void LaserScanMatcher::scanCallback (const sensor_msgs::LaserScan::ConstPtr& scan_msg)
{
// **** if first scan, cache the tf from base to the scanner
if (!initialized_)
{
createCache(scan_msg); // caches the sin and cos of all angles
// cache the static tf from base to laser
if (!getBaseToLaserTf(scan_msg->header.frame_id))
{
ROS_INFO("FRAME_ID : %s",scan_msg->header.frame_id.c_str());
ROS_WARN("Skipping laser scan");
return;
}
laserScanToLDP(scan_msg, prev_ldp_scan_);
last_icp_time_ = scan_msg->header.stamp;
initialized_ = true;
}
LDP curr_ldp_scan;
laserScanToLDP(scan_msg, curr_ldp_scan);
processScan(curr_ldp_scan, scan_msg->header.stamp);
}
/* plugin methods */
int
postgresInitMetadataLookupPlugin (void)
{
PostgresMetaCache *psqlCache;
if (plugin && plugin->initialized)
{
INFO_LOG("tried to initialize metadata lookup plugin more than once");
return EXIT_SUCCESS;
}
NEW_AND_ACQUIRE_MEMCONTEXT(CONTEXT_NAME);
memContext = getCurMemContext();
// create cache
plugin->plugin.cache = createCache();
// create postgres specific part of the cache
psqlCache = NEW(PostgresMetaCache);
psqlCache->oidToDT = NEW_MAP(Constant,Constant);
plugin->plugin.cache->cacheHook = (void *) psqlCache;
plugin->initialized = TRUE;
RELEASE_MEM_CONTEXT();
return EXIT_SUCCESS;
}
void insert(std::pair<std::string, _Type*>&& item)
{
createCache();
if (!_mutex) _mutex = new FastMutex();
FastMutex::ScopedLock _(*_mutex);
_instance->insert(std::move(item));
}
void createSearch()
{
char* cachesize = "0KB";
/* Create a tokenizer */
tok = TKCreate(FILE_CHARS, "myindex.txt");
if(tok == NULL)
{
fprintf(stderr, "Error: Could not allocate space for Tokenizer.\n");
return;
}
/* Get the file list */
files = getFilelist(tok);
if(files == NULL)
{
return;
}
/* Create a cache */
cache = createCache(cachesize);
if(cache == NULL)
{
fprintf(stderr, "Error: Could not allocate space for Cache.\n");
return;
}
/* Update the allowed characters */
adjustAllowedChars(tok, STRING_CHARS);
}
void QNetworkReplyImplPrivate::setCachingEnabled(bool enable)
{
if (!enable && !cacheEnabled)
return; // nothing to do
if (enable && cacheEnabled)
return; // nothing to do either!
if (enable) {
if (bytesDownloaded) {
// refuse to enable in this case
qCritical("QNetworkReplyImpl: backend error: caching was enabled after some bytes had been written");
return;
}
createCache();
} else {
// someone told us to turn on, then back off?
// ok... but you should make up your mind
qDebug("QNetworkReplyImpl: setCachingEnabled(true) called after setCachingEnabled(false) -- "
"backend %s probably needs to be fixed",
backend->metaObject()->className());
networkCache()->remove(url);
cacheSaveDevice = 0;
cacheEnabled = false;
}
}
void insert(std::string const& key, _Type* const& value)
{
createCache();
std::pair<std::string, _Type*> item(key, value);
if (!_mutex) _mutex = new FastMutex();
FastMutex::ScopedLock _(*_mutex);
_instance->insert(item);
}
void insert(std::string&& key, _Type*&& value)
{
createCache();
std::pair<std::string, _Type*> item(std::move(key), std::move(value));
value = NULL;
if (!_mutex) _mutex = new FastMutex();
FastMutex::ScopedLock _(*_mutex);
_instance->insert(item);
}
void insert(_Type* const& value)
{
createCache();
char buff[20] = { 0 };
sprintf(buff, "%lx", (unsigned long)value);
std::pair<std::string, _Type*> item(buff, value);
if (!_mutex) _mutex = new FastMutex();
FastMutex::ScopedLock _(*_mutex);
_instance->insert(item);
}
/*---------------------------------------------------------------------*//**
コンストラクタ
**//*---------------------------------------------------------------------*/
FileAndroidLargeAsset::FileAndroidLargeAsset(FILE_MNG_R* astmng)
: File()
, _astmng(astmng)
, _pathFileOrg(new VcString())
, _oaarr(0L)
, _offset(0)
, _numCache(NUM_OPENED_CACHE_DEFAULT)
, _isOpened(false)
{
createCache();
}
CacheFixture() : objWritten_(0), currSize_(0)
{
std::vector< uint8_t > key;
const std::string dummykey( "dummykey" );
std::copy( dummykey.begin(), dummykey.end(), back_inserter( key ) );
BOOST_TEST_MESSAGE( "Creating Cache instance" );
cache_ = createCache( "c:\\temp\\cache", key );;
BOOST_REQUIRE( cache_ );
cache_->setMaxSize( maxSize );
}
/*---------------------------------------------------------------------*//**
コンストラクタ
**//*---------------------------------------------------------------------*/
FileAndroidLargeAsset::FileAndroidLargeAsset(FILE_MNG_R* astmng, s32 numCache)
: File()
, _astmng(astmng)
, _pathFileOrg(new VcString())
, _oaarr(0L)
, _offset(0)
, _numCache(numCache)
, _isOpened(false)
{
createCache();
}
std::auto_ptr<Interfaces::IAggregate<StringTypeMap::key_type> > getKeyAggregate()
{
createCache();
if (!_mutex) _mutex = new FastMutex();
FastMutex::ScopedLock _(*_mutex);
typedef Patterns::StdContainerAggregate<StringTypeMap*,
StringTypeMap::key_type,
StringTypeMap::const_iterator,
KeyConverter,
KeyRange> result_t;
return std::auto_ptr<Interfaces::IAggregate<StringTypeMap::key_type> >(new result_t(_instance));
}
void cacheTesting(){
Cache* cache = createCache(10, 4);
int cacheLineNum;
bool compute;
cacheCall(cache, 9, &cacheLineNum, &compute);
cacheCall(cache, 0, &cacheLineNum, &compute);
cacheCall(cache, 2, &cacheLineNum, &compute);
cacheCall(cache, 2, &cacheLineNum, &compute);
cacheCall(cache, 1, &cacheLineNum, &compute);
cacheCall(cache, 9, &cacheLineNum, &compute);
cacheCall(cache, 3, &cacheLineNum, &compute);
printf("%d%d\n",cacheLineNum,compute);
}
int initKMalloc(void)
{
for(int i = 0;i < sizeof(mallocSizes)/sizeof(MallocSize);++i)
{
mallocSizes[i].cache = createCache(mallocSizes[i].size,0x0);
if(unlikely(!mallocSizes[i].cache))
{
printkInColor(0xff,0x00,0x00,"(%s) Can't get memorySize[%d].cache!",__func__,i);
return -ENOMEM;
}
}
return 0;
}
/*---------------------------------------------------------------------*//**
コンストラクタ
**//*---------------------------------------------------------------------*/
FileAndroidLargeAsset::FileAndroidLargeAsset(FILE_MNG_R* astmng, CcString pathFile, Mode mode, s32 numCache)
: File(pathFile, mode)
, _astmng(astmng)
, _pathFileOrg(new VcString())
, _oaarr(0L)
, _offset(0)
, _numCache(numCache)
, _isOpened(false)
{
if(createCache())
{
open(&pathFile, mode);
}
}
MojErr MojDbSearchCache::updateCache(const QueryKey& a_key, const IdSet& a_ids)
{
// Skip updating the cache ONLY if up-to-date cache is available.
//
if (contain(a_key) == true)
return MojErrNone;
// Delete any cache for this query/kind here.
//
MojErr err = destroyCache(a_key.getKind());
MojErrCheck(err);
// Add new cache.
//
err = createCache(a_key, a_ids);
MojErrCheck(err);
return MojErrNone;
}
std::unique_ptr<Interfaces::IAggregate<typename StringTypeMap::key_type> > getKeyAggregate()
#else
std::auto_ptr<Interfaces::IAggregate<typename StringTypeMap::key_type> > getKeyAggregate()
#endif
{
createCache();
if (!_mutex) _mutex = new FastMutex();
FastMutex::ScopedLock _(*_mutex);
typedef Patterns::StdContainerAggregate<StringTypeMap*,
typename StringTypeMap::key_type,
typename StringTypeMap::const_iterator,
KeyConverter,
KeyRange> result_t;
#if defined(PUREMVC_USES_TR1)
return std::unique_ptr<Interfaces::IAggregate<typename StringTypeMap::key_type> >(new result_t(_instance));
#else
return std::auto_ptr<Interfaces::IAggregate<typename StringTypeMap::key_type> >(new result_t(_instance));
#endif
}
void QgsValueRelationWidgetWrapper::initWidget( QWidget* editor )
{
mCache = createCache( config() );
mComboBox = qobject_cast<QComboBox*>( editor );
mListWidget = qobject_cast<QListWidget*>( editor );
mLineEdit = qobject_cast<QLineEdit*>( editor );
if ( mComboBox )
{
if ( config( "AllowNull" ).toBool() )
{
mComboBox->addItem( tr( "(no selection)" ), QVariant( field().type() ) );
}
Q_FOREACH ( const ValueRelationItem& element, mCache )
{
mComboBox->addItem( element.second, element.first );
}
connect( mComboBox, SIGNAL( currentIndexChanged( int ) ), this, SLOT( valueChanged() ) );
}
void QTextureGlyphCache::fillInPendingGlyphs()
{
if (m_pendingGlyphs.isEmpty())
return;
int requiredHeight = m_h;
int requiredWidth = m_w; // Use a minimum size to avoid a lot of initial reallocations
{
QHash<GlyphAndSubPixelPosition, Coord>::iterator iter = m_pendingGlyphs.begin();
while (iter != m_pendingGlyphs.end()) {
Coord c = iter.value();
requiredHeight = qMax(requiredHeight, c.y + c.h);
requiredWidth = qMax(requiredWidth, c.x + c.w);
++iter;
}
}
if (isNull() || requiredHeight > m_h || requiredWidth > m_w) {
if (isNull())
createCache(qt_next_power_of_two(requiredWidth), qt_next_power_of_two(requiredHeight));
else
resizeCache(qt_next_power_of_two(requiredWidth), qt_next_power_of_two(requiredHeight));
}
{
QHash<GlyphAndSubPixelPosition, Coord>::iterator iter = m_pendingGlyphs.begin();
while (iter != m_pendingGlyphs.end()) {
GlyphAndSubPixelPosition key = iter.key();
fillTexture(iter.value(), key.glyph, key.subPixelPosition);
++iter;
}
}
m_pendingGlyphs.clear();
}
CacheManager::CacheManager()
{
disabled=true;
cache=createCache();
}
void CacheManager::flushCaches()
{
delete cache;
cache=createCache();
}
CacheManager::CacheManager() :
disabled(true)
{
cache=createCache();
}
void HistogramFilter::createCellHistograms(const Mat& image, Mat& histograms, int binCount, int rowCount, int columnCount, bool interpolate) const {
if (image.channels() != 1 && image.channels() != 2 && image.channels() != 4)
throw runtime_error("HistogramFilter: image must have one, two or four channels");
if (image.depth() != CV_8U)
throw runtime_error("HistogramFilter: image must have a depth of CV_8U");
histograms = Mat::zeros(rowCount, columnCount, CV_32FC(binCount));
float factor = 1.f / 255.f;
if (interpolate) { // bilinear interpolation between cells
createCache(rowCache, image.rows, rowCount);
createCache(colCache, image.cols, columnCount);
if (image.channels() == 1) { // bin information only, no weights
for (int imageRow = 0; imageRow < image.rows; ++imageRow) {
const uchar* rowValues = image.ptr<uchar>(imageRow);
int rowIndex0 = rowCache[imageRow].index1;
int rowIndex1 = rowCache[imageRow].index2;
float rowWeight1 = rowCache[imageRow].weight2;
float rowWeight0 = rowCache[imageRow].weight1;
for (int imageCol = 0; imageCol < image.cols; ++imageCol) {
uchar bin = rowValues[imageCol];
int colIndex0 = colCache[imageCol].index1;
int colIndex1 = colCache[imageCol].index2;
float colWeight1 = colCache[imageCol].weight2;
float colWeight0 = colCache[imageCol].weight1;
if (rowIndex0 >= 0 && colIndex0 >= 0) {
float* histogramValues = histograms.ptr<float>(rowIndex0, colIndex0);
histogramValues[bin] += rowWeight0 * colWeight0;
}
if (rowIndex0 >= 0 && colIndex1 < columnCount) {
float* histogramValues = histograms.ptr<float>(rowIndex0, colIndex1);
histogramValues[bin] += rowWeight0 * colWeight1;
}
if (rowIndex1 < rowCount && colIndex0 >= 0) {
float* histogramValues = histograms.ptr<float>(rowIndex1, colIndex0);
histogramValues[bin] += rowWeight1 * colWeight0;
}
if (rowIndex1 < rowCount && colIndex1 < columnCount) {
float* histogramValues = histograms.ptr<float>(rowIndex1, colIndex1);
histogramValues[bin] += rowWeight1 * colWeight1;
}
}
}
} else if (image.channels() == 2) { // bin index and weight available
for (int imageRow = 0; imageRow < image.rows; ++imageRow) {
const Vec2b* rowValues = image.ptr<Vec2b>(imageRow);
int rowIndex0 = rowCache[imageRow].index1;
int rowIndex1 = rowCache[imageRow].index2;
float rowWeight1 = rowCache[imageRow].weight2;
float rowWeight0 = rowCache[imageRow].weight1;
for (int imageCol = 0; imageCol < image.cols; ++imageCol) {
uchar bin = rowValues[imageCol][0];
float weight = factor * rowValues[imageCol][1];
int colIndex0 = colCache[imageCol].index1;
int colIndex1 = colCache[imageCol].index2;
float colWeight1 = colCache[imageCol].weight2;
float colWeight0 = colCache[imageCol].weight1;
if (rowIndex0 >= 0 && colIndex0 >= 0) {
float* histogramValues = histograms.ptr<float>(rowIndex0, colIndex0);
histogramValues[bin] += weight * rowWeight0 * colWeight0;
}
if (rowIndex0 >= 0 && colIndex1 < columnCount) {
float* histogramValues = histograms.ptr<float>(rowIndex0, colIndex1);
histogramValues[bin] += weight * rowWeight0 * colWeight1;
}
if (rowIndex1 < rowCount && colIndex0 >= 0) {
float* histogramValues = histograms.ptr<float>(rowIndex1, colIndex0);
histogramValues[bin] += weight * rowWeight1 * colWeight0;
}
if (rowIndex1 < rowCount && colIndex1 < columnCount) {
float* histogramValues = histograms.ptr<float>(rowIndex1, colIndex1);
histogramValues[bin] += weight * rowWeight1 * colWeight1;
}
}
}
} else if (image.channels() == 4) { // two bin indices and weights available
for (int imageRow = 0; imageRow < image.rows; ++imageRow) {
const Vec4b* rowValues = image.ptr<Vec4b>(imageRow);
int rowIndex0 = rowCache[imageRow].index1;
int rowIndex1 = rowCache[imageRow].index2;
float rowWeight1 = rowCache[imageRow].weight2;
float rowWeight0 = rowCache[imageRow].weight1;
for (int imageCol = 0; imageCol < image.cols; ++imageCol) {
uchar bin1 = rowValues[imageCol][0];
float weight1 = factor * rowValues[imageCol][1];
uchar bin2 = rowValues[imageCol][2];
float weight2 = factor * rowValues[imageCol][3];
int colIndex0 = colCache[imageCol].index1;
int colIndex1 = colCache[imageCol].index2;
float colWeight1 = colCache[imageCol].weight2;
float colWeight0 = colCache[imageCol].weight1;
if (rowIndex0 >= 0 && colIndex0 >= 0) {
float* histogramValues = histograms.ptr<float>(rowIndex0, colIndex0);
histogramValues[bin1] += weight1 * rowWeight0 * colWeight0;
histogramValues[bin2] += weight2 * rowWeight0 * colWeight0;
}
if (rowIndex0 >= 0 && colIndex1 < columnCount) {
float* histogramValues = histograms.ptr<float>(rowIndex0, colIndex1);
histogramValues[bin1] += weight1 * rowWeight0 * colWeight1;
histogramValues[bin2] += weight2 * rowWeight0 * colWeight1;
}
if (rowIndex1 < rowCount && colIndex0 >= 0) {
float* histogramValues = histograms.ptr<float>(rowIndex1, colIndex0);
histogramValues[bin1] += weight1 * rowWeight1 * colWeight0;
histogramValues[bin2] += weight2 * rowWeight1 * colWeight0;
}
if (rowIndex1 < rowCount && colIndex1 < columnCount) {
float* histogramValues = histograms.ptr<float>(rowIndex1, colIndex1);
histogramValues[bin1] += weight1 * rowWeight1 * colWeight1;
histogramValues[bin2] += weight2 * rowWeight1 * colWeight1;
}
}
}
}
} else { // no bilinear interpolation between cells
float* histogramValues = histograms.ptr<float>();
if (image.channels() == 1) { // bin information only, no weights
for (int cellRow = 0; cellRow < rowCount; ++cellRow) {
for (int cellCol = 0; cellCol < columnCount; ++cellCol) {
int startRow = (cellRow * image.rows) / rowCount;
int startCol = (cellCol * image.cols) / columnCount;
int endRow = ((cellRow + 1) * image.rows) / rowCount;
int endCol = ((cellCol + 1) * image.cols) / columnCount;
for (int imageRow = startRow; imageRow < endRow; ++imageRow) {
const uchar* rowValues = image.ptr<uchar>(imageRow);
for (int imageCol = startCol; imageCol < endCol; ++imageCol)
histogramValues[rowValues[imageCol]]++;
}
histogramValues += binCount;
}
}
} else if (image.channels() == 2) { // bin index and weight available
for (int cellRow = 0; cellRow < rowCount; ++cellRow) {
for (int cellCol = 0; cellCol < columnCount; ++cellCol) {
int startRow = (cellRow * image.rows) / rowCount;
int startCol = (cellCol * image.cols) / columnCount;
int endRow = ((cellRow + 1) * image.rows) / rowCount;
int endCol = ((cellCol + 1) * image.cols) / columnCount;
for (int imageRow = startRow; imageRow < endRow; ++imageRow) {
const Vec2b* rowValues = image.ptr<Vec2b>(imageRow);
for (int imageCol = startCol; imageCol < endCol; ++imageCol) {
uchar bin = rowValues[imageCol][0];
uchar weight = rowValues[imageCol][1];
histogramValues[bin] += factor * weight;
}
}
histogramValues += binCount;
}
}
} else if (image.channels() == 4) { // two bin indices and weights available
for (int cellRow = 0; cellRow < rowCount; ++cellRow) {
for (int cellCol = 0; cellCol < columnCount; ++cellCol) {
int startRow = (cellRow * image.rows) / rowCount;
int startCol = (cellCol * image.cols) / columnCount;
int endRow = ((cellRow + 1) * image.rows) / rowCount;
int endCol = ((cellCol + 1) * image.cols) / columnCount;
for (int imageRow = startRow; imageRow < endRow; ++imageRow) {
const Vec4b* rowValues = image.ptr<Vec4b>(imageRow);
for (int imageCol = startCol; imageCol < endCol; ++imageCol) {
uchar bin1 = rowValues[imageCol][0];
uchar weight1 = rowValues[imageCol][1];
uchar bin2 = rowValues[imageCol][2];
uchar weight2 = rowValues[imageCol][3];
histogramValues[bin1] += factor * weight1;
histogramValues[bin2] += factor * weight2;
}
}
histogramValues += binCount;
}
}
}
}
}
int DcmFileProcess::readAllDcm(const char* FilePath, std::vector<float>& position)
{
//构造类对象
CStatDir statdir;
std::vector<std::string> AllDcmFile;
position.clear();
float min_axial = 1000.0f;
float max_axial = -1000.0f;
float init_x = 0.0f;
float init_y = 0.0f;
float pixelspacing = 0.0f;
int count = 0;
//设置要遍历的目录
if (!statdir.SetInitDir(FilePath))
{
puts("目录不存在");
}
//开始遍历,获取该文件夹下所有dcm格式图像，一般为一个病患
AllDcmFile.clear();
AllDcmFile = statdir.BeginBrowseFilenames("*.dcm");
//创建缓冲文件夹
createCache();
std::string dirName = GetExePath();
dirName += "\\cache\\";
cv::Mat mat[710];
// 获取原始dcm图像的前缀目录
std::string subImageName = dirName;
// 获取原始dcm图像的前缀目录的string转换为char*
const char *chSubImageName = subImageName.c_str();
for (auto iter = AllDcmFile.cbegin(); iter != AllDcmFile.cend(); iter++)
{
// 将原始dcm图像的string转换为char*
const char *chImageName = (*iter).c_str();
// 基于dcmtk实现类
THU_STD_NAMESPACE::TDcmFileFormat dcm = THU_STD_NAMESPACE::TDcmFileFormat(chImageName);
std::string result = dcm.getImagePositionPatient();
std::vector<std::string> ImagePosition;
split(result, "\\", ImagePosition);
float axial = (float)atof(ImagePosition[2].c_str());
if (axial > max_axial) max_axial = axial;
if (axial < min_axial) min_axial = axial;
// 得到世界坐标的初始值
if (count < 1)
{
init_x = (float)atof(ImagePosition[0].c_str());
init_y = (float)atof(ImagePosition[1].c_str());
// 获取dcm图像的PixelSpacing
std::string space = dcm.getPixelSpacing();
std::vector<std::string> PixelSpacing;
split(space, "\\", PixelSpacing);
pixelspacing = (float)atof(PixelSpacing[0].c_str());
}
// 获取dcm图像的InstanceNumber
int InstancePosition = dcm.getPositionNumber();
// 生成bmp图像存储路径
char BmpName[256];
sprintf_s(BmpName, "%s%06d.bmp", chSubImageName, InstancePosition);
// 进行图像转换
dcm.setWindow(715, 3478);
dcm.saveToBmp(BmpName);
//读取所有dcm图像构造三维数组
cv::Mat temp = cv::imread(BmpName, cv::IMREAD_GRAYSCALE);
mat[InstancePosition - 1] = temp;
count++;
}
position.push_back(max_axial);
position.push_back(min_axial);
position.push_back((max_axial - min_axial) / count);
position.push_back(init_x);
position.push_back(init_y);
position.push_back(pixelspacing);
//多线程需要处理的线程数，分别对应矢状位与冠状位
const int THREAD_NUM = 2;
HANDLE handle[THREAD_NUM];
//构造子线程参数结构体
ThreadInfo threadInfo;
threadInfo.dirName = dirName;
threadInfo.count = count;
for (int i = 0; i < count; i++)
threadInfo.mat[i] = mat[i];
//Coronal子线程
handle[0] = (HANDLE)_beginthreadex(NULL, 0, ThreadCoronal, &threadInfo, 0, NULL);
//Sigattal子线程
handle[1] = (HANDLE)_beginthreadex(NULL, 0, ThreadSagittal, &threadInfo, 0, NULL);
WaitForMultipleObjects(THREAD_NUM, handle, TRUE, INFINITE);
return count;
}
int train(int cacheSizeInFloats, float *input_data, int *labels,
float epsilon, float Ce, float cost, float tolerance,
int heuristic, int nPoints, int dFeatures,
float paramA, float paramB, float paramC,
size_t localInitSize, size_t globalInitSize,
int numGroups_foph1, size_t localFoph1Size, size_t globalFoph1Size,
size_t localFoph2Size, size_t globalFoph2Size,
int numGroups_soph1, size_t localSoph1Size, size_t globalSoph1Size,
size_t localSoph2Size, size_t globalSoph2Size,
int numGroups_soph3, size_t localSoph3Size, size_t globalSoph3Size,
size_t localSoph4Size, size_t globalSoph4Size,
cl_mem d_input_data, cl_mem d_input_data_colmajor, cl_mem d_labels, cl_mem d_trainingAlpha, cl_mem d_kernelDiag, cl_mem d_F,
cl_mem d_highFsFO, cl_mem d_highIndicesFO, cl_mem d_lowFsFO, cl_mem d_lowIndicesFO,
cl_mem d_highFsSO1, cl_mem d_highIndicesSO1, cl_mem d_lowFsSO3, cl_mem d_lowIndicesSO3, cl_mem d_deltaFsSO3,
cl_mem d_results, cl_mem d_cache,
cl_command_queue queue, cl_kernel init, cl_kernel step1, cl_kernel foph1, cl_kernel foph2,
cl_kernel soph1, cl_kernel soph2, cl_kernel soph3, cl_kernel soph4,
float *p_rho, int *p_nSV, int *p_iterations,
float **p_signedAlpha, float **p_supportVectors){
int numCacheLines = cacheSizeInFloats/nPoints;
Cache *cache = createCache(nPoints,numCacheLines);
Controller *progress = createController(2.0, 64 ,nPoints);
int err;
err = 0;
err |= clSetKernelArg(init, 0, sizeof(cl_mem), &d_input_data);
err |= clSetKernelArg(init, 1, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(init, 2, sizeof(cl_mem), &d_F);
err |= clSetKernelArg(init, 3, sizeof(cl_mem), &d_kernelDiag);
err |= clSetKernelArg(init, 4, sizeof(int), &nPoints);
err |= clSetKernelArg(init, 5, sizeof(int), &dFeatures);
err |= clSetKernelArg(init, 6, sizeof(float), ¶mA);
err |= clSetKernelArg(init, 7, sizeof(float), ¶mB);
err |= clSetKernelArg(init, 8, sizeof(float), ¶mC);
if (err != CL_SUCCESS){
printf("Error: Failed to set kernel arguments! %d\n", err);
return err;
}
err = clEnqueueNDRangeKernel(queue, init, 1, NULL, &globalInitSize, &localInitSize, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to execute kernel init!\n");
printf("Global Size:%zu, Local Size:%zu\n", globalInitSize, localInitSize);
return err;
}
float bLow;
float bHigh;
int iLow;
int iHigh;
float alpha1diff;
float alpha2diff;
size_t globalStep1Size = 1;
size_t localStep1Size = 1;
err = 0;
err |= clSetKernelArg(step1, 0, sizeof(cl_mem), &d_input_data);
err |= clSetKernelArg(step1, 1, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(step1, 2, sizeof(cl_mem), &d_trainingAlpha);
err |= clSetKernelArg(step1, 3, sizeof(cl_mem), &d_kernelDiag);
err |= clSetKernelArg(step1, 4, sizeof(float), &cost);
err |= clSetKernelArg(step1, 5, sizeof(int), &nPoints);
err |= clSetKernelArg(step1, 6, sizeof(int), &dFeatures);
err |= clSetKernelArg(step1, 7, sizeof(float), ¶mA);
err |= clSetKernelArg(step1, 8, sizeof(float), ¶mB);
err |= clSetKernelArg(step1, 9, sizeof(float), ¶mC);
err |= clSetKernelArg(step1, 10, sizeof(cl_mem), &d_results);
if (err != CL_SUCCESS){
printf("Error: Failed to set kernel arguments! %d\n", err);
return err;
}
err = clEnqueueNDRangeKernel(queue, step1, 1, NULL, &globalStep1Size, &localStep1Size, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to execute kernel step1!\n");
return err;
}
err = clFinish(queue);
if (err != CL_SUCCESS){
printf("Error: waiting for queue to finish failed\n");
return err;
}
float results[8];
getResults(queue, d_results, results, &bLow, &bHigh, &iLow, &iHigh, &alpha1diff, &alpha2diff);
int iteration;
bool iLowCompute;
bool iHighCompute;
int iLowCacheIndex;
int iHighCacheIndex;
int currentHeuristic;
// clock_t startTot = clock(), diffTot, diffSoph1, diffSoph2, diffSoph3, diffSoph4, diffGetResults;
for (iteration = 0; true; iteration++){
if(heuristic == 2){
currentHeuristic = progress->heuristic;
}else{
currentHeuristic = heuristic;
}
if(bLow <= bHigh + 2 * tolerance){
break;
}
if ((iteration & 0x7ff) == 0) {
printf("iteration: %d; gap: %f\n",iteration, bLow - bHigh);
}
cacheCall(cache,iHigh, &iHighCacheIndex, &iHighCompute);
cacheCall(cache,iLow, &iLowCacheIndex, &iLowCompute);
alpha1diff = labels[iHigh] * alpha1diff;
alpha2diff = labels[iLow] * alpha2diff;
if (currentHeuristic == 0){
// Set the arguments to foph1
//
err = 0;
err = clSetKernelArg(foph1, 0, sizeof(cl_mem), &d_input_data);
err |= clSetKernelArg(foph1, 1, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(foph1, 2, sizeof(cl_mem), &d_trainingAlpha);
err |= clSetKernelArg(foph1, 3, sizeof(cl_mem), &d_F);
err |= clSetKernelArg(foph1, 4, sizeof(cl_mem), &d_lowFsFO);
err |= clSetKernelArg(foph1, 5, sizeof(cl_mem), &d_highFsFO);
err |= clSetKernelArg(foph1, 6, sizeof(cl_mem), &d_lowIndicesFO);
err |= clSetKernelArg(foph1, 7, sizeof(cl_mem), &d_highIndicesFO);
err |= clSetKernelArg(foph1, 8, sizeof(int), &nPoints);
err |= clSetKernelArg(foph1, 9, sizeof(int), &dFeatures);
err |= clSetKernelArg(foph1, 10, sizeof(float), &epsilon);
err |= clSetKernelArg(foph1, 11, sizeof(float), &Ce);
err |= clSetKernelArg(foph1, 12, sizeof(int), &iHigh);
err |= clSetKernelArg(foph1, 13, sizeof(int), &iLow);
err |= clSetKernelArg(foph1, 14, sizeof(float), &alpha1diff);
err |= clSetKernelArg(foph1, 15, sizeof(float), &alpha2diff);
err |= clSetKernelArg(foph1, 16, sizeof(float), ¶mA);
err |= clSetKernelArg(foph1, 17, sizeof(float), ¶mB);
err |= clSetKernelArg(foph1, 18, sizeof(float), ¶mC);
err |= clSetKernelArg(foph1, 19, sizeof(int) * localFoph1Size, NULL);
err |= clSetKernelArg(foph1, 20, sizeof(float) * localFoph1Size, NULL);
err |= clSetKernelArg(foph1, 21, sizeof(cl_mem), &d_cache);
err |= clSetKernelArg(foph1, 22, sizeof(bool), &iHighCompute);
err |= clSetKernelArg(foph1, 23, sizeof(bool), &iLowCompute);
err |= clSetKernelArg(foph1, 24, sizeof(int), &iHighCacheIndex);
err |= clSetKernelArg(foph1, 25, sizeof(int),&iLowCacheIndex);
if (err != CL_SUCCESS){
printf("Error: Failed to set kernel arguments! %d\n", err);
return err;
}
err = clEnqueueNDRangeKernel(queue, foph1, 1, NULL, &globalFoph1Size, &localFoph1Size, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to execute kernel Foph1!\n");
return err;
}
// Set the arguments to foph2
//
err = 0;
err = clSetKernelArg(foph2, 0, sizeof(cl_mem), &d_input_data);
err |= clSetKernelArg(foph2, 1, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(foph2, 2, sizeof(cl_mem), &d_trainingAlpha);
err |= clSetKernelArg(foph2, 3, sizeof(cl_mem), &d_kernelDiag);
err |= clSetKernelArg(foph2, 4, sizeof(cl_mem), &d_lowFsFO);
err |= clSetKernelArg(foph2, 5, sizeof(cl_mem), &d_highFsFO);
err |= clSetKernelArg(foph2, 6, sizeof(cl_mem), &d_lowIndicesFO);
err |= clSetKernelArg(foph2, 7, sizeof(cl_mem), &d_highIndicesFO);
err |= clSetKernelArg(foph2, 8, sizeof(cl_mem), &d_results);
err |= clSetKernelArg(foph2, 9, sizeof(float), &cost);
err |= clSetKernelArg(foph2, 10, sizeof(int), &dFeatures);
err |= clSetKernelArg(foph2, 11, sizeof(int), &numGroups_foph1);
err |= clSetKernelArg(foph2, 12, sizeof(float), ¶mA);
err |= clSetKernelArg(foph2, 13, sizeof(float), ¶mB);
err |= clSetKernelArg(foph2, 14, sizeof(float), ¶mC);
err |= clSetKernelArg(foph2, 15, sizeof(int) * localFoph2Size, NULL);
err |= clSetKernelArg(foph2, 16, sizeof(float) * localFoph2Size, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to set kernel arguments! %d\n", err);
return err;
}
err = clEnqueueNDRangeKernel(queue, foph2, 1, NULL, &globalFoph2Size, &localFoph2Size, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to execute kernel Foph2!\n");
return err;
}
// Wait for the command queue to get serviced before reading back results
//
err = clFinish(queue);
if (err != CL_SUCCESS){
printf("Error: waiting for queue to finish failed\n");
return err;
}
getResults(queue, d_results, results, &bLow, &bHigh, &iLow, &iHigh, &alpha1diff, &alpha2diff);
}else{
// Set the arguments to soph1
//
// clock_t startSoph1 = clock();
err = 0;
err = clSetKernelArg(soph1, 0, sizeof(cl_mem), &d_input_data);
err |= clSetKernelArg(soph1, 1, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(soph1, 2, sizeof(cl_mem), &d_trainingAlpha);
err |= clSetKernelArg(soph1, 3, sizeof(cl_mem), &d_F);
err |= clSetKernelArg(soph1, 4, sizeof(cl_mem), &d_highFsSO1);
err |= clSetKernelArg(soph1, 5, sizeof(cl_mem), &d_highIndicesSO1);
err |= clSetKernelArg(soph1, 6, sizeof(int), &nPoints);
err |= clSetKernelArg(soph1, 7, sizeof(int), &dFeatures);
err |= clSetKernelArg(soph1, 8, sizeof(float), &epsilon);
err |= clSetKernelArg(soph1, 9, sizeof(float), &Ce);
err |= clSetKernelArg(soph1, 10, sizeof(int), &iHigh);
err |= clSetKernelArg(soph1, 11, sizeof(int), &iLow);
err |= clSetKernelArg(soph1, 12, sizeof(float), &alpha1diff);
err |= clSetKernelArg(soph1, 13, sizeof(float), &alpha2diff);
err |= clSetKernelArg(soph1, 14, sizeof(float), ¶mA);
err |= clSetKernelArg(soph1, 15, sizeof(float), ¶mB);
err |= clSetKernelArg(soph1, 16, sizeof(float), ¶mC);
err |= clSetKernelArg(soph1, 17, sizeof(int) * localSoph1Size, NULL);
err |= clSetKernelArg(soph1, 18, sizeof(float) * localSoph1Size, NULL);
err |= clSetKernelArg(soph1, 19, sizeof(cl_mem), &d_cache);
err |= clSetKernelArg(soph1, 20, sizeof(bool), &iHighCompute);
err |= clSetKernelArg(soph1, 21, sizeof(bool), &iLowCompute);
err |= clSetKernelArg(soph1, 22, sizeof(int), &iHighCacheIndex);
err |= clSetKernelArg(soph1, 23, sizeof(int),&iLowCacheIndex);
if (err != CL_SUCCESS){
printf("Error: Failed to set kernel arguments! %d\n", err);
return err;
}
err = clEnqueueNDRangeKernel(queue, soph1, 1, NULL, &globalSoph1Size, &localSoph1Size, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to execute kernel Soph1!\n");
return err;
}
// clFinish(queue);
// diffSoph1 = clock() - startSoph1;
// clock_t startSoph2 = clock();
// Set the arguments to soph2
//
err = 0;
err |= clSetKernelArg(soph2, 0, sizeof(cl_mem), &d_highFsSO1);
err |= clSetKernelArg(soph2, 1, sizeof(cl_mem), &d_highIndicesSO1);
err |= clSetKernelArg(soph2, 2, sizeof(cl_mem), &d_results);
err |= clSetKernelArg(soph2, 3, sizeof(int), &numGroups_soph1);
err |= clSetKernelArg(soph2, 4, sizeof(int) * localSoph2Size, NULL);
err |= clSetKernelArg(soph2, 5, sizeof(float) * localSoph2Size, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to set kernel arguments! %d\n", err);
return err;
}
err = clEnqueueNDRangeKernel(queue, soph2, 1, NULL, &globalSoph2Size, &localSoph2Size, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to execute kernel Soph2!\n");
return err;
}
// clFinish(queue);
// diffSoph2 = clock() - startSoph2;
// Set the arguments to soph3
//
clFinish(queue);
getResults(queue, d_results, results, &bLow, &bHigh, &iLow, &iHigh, &alpha1diff, &alpha2diff);
// clock_t startSoph3 = clock();
cacheCall(cache, iHigh, &iHighCacheIndex, &iHighCompute);
err = 0;
err = clSetKernelArg(soph3, 0, sizeof(cl_mem), &d_input_data);
err |= clSetKernelArg(soph3, 1, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(soph3, 2, sizeof(cl_mem), &d_trainingAlpha);
err |= clSetKernelArg(soph3, 3, sizeof(cl_mem), &d_F);
err |= clSetKernelArg(soph3, 4, sizeof(cl_mem), &d_kernelDiag);
err |= clSetKernelArg(soph3, 5, sizeof(cl_mem), &d_lowFsSO3);
err |= clSetKernelArg(soph3, 6, sizeof(cl_mem), &d_lowIndicesSO3);
err |= clSetKernelArg(soph3, 7, sizeof(cl_mem), &d_deltaFsSO3);
err |= clSetKernelArg(soph3, 8, sizeof(cl_mem), &d_results);
err |= clSetKernelArg(soph3, 9, sizeof(int), &iHigh);
err |= clSetKernelArg(soph3, 10, sizeof(float), &bHigh);
err |= clSetKernelArg(soph3, 11, sizeof(int), &nPoints);
err |= clSetKernelArg(soph3, 12, sizeof(int), &dFeatures);
err |= clSetKernelArg(soph3, 13, sizeof(float), &epsilon);
err |= clSetKernelArg(soph3, 14, sizeof(float), &Ce);
err |= clSetKernelArg(soph3, 15, sizeof(float), ¶mA);
err |= clSetKernelArg(soph3, 16, sizeof(float), ¶mB);
err |= clSetKernelArg(soph3, 17, sizeof(float), ¶mC);
err |= clSetKernelArg(soph3, 18, sizeof(int) * localSoph3Size, NULL);
err |= clSetKernelArg(soph3, 19, sizeof(float) * localSoph3Size, NULL);
err |= clSetKernelArg(soph3, 20, sizeof(cl_mem), &d_cache);
err |= clSetKernelArg(soph3, 21, sizeof(bool), &iHighCompute);
err |= clSetKernelArg(soph3, 22, sizeof(int), &iHighCacheIndex);
if (err != CL_SUCCESS){
printf("Error: Failed to set kernel arguments! %d\n", err);
return err;
}
err = clEnqueueNDRangeKernel(queue, soph3, 1, NULL, &globalSoph3Size, &localSoph3Size, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to execute kernel Soph3!\n");
return err;
}
// clFinish(queue);
// diffSoph3 = clock() - startSoph3;
// Set the arguments to soph4
//
// clock_t startSoph4 = clock();
err = 0;
err = clSetKernelArg(soph4, 0, sizeof(cl_mem), &d_input_data);
err |= clSetKernelArg(soph4, 1, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(soph4, 2, sizeof(cl_mem), &d_trainingAlpha);
err |= clSetKernelArg(soph4, 3, sizeof(cl_mem), &d_kernelDiag);
err |= clSetKernelArg(soph4, 4, sizeof(cl_mem), &d_lowFsSO3);
err |= clSetKernelArg(soph4, 5, sizeof(cl_mem), &d_lowIndicesSO3);
err |= clSetKernelArg(soph4, 6, sizeof(cl_mem), &d_deltaFsSO3);
err |= clSetKernelArg(soph4, 7, sizeof(cl_mem), &d_results);
err |= clSetKernelArg(soph4, 8, sizeof(float), &cost);
err |= clSetKernelArg(soph4, 9, sizeof(int), &dFeatures);
err |= clSetKernelArg(soph4, 10, sizeof(int), &numGroups_soph3);
err |= clSetKernelArg(soph4, 11, sizeof(float), ¶mA);
err |= clSetKernelArg(soph4, 12, sizeof(float), ¶mB);
err |= clSetKernelArg(soph4, 13, sizeof(float), ¶mC);
err |= clSetKernelArg(soph4, 14, sizeof(cl_mem), &d_F);
err |= clSetKernelArg(soph4, 15, sizeof(int) * localSoph4Size, NULL);
err |= clSetKernelArg(soph4, 16, sizeof(float) * localSoph4Size, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to set kernel arguments! %d\n", err);
return err;
}
err = clEnqueueNDRangeKernel(queue, soph4, 1, NULL, &globalSoph4Size, &localSoph4Size, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to execute kernel Soph4!\n");
return err;
}
// clFinish(queue);
// diffSoph4 = clock() - startSoph4;
// Wait for the command queue to get serviced before reading back results
//
err = clFinish(queue);
if (err != CL_SUCCESS){
printf("Error: waiting for queue to finish failed\n");
return err;
}
// clock_t startGetResults = clock();
getResults(queue, d_results, results, &bLow, &bHigh, &iLow, &iHigh, &alpha1diff, &alpha2diff);
// clFinish(queue);
// diffGetResults = clock() - startGetResults;
//printf("iLow:%d, iHigh:%d\n", (int)results[2], (int)results[3]);
}
if(heuristic == 2){
addIteration(progress, bLow - bHigh);
}
}
printf("INFO: %d iterations\n", iteration);
printf("INFO: bLow: %f, bHigh %f\n", bLow, bHigh);
// diffTot = clock() - startTot;
// float msecTot = diffTot * 1000.0 / CLOCKS_PER_SEC;
// printf("Total time taken %.5f milliseconds\n", msecTot);
// float msecSoph1 = diffSoph1 * 1000.0 / CLOCKS_PER_SEC;
// printf("Soph1 time taken %.5f milliseconds\n", msecSoph1);
// float msecSoph2 = diffSoph2 * 1000.0 / CLOCKS_PER_SEC;
// printf("Soph2 time taken %.5f milliseconds\n", msecSoph2);
// float msecSoph3 = diffSoph3 * 1000.0 / CLOCKS_PER_SEC;
// printf("Soph3 time taken %.5f milliseconds\n", msecSoph3);
// float msecSoph4 = diffSoph4 * 1000.0 / CLOCKS_PER_SEC;
// printf("Soph4 time taken %.5f milliseconds\n", msecSoph4);
// float msecGetResults = diffGetResults * 1000.0 / CLOCKS_PER_SEC;
// printf("GetResult time taken %.5f milliseconds\n", msecGetResults);
// get training alpha
float *trainingAlpha = (float *)malloc(sizeof(float) * nPoints);
err = clEnqueueReadBuffer(queue, d_trainingAlpha, CL_TRUE, 0, sizeof(float) * nPoints, trainingAlpha, 0, NULL, NULL);
if (err != CL_SUCCESS){
printf("Error: Failed to read buffer\n");
return err;
}
// save results
*p_rho = (bHigh + bLow)/2;
int nSV = 0;
for(int i = 0; i < nPoints; i++){
if (trainingAlpha[i] > epsilon){
nSV++;
}
}
int index = 0;
*p_nSV = nSV;
*p_supportVectors = (float *)malloc(sizeof(float) * nSV * dFeatures);
*p_signedAlpha = (float *)malloc(sizeof(float) * nSV);
for(int i = 0; i < nPoints; i++){
if(trainingAlpha[i] > epsilon){
(* p_signedAlpha)[index] = labels[i] * trainingAlpha[i];
for(int j = 0; j < dFeatures; j++){
(* p_supportVectors)[index*dFeatures + j] = input_data[i * dFeatures + j];
}
index ++;
}
}
// Shutdown and cleanup
//
printf("%d Cache Hits!\n",cache->hits);
printf("%d Cache Misses!\n",cache->misses);
return 0;
}
