void PhysicsHashSpace::changed(ConstFieldMaskArg whichField,
UInt32 origin,
BitVector details)
{
Inherited::changed(whichField, origin, details);
//Do not respond to changes that have a Sync origin
if(origin & ChangedOrigin::Sync)
{
return;
}
if(whichField & LevelsFieldMask)
{
dHashSpaceSetLevels(_SpaceID, (Int32)getLevels().x(), (Int32)getLevels().y());
}
}
// -----------------------------------------------------------------------------
//
// -----------------------------------------------------------------------------
void MultiOtsuThreshold::readFilterParameters(AbstractFilterParametersReader* reader, int index)
{
reader->openFilterGroup(this, index);
setSelectedCellArrayPath( reader->readDataArrayPath( "SelectedCellArrayPath", getSelectedCellArrayPath() ) );
setNewCellArrayName( reader->readString( "NewCellArrayName", getNewCellArrayName() ) );
setSaveAsNewArray( reader->readValue( "SaveAsNewArray", getSaveAsNewArray() ) );
setSlice( reader->readValue( "Slice", getSlice() ) );
setLevels( reader->readValue( "Levels", getLevels() ) );
reader->closeFilterGroup();
}
std::string Node::texLine()
{
std::ostringstream result;
result << f << " & "
<< getLevels() << " & "
<< getBranches() << " & "
<< getCount() << "\\\\"
<< std::endl;
return result.str();
}
std::string Node::csvLine()
{
std::ostringstream result;
result << f << ","
<< getLevels() << ","
<< getBranches() << ","
<< getCount() << std::endl;
result << generator() << std::endl;
return result.str();
}
void
ControlFile::ensureComponent(const char *pattern)
{
// Make sure at least one entry exists matching a pattern, if not,
// create it. Wildcard patterns cannot be created though!
ComponentIterator iter(getComponentIterator());
bool wasSeen = false;
Component *c;
while ((c = iter.next()) != NULL) {
std::unique_ptr<Component> component(c);
if (c->matches(pattern)) {
wasSeen = true;
break;
}
}
if (!wasSeen) {
(void) getLevels(pattern); // Creates it ((### ugly ###))
}
}
int DwtHaar1D::setupDwtHaar1D()
{
// signal length must be power of 2
signalLength = sampleCommon->roundToPowerOf2<cl_uint>(signalLength);
unsigned int levels = 0;
int result = getLevels(signalLength, &levels);
CHECK_ERROR(result,SDK_SUCCESS, "signalLength > 2 ^ 23 not supported");
// Allocate and init memory used by host
inData = (cl_float*)malloc(signalLength * sizeof(cl_float));
CHECK_ALLOCATION(inData, "Failed to allocate host memory. (inData)");
for(unsigned int i = 0; i < signalLength; i++)
{
inData[i] = (cl_float)(rand() % 10);
}
dOutData = (cl_float*) malloc(signalLength * sizeof(cl_float));
CHECK_ALLOCATION(dOutData, "Failed to allocate host memory. (dOutData)");
memset(dOutData, 0, signalLength * sizeof(cl_float));
dPartialOutData = (cl_float*) malloc(signalLength * sizeof(cl_float));
CHECK_ALLOCATION(dPartialOutData, "Failed to allocate host memory.(dPartialOutData)");
memset(dPartialOutData, 0, signalLength * sizeof(cl_float));
hOutData = (cl_float*)malloc(signalLength * sizeof(cl_float));
CHECK_ALLOCATION(hOutData, "Failed to allocate host memory. (hOutData)");
memset(hOutData, 0, signalLength * sizeof(cl_float));
if(!quiet)
{
sampleCommon->printArray<cl_float>("Input Signal", inData, 256, 1);
}
return SDK_SUCCESS;
}
void CampaignLevelState::init(Game* game)
{
// important!
setGame(game);
// create the menu
//mCampaignLevelMenu = new Menu("CustemLevelMenu", MOUSE, true, 4, 4);
mCampaignLevelMenu = new Menu("CustemLevelMenu", MOUSE, VER, 2, true, 4, 4);
mCampaignLevelMenu->setMenuBackground("none", 700, 450, 256, 350); // mCampaignLevelMenu->setMenuBackground("misc\\textures\\menu_bkgd.bmp", 700, 450, 256, 350);
// get the files in the map folder
std::vector<string> levelList = getLevels();
// load the campaign progress
mProgress.loadProgress("levels\\campaign\\campaign_progress.txt");
// add menu items, each level
for(int i = 0; i < levelList.size(); i++)
{
// remove .txt
if(levelList[i].find(".txt") != string::npos || levelList[i].find(".TXT") != string::npos)
levelList[i].erase(levelList[i].end()-4,levelList[i].end());
if(mProgress.getProgress(levelList[i]).playable == 0) {
mCampaignLevelMenu->addMenuItem(levelList[i], (char*)GRAY_BUTTON_NORMAL_SOURCE.c_str(), (char*)GRAY_BUTTON_HOOVER_SOURCE.c_str());
mCampaignLevelMenu->setPressable(levelList[i], false);
}
else
mCampaignLevelMenu->addMenuItem(levelList[i], "misc\\textures\\level_menu_normal.bmp", "misc\\textures\\level_menu_hoover.bmp");
}
// build the menu
mCampaignLevelMenu->buildMenu2();
mCampaignLevelMenu->connect(&CampaignLevelState::menuHandler, this);
mBackgroundTexture = gGraphics->loadTexture("misc\\textures\\epic_bkgd.bmp");
}
////////////////////////////////////////////////////////////////////////////////
//! Perform the wavelet decomposition
////////////////////////////////////////////////////////////////////////////////
int runTest( int argc, char** argv)
{
int i;
#ifdef HW
XFcuda xcore;
int Status;
Status = XFcuda_Initialize(&xcore, 0);
if (Status != XST_SUCCESS) {
printf("Initialization failed %d\r\n", Status);
return XST_FAILURE;
}
#endif
unsigned int slength = 262144;
// get the number of decompositions necessary to perform a full decomposition
unsigned int dlevels_complete = 0;
if (1 != getLevels( slength, &dlevels_complete))
{
// error message
fprintf( stderr, "Signal length not supported.\n");
return;
}
// device in data
float* d_idata = NULL;
// device out data
float* d_odata = NULL;
// device approx_final data
float* approx_final = NULL;
// The very final approximation coefficient has to be written to the output
// data, all others are reused as input data in the next global step and
// therefore have to be written to the input data again.
// The following flag indicates where to copy approx_final data
// - 0 is input, 1 is output
int approx_is_input;
// allocate device mem
const unsigned int smem_size = sizeof(float) * slength;
d_idata = (float*) malloc(smem_size);
d_odata = (float*) malloc(smem_size);
approx_final = (float*) malloc(smem_size);
// copy input data to device
memcpy(d_idata, signal, smem_size);
// clear result memory
float* tmp = (float*) malloc( smem_size);
for (i = 0; i < slength; ++i)
tmp[i] = 0.0;
memcpy(d_odata, tmp, smem_size);
free(tmp);
// total number of threads
// in the first decomposition step always one thread computes the average and
// detail signal for one pair of adjacent values
unsigned int num_threads_total_left = slength / 2;
// decomposition levels performed in the current / next step
unsigned int dlevels_step = dlevels_complete;
// 1D signal so the arrangement of elements is also 1D
dim3 block_size;
dim3 grid_size;
// number of decomposition levels left after one iteration on the device
unsigned int dlevels_left = dlevels_complete;
// if less or equal 1k elements, then the data can be processed in one block,
// this avoids the Wait-For-Idle (WFI) on host side which is necessary if the
// computation is split accross multiple SM's if enough input data
if( dlevels_complete <= 10) {
// decomposition can be performed at once
block_size.x = num_threads_total_left;
approx_is_input = 0;
} else {
// 512 threads per block
grid_size.x = (num_threads_total_left / 512);
block_size.x = 512;
// 512 threads corresponds to 10 decomposition steps
dlevels_step = 10;
dlevels_left -= 10;
approx_is_input = 1;
}
grid_size.y = 1;
grid_size.z = 1;
block_size.y = 1;
block_size.z = 1;
#ifdef HW
XFcuda_SetGriddim_y(&xcore, grid_size.y);
//XFcuda_SetGriddim_z(&xcore, grid_size.z);
//XFcuda_SetBlockdim_y(&xcore, block_size.y);
//XFcuda_SetBlockdim_z(&xcore, block_size.z);
XFcuda_SetId_addr(&xcore, (int)d_idata / sizeof(float));
XFcuda_SetOd_addr(&xcore, (int)d_odata / sizeof(float));
XFcuda_SetApprox_final_addr(&xcore, (int)approx_final / sizeof(float));
#endif
while( 0 != num_threads_total_left) {
#ifndef HW
//PS execution
dwtHaar1D(d_idata, d_odata, approx_final, dlevels_step, num_threads_total_left, block_size.x, grid_size, block_size, 1, 0);
#else
XFcuda_SetDlevels(&xcore, dlevels_step);
XFcuda_SetSlength_step_half(&xcore, num_threads_total_left);
XFcuda_SetBdim(&xcore, block_size.x);
XFcuda_SetGriddim_x(&xcore, grid_size.x);
XFcuda_SetBlockdim_x(&xcore, block_size.x);
Xil_DCacheFlush();
XFcuda_SetEn_fcuda1(&xcore, 1);
XFcuda_Start(&xcore);
while (!XFcuda_IsDone(&xcore));
#endif
// Copy approx_final to appropriate location
if (approx_is_input) {
memcpy(d_idata, approx_final, grid_size.x*4);
}
else {
memcpy(d_odata, approx_final, grid_size.x*4);
}
// update level variables
if( dlevels_left < 10) {
// approx_final = d_odata;
approx_is_input = 0;
}
// more global steps necessary
dlevels_step = (dlevels_left > 10) ? dlevels_left - 10 : dlevels_left;
dlevels_left -= 10;
// after each step only half the threads are used any longer
// therefore after 10 steps 2^10 less threads
num_threads_total_left = num_threads_total_left >> 10;
// update block and grid size
grid_size.x = (num_threads_total_left / 512)
+ (0 != (num_threads_total_left % 512)) ? 1 : 0;
if( grid_size.x <= 1) {
block_size.x = num_threads_total_left;
}
}
#ifdef VERIFY
for (i = 0; i < 10; i++)
printf("index=%d, ref=%f, fpga=%f\n", i, reference[i], d_odata[i]);
int res = compareData(reference, d_odata, slength, 0.1f);
printf("%s\n", (1 == res) ? "PASSED." : "FAILED.");
#endif
free(d_idata);
free(d_odata);
free(approx_final);
}
int
DwtHaar1D::runCLKernels(void)
{
// Calculate thread-histograms
unsigned int levels = 0;
unsigned int curLevels = 0;
unsigned int actualLevels = 0;
int result = getLevels(signalLength, &levels);
CHECK_ERROR(result, SDK_SUCCESS, "getLevels() failed");
actualLevels = levels;
//max levels on device should be decided by kernelWorkGroupSize
int tempVar = (int)(log((float)kernelInfo.kernelWorkGroupSize) / log((float)2));
maxLevelsOnDevice = tempVar + 1;
cl_float* temp = (cl_float*)malloc(signalLength * sizeof(cl_float));
memcpy(temp, inData, signalLength * sizeof(cl_float));
levelsDone = 0;
int one = 1;
while((unsigned int)levelsDone < actualLevels)
{
curLevels = (levels < maxLevelsOnDevice) ? levels : maxLevelsOnDevice;
// Set the signal length for current iteration
if(levelsDone == 0)
curSignalLength = signalLength;
else
curSignalLength = (one << levels);
// Set group size
groupSize = (1 << curLevels) / 2;
totalLevels = levels;
runDwtHaar1DKernel();
if(levels <= maxLevelsOnDevice)
{
dOutData[0] = dPartialOutData[0];
memcpy(hOutData, dOutData, (one << curLevels) * sizeof(cl_float));
memcpy(dOutData + (one << curLevels), hOutData + (one << curLevels), (signalLength - (one << curLevels)) * sizeof(cl_float));
break;
}
else
{
levels -= maxLevelsOnDevice;
memcpy(hOutData, dOutData, curSignalLength * sizeof(cl_float));
memcpy(inData, dPartialOutData, (one << levels) * sizeof(cl_float));
levelsDone += (int)maxLevelsOnDevice;
}
}
memcpy(inData, temp, signalLength * sizeof(cl_float));
free(temp);
return SDK_SUCCESS;
}
////////////////////////////////////////////////////////////////////////////////
//! Perform the wavelet decomposition
////////////////////////////////////////////////////////////////////////////////
void runTest( int argc, char** argv)
{
char* s_fname ;
char* r_fname ;
char* r_gold_fname ;
const char usage[] =
{
"\nUsage:\n"
" dwtHaar1D --signal=<signal_file> --result=<result_file> --gold=<gold_file>\n\n"
" <signal_file> Input file containing the signal\n"
" <result_file> Output file storing the result of the wavelet decomposition\n"
" <gold_file> Input file containing the reference result of the wavelet decomposition\n"
"\nExample:\n"
" bin\\win32\\release\\dwtHaar1D\n"
" --signal=projects\\dwtHaar1D\\data\\signal.dat\n"
" --result=projects\\dwtHaar1D\\data\\regression.dat\n"
" --gold=projects\\dwtHaar1D\\data\\regression.gold.dat\n"
};
char s_fname_arr[] = "data/signal_2_18.dat";
char r_gold_fname_arr[] = "data/regression_2_18.gold.dat";
char r_fname_arr[] = "regression.dat";
s_fname = s_fname_arr;
r_fname = r_fname_arr;
r_gold_fname = r_gold_fname_arr;
// read in signal
unsigned int slength = 262144;
DATATYPE* signal = NULL;
if (s_fname == 0)
{
fprintf(stderr, "Cannot find the file containing the signal.\n%s", usage);
exit(1);
}
if (readFile(s_fname, &signal) == 1) {
printf("Reading signal from %s\n", s_fname);
} else {
exit(1);
}
// get the number of decompositions necessary to perform a full decomposition
unsigned int dlevels_complete = 0;
if (1 != getLevels( slength, &dlevels_complete))
{
// error message
fprintf( stderr, "Signal length not supported.\n");
return;
}
// device in data
DATATYPE* d_idata = NULL;
// device out data
DATATYPE* d_odata = NULL;
// device approx_final data
DATATYPE* approx_final = NULL;
// The very final approximation coefficient has to be written to the output
// data, all others are reused as input data in the next global step and
// therefore have to be written to the input data again.
// The following flag indicates where to copy approx_final data
// - 0 is input, 1 is output
int approx_is_input;
// allocate device mem
const unsigned int smem_size = sizeof(DATATYPE) * slength;
d_idata = (DATATYPE*) malloc(smem_size);
d_odata = (DATATYPE*) malloc(smem_size);
approx_final = (DATATYPE*) malloc(smem_size);
memcpy(d_idata, signal, smem_size);
// clear result memory
DATATYPE* tmp = (DATATYPE*) malloc( smem_size);
int i;
for (i = 0; i < slength; ++i)
tmp[i] = 0.0;
memcpy(d_odata, tmp, smem_size);
free( tmp);
// total number of threads
// in the first decomposition step always one thread computes the average and
// detail signal for one pair of adjacent values
unsigned int num_threads_total_left = slength / 2;
// decomposition levels performed in the current / next step
unsigned int dlevels_step = dlevels_complete;
// 1D signal so the arrangement of elements is also 1D
dim3 block_size;
dim3 grid_size;
// number of decomposition levels left after one iteration on the device
unsigned int dlevels_left = dlevels_complete;
// if less or equal 1k elements, then the data can be processed in one block,
// this avoids the Wait-For-Idle (WFI) on host side which is necessary if the
// computation is split accross multiple SM's if enough input data
if( dlevels_complete <= 10) {
// decomposition can be performed at once
block_size.x = num_threads_total_left;
approx_is_input = 0;
} else {
// 512 threads per block
grid_size.x = (num_threads_total_left / 512);
block_size.x = 512;
// 512 threads corresponds to 10 decomposition steps
dlevels_step = 10;
dlevels_left -= 10;
approx_is_input = 1;
}
grid_size.y = 1;
grid_size.z = 1;
block_size.y = 1;
block_size.z = 1;
// do until full decomposition is accomplished
while( 0 != num_threads_total_left) {
// run kernel
dwtHaar1D(d_idata, d_odata, approx_final, dlevels_step, num_threads_total_left, block_size.x, grid_size, block_size, 1, 0);
// Copy approx_final to appropriate location
if (approx_is_input) {
memcpy(d_idata, approx_final, grid_size.x*4);
} else {
memcpy(d_odata, approx_final, grid_size.x*4);
}
// update level variables
if( dlevels_left < 10) {
// approx_final = d_odata;
approx_is_input = 0;
}
// more global steps necessary
dlevels_step = (dlevels_left > 10) ? dlevels_left - 10 : dlevels_left;
dlevels_left -= 10;
// after each step only half the threads are used any longer
// therefore after 10 steps 2^10 less threads
num_threads_total_left = num_threads_total_left >> 10;
// update block and grid size
grid_size.x = (num_threads_total_left / 512)
+ (0 != (num_threads_total_left % 512)) ? 1 : 0;
if( grid_size.x <= 1) {
block_size.x = num_threads_total_left;
}
}
// load the reference solution
unsigned int len_reference = 262144;
DATATYPE* reference = NULL;
if (r_gold_fname == 0) {
fprintf(stderr, "Cannot read the file containing the reference result of the wavelet decomposition.\n%s", usage);
//cudaThreadExit();
exit(1);
}
if (readFile(r_gold_fname, &reference) == 1)
printf("Reading reference result from %s\n", r_gold_fname);
else {
exit(1);
}
assert(slength == len_reference);
//compare the computed solution and the reference
int res = compareData(reference, d_odata, slength, 0.001f);
printf("%s\n", (1 == res) ? "PASSED." : "FAILED.");
free(reference);
free(d_odata);
free(d_idata);
free(approx_final);
free(signal);
}
TreeNameParser::TreeNameParser(char *filename)//Normal.txt
{
ifstream inf(filename, ios::in);
// from the name, get the layer number
QString fName(filename);
fName.remove(".out");
int idx = fName.lastIndexOf("_");
fName = fName.mid(idx+1,fName.size()-idx);
int layerNum=fName.toUInt();
//int bufsize = BUFSIZE * 3;
char lineBuf[BUFSIZE3], *token, sToken[BUFSIZE3];
char delim[]="."; // delimited by space
int iToken;
for(int i=0; i<MAX_LEVEL; i++) resetParCount(i);
while(inf)
{
inf.getline(lineBuf, BUFSIZE3);
for(int j = 0; j < BUFSIZE3; ++j)sToken[j] = '\0';
// separate the index and string token
sscanf(lineBuf, "%d %s", &iToken, sToken);
// save the length
int lev= getLevels(sToken), lc=0, wc=0;
// lc: level counter ; wc: word counter
_lev.push_back(lev); // now everything is 3
if(lev==1) break; // feature. TODO:..last line..
if(lev==2)
{
char temp[BUFSIZE3];
int i = 0;
for(int j = 0; j < BUFSIZE3; ++j)temp[j] = '\0';
for(; sToken[i] != '.'; ++i)
temp[i] = sToken[i];
temp[i] = sToken[i];
++i;
for(int j = 0; sToken[j] != '\0'; ++j, ++i)
temp[i] = sToken[j];
strcpy(sToken, temp);
lev++;
}
// split by "."
_tmpToken.clear();
token = strtok(sToken, delim);
if(lev==4)
{
// get rid of the first level
}
if(lev == layerNum)
{
_tmpToken.push_back(token); // keep what it is
pushbackLp(lc++, token, _tmpToken); // save to _lp1
}
while( (token = strtok(NULL, delim))!=NULL )
{
_tmpToken.push_back(token);
pushbackLp(lc++, token, _tmpToken); // save to _lp2, _lp3 and _lp4
};
_names.push_back(_tmpToken); // save the last one
}; // end while(inf)
int lcc= GetMaxLevel();
for(int i=0; i<lcc; i++)
{
tmp_lpc[i].push_back(parCount[i]); // save the last one
}
inf.close();
}
// -----------------------------------------------------------------------------
//
// -----------------------------------------------------------------------------
void MultiOtsuThreshold::setupFilterParameters()
{
FilterParameterVector parameters;
parameters.push_back(FilterParameter::New("Array to Process", "SelectedCellArrayPath", FilterParameterWidgetType::DataArraySelectionWidget, getSelectedCellArrayPath(), false, ""));
QStringList linkedProps;
linkedProps << "NewCellArrayName";
parameters.push_back(LinkedBooleanFilterParameter::New("Save As New Array", "SaveAsNewArray", getSaveAsNewArray(), linkedProps, false));
parameters.push_back(FilterParameter::New("Created Array Name", "NewCellArrayName", FilterParameterWidgetType::StringWidget, getNewCellArrayName(), false, ""));
parameters.push_back(FilterParameter::New("Slice at a Time", "Slice", FilterParameterWidgetType::BooleanWidget, getSlice(), false));
parameters.push_back(FilterParameter::New("Number of Levels", "Levels", FilterParameterWidgetType::IntWidget, getLevels(), false, ""));
setFilterParameters(parameters);
}
unsigned int *
ControlFile::getLevels(const char *name)
{
// these comments are all wrong...
// See if there already is some info for name. If not,
// lock the file, stat it, remap if necessary, scan again,
// and if still not there, append to it. Return pointer to mapped area.
// ### call getComponent() instead,
// ### if it does not exist, create it and return the thing
// ### get default value from pre-existing levels, or from
// ### the default level if nothing exists.
// ### (if default level does not exist, create it?)
_fileBacking.lock(_mode != READONLY);
static const char *padSpaces = " "; // 3 spaces
char buf[2000];
if (strcmp(name, "") == 0) {
name = "default";
}
// Leave space for 3 spaces and a full levels string (200 bytes should
// be enough)
snprintf(buf, sizeof buf - 200, "\n%s: ", name);
char *levels = strstr(_mapBase, buf);
if (levels) {
_fileBacking.unlock();
char *addr = levels + strlen(buf);
addr = alignLevels(addr);
return reinterpret_cast<unsigned int *>(addr);
}
char *inheritLevels = reinterpret_cast<char *>(defaultLevels());
const char *chop = strrchr(name, '.');
if (chop != NULL) {
char shorterName[2000];
strncpy(shorterName, name, chop - name);
shorterName[chop-name] = '\0';
unsigned int *inherit = getLevels(shorterName);
if (inherit != NULL) {
inheritLevels = reinterpret_cast<char *>(inherit);
}
}
// Append whatever is in buf, excluding the initial newline, and
// up to 3 more spaces to get the entire file length to be aligned.
int fileLength = _fileBacking.size();
char *appendedString = buf + 1;
int newLength = fileLength + strlen(appendedString);
unsigned int padding = static_cast<unsigned int>(-newLength) & 3u;
strcat(appendedString, &padSpaces[3 - padding]);
char *baseAddr = _mapBase + fileLength + strlen(appendedString);
strncat(appendedString, inheritLevels, Logger::NUM_LOGLEVELS*sizeof(int));
strcat(appendedString, "\n");
int len = strlen(appendedString);
int fd = _fileBacking.fd();
lseek(fd, (off_t)fileLength, SEEK_SET);
int wlen = write(_fileBacking.fd(), appendedString, len);
if (wlen != len) {
_fileBacking.unlock();
LOG(error, "Writing to control file '%s' fails (%d/%d bytes): %s",
_fileName, wlen, len, strerror(errno));
return reinterpret_cast<unsigned int *>(inheritLevels);
} else {
_fileSize += wlen;
}
if (fileLength + wlen > _mappedSize) {
if (!extendMapping()) {
_fileBacking.unlock(); // just for sure
LOG(error, "Failed to extend mapping of '%s', losing runtime "
"configurability of component '%s'", _fileName, name);
return defaultLevels();
}
}
_fileBacking.unlock();
return reinterpret_cast<unsigned int *>(baseAddr);
}
void PhysicsHashSpace::initHashSpace()
{
setLevels(getLevels());
initSpace();
}
