void AdlPrimitivesDemo::testFill2( Buffer<int2>& buf, int size, Stopwatch& sw )
{
MyFill::Data* sortData = MyFill::allocate( m_deviceData );
sw.start();
MyFill::execute( sortData, buf, make_int2(12, 13), size );
sw.stop();
MyFill::deallocate( sortData );
{
float t = sw.getMs();
sprintf_s(m_txtBuffer[m_nTxtLines++], LINE_CAPACITY, "Fill int2: %3.2fGB/s (%3.2fms)", size/t/1000/1000*8, t);
}
}
void AdlPrimitivesDemo::testSort( Buffer<SortData>& buf, int size, Stopwatch& sw )
{
MySort::Data* sortData = MySort::allocate( m_deviceData, size, RadixSortBase::SORT_ADVANCED );
sw.start();
MySort::execute( sortData, buf, size );
sw.stop();
MySort::deallocate( sortData );
{
m_nTxtLines = 0;
float t = sw.getMs();
sprintf_s(m_txtBuffer[m_nTxtLines++], LINE_CAPACITY, "%d: %3.2fms, %3.2fMKeys/s", size, t, size/t/1000);
}
}
void AdlPrimitivesDemo::test( Buffer<int2>& buf, int size, Stopwatch& sw )
{
Kernel* kernel = KernelManager::query( m_deviceData, "..\\..\\AdlDemos\\TestBed\\Demos\\AdlPrimitivesDemoKernel", "FillInt4Kernel" );
Buffer<int4> constBuffer( m_deviceData, 1, BufferBase::BUFFER_CONST );
int numGroups = (size+128*4-1)/(128*4);
Buffer<u32> workBuffer0( m_deviceData, numGroups*(16) );
Buffer<u32> workBuffer1( m_deviceData, numGroups*(16) );
Buffer<int2> sortBuffer( m_deviceData, size );
{
int2* host = new int2[size];
for(int i=0; i<size; i++)
{
host[i] = make_int2( getRandom(0, 0xf), i );
}
sortBuffer.write( host, size );
DeviceUtils::waitForCompletion( m_deviceData );
delete [] host;
}
int4 constData;
{
constData.x = size;
constData.y = 0;
constData.z = numGroups;
constData.w = 0;
}
sw.start();
int nThreads = size/4;
{
BufferInfo bInfo[] = { BufferInfo( &buf ), BufferInfo( &workBuffer0 ), BufferInfo( &workBuffer1 ) };
Launcher launcher( m_deviceData, kernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(Launcher::BufferInfo) );
launcher.setConst( constBuffer, constData );
launcher.launch1D( nThreads, 128 );
}
sw.split();
{
constData.w = 1;
int nThreads = size/4;
BufferInfo bInfo[] = { BufferInfo( &buf ), BufferInfo( &workBuffer0 ), BufferInfo( &workBuffer1 ) };
Launcher launcher( m_deviceData, kernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(Launcher::BufferInfo) );
launcher.setConst( constBuffer, constData );
launcher.launch1D( nThreads, 128 );
}
sw.split();
{
constData.w = 2;
int nThreads = size/4;
BufferInfo bInfo[] = { BufferInfo( &sortBuffer ), BufferInfo( &workBuffer0 ), BufferInfo( &workBuffer1 ) };
Launcher launcher( m_deviceData, kernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(Launcher::BufferInfo) );
launcher.setConst( constBuffer, constData );
launcher.launch1D( nThreads, 128 );
}
sw.stop();
{
int2* host = new int2[size];
buf.read( host, size );
DeviceUtils::waitForCompletion( m_deviceData );
for(int i=0; i<128*4-1; i++)
{
ADLASSERT( host[i].x <= host[i+1].x );
}
delete [] host;
}
{
float t[3];
sw.getMs(t, 3);
// (byte * nElems)
sprintf_s(m_txtBuffer[m_nTxtLines++], LINE_CAPACITY, "LoadStore: %3.2fGB/s (%3.2fns)", (4*8*2)*nThreads/t[0]/1000/1000, t[0]*1000.f);
sprintf_s(m_txtBuffer[m_nTxtLines++], LINE_CAPACITY, "GenHistog: %3.2fGB/s (%3.2fns)", (4*(8*2+2))*nThreads/t[1]/1000/1000, t[1]*1000.f);
sprintf_s(m_txtBuffer[m_nTxtLines++], LINE_CAPACITY, "FullSort: %3.2fGB/s (%3.2fns)", (4*(8*2+2))*nThreads/t[2]/1000/1000, t[2]*1000.f);
}
}