Array< uint32_t > & allocSizes, const uint32_t repeatCount, const bool threadSafe )
{
const size_t numAllocs = allocSizes.GetSize();
Array< void * > allocs( numAllocs, false );
Timer timer;
if ( threadSafe == false )
{
SmallBlockAllocator::SetSingleThreadedMode( true );
}
for ( size_t r=0; r<repeatCount; ++r )
{
// Use ALLOC
for ( uint32_t i=0; i<numAllocs; ++i )
{
uint32_t * mem = (uint32_t *)ALLOC( allocSizes[i] );
allocs.Append( mem );
}
// Use FREE
for ( uint32_t i=0; i<numAllocs; ++i )
{
void * mem = allocs[ i ];
FREE( mem );
}
allocs.Clear();
}
if ( threadSafe == false )
{
SmallBlockAllocator::SetSingleThreadedMode( false );
}
return timer.GetElapsed();
}
//=================================================================================================================================
/// The main function.
//=================================================================================================================================
int main(int argc, char* argv[])
{
// initialize settings to defaults
TootleSettings settings;
settings.pMeshName = NULL;
settings.pViewpointName = NULL;
settings.nClustering = 0;
settings.nCacheSize = TOOTLE_DEFAULT_VCACHE_SIZE;
settings.eWinding = TOOTLE_CW;
settings.algorithmChoice = TOOTLE_OPTIMIZE;
settings.eVCacheOptimizer = TOOTLE_VCACHE_AUTO; // the auto selection as the default to optimize vertex cache
settings.bOptimizeVertexMemory = true; // default value is to optimize the vertex memory
settings.bMeasureOverdraw = true; // default is to measure overdraw
// parse the command line
ParseCommandLine(argc, argv, &settings);
// ***************************************************
// Load the mesh
// ***************************************************
// read the mesh from the OBJ file
std::vector<ObjVertexFinal> objVertices;
std::vector<ObjFace> objFaces;
ObjLoader loader;
if (!loader.LoadGeometry(settings.pMeshName, objVertices, objFaces))
{
std::cerr << "Error loading mesh file: " << settings.pMeshName << std::endl;
return 1;
}
// build buffers containing only the vertex positions and indices, since this is what Tootle requires
std::vector<ObjVertex3D> vertices;
vertices.resize(objVertices.size());
for (unsigned int i = 0; i < vertices.size(); i++)
{
vertices[i] = objVertices[i].pos;
}
std::vector<unsigned int> indices;
indices.resize(objFaces.size() * 3);
for (unsigned int i = 0; i < indices.size(); i++)
{
indices[i] = objFaces[ i / 3 ].finalVertexIndices[ i % 3 ];
}
// ******************************************
// Load viewpoints if necessary
// ******************************************
// read viewpoints if needed
std::vector<ObjVertex3D> viewpoints;
if (settings.pViewpointName != NULL)
{
if (!LoadViewpoints(settings.pViewpointName, viewpoints))
{
std::cerr << "Unable to load viewpoints from file: " << settings.pViewpointName;
return 1;
}
}
// if we didn't get any viewpoints, then use a NULL array
const float* pViewpoints = NULL;
unsigned int nViewpoints = (unsigned int) viewpoints.size();
if (viewpoints.size() > 0)
{
pViewpoints = (const float*) &viewpoints[0];
}
// *****************************************************************
// Prepare the mesh and initialize stats variables
// *****************************************************************
unsigned int nFaces = (unsigned int) indices.size() / 3;
unsigned int nVertices = (unsigned int) vertices.size();
float* pfVB = (float*) &vertices[0];
unsigned int* pnIB = (unsigned int*) &indices[0];
unsigned int nStride = 3 * sizeof(float);
TootleStats stats;
// initialize the timing variables
stats.fOptimizeVCacheTime = INVALID_TIME;
stats.fClusterMeshTime = INVALID_TIME;
stats.fVCacheClustersTime = INVALID_TIME;
stats.fOptimizeVCacheAndClusterMeshTime = INVALID_TIME;
stats.fOptimizeOverdrawTime = INVALID_TIME;
stats.fTootleOptimizeTime = INVALID_TIME;
stats.fTootleFastOptimizeTime = INVALID_TIME;
stats.fMeasureOverdrawTime = INVALID_TIME;
stats.fOptimizeVertexMemoryTime = INVALID_TIME;
TootleResult result;
// initialize Tootle
result = TootleInit();
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
// measure input VCache efficiency
result = TootleMeasureCacheEfficiency(pnIB, nFaces, settings.nCacheSize, &stats.fVCacheIn);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
if (settings.bMeasureOverdraw)
{
// measure input overdraw. Note that we assume counter-clockwise vertex winding.
result = TootleMeasureOverdraw(pfVB, pnIB, nVertices, nFaces, nStride, pViewpoints, nViewpoints, settings.eWinding,
&stats.fOverdrawIn, &stats.fMaxOverdrawIn);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
}
// allocate an array to hold the cluster ID for each face
std::vector<unsigned int> faceClusters;
faceClusters.resize(nFaces + 1);
unsigned int nNumClusters;
Timer timer;
timer.Reset();
// **********************************************************************************************************************
// Optimize the mesh:
//
// The following cases show five examples for developers on how to use the library functions in Tootle.
// 1. If you are interested in optimizing vertex cache only, see the TOOTLE_VCACHE_ONLY case.
// 2. If you are interested to optimize vertex cache and overdraw, see either TOOTLE_CLUSTER_VCACHE_OVERDRAW
// or TOOTLE_OPTIMIZE cases. The former uses three separate function calls while the latter uses a single
// utility function.
// 3. To use the algorithm from SIGGRAPH 2007 (v2.0), see TOOTLE_FAST_VCACHECLUSTER_OVERDRAW or TOOTLE_FAST_OPTIMIZE
// cases. The former uses two separate function calls while the latter uses a single utility function.
//
// Note the algorithm from SIGGRAPH 2007 (v2.0) is very fast but produces less quality results especially for the
// overdraw optimization. During our experiments with some medium size models, we saw an improvement of 1000x in
// running time (from 20+ minutes to less than 1 second) for using v2.0 calls vs v1.2 calls. The resulting vertex
// cache optimization is very similar while the overdraw optimization drops from 3.8x better to 2.5x improvement over
// the input mesh.
// Developers should always run the overdraw optimization using the fast algorithm from SIGGRAPH initially.
// If they require a better result, then re-run the overdraw optimization using the old v1.2 path (TOOTLE_OVERDRAW_AUTO).
// Passing TOOTLE_OVERDRAW_AUTO to the algorithm will let the algorithm choose between Direct3D or raytracing path
// depending on the total number of clusters (less than 200 clusters, it will use Direct3D path otherwise it will
// use raytracing path since the raytracing path will be faster than the Direct3D path at that point).
//
// Tips: use v2.0 for fast optimization, and v1.2 to further improve the result by mix-matching the calls.
// **********************************************************************************************************************
switch (settings.algorithmChoice)
{
case TOOTLE_VCACHE_ONLY:
// *******************************************************************************************************************
// Perform Vertex Cache Optimization ONLY
// *******************************************************************************************************************
stats.nClusters = 1;
// Optimize vertex cache
result = TootleOptimizeVCache(pnIB, nFaces, nVertices, settings.nCacheSize,
pnIB, NULL, settings.eVCacheOptimizer);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
stats.fOptimizeVCacheTime = timer.GetElapsed();
break;
case TOOTLE_CLUSTER_VCACHE_OVERDRAW:
// *******************************************************************************************************************
// An example of calling clustermesh, vcacheclusters and optimize overdraw individually.
// This case demonstrate mix-matching v1.2 clustering with v2.0 overdraw optimization.
// *******************************************************************************************************************
// Cluster the mesh, and sort faces by cluster.
result = TootleClusterMesh(pfVB, pnIB, nVertices, nFaces, nStride, settings.nClustering, pnIB, &faceClusters[0], NULL);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
stats.fClusterMeshTime = timer.GetElapsed();
timer.Reset();
// The last entry of of faceClusters store the total number of clusters.
stats.nClusters = faceClusters[ nFaces ];
// Perform vertex cache optimization on the clustered mesh.
result = TootleVCacheClusters(pnIB, nFaces, nVertices, settings.nCacheSize, &faceClusters[0],
pnIB, NULL, settings.eVCacheOptimizer);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
stats.fVCacheClustersTime = timer.GetElapsed();
timer.Reset();
// Optimize the draw order (using v1.2 path: TOOTLE_OVERDRAW_AUTO, the default path is from v2.0--SIGGRAPH version).
result = TootleOptimizeOverdraw(pfVB, pnIB, nVertices, nFaces, nStride, pViewpoints, nViewpoints, settings.eWinding,
&faceClusters[0], pnIB, NULL, TOOTLE_OVERDRAW_AUTO);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
stats.fOptimizeOverdrawTime = timer.GetElapsed();
break;
case TOOTLE_FAST_VCACHECLUSTER_OVERDRAW:
// *******************************************************************************************************************
// An example of calling v2.0 optimize vertex cache and clustering mesh with v1.2 overdraw optimization.
// *******************************************************************************************************************
// Optimize vertex cache and create cluster
// The algorithm from SIGGRAPH combine the vertex cache optimization and clustering mesh into a single step
result = TootleFastOptimizeVCacheAndClusterMesh(pnIB, nFaces, nVertices, settings.nCacheSize, pnIB,
&faceClusters[0], &nNumClusters, TOOTLE_DEFAULT_ALPHA);
if (result != TOOTLE_OK)
{
// an error detected
DisplayTootleErrorMessage(result);
return 1;
}
stats.fOptimizeVCacheAndClusterMeshTime = timer.GetElapsed();
timer.Reset();
stats.nClusters = nNumClusters;
// In this example, we use TOOTLE_OVERDRAW_AUTO to show that we can mix-match the clustering and
// vcache computation from the new library with the overdraw optimization from the old library.
// TOOTLE_OVERDRAW_AUTO will choose between using Direct3D or CPU raytracing path. This path is
// much slower than TOOTLE_OVERDRAW_FAST but usually produce 2x better results.
result = TootleOptimizeOverdraw(pfVB, pnIB, nVertices, nFaces, nStride, NULL, 0,
settings.eWinding, &faceClusters[0], pnIB, NULL, TOOTLE_OVERDRAW_AUTO);
if (result != TOOTLE_OK)
{
// an error detected
DisplayTootleErrorMessage(result);
return 1;
}
stats.fOptimizeOverdrawTime = timer.GetElapsed();
break;
case TOOTLE_OPTIMIZE:
// *******************************************************************************************************************
// An example of using a single utility function to perform v1.2 optimizations.
// *******************************************************************************************************************
// This function will compute the entire optimization (cluster mesh, vcache per cluster, and optimize overdraw).
// It will use TOOTLE_OVERDRAW_FAST as the default overdraw optimization
result = TootleOptimize(pfVB, pnIB, nVertices, nFaces, nStride, settings.nCacheSize,
pViewpoints, nViewpoints, settings.eWinding, pnIB, &nNumClusters, settings.eVCacheOptimizer);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
stats.fTootleOptimizeTime = timer.GetElapsed();
stats.nClusters = nNumClusters;
break;
case TOOTLE_FAST_OPTIMIZE:
// *******************************************************************************************************************
// An example of using a single utility function to perform v2.0 optimizations.
// *******************************************************************************************************************
// This function will compute the entire optimization (optimize vertex cache, cluster mesh, and optimize overdraw).
// It will use TOOTLE_OVERDRAW_FAST as the default overdraw optimization
result = TootleFastOptimize(pfVB, pnIB, nVertices, nFaces, nStride, settings.nCacheSize,
settings.eWinding, pnIB, &nNumClusters, TOOTLE_DEFAULT_ALPHA);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
stats.fTootleFastOptimizeTime = timer.GetElapsed();
stats.nClusters = nNumClusters;
break;
default:
// wrong algorithm choice
break;
}
// measure output VCache efficiency
result = TootleMeasureCacheEfficiency(pnIB, nFaces, settings.nCacheSize, &stats.fVCacheOut);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
if (settings.bMeasureOverdraw)
{
// measure output overdraw
timer.Reset();
result = TootleMeasureOverdraw(pfVB, pnIB, nVertices, nFaces, nStride, pViewpoints, nViewpoints, settings.eWinding,
&stats.fOverdrawOut, &stats.fMaxOverdrawOut);
stats.fMeasureOverdrawTime = timer.GetElapsed();
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
}
//-----------------------------------------------------------------------------------------------------
// PERFORM VERTEX MEMORY OPTIMIZATION (rearrange memory layout for vertices based on the final indices
// to exploit vertex cache prefetch).
// We want to optimize the vertex memory locations based on the final optimized index buffer that will
// be in the output file.
// Thus, in this sample code, we recompute a copy of the indices that point to the original vertices
// (pnIBTmp) to be passed into the function TootleOptimizeVertexMemory. If we use the array pnIB, we
// will optimize for the wrong result since the array pnIB is based on the rehashed vertex location created
// by the function ObjLoader.
//-----------------------------------------------------------------------------------------------------
timer.Reset();
std::vector<unsigned int> pnVertexRemapping;
unsigned int nReferencedVertices = 0; // The actual total number of vertices referenced by the indices
if (settings.bOptimizeVertexMemory)
{
std::vector<unsigned int> pnIBTmp;
pnIBTmp.resize(nFaces * 3);
// compute the indices to be optimized for (the original pointed by the obj file).
for (unsigned int i = 0; i < indices.size(); i += 3)
{
for (int j = 0; j < 3; j++)
{
const ObjVertexFinal& rVertex = objVertices[ pnIB[ i + j ] ];
pnIBTmp[ i + j ] = rVertex.nVertexIndex - 1; // index is off by 1
// compute the max vertices
if (rVertex.nVertexIndex > nReferencedVertices)
{
nReferencedVertices = rVertex.nVertexIndex;
}
}
}
pnVertexRemapping.resize(nReferencedVertices);
// For this sample code, we are just going to use vertexRemapping array result. This is to support general obj
// file input and output.
// In fact, we are sending the wrong vertex buffer here (it should be based on the original file instead of the
// rehashed vertices). But, it is ok because we do not request the reordered vertex buffer as an output.
result = TootleOptimizeVertexMemory(pfVB, &pnIBTmp[0], nReferencedVertices, nFaces, nStride, NULL, NULL,
&pnVertexRemapping[0]);
if (result != TOOTLE_OK)
{
DisplayTootleErrorMessage(result);
return 1;
}
stats.fOptimizeVertexMemoryTime = timer.GetElapsed();
}
// clean up tootle
TootleCleanup();
// print tootle statistics to stdout and stderr
// display the current test case
PrintAlgorithm(stderr, settings.eVCacheOptimizer, settings.algorithmChoice, settings.nCacheSize, stats.nClusters);
PrintAlgorithm(stdout, settings.eVCacheOptimizer, settings.algorithmChoice, settings.nCacheSize, stats.nClusters);
PrintStats(stdout, &stats);
PrintStats(stderr, &stats);
// emit a modified .OBJ file
std::ifstream inputStream(settings.pMeshName);
bool bResult;
if (settings.bOptimizeVertexMemory)
{
bResult = EmitModifiedObj(inputStream, std::cout, objVertices, indices, &pnVertexRemapping[0], nReferencedVertices);
}
else
{
bResult = EmitModifiedObj(inputStream, std::cout, objVertices, indices, NULL, 0);
}
if (bResult)
{
return 1;
}
return 0;
}
// CompareHashTimes_Small
//------------------------------------------------------------------------------
void TestHash::CompareHashTimes_Small() const
{
// some different strings to hash
Array< AString > strings( 32, true );
strings.Append( AString( " " ) );
strings.Append( AString( "short" ) );
strings.Append( AString( "mediumstringmediumstring123456789" ) );
strings.Append( AString( "longstring_98274ncoif834jodhiorhmwe8r8wy48on87h8mhwejrijrdierwurd9j,8chm8hiuorciwriowjri" ) );
strings.Append( AString( "c:\\files\\subdir\\project\\thing\\stuff.cpp" ) );
const size_t numStrings = strings.GetSize();
const size_t numIterations = 102400;
// calc datasize
size_t dataSize( 0 );
for ( size_t i=0; i<numStrings; ++i )
{
dataSize += strings[ i ].GetLength();
}
dataSize *= numIterations;
// xxHash - 32
{
Timer t;
uint32_t crc( 0 );
for ( size_t j=0; j<numIterations; ++j )
{
for ( size_t i=0; i<numStrings; ++i )
{
crc += xxHash::Calc32( strings[ i ].Get(), strings[ i ].GetLength() );
}
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "xxHash-32 : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// xxHash - 64
{
Timer t;
uint64_t crc( 0 );
for ( size_t j=0; j<numIterations; ++j )
{
for ( size_t i=0; i<numStrings; ++i )
{
crc += xxHash::Calc64( strings[ i ].Get(), strings[ i ].GetLength() );
}
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "xxHash-64 : %2.3fs @ %6.3f GiB/s (hash: %016llx)\n", time, speed, crc );
}
// Murmur3 - 32
{
Timer t;
uint32_t crc( 0 );
for ( size_t j=0; j<numIterations; ++j )
{
for ( size_t i=0; i<numStrings; ++i )
{
crc += Murmur3::Calc32( strings[ i ].Get(), strings[ i ].GetLength() );
}
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "Murmur3-32 : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// Murmur3 - 128
{
Timer t;
uint64_t hashB( 0 );
uint64_t hashA( 0 );
for ( size_t j=0; j<numIterations; ++j )
{
for ( size_t i=0; i<numStrings; ++i )
{
hashA += Murmur3::Calc128( strings[ i ].Get(), strings[ i ].GetLength(), hashB );
}
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "Murmur3-128 : %2.3fs @ %6.3f GiB/s (%016llx, %016llx)\n", time, speed, hashA, hashB );
}
// CRC32 - 8x8 slicing
{
Timer t;
uint32_t crc( 0 );
for ( size_t j=0; j<numIterations; ++j )
{
for ( size_t i=0; i<numStrings; ++i )
{
crc += CRC32::Calc( strings[ i ].Get(), strings[ i ].GetLength() );
}
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "CRC32 8x8 : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// CRC32 - "standard" algorithm
{
Timer t;
uint32_t crc( 0 );
for ( size_t j=0; j<numIterations; ++j )
{
for ( size_t i=0; i<numStrings; ++i )
{
crc += CRC32::Start();
crc += CRC32::Update( crc, strings[ i ].Get(), strings[ i ].GetLength() );
crc += CRC32::Stop( crc );
}
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "CRC32 : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// CRC32Lower
{
Timer t;
uint32_t crc( 0 );
for ( size_t j=0; j<numIterations; ++j )
{
for ( size_t i=0; i<numStrings; ++i )
{
crc += CRC32::CalcLower( strings[ i ].Get(), strings[ i ].GetLength() );
}
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "CRC32Lower : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// Murmur3 - 32 Lower
{
Timer t;
// lower-case and hash it
uint32_t crc( 0 );
for ( size_t j=0; j<numIterations; ++j )
{
for ( size_t i=0; i<numStrings; ++i )
{
const AString & str( strings[ i ] );
AStackString<> tmp;
const size_t strLen( str.GetLength() );
tmp.SetLength( (uint32_t)strLen );
for ( size_t k=0; k<strLen; ++k )
{
char c = str[ (uint32_t)k ];
tmp[ (uint32_t)k ] = ( ( c >= 'A' ) && ( c <= 'Z' ) ) ? 'a' + ( c - 'A' ) : c;
}
crc += Murmur3::Calc32( tmp.Get(), tmp.GetLength() );
}
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "Murmur3-32-Lower: %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
}
int main(int argc, char *argv[])
{
TestStructs();
#ifdef _CELL
InitPPECallbacks();
InitSPEs();
#endif
Init();
if(parseArgs(argc, argv) != 0) return 1;
/* Init framebuffer and Scene */
printf("lzrt %ix%i (" ARCH_STR ")\n", args.width, args.height);
ImageBuffer *imgbuf = new FrameBuffer(args.width, args.height, args.fullscreen, "lzrt %ix%i (" ARCH_STR ")", args.width, args.height);
Scene scene(imgbuf, args.animate, args.renderer);
// Execute Lua script which sets up the scene
lua_State *L = 0;
L = InitLua();
if(lua_dofile(L, args.luascript) != 0)
{
printf("Error: Couldn't execute Lua script '%s'\n", args.luascript);
lua_close(L);
return 1;
}
threadpool = new ThreadPool(args.nthreads);
// Setup Scene
scene.Setup(threadpool);
scene.imgbuf->SetClearColor(0.2f, 0.2f, 0.3f);
//scene.imgbuf->SetClearColor(1.0f, 1.0f, 1.0f);
Timer timer;
while(run)
{
SDLevent();
threadpool->SetNumThreads(args.nthreads);
timer.Mark();
// Call Lua loop function
if(L != 0 && scene.animate) call_function<void>(L, "lzrtloop");
scene.camera->Move(firstframe, camerapos.x, camerapos.y, camerapos.z);
scene.imgbuf->Clear();
scene.Render(threadpool, args.njobs);
args.njobs = scene.njobs;
rendertime = timer.GetElapsed();
PrintStats(imgbuf, &scene);
if(scene.animate || firstframe) frame++;
imgbuf->Update();
firstframe = false;
}
}
// client
TCPConnectionPool client;
const ConnectionInfo * ci = client.Connect( AStackString<>( "127.0.0.1" ), testPort );
TEST_ASSERT( ci );
size_t sendSize = 31;
while ( sendSize <= maxSendSize )
{
server.m_ReceivedBytes = 0;
server.m_DataSize = sendSize;
Timer timer;
size_t totalSent = 0;
while ( timer.GetElapsed() < 0.1f )
{
// client sends some know data to the server
TEST_ASSERT( client.Send( ci, data.Get(), sendSize ) );
totalSent += sendSize;
}
while( server.m_ReceivedBytes < totalSent )
{
Thread::Sleep( 1 );
}
const float speedMBs = ( float( totalSent ) / timer.GetElapsed() ) / float( 1024 * 1024 );
OUTPUT( "Speed: %2.1f MiB/s, SendSize: %u\n", speedMBs, (uint32_t)sendSize );
sendSize = ( sendSize * 2 ) + 33; // +33 to avoid powers of 2
// MainCommon
//------------------------------------------------------------------------------
int MainCommon( const AString & args, void * hInstance )
{
// don't buffer output
VERIFY( setvbuf(stdout, nullptr, _IONBF, 0) == 0 );
VERIFY( setvbuf(stderr, nullptr, _IONBF, 0) == 0 );
// process cmd line args
FBuildWorkerOptions options;
if ( options.ProcessCommandLine( args ) == false )
{
return -3;
}
// only allow 1 worker per system
Timer t;
while ( g_OneProcessMutex.TryLock() == false )
{
// retry for upto 2 seconds, to allow some time for old worker to close
if ( t.GetElapsed() > 5.0f )
{
ShowMsgBox( "An FBuildWorker is already running!" );
return -1;
}
Thread::Sleep(100);
}
#if defined( __WINDOWS__ )
if ( options.m_UseSubprocess && !options.m_IsSubprocess )
{
return LaunchSubProcess( args );
}
#endif
// prevent popups when launching tools with missing dlls
#if defined( __WINDOWS__ )
::SetErrorMode( SEM_FAILCRITICALERRORS );
#else
// TODO:MAC SetErrorMode equivalent
// TODO:LINUX SetErrorMode equivalent
#endif
#if defined( __WINDOWS__ )
VERIFY( SetPriorityClass( GetCurrentProcess(), BELOW_NORMAL_PRIORITY_CLASS ) );
#else
// TODO:MAC SetPriorityClass equivalent
// TODO:LINUX SetPriorityClass equivalent
#endif
// start the worker and wait for it to be closed
int ret;
{
Worker worker( hInstance, args );
if ( options.m_OverrideCPUAllocation )
{
WorkerSettings::Get().SetNumCPUsToUse( options.m_CPUAllocation );
}
if ( options.m_OverrideWorkMode )
{
WorkerSettings::Get().SetMode( options.m_WorkMode );
}
ret = worker.Work();
}
MEMTRACKER_DUMP_ALLOCATIONS
return ret;
}
REGISTER_TESTS_END
// CompareHashTimes
//------------------------------------------------------------------------------
void TestHash::CompareHashTimes_Large() const
{
// use pseudo-random (but deterministic) data
const uint32_t seed = 0xB1234567;
Random r( seed );
// fill a buffer to use for tests
const size_t dataSize( 64 * 1024 * 1024 );
AutoPtr< uint32_t > data( (uint32_t *)ALLOC( dataSize ) );
for ( size_t i=0; i<dataSize / sizeof( uint32_t ); ++i )
{
data.Get()[ i ] = r.GetRand();
}
// baseline - sum 64 bits
{
Timer t;
uint64_t sum( 0 );
uint64_t * it = (uint64_t *)data.Get();
uint64_t * end = it + ( dataSize / sizeof( uint64_t ) );
while ( it != end )
{
sum += *it;
++it;
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "Sum64 : %2.3fs @ %2.3f GiB/s (sum: %016llx)\n", time, speed, sum );
}
// baseline - sum 32 bits
{
Timer t;
uint32_t sum( 0 );
uint32_t * it = data.Get();
uint32_t * end = it + ( dataSize / sizeof( uint32_t ) );
while ( it != end )
{
sum += *it;
++it;
}
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "Sum32 : %2.3fs @ %6.3f GiB/s (sum: 0x%x)\n", time, speed, sum );
}
// xxHash32
{
Timer t;
uint32_t crc = xxHash::Calc32( data.Get(), dataSize );
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "xxHash-32 : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// xxHash64
{
Timer t;
uint64_t crc = xxHash::Calc64( data.Get(), dataSize );
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "xxHash-64 : %2.3fs @ %6.3f GiB/s (hash: %016llx)\n", time, speed, crc );
}
// Murmur3 - 32
{
Timer t;
uint32_t crc = Murmur3::Calc32( data.Get(), dataSize );
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "Murmur3-32 : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// Murmur3 - 128
{
Timer t;
uint64_t hashB( 0 );
uint64_t hashA = Murmur3::Calc128( data.Get(), dataSize, hashB );
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "Murmur3-128 : %2.3fs @ %6.3f GiB/s (%016llx, %016llx)\n", time, speed, hashA, hashB );
}
// CRC32 - 8x8 slicing
{
Timer t;
uint32_t crc = CRC32::Calc( data.Get(), dataSize );
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "CRC32 8x8 : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// CRC32 - "standard" algorithm
{
Timer t;
uint32_t crc = CRC32::Start();
crc = CRC32::Update( crc, data.Get(), dataSize );
crc = CRC32::Stop( crc );
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "CRC32 : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// CRC32Lower
{
Timer t;
uint32_t crc = CRC32::CalcLower( data.Get(), dataSize );
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "CRC32Lower : %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
// Murmur3 - 32 Lower
{
Timer t;
// lower-case the data into a copy
AutoPtr< uint32_t > dataCopy( (uint32_t *)ALLOC( dataSize ) );
const char * src = (const char * )data.Get();
const char * end( src + ( dataSize / sizeof( uint32_t ) ) );
char * dst = (char *)dataCopy.Get();
while ( src < end )
{
char c = *src;
*dst = ( ( c >= 'A' ) && ( c <= 'Z' ) ) ? 'a' + c - 'A' : c ;
++src;
++dst;
}
// hash it
uint32_t crc = Murmur3::Calc32( dataCopy.Get(), dataSize );
float time = t.GetElapsed();
float speed = ( (float)dataSize / (float)( 1024 * 1024 * 1024 ) ) / time;
OUTPUT( "Murmur3-32-Lower: %2.3fs @ %6.3f GiB/s (hash: 0x%x)\n", time, speed, crc );
}
}
// CompressHelper
//------------------------------------------------------------------------------
void TestCompressor::CompressHelper( const char * fileName ) const
{
// read some test data into a file
AutoPtr< void > data;
size_t dataSize;
{
FileStream fs;
TEST_ASSERT( fs.Open( fileName ) );
dataSize = (size_t)fs.GetFileSize();
data = (char *)ALLOC( dataSize );
TEST_ASSERT( (uint32_t)fs.Read( data.Get(), dataSize ) == dataSize );
}
OUTPUT( "File : %s\n", fileName );
OUTPUT( "Size : %u\n", (uint32_t)dataSize );
// compress the data to obtain size
Compressor comp;
comp.Compress( data.Get(), dataSize );
size_t compressedSize = comp.GetResultSize();
AutoPtr< char > compressedData( (char *)ALLOC( compressedSize ) );
memcpy( compressedData.Get(), comp.GetResult(), compressedSize );
float compressedPerc = ( (float)compressedSize / (float)dataSize ) * 100.0f;
OUTPUT( "Compressed Size: %u (%2.1f%% of original)\n", (uint32_t)compressedSize, compressedPerc );
// decompress to check we get original data back
Compressor decomp;
decomp.Decompress( compressedData.Get() );
size_t uncompressedSize = decomp.GetResultSize();
TEST_ASSERT( uncompressedSize == dataSize );
for ( size_t i=0; i<uncompressedSize; ++i )
{
TEST_ASSERT( ( (char *)data.Get() )[ i ] == ( (char *)decomp.GetResult() )[ i ] );
}
// speed checks
//--------------
const size_t NUM_REPEATS( 100 );
// compress the data several times to get more stable throughput value
Timer t;
for ( size_t i=0; i<NUM_REPEATS; ++i )
{
Compressor c;
c.Compress( data.Get(), dataSize );
TEST_ASSERT( c.GetResultSize() == compressedSize );
}
float compressTimeTaken = t.GetElapsed();
double compressThroughputMBs = ( ( (double)dataSize / 1024.0 * (double)NUM_REPEATS ) / compressTimeTaken ) / 1024.0;
OUTPUT( " Comp Speed: %2.1f MB/s - %2.3fs (%u repeats)\n", (float)compressThroughputMBs, compressTimeTaken, NUM_REPEATS );
// decompress the data
Timer t2;
for ( size_t i=0; i<NUM_REPEATS; ++i )
{
Compressor d;
d.Decompress( compressedData.Get() );
TEST_ASSERT( d.GetResultSize() == dataSize );
}
float decompressTimeTaken = t2.GetElapsed();
double decompressThroughputMBs = ( ( (double)dataSize / 1024.0 * (double)NUM_REPEATS ) / decompressTimeTaken ) / 1024.0;
OUTPUT( " Decomp Speed: %2.1f MB/s - %2.3fs (%u repeats)\n", (float)decompressThroughputMBs, decompressTimeTaken, NUM_REPEATS );
// time memcpy to compare with
Timer t0;
for ( size_t i=0; i<NUM_REPEATS; ++i )
{
char * mem = (char *)ALLOC( dataSize );
memcpy( mem, data.Get(), dataSize );
FREE( mem );
}
float memcpyTimeTaken = t0.GetElapsed();
double memcpyThroughputMBs = ( ( (double)dataSize / 1024.0 * (double)NUM_REPEATS ) / memcpyTimeTaken ) / 1024.0;
OUTPUT( " MemCpy Speed: %2.1f MB/s - %2.3fs (%u repeats)\n", (float)memcpyThroughputMBs, memcpyTimeTaken, NUM_REPEATS );
}