/**
* Compare results
*
* @param argv Commmand line parameters
*
* @relates Compare
* @ingroup pwglf_forward_tracklets
*/
void CompareResults(const char** argv)
{
TString newFile;
TString oldFile;
TString newTit("");
TString oldTit("");
const char** ptr = argv;
while ((*ptr)) {
TString argi = *ptr;
ptr++;
if (argi.Contains("help")) {
Printf("Usage: CompareResults AFILE BFILE [ATITLTE [BTITLE]]");
return;
}
if (argi.Contains("CompareResults.C")) continue;
if (argi.BeginsWith("-")) continue;
if (argi.EndsWith(".root")) {
if (newFile.IsNull()) newFile = argi;
else oldFile = argi;
}
else {
if (newTit.IsNull()) newTit = argi;
else oldTit = argi;
}
}
if (newTit.IsNull()) newTit = "New";
if (oldTit.IsNull()) oldTit = "Old";
CompareResults(newFile, oldFile, newTit, oldTit);
}
void CIndex::SortResults(CSearchObject& SearchObject, CSortType SortType, int l, int r) {
int i = l;
int j = r;
int pivot = SearchObject.m_SearchData.m_Results[(l+r)/2];
while (i<=j) {
while(CompareResults(SortType, SearchObject.m_SearchData.m_Results[i], pivot) < 0) i++;
while(CompareResults(SortType, SearchObject.m_SearchData.m_Results[j], pivot) > 0) j--;
if (i<=j){
int t = SearchObject.m_SearchData.m_Results[i];
int tt = SearchObject.m_SearchData.m_ResultsPositions[i];
SearchObject.m_SearchData.m_Results[i] = SearchObject.m_SearchData.m_Results[j];
SearchObject.m_SearchData.m_ResultsPositions[i] = SearchObject.m_SearchData.m_ResultsPositions[j];
SearchObject.m_SearchData.m_Results[j] = t;
SearchObject.m_SearchData.m_ResultsPositions[j] = tt;
i++;
j--;
}
}
if (l < j) SortResults(SearchObject, SortType, l, j);
if (i < r) SortResults(SearchObject, SortType, i, r);
}
//____________________________________________________________________
Int_t multidimfit(bool doFit = true)
{
cout << "*************************************************" << endl;
cout << "* Multidimensional Fit *" << endl;
cout << "* *" << endl;
cout << "* By Christian Holm <*****@*****.**> 14/10/00 *" << endl;
cout << "*************************************************" << endl;
cout << endl;
// Initialize global TRannom object.
gRandom = new TRandom();
// Open output file
TFile* output = new TFile("mdf.root", "RECREATE");
// Global data parameters
Int_t nVars = 4;
Int_t nData = 500;
Double_t x[4];
// make fit object and set parameters on it.
TMultiDimFit* fit = new TMultiDimFit(nVars, TMultiDimFit::kMonomials,"v");
Int_t mPowers[] = { 6 , 6, 6, 6 };
fit->SetMaxPowers(mPowers);
fit->SetMaxFunctions(1000);
fit->SetMaxStudy(1000);
fit->SetMaxTerms(30);
fit->SetPowerLimit(1);
fit->SetMinAngle(10);
fit->SetMaxAngle(10);
fit->SetMinRelativeError(.01);
// variables to hold the temporary input data
Double_t d;
Double_t e;
// Print out the start parameters
fit->Print("p");
printf("======================================\n");
// Create training sample
Int_t i;
for (i = 0; i < nData ; i++) {
// Make some data
makeData(x,d,e);
// Add the row to the fit object
fit->AddRow(x,d,e);
}
// Print out the statistics
fit->Print("s");
// Book histograms
fit->MakeHistograms();
// Find the parameterization
fit->FindParameterization();
// Print coefficents
fit->Print("rc");
// Get the min and max of variables from the training sample, used
// for cuts in test sample.
Double_t *xMax = new Double_t[nVars];
Double_t *xMin = new Double_t[nVars];
for (i = 0; i < nVars; i++) {
xMax[i] = (*fit->GetMaxVariables())(i);
xMin[i] = (*fit->GetMinVariables())(i);
}
nData = fit->GetNCoefficients() * 100;
Int_t j;
// Create test sample
for (i = 0; i < nData ; i++) {
// Make some data
makeData(x,d,e);
for (j = 0; j < nVars; j++)
if (x[j] < xMin[j] || x[j] > xMax[j])
break;
// If we get through the loop above, all variables are in range
if (j == nVars)
// Add the row to the fit object
fit->AddTestRow(x,d,e);
else
i--;
}
//delete gRandom;
// Test the parameterizatio and coefficents using the test sample.
if (doFit)
fit->Fit("M");
// Print result
fit->Print("fc v");
// Write code to file
fit->MakeCode();
// Write histograms to disk, and close file
output->Write();
output->Close();
delete output;
// Compare results with reference run
Int_t compare = CompareResults(fit, doFit);
if (!compare) {
printf("\nmultidimfit .............................................. OK\n");
} else {
printf("\nmultidimfit .............................................. fails case %d\n",compare);
}
// We're done
delete fit;
return compare;
}
int main(int argc,
char **argv)
{
l_int32 i, j, k, w, h, w2, w4, w8, w16, w32, wpl;
l_int32 count1, count2, count3;
l_uint32 val32, val1, val2;
l_uint32 *data1, *line1, *data2, *line2;
void **lines1, **linet1, **linet2;
PIX *pixs, *pix1, *pix2;
L_REGPARAMS *rp;
if (regTestSetup(argc, argv, &rp))
return 1;
pixs = pixRead("feyn-fract.tif");
pixGetDimensions(pixs, &w, &h, NULL);
data1 = pixGetData(pixs);
wpl = pixGetWpl(pixs);
lines1 = pixGetLinePtrs(pixs, NULL);
/* Get timing for the 3 different methods */
startTimer();
for (k = 0; k < 10; k++) {
count1 = 0;
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
if (GET_DATA_BIT(lines1[i], j))
count1++;
}
}
}
fprintf(stderr, "Time with line ptrs = %5.3f sec, count1 = %d\n",
stopTimer(), count1);
startTimer();
for (k = 0; k < 10; k++) {
count2 = 0;
for (i = 0; i < h; i++) {
line1 = data1 + i * wpl;
for (j = 0; j < w; j++) {
if (l_getDataBit(line1, j))
count2++;
}
}
}
fprintf(stderr, "Time with l_get* = %5.3f sec, count2 = %d\n",
stopTimer(), count2);
startTimer();
for (k = 0; k < 10; k++) {
count3 = 0;
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
pixGetPixel(pixs, j, i, &val32);
count3 += val32;
}
}
}
fprintf(stderr, "Time with pixGetPixel() = %5.3f sec, count3 = %d\n",
stopTimer(), count3);
pix1 = pixCreateTemplate(pixs);
linet1 = pixGetLinePtrs(pix1, NULL);
pix2 = pixCreateTemplate(pixs);
data2 = pixGetData(pix2);
linet2 = pixGetLinePtrs(pix2, NULL);
/* ------------------------------------------------- */
/* Test different methods for 1 bpp */
/* ------------------------------------------------- */
count1 = 0;
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
val1 = GET_DATA_BIT(lines1[i], j);
count1 += val1;
if (val1) SET_DATA_BIT(linet1[i], j);
}
}
count2 = 0;
for (i = 0; i < h; i++) {
line1 = data1 + i * wpl;
line2 = data2 + i * wpl;
for (j = 0; j < w; j++) {
val2 = l_getDataBit(line1, j);
count2 += val2;
if (val2) l_setDataBit(line2, j);
}
}
CompareResults(pixs, pix1, pix2, count1, count2, "1 bpp", rp);
/* ------------------------------------------------- */
/* Test different methods for 2 bpp */
/* ------------------------------------------------- */
count1 = 0;
w2 = w / 2;
for (i = 0; i < h; i++) {
for (j = 0; j < w2; j++) {
val1 = GET_DATA_DIBIT(lines1[i], j);
count1 += val1;
val1 += 0xbbbbbbbc;
SET_DATA_DIBIT(linet1[i], j, val1);
}
}
count2 = 0;
for (i = 0; i < h; i++) {
line1 = data1 + i * wpl;
line2 = data2 + i * wpl;
for (j = 0; j < w2; j++) {
val2 = l_getDataDibit(line1, j);
count2 += val2;
val2 += 0xbbbbbbbc;
l_setDataDibit(line2, j, val2);
}
}
CompareResults(pixs, pix1, pix2, count1, count2, "2 bpp", rp);
/* ------------------------------------------------- */
/* Test different methods for 4 bpp */
/* ------------------------------------------------- */
count1 = 0;
w4 = w / 4;
for (i = 0; i < h; i++) {
for (j = 0; j < w4; j++) {
val1 = GET_DATA_QBIT(lines1[i], j);
count1 += val1;
val1 += 0xbbbbbbb0;
SET_DATA_QBIT(linet1[i], j, val1);
}
}
count2 = 0;
for (i = 0; i < h; i++) {
line1 = data1 + i * wpl;
line2 = data2 + i * wpl;
for (j = 0; j < w4; j++) {
val2 = l_getDataQbit(line1, j);
count2 += val2;
val2 += 0xbbbbbbb0;
l_setDataQbit(line2, j, val2);
}
}
CompareResults(pixs, pix1, pix2, count1, count2, "4 bpp", rp);
/* ------------------------------------------------- */
/* Test different methods for 8 bpp */
/* ------------------------------------------------- */
count1 = 0;
w8 = w / 8;
for (i = 0; i < h; i++) {
for (j = 0; j < w8; j++) {
val1 = GET_DATA_BYTE(lines1[i], j);
count1 += val1;
val1 += 0xbbbbbb00;
SET_DATA_BYTE(linet1[i], j, val1);
}
}
count2 = 0;
for (i = 0; i < h; i++) {
line1 = data1 + i * wpl;
line2 = data2 + i * wpl;
for (j = 0; j < w8; j++) {
val2 = l_getDataByte(line1, j);
count2 += val2;
val2 += 0xbbbbbb00;
l_setDataByte(line2, j, val2);
}
}
CompareResults(pixs, pix1, pix2, count1, count2, "8 bpp", rp);
/* ------------------------------------------------- */
/* Test different methods for 16 bpp */
/* ------------------------------------------------- */
count1 = 0;
w16 = w / 16;
for (i = 0; i < h; i++) {
for (j = 0; j < w16; j++) {
val1 = GET_DATA_TWO_BYTES(lines1[i], j);
count1 += val1;
val1 += 0xbbbb0000;
SET_DATA_TWO_BYTES(linet1[i], j, val1);
}
}
count2 = 0;
for (i = 0; i < h; i++) {
line1 = data1 + i * wpl;
line2 = data2 + i * wpl;
for (j = 0; j < w16; j++) {
val2 = l_getDataTwoBytes(line1, j);
count2 += val2;
val2 += 0xbbbb0000;
l_setDataTwoBytes(line2, j, val2);
}
}
CompareResults(pixs, pix1, pix2, count1, count2, "16 bpp", rp);
/* ------------------------------------------------- */
/* Test different methods for 32 bpp */
/* ------------------------------------------------- */
count1 = 0;
w32 = w / 32;
for (i = 0; i < h; i++) {
for (j = 0; j < w32; j++) {
val1 = GET_DATA_FOUR_BYTES(lines1[i], j);
count1 += val1 & 0xfff;
SET_DATA_FOUR_BYTES(linet1[i], j, val1);
}
}
count2 = 0;
for (i = 0; i < h; i++) {
line1 = data1 + i * wpl;
line2 = data2 + i * wpl;
for (j = 0; j < w32; j++) {
val2 = l_getDataFourBytes(line1, j);
count2 += val2 & 0xfff;
l_setDataFourBytes(line2, j, val2);
}
}
CompareResults(pixs, pix1, pix2, count1, count2, "32 bpp", rp);
pixDestroy(&pixs);
pixDestroy(&pix1);
pixDestroy(&pix2);
lept_free(lines1);
lept_free(linet1);
lept_free(linet2);
return regTestCleanup(rp);
}
// Main program
//*****************************************************************************
int main(int argc, char** argv)
{
// Locals used with command line args
int p = 256; // workgroup X dimension
int q = 1; // workgroup Y dimension
pArgc = &argc;
pArgv = argv;
shrQAStart(argc, argv);
// latch the executable path for other funcs to use
cExecutablePath = argv[0];
// start logs and show command line help
shrSetLogFileName ("oclNbody.txt");
shrLog("%s Starting...\n\n", cExecutablePath);
shrLog("Command line switches:\n");
shrLog(" --qatest\t\tCheck correctness of GPU execution and measure performance)\n");
shrLog(" --noprompt\t\tQuit simulation automatically after a brief period\n");
shrLog(" --n=<numbodies>\tSpecify # of bodies to simulate (default = %d)\n", numBodies);
shrLog(" --double\t\tUse double precision floating point values for simulation\n");
shrLog(" --p=<workgroup X dim>\tSpecify X dimension of workgroup (default = %d)\n", p);
shrLog(" --q=<workgroup Y dim>\tSpecify Y dimension of workgroup (default = %d)\n\n", q);
// Get command line arguments if there are any and set vars accordingly
if (argc > 0)
{
shrGetCmdLineArgumenti(argc, (const char**)argv, "p", &p);
shrGetCmdLineArgumenti(argc, (const char**)argv, "q", &q);
shrGetCmdLineArgumenti(argc, (const char**)argv, "n", &numBodies);
bDouble = (shrTRUE == shrCheckCmdLineFlag(argc, (const char**)argv, "double"));
bNoPrompt = shrCheckCmdLineFlag(argc, (const char**)argv, "noprompt");
bQATest = shrCheckCmdLineFlag(argc, (const char**)argv, "qatest");
}
//Get the NVIDIA platform
cl_int ciErrNum = oclGetPlatformID(&cpPlatform);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
shrLog("clGetPlatformID...\n\n");
if (bDouble)
{
shrLog("Double precision execution...\n\n");
}
else
{
shrLog("Single precision execution...\n\n");
}
flopsPerInteraction = bDouble ? 30 : 20;
//Get all the devices
shrLog("Get the Device info and select Device...\n");
ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, 0, NULL, &uiNumDevices);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
cdDevices = (cl_device_id *)malloc(uiNumDevices * sizeof(cl_device_id) );
ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, uiNumDevices, cdDevices, NULL);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Set target device and Query number of compute units on uiTargetDevice
shrLog(" # of Devices Available = %u\n", uiNumDevices);
if(shrGetCmdLineArgumentu(argc, (const char**)argv, "device", &uiTargetDevice)== shrTRUE)
{
uiTargetDevice = CLAMP(uiTargetDevice, 0, (uiNumDevices - 1));
}
shrLog(" Using Device %u, ", uiTargetDevice);
oclPrintDevName(LOGBOTH, cdDevices[uiTargetDevice]);
cl_uint uiNumComputeUnits;
clGetDeviceInfo(cdDevices[uiTargetDevice], CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(uiNumComputeUnits), &uiNumComputeUnits, NULL);
shrLog(" # of Compute Units = %u\n", uiNumComputeUnits);
//Create the context
shrLog("clCreateContext...\n");
cxContext = clCreateContext(0, uiNumDevsUsed, &cdDevices[uiTargetDevice], NULL, NULL, &ciErrNum);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Create a command-queue
shrLog("clCreateCommandQueue...\n\n");
cqCommandQueue = clCreateCommandQueue(cxContext, cdDevices[uiTargetDevice], CL_QUEUE_PROFILING_ENABLE, &ciErrNum);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Log and config for number of bodies
shrLog("Number of Bodies = %d\n", numBodies);
switch (numBodies)
{
case 1024:
activeParams.m_clusterScale = 1.52f;
activeParams.m_velocityScale = 2.f;
break;
case 2048:
activeParams.m_clusterScale = 1.56f;
activeParams.m_velocityScale = 2.64f;
break;
case 4096:
activeParams.m_clusterScale = 1.68f;
activeParams.m_velocityScale = 2.98f;
break;
case 7680:
case 8192:
activeParams.m_clusterScale = 1.98f;
activeParams.m_velocityScale = 2.9f;
break;
default:
case 15360:
case 16384:
activeParams.m_clusterScale = 1.54f;
activeParams.m_velocityScale = 8.f;
break;
case 30720:
case 32768:
activeParams.m_clusterScale = 1.44f;
activeParams.m_velocityScale = 11.f;
break;
}
if ((q * p) > 256)
{
p = 256 / q;
shrLog("Setting p=%d to maintain %d threads per block\n", p, 256);
}
if ((q == 1) && (numBodies < p))
{
p = numBodies;
shrLog("Setting p=%d because # of bodies < p\n", p);
}
shrLog("Workgroup Dims = (%d x %d)\n\n", p, q);
// Initialize OpenGL items if using GL
if (bQATest == shrFALSE)
{
assert(0);
/*
shrLog("Calling InitGL...\n");
InitGL(&argc, argv);
*/
}
else
{
shrLog("Skipping InitGL...\n");
}
// CL/GL interop disabled
bUsePBO = (false && (bQATest == shrFALSE));
InitNbody(cdDevices[uiTargetDevice], cxContext, cqCommandQueue, numBodies, p, q, bUsePBO, bDouble);
ResetSim(nbody, numBodies, NBODY_CONFIG_SHELL, bUsePBO);
// init timers
shrDeltaT(DEMOTIME); // timer 0 is for timing demo periods
shrDeltaT(FUNCTIME); // timer 1 is for logging function delta t's
shrDeltaT(FPSTIME); // timer 2 is for fps measurement
// Standard simulation
if (bQATest == shrFALSE)
{
assert(0);
/*
shrLog("Running standard oclNbody simulation...\n\n");
glutDisplayFunc(DisplayGL);
glutReshapeFunc(ReshapeGL);
glutMouseFunc(MouseGL);
glutMotionFunc(MotionGL);
glutKeyboardFunc(KeyboardGL);
glutSpecialFunc(SpecialGL);
glutIdleFunc(IdleGL);
glutMainLoop();
*/
}
// Compare to host, profile and write out file for regression analysis
if (bQATest == shrTRUE) {
bool bTestResults = false;
shrLog("Running oclNbody Results Comparison...\n\n");
bTestResults = CompareResults(numBodies);
//shrLog("Profiling oclNbody...\n\n");
//RunProfiling(100, (unsigned int)(p * q)); // 100 iterations
shrQAFinish(argc, (const char **)argv, bTestResults ? QA_PASSED : QA_FAILED);
} else {
// Cleanup/exit
bNoPrompt = shrTRUE;
shrQAFinish2(false, *pArgc, (const char **)pArgv, QA_PASSED);
}
Cleanup(EXIT_SUCCESS);
}
void NonTemporalStoreTimed(void)
{
ThreadTimer::SetThreadAffinityMask();
const int num_it = 500;
const int num_alg = 3;
const double et_scale = 1.0e6;
double et[num_it][num_alg];
ThreadTimer tt;
const int n = 1000000;
const int align = 16;
float* a = (float*)_aligned_malloc(n * sizeof(float), align);
float* b = (float*)_aligned_malloc(n * sizeof(float), align);
float* c1 = (float*)_aligned_malloc(n * sizeof(float), align);
float* c2a = (float*)_aligned_malloc(n * sizeof(float), align);
float* c2b = (float*)_aligned_malloc(n * sizeof(float), align);
srand(67);
for (int i = 0; i < n; i++)
{
a[i] = (float)(rand() % 100);
b[i] = (float)(rand() % 100);
}
for (int i = 0; i < num_it; i++)
{
tt.Start();
CalcResultCpp(c1, a, b, n);
tt.Stop();
et[i][0] = tt.GetElapsedTime() * et_scale;
}
for (int i = 0; i < num_it; i++)
{
tt.Start();
CalcResultA_(c2a, a, b, n);
tt.Stop();
et[i][1] = tt.GetElapsedTime() * et_scale;
}
for (int i = 0; i < num_it; i++)
{
tt.Start();
CalcResultB_(c2b, a, b, n);
tt.Stop();
et[i][2] = tt.GetElapsedTime() * et_scale;
}
#ifdef _WIN64
const char* fn = "__NonTemporalStore64.csv";
#else
const char* fn = "__NonTemporalStore32.csv";
#endif
if (!CompareResults(c1, c2a, c2b, n, false))
printf("NonTemporalStoreTimed() - array compare FAILED\n");
ThreadTimer::SaveElapsedTimeMatrix(fn, (double*)et, num_it, num_alg);
printf("\nBenchmark times saved to file %s\n", fn);
_aligned_free(a);
_aligned_free(b);
_aligned_free(c1);
_aligned_free(c2a);
_aligned_free(c2b);
}
/**
* Compare results
*
* @relates Compare
* @ingroup pwglf_forward_tracklets
*/
void CompareResults()
{
CompareResults(const_cast<const char**>(&(gApplication->Argv()[1])));
gApplication->ClearInputFiles();
}
