void S<T>::test()
{
#pragma omp parallel num_threads(n) // { dg-error "must be integral" }
work();
}
/*
* Receive callback for the /camera/depth_registered/points subscription
*/
std::vector<suturo_perception_msgs::PerceivedObject> SuturoPerceptionKnowledgeROSNode::receive_image_and_cloud(const sensor_msgs::ImageConstPtr& inputImage, const sensor_msgs::PointCloud2ConstPtr& inputCloud)
{
// process only one cloud
pcl::PointCloud<pcl::PointXYZRGB>::Ptr cloud_in (new pcl::PointCloud<pcl::PointXYZRGB>());
pcl::fromROSMsg(*inputCloud,*cloud_in);
logger.logInfo((boost::format("Received a new point cloud: size = %s") % cloud_in->points.size()).str());
// Gazebo sends us unorganized pointclouds!
// Reorganize them to be able to compute the ROI of the objects
// This workaround is only tested for gazebo 1.9!
if(!cloud_in->isOrganized ())
{
logger.logInfo((boost::format("Received an unorganized PointCloud: %d x %d .Convert it to a organized one ...") % cloud_in->width % cloud_in->height ).str());
pcl::PointCloud<pcl::PointXYZRGB>::Ptr org_cloud (new pcl::PointCloud<pcl::PointXYZRGB>());
org_cloud->width = 640;
org_cloud->height = 480;
org_cloud->is_dense = false;
org_cloud->points.resize(640 * 480);
for (int i = 0; i < cloud_in->points.size(); i++) {
pcl::PointXYZRGB result;
result.x = 0;
result.y = 0;
result.z = 0;
org_cloud->points[i]=cloud_in->points[i];
}
cloud_in = org_cloud;
}
cv_bridge::CvImagePtr cv_ptr;
cv_ptr = cv_bridge::toCvCopy(inputImage, enc::BGR8);
// Make a deep copy of the passed cv::Mat and set a new
// boost pointer to it.
boost::shared_ptr<cv::Mat> img(new cv::Mat(cv_ptr->image.clone()));
sp.setOriginalRGBImage(img);
logger.logInfo("processing...");
sp.setOriginalCloud(cloud_in);
sp.processCloudWithProjections(cloud_in);
logger.logInfo("Cloud processed. Lock buffer and return the results");
mutex.lock();
perceivedObjects = sp.getPerceivedObjects();
if(sp.getOriginalRGBImage()->cols != sp.getOriginalCloud()->width
&& sp.getOriginalRGBImage()->rows != sp.getOriginalCloud()->height)
{
// Adjust the ROI if the image is at 1280x1024 and the pointcloud is at 640x480
if(sp.getOriginalRGBImage()->cols == 1280 && sp.getOriginalRGBImage()->rows == 1024)
{
for (int i = 0; i < perceivedObjects.size(); i++) {
ROI roi = perceivedObjects.at(i).get_c_roi();
roi.origin.x*=2;
roi.origin.y*=2;
roi.width*=2;
roi.height*=2;
perceivedObjects.at(i).set_c_roi(roi);
}
}
else
{
logger.logError("UNSUPPORTED MIXTURE OF IMAGE AND POINTCLOUD DIMENSIONS");
}
}
// Execution pipeline
// Each capability provides an enrichment for the
// returned PerceivedObject
// initialize threadpool
boost::asio::io_service ioService;
boost::thread_group threadpool;
std::auto_ptr<boost::asio::io_service::work> work(new boost::asio::io_service::work(ioService));
// Add worker threads to threadpool
for(int i = 0; i < numThreads; ++i)
{
threadpool.create_thread(
boost::bind(&boost::asio::io_service::run, &ioService)
);
}
for (int i = 0; i < perceivedObjects.size(); i++)
{
// Initialize Capabilities
ColorAnalysis ca(perceivedObjects[i]);
ca.setLowerSThreshold(color_analysis_lower_s);
ca.setUpperSThreshold(color_analysis_upper_s);
ca.setLowerVThreshold(color_analysis_lower_v);
ca.setUpperVThreshold(color_analysis_upper_v);
suturo_perception_shape_detection::RandomSampleConsensus sd(perceivedObjects[i]);
//suturo_perception_vfh_estimation::VFHEstimation vfhe(perceivedObjects[i]);
// suturo_perception_3d_capabilities::CuboidMatcherAnnotator cma(perceivedObjects[i]);
// Init the cuboid matcher with the table coefficients
suturo_perception_3d_capabilities::CuboidMatcherAnnotator cma(perceivedObjects[i], sp.getTableCoefficients() );
// post work to threadpool
ioService.post(boost::bind(&ColorAnalysis::execute, ca));
ioService.post(boost::bind(&suturo_perception_shape_detection::RandomSampleConsensus::execute, sd));
//ioService.post(boost::bind(&suturo_perception_vfh_estimation::VFHEstimation::execute, vfhe));
ioService.post(boost::bind(&suturo_perception_3d_capabilities::CuboidMatcherAnnotator::execute, cma));
// Is 2d recognition enabled?
if(!recognitionDir.empty())
{
// perceivedObjects[i].c_recognition_label_2d="";
suturo_perception_2d_capabilities::LabelAnnotator2D la(perceivedObjects[i], sp.getOriginalRGBImage(), object_matcher_);
la.execute();
}
else
{
// Set an empty label
perceivedObjects[i].set_c_recognition_label_2d("");
}
}
//boost::this_thread::sleep(boost::posix_time::microseconds(1000));
// wait for thread completion.
// destroy the work object to wait for all queued tasks to finish
work.reset();
ioService.run();
threadpool.join_all();
std::vector<suturo_perception_msgs::PerceivedObject> perceivedObjs = *convertPerceivedObjects(&perceivedObjects); // TODO handle images in this method
mutex.unlock();
return perceivedObjs;
}
int main()
{
init();
work();
return 0;
}
void HibernateBoot(char *image_filename)
{
long long size, imageSize, codeSize, allocSize;
long mem_base;
IOHibernateImageHeader _header;
IOHibernateImageHeader * header = &_header;
long buffer;
size = ReadFileAtOffset (image_filename, header, 0, sizeof(IOHibernateImageHeader));
printf("header read size %x\n", size);
imageSize = header->image1Size;
codeSize = header->restore1PageCount << 12;
if (kIOHibernateHeaderSignature != header->signature)
{
printf ("Incorrect image signature\n");
return;
}
if (header->encryptStart)
{
printf ("Resuming from Encrypted image is unsupported.\n"
"Uncheck \"Use secure virtual memory\" in \"Security\" pane on system preferences.\n"
"Press any key to proceed with normal boot.\n");
getc ();
return;
}
// depends on NVRAM
#if 0
{
uint32_t machineSignature;
size = GetProp(gChosenPH, kIOHibernateMachineSignatureKey,
(char *)&machineSignature, sizeof(machineSignature));
if (size != sizeof(machineSignature)) machineSignature = 0;
if (machineSignature != header->machineSignature)
break;
}
#endif
allocSize = imageSize + ((4095 + sizeof(hibernate_graphics_t)) & ~4095);
mem_base = getmemorylimit() - allocSize;//TODO: lower this
printf("mem_base %x\n", mem_base);
if (!((long long)mem_base+allocSize<1024*bootInfo->extmem+0x100000))
{
printf ("Not enough space to restore image. Press any key to proceed with normal boot.\n");
getc ();
return;
}
bcopy(header, (void *) mem_base, sizeof(IOHibernateImageHeader));
header = (IOHibernateImageHeader *) mem_base;
imageSize -= sizeof(IOHibernateImageHeader);
buffer = (long)(header + 1);
if (header->previewSize)
{
uint64_t preview_offset = header->fileExtentMapSize - sizeof(header->fileExtentMap) + codeSize;
uint8_t progressSaveUnder[kIOHibernateProgressCount][kIOHibernateProgressSaveUnderSize];
ReadFileAtOffset (image_filename, (char *)buffer, sizeof(IOHibernateImageHeader), preview_offset+header->previewSize);
drawPreview ((void *)(long)(buffer+preview_offset + header->previewPageListSize), &(progressSaveUnder[0][0]));
previewTotalSectors = (imageSize-(preview_offset+header->previewSize))/512;
previewLoadedSectors = 0;
previewSaveunder = &(progressSaveUnder[0][0]);
if (preview_offset+header->previewSize<imageSize)
ReadFileAtOffset (image_filename, (char *)(long)(buffer+preview_offset+header->previewSize),
sizeof(IOHibernateImageHeader)+preview_offset+header->previewSize,
imageSize-(preview_offset+header->previewSize));
previewTotalSectors = 0;
previewLoadedSectors = 0;
previewSaveunder = 0;
#if 0
AsereBLN:
check_vga_nvidia() didn't work as expected (recursion level > 0 & return value).
Unforutnaltely I cannot find a note why to switch back to text mode for nVidia cards only
and because it check_vga_nvidia does not work (cards normally are behind a bridge) I will
remove it completely
setVideoMode( VGA_TEXT_MODE, 0 );
#endif
}
// a new aggregate is to be inserted into the work queue
inline void new_work_agg(db::node *node, db::simple_tuple *stpl)
{
process::work work(node, stpl, process::mods::LOCAL_TUPLE | process::mods::FORCE_AGGREGATE);
new_agg(work);
}
void EigenValuesAdvection::v_DoSolve()
{
int nvariables = 1;
int i,dofs = GetNcoeffs();
//bool UseContCoeffs = false;
Array<OneD, Array<OneD, NekDouble> > inarray(nvariables);
Array<OneD, Array<OneD, NekDouble> > tmp(nvariables);
Array<OneD, Array<OneD, NekDouble> > outarray(nvariables);
Array<OneD, Array<OneD, NekDouble> > WeakAdv(nvariables);
int npoints = GetNpoints();
int ncoeffs = GetNcoeffs();
switch (m_projectionType)
{
case MultiRegions::eDiscontinuous:
{
dofs = ncoeffs;
break;
}
case MultiRegions::eGalerkin:
case MultiRegions::eMixed_CG_Discontinuous:
{
//dofs = GetContNcoeffs();
//UseContCoeffs = true;
break;
}
}
cout << endl;
cout << "Num Phys Points = " << npoints << endl; // phisical points
cout << "Num Coeffs = " << ncoeffs << endl; //
cout << "Num Cont Coeffs = " << dofs << endl;
inarray[0] = Array<OneD, NekDouble>(npoints,0.0);
outarray[0] = Array<OneD, NekDouble>(npoints,0.0);
tmp[0] = Array<OneD, NekDouble>(npoints,0.0);
WeakAdv[0] = Array<OneD, NekDouble>(ncoeffs,0.0);
Array<OneD, NekDouble> MATRIX(npoints*npoints,0.0);
for (int j = 0; j < npoints; j++)
{
inarray[0][j] = 1.0;
/// Feeding the weak Advection oprator with a vector (inarray)
/// Looping on inarray and changing the position of the only non-zero entry
/// we simulate the multiplication by the identity matrix.
/// The results stored in outarray is one of the columns of the weak advection oprators
/// which are then stored in MATRIX for the futher eigenvalues calculation.
switch (m_projectionType)
{
case MultiRegions::eDiscontinuous:
{
WeakDGAdvection(inarray, WeakAdv,true,true,1);
m_fields[0]->MultiplyByElmtInvMass(WeakAdv[0],WeakAdv[0]);
m_fields[0]->BwdTrans(WeakAdv[0],outarray[0]);
Vmath::Neg(npoints,outarray[0],1);
break;
}
case MultiRegions::eGalerkin:
case MultiRegions::eMixed_CG_Discontinuous:
{
// Calculate -V\cdot Grad(u);
for(i = 0; i < nvariables; ++i)
{
//Projection
m_fields[i]->FwdTrans(inarray[i],WeakAdv[i]);
m_fields[i]->BwdTrans_IterPerExp(WeakAdv[i],tmp[i]);
//Advection operator
AdvectionNonConservativeForm(m_velocity,tmp[i],outarray[i]);
Vmath::Neg(npoints,outarray[i],1);
//m_fields[i]->MultiplyByInvMassMatrix(WeakAdv[i],WeakAdv[i]);
//Projection
m_fields[i]->FwdTrans(outarray[i],WeakAdv[i]);
m_fields[i]->BwdTrans_IterPerExp(WeakAdv[i],outarray[i]);
}
break;
}
}
/// The result is stored in outarray (is the j-th columns of the weak advection operator).
/// We now store it in MATRIX(j)
Vmath::Vcopy(npoints,&(outarray[0][0]),1,&(MATRIX[j]),npoints);
/// Set the j-th entry of inarray back to zero
inarray[0][j] = 0.0;
}
////////////////////////////////////////////////////////////////////////////////
/// Calulating the eigenvalues of the weak advection operator stored in (MATRIX)
/// using Lapack routines
char jobvl = 'N';
char jobvr = 'N';
int info = 0, lwork = 3*npoints;
NekDouble dum;
Array<OneD, NekDouble> EIG_R(npoints);
Array<OneD, NekDouble> EIG_I(npoints);
Array<OneD, NekDouble> work(lwork);
Lapack::Dgeev(jobvl,jobvr,npoints,MATRIX.get(),npoints,EIG_R.get(),EIG_I.get(),&dum,1,&dum,1,&work[0],lwork,info);
////////////////////////////////////////////////////////
//Print Matrix
FILE *mFile;
mFile = fopen ("WeakAdvMatrix.txt","w");
for(int j = 0; j<npoints; j++)
{
for(int k = 0; k<npoints; k++)
{
fprintf(mFile,"%e ",MATRIX[j*npoints+k]);
}
fprintf(mFile,"\n");
}
fclose (mFile);
////////////////////////////////////////////////////////
//Output of the EigenValues
FILE *pFile;
pFile = fopen ("Eigenvalues.txt","w");
for(int j = 0; j<npoints; j++)
{
fprintf(pFile,"%e %e\n",EIG_R[j],EIG_I[j]);
}
fclose (pFile);
cout << "\nEigenvalues : " << endl;
for(int j = 0; j<npoints; j++)
{
cout << EIG_R[j] << "\t" << EIG_I[j] << endl;
}
cout << endl;
}
static GstFlowReturn
gst_bml_transform_transform_ip_mono (GstBaseTransform * base,
GstBuffer * outbuf)
{
GstMapInfo info;
GstBMLTransform *bml_transform = GST_BML_TRANSFORM (base);
GstBMLTransformClass *klass = GST_BML_TRANSFORM_GET_CLASS (bml_transform);
GstBML *bml = GST_BML (bml_transform);
GstBMLClass *bml_class = GST_BML_CLASS (klass);
BMLData *data, *seg_data;
gpointer bm = bml->bm;
guint todo, seg_size, samples_per_buffer;
gboolean has_data;
guint mode = 3; /*WM_READWRITE */
bml->running_time =
gst_segment_to_stream_time (&base->segment, GST_FORMAT_TIME,
GST_BUFFER_TIMESTAMP (outbuf));
if (GST_BUFFER_FLAG_IS_SET (outbuf, GST_BUFFER_FLAG_DISCONT)) {
bml->subtick_count = (!bml->reverse) ? bml->subticks_per_tick : 1;
}
/* TODO(ensonic): sync on subticks ? */
if (bml->subtick_count >= bml->subticks_per_tick) {
bml (gstbml_reset_triggers (bml, bml_class));
bml (gstbml_sync_values (bml, bml_class, GST_BUFFER_TIMESTAMP (outbuf)));
bml (tick (bm));
bml->subtick_count = 1;
} else {
bml->subtick_count++;
}
/* don't process data in passthrough-mode */
if (gst_base_transform_is_passthrough (base))
return GST_FLOW_OK;
if (!gst_buffer_map (outbuf, &info, GST_MAP_READ | GST_MAP_WRITE)) {
GST_WARNING_OBJECT (base, "unable to map buffer for read & write");
return GST_FLOW_ERROR;
}
data = (BMLData *) info.data;
samples_per_buffer = info.size / sizeof (BMLData);
/* if buffer has only silence process with different mode */
if (GST_BUFFER_FLAG_IS_SET (outbuf, GST_BUFFER_FLAG_GAP)) {
mode = 2; /* WM_WRITE */
} else {
// buzz generates loud output
gfloat fc = 32768.0;
orc_scalarmultiply_f32_ns (data, data, fc, samples_per_buffer);
}
GST_DEBUG_OBJECT (bml_transform, " calling work(%d,%d)", samples_per_buffer,
mode);
todo = samples_per_buffer;
seg_data = data;
has_data = FALSE;
while (todo) {
// 256 is MachineInterface.h::MAX_BUFFER_LENGTH
seg_size = (todo > 256) ? 256 : todo;
has_data |= bml (work (bm, seg_data, (int) seg_size, mode));
seg_data = &seg_data[seg_size];
todo -= seg_size;
}
if (gstbml_fix_data ((GstElement *) bml_transform, &info, has_data)) {
GST_BUFFER_FLAG_SET (outbuf, GST_BUFFER_FLAG_GAP);
} else {
GST_BUFFER_FLAG_UNSET (outbuf, GST_BUFFER_FLAG_GAP);
}
gst_buffer_unmap (outbuf, &info);
return GST_FLOW_OK;
}
int main()
{
while (work());
return 0;
}
void Solve::solve(FILE *fin, FILE *fout)
{
int cnt = 1;
for(int i = 0; i <= 10; i ++, cnt *= 2)
lb[cnt] = i;
fscanf(fin, "%d%d", &n, &m);
for(int i = 1; i <= n; i ++)
for(int j = 1; j <= m; j ++)
fscanf(fin, "%d", &object[i][j].a);
for(int i = 1; i <= n; i ++)
for(int j = 1; j <= m; j ++)
fscanf(fin, "%d", &object[i][j].d);
for(int i = 1; i <= n; i ++)
for(int j = 1; j <= m; j ++)
fscanf(fin, "%d", &object[i][j].hp);
fscanf(fin, "%d%d%d", &llx.a, &llx.d, &llx.hp);
fscanf(fin, "%d", &nBaby);
for(int i = 1; i <= nBaby; i ++)
fscanf(fin, "%d%d%d", &baby[i].a, &baby[i].d, &baby[i].hp);
for(int i = 1; i <= n; i ++)
for(int j = 1; j <= m; j ++)
{
Stuff &lyd = object[i][j];
if(llx.a <= lyd.d)
w[i][j][0] = INFINITY;
else
{
int t1 = ceilDiv(lyd.hp, llx.a-lyd.d);
int tmp = (t1 - 1) * MAX(0, lyd.a - llx.d);
if(tmp >= llx.hp)
w[i][j][0] = INFINITY;
else
w[i][j][0] = tmp;
}
for(int k = 1; k <= nBaby; k ++)
{
Stuff &bb = baby[k];
if(bb.a <= lyd.d)
w[i][j][k] = INFINITY;
else
{
int t1 = ceilDiv(lyd.hp, bb.a - lyd.d);
int tmp = (t1 - 1) * MAX(0, lyd.a - bb.d);
if(tmp >= bb.hp)
{
int t2 = ceilDiv(bb.hp, lyd.a - bb.d);
tmp = t2 * MAX(0, bb.a - lyd.d);
//baby died
if(llx.a <= lyd.d)
w[i][j][k] = INFINITY;
else
{
int t1 = ceilDiv(lyd.hp - tmp, llx.a-lyd.d);
tmp = (t1 - 1) * MAX(0, lyd.a - llx.d);
if(tmp >= llx.hp)
w[i][j][k] = INFINITY;
else
w[i][j][k] = tmp;
}
}
else
w[i][j][k] = 0;
}
}
}
upperlim = (1 << nBaby) - 1;
work(fin, fout);
}
int main(int argc, char **argv){
// Add some plugin searhc paths
plugin_search_path=list_new(free);
const char *infilename=NULL;
const char *outfilename=NULL;
char tmp[256];
char *assetfilename="assets.h";
int i;
for (i=1;i<argc;i++){
if (strcmp(argv[i], "--help")==0){
help(NULL);
return 0;
}
else if ((strcmp(argv[i], "--templatetagsdir")==0) || (strcmp(argv[i], "-t")==0)){
i++;
if (argc<=i){
help("Missing templatedir name");
return 3;
}
snprintf(tmp, sizeof(tmp), "%s/lib%%s.so", argv[i]);
ONION_DEBUG("Added templatedir %s", tmp);
list_add(plugin_search_path, strdup(tmp)); // dup, remember to free later.
}
else if ((strcmp(argv[i], "--no-orig-lines")==0) || (strcmp(argv[i], "-n")==0)){
use_orig_line_numbers=0;
ONION_DEBUG("Disable original line numbers");
}
else if ((strcmp(argv[i], "--asset-file")==0) || (strcmp(argv[i], "-a")==0)){
i++;
if (argc<=i){
help("Missing assets file name");
return 3;
}
assetfilename=argv[i];
ONION_DEBUG("Assets file: %s", assetfilename);
}
else{
if (infilename){
if (outfilename){
help("Too many arguments");
return 1;
}
outfilename=argv[i];
ONION_DEBUG("Set outfilename %s", outfilename);
}
else{
infilename=argv[i];
ONION_DEBUG("Set infilename %s", infilename);
}
}
}
if (!infilename || !outfilename){
help("Missing input or output filename");
return 2;
}
if (strcmp(infilename,"-")==0){
infilename="";
}
else{
char tmp2[256];
strncpy(tmp2, argv[1], sizeof(tmp2)-1);
snprintf(tmp, sizeof(tmp), "%s/lib%%s.so", dirname(tmp2));
list_add(plugin_search_path, strdup(tmp));
strncpy(tmp2, argv[1], sizeof(tmp2)-1);
snprintf(tmp, sizeof(tmp), "%s/templatetags/lib%%s.so", dirname(tmp2));
list_add(plugin_search_path, strdup(tmp));
}
// Default template dirs
list_add_with_flags(plugin_search_path, "lib%s.so", LIST_ITEM_NO_FREE);
list_add_with_flags(plugin_search_path, "templatetags/lib%s.so", LIST_ITEM_NO_FREE);
char tmp2[256];
strncpy(tmp2, argv[0], sizeof(tmp2)-1);
snprintf(tmp, sizeof(tmp), "%s/templatetags/lib%%s.so", dirname(tmp2));
list_add(plugin_search_path, strdup(tmp)); // dupa is ok, as im at main.
strncpy(tmp2, argv[0], sizeof(tmp2)-1);
snprintf(tmp, sizeof(tmp), "%s/lib%%s.so", dirname(tmp2));
list_add(plugin_search_path, strdup(tmp)); // dupa is ok, as im at main.
list_add_with_flags(plugin_search_path, "/usr/local/lib/otemplate/templatetags/lib%s.so", LIST_ITEM_NO_FREE);
list_add_with_flags(plugin_search_path, "/usr/lib/otemplate/templatetags/lib%s.so", LIST_ITEM_NO_FREE);
onion_assets_file *assetsfile=onion_assets_file_new(assetfilename);
int error=work(infilename, outfilename, assetsfile);
onion_assets_file_free(assetsfile);
list_free(plugin_search_path);
return error;
}
int main(int argc, char** argv)
{
boost::program_options::options_description desc("options");
desc.add_options()
("help", "produce help message")
("topic", boost::program_options::value<std::string>(), "topic")
("broker", boost::program_options::value<std::string>(), "broker")
("schema_registry", boost::program_options::value<std::string>(), "schema_registry")
("schema_registry_port", boost::program_options::value<int>()->default_value(8081), "schema_registry_port")
;
boost::program_options::variables_map vm;
boost::program_options::store(boost::program_options::parse_command_line(argc, argv, desc), vm);
boost::program_options::notify(vm);
boost::log::core::get()->set_filter(boost::log::trivial::severity >= boost::log::trivial::info);
if (vm.count("help"))
{
std::cout << desc << std::endl;
return 0;
}
std::string topic;
if (vm.count("topic"))
{
topic = vm["topic"].as<std::string>();
}
else
{
std::cout << "--topic must be specified" << std::endl;
return 0;
}
int32_t kafka_port = 9092;
std::vector<csi::kafka::broker_address> kafka_brokers;
if (vm.count("broker"))
{
std::string s = vm["broker"].as<std::string>();
size_t last_colon = s.find_last_of(':');
if (last_colon != std::string::npos)
kafka_port = atoi(s.substr(last_colon + 1).c_str());
s = s.substr(0, last_colon);
// now find the brokers...
size_t last_separator = s.find_last_of(',');
while (last_separator != std::string::npos)
{
std::string host = s.substr(last_separator + 1);
kafka_brokers.push_back(csi::kafka::broker_address(host, kafka_port));
s = s.substr(0, last_separator);
last_separator = s.find_last_of(',');
}
kafka_brokers.push_back(csi::kafka::broker_address(s, kafka_port));
}
else
{
std::cout << "--broker must be specified" << std::endl;
return 0;
}
int32_t schema_registry_port = 8081;
std::vector<csi::kafka::broker_address> schema_registrys;
std::string used_schema_registry;
if (vm.count("schema_registry_port"))
{
schema_registry_port = vm["schema_registry_port"].as<int>();
}
if (vm.count("schema_registry"))
{
std::string s = vm["schema_registry"].as<std::string>();
size_t last_colon = s.find_last_of(':');
if (last_colon != std::string::npos)
schema_registry_port = atoi(s.substr(last_colon + 1).c_str());
s = s.substr(0, last_colon);
// now find the brokers...
size_t last_separator = s.find_last_of(',');
while (last_separator != std::string::npos)
{
std::string host = s.substr(last_separator + 1);
schema_registrys.push_back(csi::kafka::broker_address(host, schema_registry_port));
s = s.substr(0, last_separator);
last_separator = s.find_last_of(',');
}
schema_registrys.push_back(csi::kafka::broker_address(s, schema_registry_port));
}
else
{
// default - assume registry is running on all kafka brokers
for (std::vector<csi::kafka::broker_address>::const_iterator i = kafka_brokers.begin(); i != kafka_brokers.end(); ++i)
{
schema_registrys.push_back(csi::kafka::broker_address(i->host_name, schema_registry_port));
}
}
// right now the schema registry class cannot handle severel hosts so just stick to the first one.
used_schema_registry = schema_registrys[0].host_name + ":" + std::to_string(schema_registrys[0].port);
std::string kafka_broker_str = "";
for (std::vector<csi::kafka::broker_address>::const_iterator i = kafka_brokers.begin(); i != kafka_brokers.end(); ++i)
{
kafka_broker_str += i->host_name + ":" + std::to_string(i->port);
if (i != kafka_brokers.end() - 1)
kafka_broker_str += ", ";
}
BOOST_LOG_TRIVIAL(info) << "kafka broker(s): " << kafka_broker_str;
BOOST_LOG_TRIVIAL(info) << "topic : " << topic;
std::string schema_registrys_info;
for (std::vector<csi::kafka::broker_address>::const_iterator i = schema_registrys.begin(); i != schema_registrys.end(); ++i)
{
schema_registrys_info += i->host_name + ":" + std::to_string(i->port);
if (i != schema_registrys.end() - 1)
schema_registrys_info += ", ";
}
BOOST_LOG_TRIVIAL(info) << "schema_registry(s) : " << schema_registrys_info;
BOOST_LOG_TRIVIAL(info) << "used schema registry: " << used_schema_registry;
int64_t total = 0;
boost::asio::io_service fg_ios;
std::auto_ptr<boost::asio::io_service::work> work(new boost::asio::io_service::work(fg_ios));
boost::thread fg(boost::bind(&boost::asio::io_service::run, &fg_ios));
csi::kafka::highlevel_producer producer(fg_ios, topic, -1, 200, 1000000);
confluent::registry registry(fg_ios, used_schema_registry);
confluent::codec avro_codec(registry);
producer.connect(kafka_brokers);
BOOST_LOG_TRIVIAL(info) << "connected to kafka";
producer.connect_forever(kafka_brokers);
boost::thread do_log([&producer]
{
while (true)
{
boost::this_thread::sleep(boost::posix_time::seconds(1));
std::vector<csi::kafka::highlevel_producer::metrics> metrics = producer.get_metrics();
size_t total_queue = 0;
uint32_t tx_msg_sec_total = 0;
uint32_t tx_kb_sec_total = 0;
for (std::vector<csi::kafka::highlevel_producer::metrics>::const_iterator i = metrics.begin(); i != metrics.end(); ++i)
{
total_queue += (*i).msg_in_queue;
tx_msg_sec_total += (*i).tx_msg_sec;
tx_kb_sec_total += (*i).tx_kb_sec;
}
BOOST_LOG_TRIVIAL(info) << "\t \tqueue:" << total_queue << "\t" << tx_msg_sec_total << " msg/s \t" << (tx_kb_sec_total / 1024) << "MB/s";
}
});
std::cerr << "registring schemas" << std::endl;
auto key_res = avro_codec.put_schema("sample.contact_info_key", sample::contact_info_key::valid_schema());
if (key_res.first!=0)
{
BOOST_LOG_TRIVIAL(error) << "registring sample.contact_info_key failed";
return -1;
}
auto val_res = avro_codec.put_schema("sample.contact_info", sample::contact_info::valid_schema());
if (val_res.first!=0)
{
BOOST_LOG_TRIVIAL(error) << "registring sample.contact_info failed";
return -1;
}
BOOST_LOG_TRIVIAL(info) << "registring schemas done";
//produce messages
std::vector<boost::thread*> threads;
for (int i = 0; i != 10; ++i)
{
threads.emplace_back(new boost::thread([&avro_codec, key_res, val_res, &producer, i]
{
send_messages(avro_codec, key_res.second, val_res.second, producer, i);
}));
}
while (true)
{
boost::this_thread::sleep(boost::posix_time::seconds(1));
}
work.reset();
fg_ios.stop();
return EXIT_SUCCESS;
}
void bob::math::eig_(const blitz::Array<double,2>& A,
blitz::Array<std::complex<double>,2>& V,
blitz::Array<std::complex<double>,1>& D)
{
// Size variable
const int N = A.extent(0);
// Prepares to call LAPACK function
// Initialises LAPACK variables
const char jobvl = 'N'; // Do NOT compute left eigen-vectors
const char jobvr = 'V'; // Compute right eigen-vectors
int info = 0;
const int lda = N;
const int ldvr = N;
double VL = 0; // notice we don't compute the left eigen-values
const int ldvl = 1;
// Initialises LAPACK arrays
blitz::Array<double,2> A_lapack = bob::core::array::ccopy(const_cast<blitz::Array<double,2>&>(A).transpose(1,0));
// temporary arrays to receive LAPACK's eigen-values and eigen-vectors
blitz::Array<double,1> WR(D.shape()); //real part
blitz::Array<double,1> WI(D.shape()); //imaginary part
blitz::Array<double,2> VR(A.shape()); //right eigen-vectors
// Calls the LAPACK function
// A/ Queries the optimal size of the working arrays
const int lwork_query = -1;
double work_query;
dgeev_( &jobvl, &jobvr, &N, A_lapack.data(), &lda, WR.data(), WI.data(),
&VL, &ldvl, VR.data(), &ldvr, &work_query, &lwork_query, &info);
// B/ Computes the eigenvalue decomposition
const int lwork = static_cast<int>(work_query);
boost::shared_array<double> work(new double[lwork]);
dgeev_( &jobvl, &jobvr, &N, A_lapack.data(), &lda, WR.data(), WI.data(),
&VL, &ldvl, VR.data(), &ldvr, work.get(), &lwork, &info);
// Checks info variable
if (info != 0) {
throw std::runtime_error("the QR algorithm failed to compute all the eigenvalues, and no eigenvectors have been computed.");
}
// Copy results back from WR, WI => D
blitz::real(D) = WR;
blitz::imag(D) = WI;
// Copy results back from VR => V, with two rules:
// 1) If the j-th eigenvalue is real, then v(j) = VR(:,j), the j-th column of
// VR.
// 2) If the j-th and (j+1)-st eigenvalues form a complex conjugate pair,
// then v(j) = VR(:,j) + i*VR(:,j+1) and v(j+1) = VR(:,j) - i*VR(:,j+1).
blitz::Range a = blitz::Range::all();
int i=0;
while (i<N) {
if (std::imag(D(i)) == 0.) { //real eigen-value, consume 1
blitz::real(V(a,i)) = VR(i,a);
blitz::imag(V(a,i)) = 0.;
++i;
}
else { //complex eigen-value, consume 2
blitz::real(V(a,i)) = VR(i,a);
blitz::imag(V(a,i)) = VR(i+1,a);
blitz::real(V(a,i+1)) = VR(i,a);
blitz::imag(V(a,i+1)) = -VR(i+1,a);
i += 2;
}
}
}
void bob::math::eigSym_(const blitz::Array<double,2>& A, const blitz::Array<double,2>& B,
blitz::Array<double,2>& V, blitz::Array<double,1>& D)
{
// Size variable
const int N = A.extent(0);
// Prepares to call LAPACK function
// Initialises LAPACK variables
const int itype = 1;
const char jobz = 'V'; // Get both the eigenvalues and the eigenvectors
const char uplo = 'U';
int info = 0;
const int lda = N;
const int ldb = N;
// Initialises LAPACK arrays
blitz::Array<double,2> A_blitz_lapack;
// Tries to use V directly
blitz::Array<double,2> Vt = V.transpose(1,0);
const bool V_direct_use = bob::core::array::isCZeroBaseContiguous(Vt);
if (V_direct_use)
{
A_blitz_lapack.reference(Vt);
// Ugly fix for non-const transpose
A_blitz_lapack = const_cast<blitz::Array<double,2>&>(A).transpose(1,0);
}
else
// Ugly fix for non-const transpose
A_blitz_lapack.reference(
bob::core::array::ccopy(const_cast<blitz::Array<double,2>&>(A).transpose(1,0)));
double *A_lapack = A_blitz_lapack.data();
// Ugly fix for non-const transpose
blitz::Array<double,2> B_blitz_lapack(
bob::core::array::ccopy(const_cast<blitz::Array<double,2>&>(B).transpose(1,0)));
double *B_lapack = B_blitz_lapack.data();
blitz::Array<double,1> D_blitz_lapack;
const bool D_direct_use = bob::core::array::isCZeroBaseContiguous(D);
if (D_direct_use)
D_blitz_lapack.reference(D);
else
D_blitz_lapack.resize(D.shape());
double *D_lapack = D_blitz_lapack.data();
// Calls the LAPACK function
// A/ Queries the optimal size of the working arrays
const int lwork_query = -1;
double work_query;
const int liwork_query = -1;
int iwork_query;
dsygvd_( &itype, &jobz, &uplo, &N, A_lapack, &lda, B_lapack, &ldb, D_lapack,
&work_query, &lwork_query, &iwork_query, &liwork_query, &info);
// B/ Computes the generalized eigenvalue decomposition
const int lwork = static_cast<int>(work_query);
boost::shared_array<double> work(new double[lwork]);
const int liwork = static_cast<int>(iwork_query);
boost::shared_array<int> iwork(new int[liwork]);
dsygvd_( &itype, &jobz, &uplo, &N, A_lapack, &lda, B_lapack, &ldb, D_lapack,
work.get(), &lwork, iwork.get(), &liwork, &info);
// Checks info variable
if (info != 0)
throw std::runtime_error("The LAPACK function 'dsygvd' returned a non-zero value. This might be caused by a non-positive definite B matrix.");
// Copy singular vectors back to V if required
if (!V_direct_use)
V = A_blitz_lapack.transpose(1,0);
// Copy result back to sigma if required
if (!D_direct_use)
D = D_blitz_lapack;
}
int main()
{
read();
work();
return 0;
}
Basis_HGRAD_LINE_Cn_FEM<SpT,OT,PT>::
Basis_HGRAD_LINE_Cn_FEM( const ordinal_type order,
const EPointType pointType ) {
this->basisCardinality_ = order+1;
this->basisDegree_ = order;
this->basisCellTopology_ = shards::CellTopology(shards::getCellTopologyData<shards::Line<2> >() );
this->basisType_ = BASIS_FEM_FIAT;
this->basisCoordinates_ = COORDINATES_CARTESIAN;
const ordinal_type card = this->basisCardinality_;
// points are computed in the host and will be copied
Kokkos::DynRankView<typename scalarViewType::value_type,typename SpT::array_layout,Kokkos::HostSpace>
dofCoords("Hgrad::Line::Cn::dofCoords", card, 1);
switch (pointType) {
case POINTTYPE_EQUISPACED:
case POINTTYPE_WARPBLEND: {
// lattice ordering
{
const ordinal_type offset = 0;
PointTools::getLattice( dofCoords,
this->basisCellTopology_,
order, offset,
pointType );
}
// topological order
// {
// // two vertices
// dofCoords(0,0) = -1.0;
// dofCoords(1,0) = 1.0;
// // internal points
// typedef Kokkos::pair<ordinal_type,ordinal_type> range_type;
// auto pts = Kokkos::subview(dofCoords, range_type(2, card), Kokkos::ALL());
// const auto offset = 1;
// PointTools::getLattice( pts,
// this->basisCellTopology_,
// order, offset,
// pointType );
// }
break;
}
case POINTTYPE_GAUSS: {
// internal points only
PointTools::getGaussPoints( dofCoords,
order );
break;
}
default: {
INTREPID2_TEST_FOR_EXCEPTION( !isValidPointType(pointType),
std::invalid_argument ,
">>> ERROR: (Intrepid2::Basis_HGRAD_LINE_Cn_FEM) invalid pointType." );
}
}
this->dofCoords_ = Kokkos::create_mirror_view(typename SpT::memory_space(), dofCoords);
Kokkos::deep_copy(this->dofCoords_, dofCoords);
// form Vandermonde matrix; actually, this is the transpose of the VDM,
// this matrix is used in LAPACK so it should be column major and left layout
const ordinal_type lwork = card*card;
Kokkos::DynRankView<typename scalarViewType::value_type,Kokkos::LayoutLeft,Kokkos::HostSpace>
vmat("Hgrad::Line::Cn::vmat", card, card),
work("Hgrad::Line::Cn::work", lwork),
ipiv("Hgrad::Line::Cn::ipiv", card);
const double alpha = 0.0, beta = 0.0;
Impl::Basis_HGRAD_LINE_Cn_FEM_JACOBI::
getValues<Kokkos::HostSpace::execution_space,Parameters::MaxNumPtsPerBasisEval>
(vmat, dofCoords, order, alpha, beta, OPERATOR_VALUE);
ordinal_type info = 0;
Teuchos::LAPACK<ordinal_type,typename scalarViewType::value_type> lapack;
lapack.GETRF(card, card,
vmat.data(), vmat.stride_1(),
(ordinal_type*)ipiv.data(),
&info);
INTREPID2_TEST_FOR_EXCEPTION( info != 0,
std::runtime_error ,
">>> ERROR: (Intrepid2::Basis_HGRAD_LINE_Cn_FEM) lapack.GETRF returns nonzero info." );
lapack.GETRI(card,
vmat.data(), vmat.stride_1(),
(ordinal_type*)ipiv.data(),
work.data(), lwork,
&info);
INTREPID2_TEST_FOR_EXCEPTION( info != 0,
std::runtime_error ,
">>> ERROR: (Intrepid2::Basis_HGRAD_LINE_Cn_FEM) lapack.GETRI returns nonzero info." );
// create host mirror
Kokkos::DynRankView<typename scalarViewType::value_type,typename SpT::array_layout,Kokkos::HostSpace>
vinv("Hgrad::Line::Cn::vinv", card, card);
for (ordinal_type i=0;i<card;++i)
for (ordinal_type j=0;j<card;++j)
vinv(i,j) = vmat(j,i);
this->vinv_ = Kokkos::create_mirror_view(typename SpT::memory_space(), vinv);
Kokkos::deep_copy(this->vinv_ , vinv);
// initialize tags
{
const bool is_vertex_included = (pointType != POINTTYPE_GAUSS);
// Basis-dependent initializations
const ordinal_type tagSize = 4; // size of DoF tag, i.e., number of fields in the tag
const ordinal_type posScDim = 0; // position in the tag, counting from 0, of the subcell dim
const ordinal_type posScOrd = 1; // position in the tag, counting from 0, of the subcell ordinal
const ordinal_type posDfOrd = 2; // position in the tag, counting from 0, of DoF ordinal relative to the subcell
ordinal_type tags[Parameters::MaxOrder+1][4];
// now we check the points for association
if (is_vertex_included) {
// lattice order
{
const auto v0 = 0;
tags[v0][0] = 0; // vertex dof
tags[v0][1] = 0; // vertex id
tags[v0][2] = 0; // local dof id
tags[v0][3] = 1; // total number of dofs in this vertex
const ordinal_type iend = card - 2;
for (ordinal_type i=0;i<iend;++i) {
const auto e = i + 1;
tags[e][0] = 1; // edge dof
tags[e][1] = 0; // edge id
tags[e][2] = i; // local dof id
tags[e][3] = iend; // total number of dofs in this edge
}
const auto v1 = card -1;
tags[v1][0] = 0; // vertex dof
tags[v1][1] = 1; // vertex id
tags[v1][2] = 0; // local dof id
tags[v1][3] = 1; // total number of dofs in this vertex
}
// topological order
// {
// tags[0][0] = 0; // vertex dof
// tags[0][1] = 0; // vertex id
// tags[0][2] = 0; // local dof id
// tags[0][3] = 1; // total number of dofs in this vertex
// tags[1][0] = 0; // vertex dof
// tags[1][1] = 1; // vertex id
// tags[1][2] = 0; // local dof id
// tags[1][3] = 1; // total number of dofs in this vertex
// const ordinal_type iend = card - 2;
// for (ordinal_type i=0;i<iend;++i) {
// const auto ii = i + 2;
// tags[ii][0] = 1; // edge dof
// tags[ii][1] = 0; // edge id
// tags[ii][2] = i; // local dof id
// tags[ii][3] = iend; // total number of dofs in this edge
// }
// }
} else {
for (ordinal_type i=0;i<card;++i) {
tags[i][0] = 1; // edge dof
tags[i][1] = 0; // edge id
tags[i][2] = i; // local dof id
tags[i][3] = card; // total number of dofs in this edge
}
}
ordinal_type_array_1d_host tagView(&tags[0][0], card*4);
// Basis-independent function sets tag and enum data in tagToOrdinal_ and ordinalToTag_ arrays:
// tags are constructed on host
this->setOrdinalTagData(this->tagToOrdinal_,
this->ordinalToTag_,
tagView,
this->basisCardinality_,
tagSize,
posScDim,
posScOrd,
posDfOrd);
}
}
void PBCmgr::maintain (const int_t step ,
const Field* P ,
const AuxField** Us ,
const AuxField** Uf ,
const bool timedep)
// ---------------------------------------------------------------------------
// Update storage for evaluation of high-order pressure boundary
// condition. Storage order for each edge represents a CCW traverse
// of element boundaries.
//
// If the velocity field varies in time on HOPB field boundaries
// (e.g. due to time-varying BCs) the local fluid acceleration will be
// estimated from input velocity fields by explicit extrapolation if
// timedep is true. This correction cannot be carried out at the
// first timestep, since the required extrapolation cannot be done.
// If the acceleration is known, (for example, a known reference frame
// acceleration) it is probably better to leave timedep unset, and to
// use PBCmgr::accelerate() to add in the accelerative term. Note
// also that since grad P is dotted with n, the unit outward normal,
// at a later stage, timedep only needs to be set if there are
// wall-normal accelerative terms. NB: The default value of timedep
// is 1.
//
// Field* master gives a list of pressure boundary conditions with
// which to traverse storage areas (note this assumes equal-order
// interpolations).
//
// No smoothing is done to high-order spatial derivatives computed here.
// ---------------------------------------------------------------------------
{
const real_t nu = Femlib::value ("KINVIS");
const real_t invDt = 1.0 / Femlib::value ("D_T");
const int_t nTime = Femlib::ivalue ("N_TIME");
const int_t nEdge = P -> _nbound;
const int_t nZ = P -> _nz;
const int_t nP = Geometry::nP();
const int_t base = Geometry::baseMode();
const int_t nMode = Geometry::nModeProc();
const int_t mLo = (Geometry::procID() == 0) ? 1 : 0;
const AuxField* Ux = Us[0];
const AuxField* Uy = Us[1];
const AuxField* Uz = (nZ > 1) ? Us[2] : 0;
const AuxField* Nx = Uf[0];
const AuxField* Ny = Uf[1];
const vector<Boundary*>& BC = P -> _bsys -> BCs (0);
register Boundary* B;
register int_t i, k, q;
int_t m, offset, skip, Je;
// -- Roll grad P storage area up, load new level of nonlinear terms Uf.
rollv (_Pnx, nTime);
rollv (_Pny, nTime);
for (i = 0; i < nEdge; i++) {
B = BC[i];
offset = B -> dOff ();
skip = B -> dSkip();
for (k = 0; k < nZ; k++) {
ROOTONLY if (k == 1) continue;
Veclib::copy (nP, Nx -> _plane[k] + offset, skip, _Pnx[0][i][k], 1);
Veclib::copy (nP, Ny -> _plane[k] + offset, skip, _Pny[0][i][k], 1);
// -- For cylindrical coordinates, N_ are radius-premultiplied. Cancel.
if (Geometry::cylindrical()) {
B -> divY (_Pnx[0][i][k]);
B -> divY (_Pny[0][i][k]);
}
}
}
// -- Add in -nu * curl curl u.
vector<real_t> work (5 * sqr(nP) + 7 * nP + Integration::OrderMax + 1);
real_t *UxRe, *UxIm, *UyRe, *UyIm, *UzRe, *UzIm, *tmp;
real_t* wrk = &work[0];
real_t* xr = wrk + 5*sqr(nP) + 3*nP;
real_t* xi = xr + nP;
real_t* yr = xi + nP;
real_t* yi = yr + nP;
real_t* alpha = yi + nP;
for (i = 0; i < nEdge; i++) {
B = BC[i];
offset = B -> dOff ();
skip = B -> dSkip();
ROOTONLY { // -- Deal with 2D/zero Fourier mode terms.
UxRe = Ux -> _plane[0];
UyRe = Uy -> _plane[0];
B -> curlCurl (0,UxRe,0,UyRe,0,0,0,xr,0,yr,0,wrk);
Blas::axpy (nP, -nu, xr, 1, _Pnx[0][i][0], 1);
Blas::axpy (nP, -nu, yr, 1, _Pny[0][i][0], 1);
}
for (m = mLo; m < nMode; m++) { // -- Higher modes.
UxRe = Ux -> _plane[2 * m] ;
UxIm = Ux -> _plane[2 * m + 1];
UyRe = Uy -> _plane[2 * m];
UyIm = Uy -> _plane[2 * m + 1];
UzRe = Uz -> _plane[2 * m];
UzIm = Uz -> _plane[2 * m + 1];
B -> curlCurl (m+base,UxRe,UxIm,UyRe,UyIm,UzRe,UzIm,xr,xi,yr,yi,wrk);
Blas::axpy (nP, -nu, xr, 1, _Pnx[0][i][2 * m], 1);
Blas::axpy (nP, -nu, xi, 1, _Pnx[0][i][2 * m + 1], 1);
Blas::axpy (nP, -nu, yr, 1, _Pny[0][i][2 * m], 1);
Blas::axpy (nP, -nu, yi, 1, _Pny[0][i][2 * m + 1], 1);
}
}
if (timedep) {
// -- Estimate -du / dt by backwards differentiation and add in.
if (step > 1) {
Je = min (step - 1, nTime);
tmp = xr;
Integration::StifflyStable (Je, alpha);
for (i = 0; i < nEdge; i++) {
B = BC[i];
offset = B -> dOff ();
skip = B -> dSkip();
for (k = 0; k < nZ; k++) {
ROOTONLY if (k == 1) continue;
Veclib::copy (nP, Ux -> _plane[k] + offset, skip, tmp, 1);
Blas::scal (nP, alpha[0], tmp, 1);
for (q = 0; q < Je; q++)
Blas::axpy (nP, alpha[q + 1], _Unx[q][i][k], 1, tmp, 1);
Blas::axpy (nP, -invDt, tmp, 1, _Pnx[0][i][k], 1);
Veclib::copy (nP, Uy -> _plane[k] + offset, skip, tmp, 1);
Blas::scal (nP, alpha[0], tmp, 1);
for (q = 0; q < Je; q++)
Blas::axpy (nP, alpha[q + 1], _Uny[q][i][k], 1, tmp, 1);
Blas::axpy (nP, -invDt, tmp, 1, _Pny[0][i][k], 1);
}
}
}
// -- Roll velocity storage area up, load new level.
rollv (_Unx, nTime);
rollv (_Uny, nTime);
for (i = 0; i < nEdge; i++) {
B = BC[i];
offset = B -> dOff ();
skip = B -> dSkip();
for (k = 0; k < nZ; k++) {
ROOTONLY if (k == 1) continue;
Veclib::copy (nP, Ux -> _plane[k] + offset, skip, _Unx[0][i][k], 1);
Veclib::copy (nP, Uy -> _plane[k] + offset, skip, _Uny[0][i][k], 1);
}
}
}
void AutomaticThread::start(const LockHolder&)
{
RELEASE_ASSERT(m_isRunning);
RefPtr<AutomaticThread> preserveThisForThread = this;
m_hasUnderlyingThread = true;
ThreadIdentifier thread = createThread(
"WTF::AutomaticThread",
[=] () {
if (verbose)
dataLog(RawPointer(this), ": Running automatic thread!\n");
RefPtr<AutomaticThread> thread = preserveThisForThread;
thread->threadDidStart();
if (!ASSERT_DISABLED) {
LockHolder locker(*m_lock);
ASSERT(m_condition->contains(locker, this));
}
auto stopImpl = [&] (const LockHolder& locker) {
thread->threadIsStopping(locker);
thread->m_hasUnderlyingThread = false;
};
auto stopPermanently = [&] (const LockHolder& locker) {
m_isRunning = false;
m_isRunningCondition.notifyAll();
stopImpl(locker);
};
auto stopForTimeout = [&] (const LockHolder& locker) {
stopImpl(locker);
};
for (;;) {
{
LockHolder locker(*m_lock);
for (;;) {
PollResult result = poll(locker);
if (result == PollResult::Work)
break;
if (result == PollResult::Stop)
return stopPermanently(locker);
RELEASE_ASSERT(result == PollResult::Wait);
// Shut the thread down after one second.
m_isWaiting = true;
bool awokenByNotify =
m_waitCondition.waitFor(*m_lock, 1_s);
if (verbose && !awokenByNotify && !m_isWaiting)
dataLog(RawPointer(this), ": waitFor timed out, but notified via m_isWaiting flag!\n");
if (m_isWaiting) {
m_isWaiting = false;
if (verbose)
dataLog(RawPointer(this), ": Going to sleep!\n");
// It's important that we don't release the lock until we have completely
// indicated that the thread is kaput. Otherwise we'll have a a notify
// race that manifests as a deadlock on VM shutdown.
return stopForTimeout(locker);
}
}
}
WorkResult result = work();
if (result == WorkResult::Stop) {
LockHolder locker(*m_lock);
return stopPermanently(locker);
}
RELEASE_ASSERT(result == WorkResult::Continue);
}
});
detachThread(thread);
}
void
o3d3xx::FrameGrabber::Run()
{
boost::asio::io_service::work work(this->io_service_);
//
// setup the camera for image acquistion
//
std::string cam_ip;
int cam_port;
try
{
cam_ip = this->cam_->GetIP();
cam_port = std::stoi(this->cam_->GetParameter("PcicTcpPort"));
}
catch (const o3d3xx::error_t& ex)
{
LOG(ERROR) << "Could not get IP/Port of the camera: "
<< ex.what();
return;
}
LOG(INFO) << "Camera connection info: ip=" << cam_ip
<< ", port=" << cam_port;
try
{
this->cam_->RequestSession();
this->cam_->SetOperatingMode(o3d3xx::Camera::operating_mode::RUN);
this->cam_->CancelSession();
}
catch (const o3d3xx::error_t& ex)
{
LOG(ERROR) << "Failed to setup camera for image acquisition: "
<< ex.what();
return;
}
//
// init the asio structures
//
boost::asio::ip::tcp::socket sock(this->io_service_);
boost::asio::ip::tcp::endpoint endpoint(
boost::asio::ip::address::from_string(cam_ip), cam_port);
//
// Forward declare our read handlers (because they need to call
// eachother).
//
o3d3xx::FrameGrabber::WriteHandler result_schema_write_handler;
o3d3xx::FrameGrabber::ReadHandler ticket_handler;
o3d3xx::FrameGrabber::ReadHandler image_handler;
//
// image data callback
//
std::size_t bytes_read = 0;
std::size_t buff_sz = 0; // bytes
image_handler =
[&, this]
(const boost::system::error_code& ec, std::size_t bytes_transferred)
{
if (ec) { throw o3d3xx::error_t(ec.value()); }
bytes_read += bytes_transferred;
//DLOG(INFO) << "Read " << bytes_read << " image bytes of "
// << buff_sz;
if (bytes_read == buff_sz)
{
DLOG(INFO) << "Got full image!";
bytes_read = 0;
// 1. verify the data
if (o3d3xx::verify_image_buffer(this->back_buffer_))
{
DLOG(INFO) << "Image OK";
// 2. move the data to the front buffer in O(1) time complexity
this->front_buffer_mutex_.lock();
this->back_buffer_.swap(this->front_buffer_);
this->front_buffer_mutex_.unlock();
// 3. notify waiting clients
this->front_buffer_cv_.notify_all();
}
else
{
LOG(WARNING) << "Bad image!";
}
// read another ticket
sock.async_read_some(
boost::asio::buffer(this->ticket_buffer_.data(),
o3d3xx::IMG_TICKET_SZ),
ticket_handler);
return;
}
sock.async_read_some(
boost::asio::buffer(&this->back_buffer_[bytes_read],
buff_sz - bytes_read),
image_handler);
};
//
// ticket callback
//
std::size_t ticket_bytes_read = 0;
std::size_t ticket_buff_sz = o3d3xx::IMG_TICKET_SZ;
this->ticket_buffer_.resize(ticket_buff_sz);
ticket_handler =
[&, this]
(const boost::system::error_code& ec, std::size_t bytes_transferred)
{
if (ec) { throw o3d3xx::error_t(ec.value()); }
ticket_bytes_read += bytes_transferred;
DLOG(INFO) << "Read " << ticket_bytes_read
<< " ticket bytes of " << ticket_buff_sz;
if (ticket_bytes_read == ticket_buff_sz)
{
DLOG(INFO) << "Got full ticket!";
ticket_bytes_read = 0;
if (o3d3xx::verify_ticket_buffer(this->ticket_buffer_))
{
DLOG(INFO) << "Ticket OK";
buff_sz = o3d3xx::get_image_buffer_size(this->ticket_buffer_);
DLOG(INFO) << "Image buffer size: " << buff_sz;
this->back_buffer_.resize(buff_sz);
sock.async_read_some(
boost::asio::buffer(this->back_buffer_.data(),
buff_sz),
image_handler);
return;
}
LOG(WARNING) << "Bad ticket!";
}
sock.async_read_some(
boost::asio::buffer(&this->ticket_buffer_[ticket_bytes_read],
ticket_buff_sz - ticket_bytes_read),
ticket_handler);
};
//
// Check that our request to set the result schema was successful
//
result_schema_write_handler =
[&, this]
(const boost::system::error_code& ec, std::size_t bytes_transferred)
{
if (ec) { throw o3d3xx::error_t(ec.value()); }
DLOG(INFO) << "Wrote: " << bytes_transferred << " bytes to camera";
std::size_t c_buff_sz = 16 + 7;
std::uint8_t resp_buff[c_buff_sz];
std::size_t resp_bytes_read =
boost::asio::read(sock, boost::asio::buffer(resp_buff, c_buff_sz));
if (resp_bytes_read < c_buff_sz)
{
LOG(ERROR) << "Error getting c_command response!";
throw o3d3xx::error_t(O3D3XX_IO_ERROR);
}
if (resp_buff[20] != '*')
{
LOG(ERROR) << "Got back bad response from camera: '"
<< resp_buff[20] << "'";
throw o3d3xx::error_t(O3D3XX_PCIC_BAD_REPLY);
}
sock.async_read_some(
boost::asio::buffer(
this->ticket_buffer_.data(), ticket_buff_sz),
ticket_handler);
};
//
// connect to the sensor and start streaming in image data
//
try
{
sock.async_connect(endpoint,
[&, this]
(const boost::system::error_code& ec)
{
if (ec) { throw o3d3xx::error_t(ec.value()); }
boost::asio::async_write(
sock,
boost::asio::buffer(this->schema_buffer_.data(),
this->schema_buffer_.size()),
result_schema_write_handler);
});
this->io_service_.run();
}
catch (const std::exception& ex)
{
//
// In here we should discern why the exception with thrown.
//
// Special case the "Stop()" request from the control thread
//
LOG(WARNING) << "Exception: " << ex.what();
}
LOG(INFO) << "Framegrabber thread done.";
}
int main(int argc, char *argv[])
{
const char *url;
int i, threads;
pthread_t *t;
int *args;
lList *answer_list = NULL;
lListElem *spooling_context;
DENTER_MAIN(TOP_LAYER, "test_berkeleydb_mt");
/* parse commandline parameters */
if (argc < 3) {
ERROR((SGE_EVENT, "usage: test_berkeleydb_mt <url> <threads> [<delay>]\n"));
ERROR((SGE_EVENT, " <url> = path or host:database\n"));
ERROR((SGE_EVENT, " <threads> = number of threads\n"));
ERROR((SGE_EVENT, " <delay> = delay after writing [ms]\n"));
SGE_EXIT(NULL, 1);
}
url = argv[1];
threads = atoi(argv[2]);
if (argc > 3) {
delay = atoi(argv[3]);
}
/* allocate memory for pthreads and arguments */
t = (pthread_t *)malloc(threads * sizeof(pthread_t));
args = (int *)malloc(threads * sizeof(int));
DPRINTF(("writing to database %s from %d threads\n", url, threads));
/* initialize spooling */
spooling_context = spool_create_dynamic_context(&answer_list, NULL, url, NULL);
answer_list_output(&answer_list);
if (spooling_context == NULL) {
SGE_EXIT(NULL, EXIT_FAILURE);
}
spool_set_default_context(spooling_context);
if (!spool_startup_context(&answer_list, spooling_context, true)) {
answer_list_output(&answer_list);
SGE_EXIT(NULL, EXIT_FAILURE);
}
answer_list_output(&answer_list);
/* let n threads to parallel spooling */
for (i = 0; i < threads; i++) {
args[i] = i + 1;
pthread_create(&(t[i]), NULL, work, (void*)(&args[i]));
}
/* also work in current thread */
work((void *)0);
/* wait for termination of all threads */
for (i = 0; i < threads; i++) {
pthread_join(t[i], NULL);
}
/* shutdown spooling */
spool_shutdown_context(&answer_list, spooling_context);
answer_list_output(&answer_list);
sge_free(&t);
DEXIT;
return EXIT_SUCCESS;
}
int sumNumbers(TreeNode *root) {
int ret = 0;
//travel all the path
work(root, ret, 0);
return ret;
}
magma_int_t magma_ztrevc3(
magma_side_t side, magma_vec_t howmany,
magma_int_t *select, // logical in Fortran
magma_int_t n,
magmaDoubleComplex *T, magma_int_t ldt,
magmaDoubleComplex *VL, magma_int_t ldvl,
magmaDoubleComplex *VR, magma_int_t ldvr,
magma_int_t mm, magma_int_t *mout,
magmaDoubleComplex *work, magma_int_t lwork,
double *rwork, magma_int_t *info )
{
#define T(i,j) ( T + (i) + (j)*ldt )
#define VL(i,j) (VL + (i) + (j)*ldvl)
#define VR(i,j) (VR + (i) + (j)*ldvr)
#define work(i,j) (work + (i) + (j)*n)
// .. Parameters ..
const magmaDoubleComplex c_zero = MAGMA_Z_ZERO;
const magmaDoubleComplex c_one = MAGMA_Z_ONE;
const magma_int_t nbmin = 16, nbmax = 128;
const magma_int_t ione = 1;
// .. Local Scalars ..
magma_int_t allv, bothv, leftv, over, rightv, somev;
magma_int_t i, ii, is, j, k, ki, iv, n2, nb, nb2, version;
double ovfl, remax, scale, smin, smlnum, ulp, unfl;
// Decode and test the input parameters
bothv = (side == MagmaBothSides);
rightv = (side == MagmaRight) || bothv;
leftv = (side == MagmaLeft ) || bothv;
allv = (howmany == MagmaAllVec);
over = (howmany == MagmaBacktransVec);
somev = (howmany == MagmaSomeVec);
// Set mout to the number of columns required to store the selected
// eigenvectors.
if ( somev ) {
*mout = 0;
for( j=0; j < n; ++j ) {
if ( select[j] ) {
*mout += 1;
}
}
}
else {
*mout = n;
}
*info = 0;
if ( ! rightv && ! leftv )
*info = -1;
else if ( ! allv && ! over && ! somev )
*info = -2;
else if ( n < 0 )
*info = -4;
else if ( ldt < max( 1, n ) )
*info = -6;
else if ( ldvl < 1 || ( leftv && ldvl < n ) )
*info = -8;
else if ( ldvr < 1 || ( rightv && ldvr < n ) )
*info = -10;
else if ( mm < *mout )
*info = -11;
else if ( lwork < max( 1, 2*n ) )
*info = -14;
if ( *info != 0 ) {
magma_xerbla( __func__, -(*info) );
return *info;
}
// Quick return if possible.
if ( n == 0 ) {
return *info;
}
// Use blocked version (2) if sufficient workspace.
// Requires 1 vector to save diagonal elements, and 2*nb vectors for x and Q*x.
// (Compared to dtrevc3, rwork stores 1-norms.)
// Zero-out the workspace to avoid potential NaN propagation.
nb = 2;
if ( lwork >= n + 2*n*nbmin ) {
version = 2;
nb = (lwork - n) / (2*n);
nb = min( nb, nbmax );
nb2 = 1 + 2*nb;
lapackf77_zlaset( "F", &n, &nb2, &c_zero, &c_zero, work, &n );
}
else {
version = 1;
}
// Set the constants to control overflow.
unfl = lapackf77_dlamch( "Safe minimum" );
ovfl = 1. / unfl;
lapackf77_dlabad( &unfl, &ovfl );
ulp = lapackf77_dlamch( "Precision" );
smlnum = unfl*( n / ulp );
// Store the diagonal elements of T in working array work.
for( i=0; i < n; ++i ) {
*work(i,0) = *T(i,i);
}
// Compute 1-norm of each column of strictly upper triangular
// part of T to control overflow in triangular solver.
rwork[0] = 0.;
for( j=1; j < n; ++j ) {
rwork[j] = cblas_dzasum( j, T(0,j), ione );
}
magma_timer_t time_total=0, time_trsv=0, time_gemm=0, time_gemv=0, time_trsv_sum=0, time_gemm_sum=0, time_gemv_sum=0;
timer_start( time_total );
if ( rightv ) {
// ============================================================
// Compute right eigenvectors.
// iv is index of column in current block.
// Non-blocked version always uses iv=1;
// blocked version starts with iv=nb, goes down to 1.
// (Note the "0-th" column is used to store the original diagonal.)
iv = 1;
if ( version == 2 ) {
iv = nb;
}
timer_start( time_trsv );
is = *mout - 1;
for( ki=n-1; ki >= 0; --ki ) {
if ( somev ) {
if ( ! select[ki] ) {
continue;
}
}
smin = max( ulp*( MAGMA_Z_ABS1( *T(ki,ki) ) ), smlnum );
// --------------------------------------------------------
// Complex right eigenvector
*work(ki,iv) = c_one;
// Form right-hand side.
for( k=0; k < ki; ++k ) {
*work(k,iv) = -(*T(k,ki));
}
// Solve upper triangular system:
// [ T(1:ki-1,1:ki-1) - T(ki,ki) ]*X = scale*work.
for( k=0; k < ki; ++k ) {
*T(k,k) -= *T(ki,ki);
if ( MAGMA_Z_ABS1( *T(k,k) ) < smin ) {
*T(k,k) = MAGMA_Z_MAKE( smin, 0. );
}
}
if ( ki > 0 ) {
lapackf77_zlatrs( "Upper", "No transpose", "Non-unit", "Y",
&ki, T, &ldt,
work(0,iv), &scale, rwork, info );
*work(ki,iv) = MAGMA_Z_MAKE( scale, 0. );
}
// Copy the vector x or Q*x to VR and normalize.
if ( ! over ) {
// ------------------------------
// no back-transform: copy x to VR and normalize
n2 = ki+1;
blasf77_zcopy( &n2, work(0,iv), &ione, VR(0,is), &ione );
ii = blasf77_izamax( &n2, VR(0,is), &ione ) - 1;
remax = 1. / MAGMA_Z_ABS1( *VR(ii,is) );
blasf77_zdscal( &n2, &remax, VR(0,is), &ione );
for( k=ki+1; k < n; ++k ) {
*VR(k,is) = c_zero;
}
}
else if ( version == 1 ) {
// ------------------------------
// version 1: back-transform each vector with GEMV, Q*x.
time_trsv_sum += timer_stop( time_trsv );
timer_start( time_gemv );
if ( ki > 0 ) {
blasf77_zgemv( "n", &n, &ki, &c_one,
VR, &ldvr,
work(0, iv), &ione,
work(ki,iv), VR(0,ki), &ione );
}
time_gemv_sum += timer_stop( time_gemv );
ii = blasf77_izamax( &n, VR(0,ki), &ione ) - 1;
remax = 1. / MAGMA_Z_ABS1( *VR(ii,ki) );
blasf77_zdscal( &n, &remax, VR(0,ki), &ione );
timer_start( time_trsv );
}
else if ( version == 2 ) {
// ------------------------------
// version 2: back-transform block of vectors with GEMM
// zero out below vector
for( k=ki+1; k < n; ++k ) {
*work(k,iv) = c_zero;
}
// Columns iv:nb of work are valid vectors.
// When the number of vectors stored reaches nb,
// or if this was last vector, do the GEMM
if ( (iv == 1) || (ki == 0) ) {
time_trsv_sum += timer_stop( time_trsv );
timer_start( time_gemm );
nb2 = nb-iv+1;
n2 = ki+nb-iv+1;
blasf77_zgemm( "n", "n", &n, &nb2, &n2, &c_one,
VR, &ldvr,
work(0,iv ), &n, &c_zero,
work(0,nb+iv), &n );
time_gemm_sum += timer_stop( time_gemm );
// normalize vectors
// TODO if somev, should copy vectors individually to correct location.
for( k = iv; k <= nb; ++k ) {
ii = blasf77_izamax( &n, work(0,nb+k), &ione ) - 1;
remax = 1. / MAGMA_Z_ABS1( *work(ii,nb+k) );
blasf77_zdscal( &n, &remax, work(0,nb+k), &ione );
}
lapackf77_zlacpy( "F", &n, &nb2, work(0,nb+iv), &n, VR(0,ki), &ldvr );
iv = nb;
timer_start( time_trsv );
}
else {
iv -= 1;
}
} // blocked back-transform
// Restore the original diagonal elements of T.
for( k=0; k <= ki - 1; ++k ) {
*T(k,k) = *work(k,0);
}
is -= 1;
}
}
timer_stop( time_trsv );
timer_stop( time_total );
timer_printf( "trevc trsv %.4f, gemm %.4f, gemv %.4f, total %.4f\n",
time_trsv_sum, time_gemm_sum, time_gemv_sum, time_total );
if ( leftv ) {
// ============================================================
// Compute left eigenvectors.
// iv is index of column in current block.
// Non-blocked version always uses iv=1;
// blocked version starts with iv=1, goes up to nb.
// (Note the "0-th" column is used to store the original diagonal.)
iv = 1;
is = 0;
for( ki=0; ki < n; ++ki ) {
if ( somev ) {
if ( ! select[ki] ) {
continue;
}
}
smin = max( ulp*MAGMA_Z_ABS1( *T(ki,ki) ), smlnum );
// --------------------------------------------------------
// Complex left eigenvector
*work(ki,iv) = c_one;
// Form right-hand side.
for( k = ki + 1; k < n; ++k ) {
*work(k,iv) = -MAGMA_Z_CNJG( *T(ki,k) );
}
// Solve conjugate-transposed triangular system:
// [ T(ki+1:n,ki+1:n) - T(ki,ki) ]**H * X = scale*work.
for( k = ki + 1; k < n; ++k ) {
*T(k,k) -= *T(ki,ki);
if ( MAGMA_Z_ABS1( *T(k,k) ) < smin ) {
*T(k,k) = MAGMA_Z_MAKE( smin, 0. );
}
}
if ( ki < n-1 ) {
n2 = n-ki-1;
lapackf77_zlatrs( "Upper", "Conjugate transpose", "Non-unit", "Y",
&n2, T(ki+1,ki+1), &ldt,
work(ki+1,iv), &scale, rwork, info );
*work(ki,iv) = MAGMA_Z_MAKE( scale, 0. );
}
// Copy the vector x or Q*x to VL and normalize.
if ( ! over ) {
// ------------------------------
// no back-transform: copy x to VL and normalize
n2 = n-ki;
blasf77_zcopy( &n2, work(ki,iv), &ione, VL(ki,is), &ione );
ii = blasf77_izamax( &n2, VL(ki,is), &ione ) + ki - 1;
remax = 1. / MAGMA_Z_ABS1( *VL(ii,is) );
blasf77_zdscal( &n2, &remax, VL(ki,is), &ione );
for( k=0; k < ki; ++k ) {
*VL(k,is) = c_zero;
}
}
else if ( version == 1 ) {
// ------------------------------
// version 1: back-transform each vector with GEMV, Q*x.
if ( ki < n-1 ) {
n2 = n-ki-1;
blasf77_zgemv( "n", &n, &n2, &c_one,
VL(0,ki+1), &ldvl,
work(ki+1,iv), &ione,
work(ki, iv), VL(0,ki), &ione );
}
ii = blasf77_izamax( &n, VL(0,ki), &ione ) - 1;
remax = 1. / MAGMA_Z_ABS1( *VL(ii,ki) );
blasf77_zdscal( &n, &remax, VL(0,ki), &ione );
}
else if ( version == 2 ) {
// ------------------------------
// version 2: back-transform block of vectors with GEMM
// zero out above vector
// could go from (ki+1)-NV+1 to ki
for( k=0; k < ki; ++k ) {
*work(k,iv) = c_zero;
}
// Columns 1:iv of work are valid vectors.
// When the number of vectors stored reaches nb,
// or if this was last vector, do the GEMM
if ( (iv == nb) || (ki == n-1) ) {
n2 = n-(ki+1)+iv;
blasf77_zgemm( "n", "n", &n, &iv, &n2, &c_one,
VL(0,ki-iv+1), &ldvl,
work(ki-iv+1,1 ), &n, &c_zero,
work(0, nb+1), &n );
// normalize vectors
for( k=1; k <= iv; ++k ) {
ii = blasf77_izamax( &n, work(0,nb+k), &ione ) - 1;
remax = 1. / MAGMA_Z_ABS1( *work(ii,nb+k) );
blasf77_zdscal( &n, &remax, work(0,nb+k), &ione );
}
lapackf77_zlacpy( "F", &n, &iv, work(0,nb+1), &n, VL(0,ki-iv+1), &ldvl );
iv = 1;
}
else {
iv += 1;
}
} // blocked back-transform
// Restore the original diagonal elements of T.
for( k = ki + 1; k < n; ++k ) {
*T(k,k) = *work(k,0);
}
is += 1;
}
}
return *info;
} // End of ZTREVC
void ISVDMultiCD::makePass() {
Epetra_LAPACK lapack;
Epetra_BLAS blas;
bool firstPass = (curRank_ == 0);
const int numCols = A_->NumVectors();
TEUCHOS_TEST_FOR_EXCEPTION( !firstPass && (numProc_ != numCols), std::logic_error,
"RBGen::ISVDMultiCD::makePass(): after first pass, numProc should be numCols");
// compute W = I - Z T Z^T from current V_
Teuchos::RCP<Epetra_MultiVector> lclAZT, lclZ;
double *Z_A, *AZT_A;
int Z_LDA, AZT_LDA;
int oldRank = 0;
double Rerr = 0.0;
if (!firstPass) {
// copy V_ into workZ_
lclAZT = Teuchos::rcp( new Epetra_MultiVector(::View,*workAZT_,0,curRank_) );
lclZ = Teuchos::rcp( new Epetra_MultiVector(::View,*workZ_,0,curRank_) );
{
Epetra_MultiVector lclV(::View,*V_,0,curRank_);
*lclZ = lclV;
}
// compute the Householder QR factorization of the current right basis
// Vhat = W*R
int info, lwork = curRank_;
std::vector<double> tau(curRank_), work(lwork);
info = lclZ->ExtractView(&Z_A,&Z_LDA);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0, std::logic_error,
"RBGen::ISVDMultiCD::makePass(): error calling ExtractView on Epetra_MultiVector Z.");
lapack.GEQRF(numCols,curRank_,Z_A,Z_LDA,&tau[0],&work[0],lwork,&info);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0, std::logic_error,
"RBGen::ISVDMultiCD::makePass(): error calling GEQRF on current right basis while constructing next pass coefficients.");
if (debug_) {
// we just took the QR factorization of a set of orthonormal vectors
// they should have an R factor which is diagonal, with unit elements (\pm 1)
// check it
Rerr = 0.0;
for (int j=0; j<curRank_; j++) {
for (int i=0; i<j; i++) {
Rerr += abs(Z_A[j*Z_LDA+i]);
}
Rerr += abs(abs(Z_A[j*Z_LDA+j]) - 1.0);
}
}
// compute the block representation
// W = I - Z T Z^T
lapack.LARFT('F','C',numCols,curRank_,Z_A,Z_LDA,&tau[0],workT_->A(),workT_->LDA());
// LARFT left upper tri block of Z unchanged
// note: it should currently contain R factor of V_, which is very close to
// diag(\pm 1, ..., \pm 1)
//
// we need to set it to:
// [1 0 0 ... 0]
// [ 1 0 ... 0]
// [ .... ]
// [ 1]
//
// see documentation for LARFT
//
for (int j=0; j<curRank_; j++) {
Z_A[j*Z_LDA+j] = 1.0;
for (int i=0; i<j; i++) {
Z_A[j*Z_LDA+i] = 0.0;
}
}
// compute part of A W: A Z T
// put this in workAZT_
// first, A Z
info = lclAZT->Multiply('N','N',1.0,*A_,*lclZ,0.0);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0,std::logic_error,
"RBGen::ISVDMultiCD::makePass(): Error calling Epetra_MultiVector::Multiply() for A*Z");
// second, (A Z) T (in situ, as T is upper triangular)
info = lclAZT->ExtractView(&AZT_A,&AZT_LDA);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0, std::logic_error,
"RBGen::ISVDMultiCD::makePass(): error calling ExtractView on Epetra_MultiVector AZ.");
blas.TRMM('R','U','N','N',numCols,curRank_,1.0,workT_->A(),workT_->LDA(),AZT_A,AZT_LDA);
// save oldRank: it tells us the width of Z
oldRank = curRank_;
curRank_ = 0;
numProc_ = 0;
}
else { // firstPass == true
curRank_ = 0;
numProc_ = 0;
}
while (numProc_ < numCols) {
//
// determine lup
//
// want lup >= lmin
// lup <= lmax
// need lup <= numCols - numProc
// lup <= maxBasisSize - curRank
//
int lup;
if (curRank_ == 0) {
// first step uses startRank_
// this is not affected by lmin,lmax
lup = startRank_;
}
else {
// this value minimizes overall complexity, assuming fixed rank
lup = (int)(curRank_ / Teuchos::ScalarTraits<double>::squareroot(2.0));
// contrain to [lmin,lmax]
lup = (lup < lmin_ ? lmin_ : lup);
lup = (lup > lmax_ ? lmax_ : lup);
}
//
// now cap lup via maxBasisSize and the available data
// these caps apply to all lup, as a result of memory and data constraints
//
// available data
lup = (lup > numCols - numProc_ ? numCols - numProc_ : lup);
// available memory
lup = (lup > maxBasisSize_ - curRank_ ? maxBasisSize_ - curRank_ : lup);
// get view of new vectors
{
const Epetra_MultiVector Aplus(::View,*A_,numProc_,lup);
Epetra_MultiVector Unew(::View,*U_,curRank_,lup);
// put them in U
if (firstPass) {
// new vectors are just Aplus
Unew = Aplus;
}
else {
// new vectors are Aplus - (A Z T) Z_i^T
// specifically, Aplus - (A Z T) Z(numProc:numProc+lup-1,1:oldRank)^T
Epetra_LocalMap lclmap(lup,0,A_->Comm());
Epetra_MultiVector Zi(::View,lclmap,&Z_A[numProc_],Z_LDA,oldRank);
Unew = Aplus;
int info = Unew.Multiply('N','T',-1.0,*lclAZT,Zi,1.0);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0,std::logic_error,
"RBGen::ISVDMultiCD::makePass(): Error calling Epetra_MultiVector::Multiply() for A*Wi");
}
}
// perform the incremental step
incStep(lup);
}
// compute W V = V - Z T Z^T V
// Z^T V is oldRank x curRank
// T Z^T V is oldRank x curRank
// we need T Z^T V in a local Epetra_MultiVector
if (!firstPass) {
Teuchos::RCP<Epetra_MultiVector> lclV;
double *TZTV_A;
int TZTV_LDA;
int info;
Epetra_LocalMap lclmap(oldRank,0,A_->Comm());
// get pointer to current V
lclV = Teuchos::rcp( new Epetra_MultiVector(::View,*V_,0,curRank_) );
// create space for T Z^T V
Epetra_MultiVector TZTV(lclmap,curRank_,false);
// multiply Z^T V
info = TZTV.Multiply('T','N',1.0,*lclZ,*lclV,0.0);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0,std::logic_error,
"RBGen::ISVDMultiCD::makePass(): Error calling Epetra_MultiVector::Multiply() for Z^T V.");
// get pointer to data in Z^T V
info = TZTV.ExtractView(&TZTV_A,&TZTV_LDA);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0, std::logic_error,
"RBGen::ISVDMultiCD::makePass(): error calling ExtractView on Epetra_MultiVector TZTV.");
// multiply T (Z^T V)
blas.TRMM('L','U','N','N',oldRank,curRank_,1.0,workT_->A(),workT_->LDA(),TZTV_A,TZTV_LDA);
// multiply V - Z (T Z^T V)
info = lclV->Multiply('N','N',-1.0,*lclZ,TZTV,1.0);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0,std::logic_error,
"RBGen::ISVDMultiCD::makePass(): Error calling Epetra_MultiVector::Multiply() for W V.");
}
//
// compute the new residuals
// we know that A V = U S
// if, in addition, A^T U = V S, then have singular subspaces
// check residuals A^T U - V S, scaling the i-th column by sigma[i]
//
{
// make these static, because makePass() will be likely be called again
static Epetra_LocalMap lclmap(A_->NumVectors(),0,A_->Comm());
static Epetra_MultiVector ATU(lclmap,maxBasisSize_,false);
// we know that A V = U S
// if, in addition, A^T U = V S, then have singular subspaces
// check residuals A^T U - V S, scaling the i-th column by sigma[i]
Epetra_MultiVector ATUlcl(::View,ATU,0,curRank_);
Epetra_MultiVector Ulcl(::View,*U_,0,curRank_);
Epetra_MultiVector Vlcl(::View,*V_,0,curRank_);
// compute A^T U
int info = ATUlcl.Multiply('T','N',1.0,*A_,Ulcl,0.0);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0, std::logic_error,
"RBGen::ISVDMultiCD::makePass(): Error calling Epetra_MultiVector::Multiply for A^T U.");
Epetra_LocalMap rankmap(curRank_,0,A_->Comm());
Epetra_MultiVector S(rankmap,curRank_,true);
for (int i=0; i<curRank_; i++) {
S[i][i] = sigma_[i];
}
// subtract V S from A^T U
info = ATUlcl.Multiply('N','N',-1.0,Vlcl,S,1.0);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0, std::logic_error,
"RBGen::ISVDMultiCD::computeBasis(): Error calling Epetra_MultiVector::Multiply for V S.");
resNorms_.resize(curRank_);
ATUlcl.Norm2(&resNorms_[0]);
// scale by sigmas
for (int i=0; i<curRank_; i++) {
if (sigma_[i] != 0.0) {
resNorms_[i] /= sigma_[i];
}
}
}
// debugging checks
std::vector<double> errnorms(curRank_);
if (debug_) {
int info;
// Check that A V = U Sigma
// get pointers to current U and V, create workspace for A V - U Sigma
Epetra_MultiVector work(U_->Map(),curRank_,false),
curU(::View,*U_,0,curRank_),
curV(::View,*V_,0,curRank_);
// create local MV for sigmas
Epetra_LocalMap lclmap(curRank_,0,A_->Comm());
Epetra_MultiVector curS(lclmap,curRank_,true);
for (int i=0; i<curRank_; i++) {
curS[i][i] = sigma_[i];
}
info = work.Multiply('N','N',1.0,curU,curS,0.0);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0,std::logic_error,
"RBGen::ISVDMultiCD::makePass(): Error calling Epetra_MultiVector::Multiply() for debugging U S.");
info = work.Multiply('N','N',-1.0,*A_,curV,1.0);
TEUCHOS_TEST_FOR_EXCEPTION(info != 0,std::logic_error,
"RBGen::ISVDMultiCD::makePass(): Error calling Epetra_MultiVector::Multiply() for debugging U S - A V.");
work.Norm2(&errnorms[0]);
for (int i=0; i<curRank_; i++) {
if (sigma_[i] != 0.0) {
errnorms[i] /= sigma_[i];
}
}
}
// update pass counter
curNumPasses_++;
// print out some info
const Epetra_Comm *comm = &A_->Comm();
if (comm->MyPID() == 0 && verbLevel_ >= 1) {
std::cout
<< "------------- ISVDMultiCD::makePass() -----------" << std::endl
<< "| Number of passes: " << curNumPasses_ << std::endl
<< "| Current rank: " << curRank_ << std::endl
<< "| Current sigmas: " << std::endl;
for (int i=0; i<curRank_; i++) {
std::cout << "| " << sigma_[i] << std::endl;
}
if (debug_) {
std::cout << "|DBG US-AV norms: " << std::endl;
for (int i=0; i<curRank_; i++) {
std::cout << "|DBG " << errnorms[i] << std::endl;
}
if (!firstPass) {
std::cout << "|DBG R-I norm: " << Rerr << std::endl;
}
}
}
return;
}
// a new work was created for the current executing node
inline void new_work_self(db::node *node, db::simple_tuple *stpl, const process::work_modifier mod = process::mods::NOTHING)
{
process::work work(node, stpl, process::mods::LOCAL_TUPLE | mod);
new_work(node, work);
}
// DGELSS computes minimum norm solution to a real linear
// least squares problem: Minimize 2-norm(| b - A*x |).
// using the singular value decomposition (SVD) of A.
// A is an M-by-N matrix which may be rank-deficient.
//---------------------------------------------------------
void umSOLVE_LS(const DMat& mat, const DMat& B, DMat& X)
//---------------------------------------------------------
{
if (!mat.ok()) {umWARNING("umSOLVE_LS()", "system is empty"); return;}
DMat A(mat); // work with copy of input.
int rows=A.num_rows(), cols=A.num_cols(), mmn=A.min_mn();
int LDB=A.max_mn(), NRHS=B.num_cols();
if (rows!=B.num_rows()) {umERROR("umSOLVE_LS(A,B)", "Inconsistant matrix sizes.");}
DVec s(mmn); // allocate array for singular values
// X must be big enough to store various results.
// Resize X so that its leading dimension = max(M,N),
// then load the set of right hand sides.
X.resize(LDB,NRHS, true, 0.0);
for (int j=1; j<=NRHS; ++j) // loop across colums
for (int i=1; i<=rows; ++i) // loop down rows
X(i,j) = B(i,j);
// RCOND is used to determine the effective rank of A.
// Singular values S(i) <= RCOND*S(1) are treated as zero.
// If RCOND < 0, machine precision is used instead.
//double rcond = 1.0 / 1.0e16;
double rcond = -1.0;
// NBN: ACML does not use the work vector.
int mnLo=A.min_mn(), mnHi=A.max_mn(), rank=1, info=1;
int lwork = 10*mnLo + std::max(2*mnLo, std::max(mnHi, NRHS));
DVec work(lwork);
// Solve the system
GELSS (rows, cols, NRHS, A.data(), rows, X.data(), LDB, s.data(), rcond, rank, work.data(), lwork, info);
//---------------------------------------------
// Report:
//---------------------------------------------
if (info == 0) {
umLOG(1, "umSOLVE_LS reports successful LS-solution."
"\nRCOND = %0.6e, "
"\nOptimal length of work array was %d\n", rcond, lwork);
}
else
{
if (info < 0) {
X = 0.0;
umERROR("umSOLVE_LS(DMat&, DMat&)",
"Error in input argument (%d)\nNo solution or error bounds computed.", -info);
} else if (info > 0) {
X = 0.0;
umERROR("umSOLVE_LS(DMat&, DMat&)",
"\nThe algorithm for computing the SVD failed to converge.\n"
"\n%d off-diagonal elements of an intermediate "
"\nbidiagonal form did not converge to zero.\n "
"\nRCOND = %0.6e, "
"\nOptimal length of work array was %d.\n", info, rcond, lwork);
}
}
}
int main()
{
freopen("input.in","r",stdin);
work();
return 0;
}
//************************************************************************
//Обращение матрицы, заданной LU-разложением
//
//Входные параметры:
// A - LU-разложение матрицы (результат работы подпрограммы
// LUDecomposition).
// Pivots - таблица перестановок, произведенных в ходе LU-разложения.
// (результат работы подпрограммы LUDecomposition).
// N - размерность матрицы
//
//Выходные параметры:
// A - матрица, обратная к исходной. Массив с нумерацией
// элементов [1..N, 1..N]
//
//Результат:
// True, если исходная матрица невырожденная.
// False, если исходная матрица вырожденная.
//
// -- LAPACK routine (version 3.0) --
// Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
// Courant Institute, Argonne National Lab, and Rice University
// February 29, 1992
//************************************************************************
bool inverselu(ap::real_2d_array& a,
const ap::integer_1d_array& pivots,
int n)
{
bool result;
ap::real_1d_array work;
int i;
int iws;
int j;
int jb;
int jj;
int jp;
int jp1;
double v;
result = true;
//
// Quick return if possible
//
if( n==0 )
{
return result;
}
work.setbounds(1, n);
//
// Form inv(U)
//
if( !invtriangular(a, n, true, false) )
{
result = false;
return result;
}
//
// Solve the equation inv(A)*L = inv(U) for inv(A).
//
for(j = n; j >= 1; j--)
{
//
// Copy current column of L to WORK and replace with zeros.
//
for(i = j+1; i <= n; i++)
{
work(i) = a(i,j);
a(i,j) = 0;
}
//
// Compute current column of inv(A).
//
if( j<n )
{
jp1 = j+1;
for(i = 1; i <= n; i++)
{
v = ap::vdotproduct(a.getrow(i, jp1, n), work.getvector(jp1, n));
a(i,j) = a(i,j)-v;
}
}
}
//
// Apply column interchanges.
//
for(j = n-1; j >= 1; j--)
{
jp = pivots(j);
if( jp!=j )
{
ap::vmove(work.getvector(1, n), a.getcolumn(j, 1, n));
ap::vmove(a.getcolumn(j, 1, n), a.getcolumn(jp, 1, n));
ap::vmove(a.getcolumn(jp, 1, n), work.getvector(1, n));
}
}
return result;
}
void operator()(){
work();
}
void Emulator::WorkerRun() {
boost::asio::io_service::work work(mIo);
mIo.run();
}
SEXP dieharder(SEXP genS, SEXP testS, SEXP seedS, SEXP psamplesS, SEXP verbS, SEXP infileS, SEXP ntupleS) {
int verb, testarg;
unsigned int i;
SEXP result = NULL, vec, pv, name, desc, nkps;
char *inputfile;
/* Setup argv to allow call of parsecl() to let dieharder set globals */
char *argv[] = { "dieharder" };
optind = 0;
parsecl(1, argv);
/* Parse 'our' parameters from R */
generator = INTEGER_VALUE(genS);
testarg = INTEGER_VALUE(testS);
diehard = rgb = sts = user = 0;
if (testarg < 100) {
diehard = testarg;
} else if (testarg < 200) {
rgb = testarg - 100;
} else if (testarg < 300) {
sts = testarg - 200;
} else {
user = testarg - 300;
}
Seed = (unsigned long int) INTEGER_VALUE(seedS); /* (user-select) Seed, not (save switch) seed */
psamples = INTEGER_VALUE(psamplesS);
verb = INTEGER_VALUE(verbS);
inputfile = (char*) CHARACTER_VALUE(infileS);
ntuple = INTEGER_VALUE(ntupleS);
rdh_testptr = NULL;
rdh_dtestptr = NULL; /* to be safe, explicitly flag as NULL; cf test in output.c */
if (strcmp(inputfile, "") != 0) {
strncpy(filename, inputfile, 128);
fromfile = 1; /* flag this as file input */
}
if (Seed == 0) {
seed = random_seed();
} else {
seed = (unsigned long int) Seed;
}
if (verb) {
Rprintf("Dieharder called with gen=%d test=%d seed=%lu\n", generator, diehard, seed);
quiet = 0;
hist_flag = 1;
} else {
quiet = 1; /* override dieharder command-line default */
hist_flag = 0;
}
/* Now do the work that dieharder.c does */
startup();
work();
gsl_rng_free(rng);
reset_bit_buffers();
/* And then bring our results back to R */
/* create vector of size four: [0] is vector (!!) ks_pv, [1] is pvalues vec, [2] name, [3] nkps */
PROTECT(result = allocVector(VECSXP, 4));
/* alloc vector and scalars, and set it */
PROTECT(pv = allocVector(REALSXP, rdh_dtestptr->nkps));
PROTECT(name = allocVector(STRSXP, 1));
PROTECT(nkps = allocVector(INTSXP, 1));
if (rdh_testptr != NULL && rdh_dtestptr != NULL) {
for (i=0; i<rdh_dtestptr->nkps; i++) { /* there can be nkps p-values per test */
REAL(pv)[i] = rdh_testptr[i]->ks_pvalue;
}
PROTECT(vec = allocVector(REALSXP, rdh_testptr[0]->psamples)); /* alloc vector and set it */
for (i = 0; i < rdh_testptr[0]->psamples; i++) {
REAL(vec)[i] = rdh_testptr[0]->pvalues[i];
}
SET_STRING_ELT(name, 0, mkChar(rdh_dtestptr->name));
INTEGER(nkps)[0] = rdh_dtestptr->nkps; /* nb of Kuiper KS p-values in pv vector */
} else {
PROTECT(vec = allocVector(REALSXP, 1));
REAL(pv)[0] = R_NaN;
REAL(vec)[0] = R_NaN;
SET_STRING_ELT(name, 0, mkChar(""));
INTEGER(nkps)[0] = R_NaN;
}
/* insert vectors and scalars into result vector */
SET_VECTOR_ELT(result, 0, pv);
SET_VECTOR_ELT(result, 1, vec);
SET_VECTOR_ELT(result, 2, name);
SET_VECTOR_ELT(result, 3, nkps);
UNPROTECT(5);
return result;
}
extern "C" magma_int_t
magma_zgetrf2_gpu(
magma_int_t m, magma_int_t n,
magmaDoubleComplex_ptr dA, size_t dA_offset, magma_int_t ldda,
magma_int_t *ipiv,
magma_queue_t queues[2],
magma_int_t *info )
{
/* -- clMAGMA (version 1.3.0) --
Univ. of Tennessee, Knoxville
Univ. of California, Berkeley
Univ. of Colorado, Denver
@date November 2014
Purpose
=======
ZGETRF computes an LU factorization of a general M-by-N matrix A
using partial pivoting with row interchanges.
The factorization has the form
A = P * L * U
where P is a permutation matrix, L is lower triangular with unit
diagonal elements (lower trapezoidal if m > n), and U is upper
triangular (upper trapezoidal if m < n).
This is the right-looking Level 3 BLAS version of the algorithm.
Arguments
=========
M (input) INTEGER
The number of rows of the matrix A. M >= 0.
N (input) INTEGER
The number of columns of the matrix A. N >= 0.
A (input/output) COMPLEX_16 array on the GPU, dimension (LDDA,N).
On entry, the M-by-N matrix to be factored.
On exit, the factors L and U from the factorization
A = P*L*U; the unit diagonal elements of L are not stored.
LDDA (input) INTEGER
The leading dimension of the array A. LDDA >= max(1,M).
IPIV (output) INTEGER array, dimension (min(M,N))
The pivot indices; for 1 <= i <= min(M,N), row i of the
matrix was interchanged with row IPIV(i).
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
or another error occured, such as memory allocation failed.
> 0: if INFO = i, U(i,i) is exactly zero. The factorization
has been completed, but the factor U is exactly
singular, and division by zero will occur if it is used
to solve a system of equations.
===================================================================== */
#define dA(i_, j_) dA, dA_offset + (i_)*nb + (j_)*nb*ldda
#define dAT(i_, j_) dAT, dAT_offset + (i_)*nb*lddat + (j_)*nb
#define dAP(i_, j_) dAP, (i_) + (j_)*maxm
#define work(i_) (work + (i_))
magmaDoubleComplex c_one = MAGMA_Z_ONE;
magmaDoubleComplex c_neg_one = MAGMA_Z_NEG_ONE;
magma_int_t iinfo, nb;
magma_int_t maxm, maxn, mindim;
magma_int_t i, j, rows, s, lddat, ldwork;
magmaDoubleComplex_ptr dAT, dAP;
magmaDoubleComplex *work;
size_t dAT_offset;
/* Check arguments */
*info = 0;
if (m < 0)
*info = -1;
else if (n < 0)
*info = -2;
else if (ldda < max(1,m))
*info = -4;
if (*info != 0) {
magma_xerbla( __func__, -(*info) );
return *info;
}
/* Quick return if possible */
if (m == 0 || n == 0)
return *info;
/* Function Body */
mindim = min(m, n);
nb = magma_get_zgetrf_nb(m);
s = mindim / nb;
if (nb <= 1 || nb >= min(m,n)) {
/* Use CPU code. */
if ( MAGMA_SUCCESS != magma_zmalloc_cpu( &work, m*n )) {
*info = MAGMA_ERR_HOST_ALLOC;
return *info;
}
magma_zgetmatrix( m, n, dA(0,0), ldda, work(0), m, queues[0] );
lapackf77_zgetrf( &m, &n, work, &m, ipiv, info );
magma_zsetmatrix( m, n, work(0), m, dA(0,0), ldda, queues[0] );
magma_free_cpu( work );
}
else {
/* Use hybrid blocked code. */
maxm = ((m + 31)/32)*32;
maxn = ((n + 31)/32)*32;
if ( MAGMA_SUCCESS != magma_zmalloc( &dAP, nb*maxm )) {
*info = MAGMA_ERR_DEVICE_ALLOC;
return *info;
}
// square matrices can be done in place;
// rectangular requires copy to transpose
if ( m == n ) {
dAT = dA;
dAT_offset = dA_offset;
lddat = ldda;
magmablas_ztranspose_inplace( m, dAT(0,0), lddat, queues[0] );
}
else {
lddat = maxn; // N-by-M
dAT_offset = 0;
if ( MAGMA_SUCCESS != magma_zmalloc( &dAT, lddat*maxm )) {
magma_free( dAP );
*info = MAGMA_ERR_DEVICE_ALLOC;
return *info;
}
magmablas_ztranspose( m, n, dA(0,0), ldda, dAT(0,0), lddat, queues[0] );
}
ldwork = maxm;
/*
if ( MAGMA_SUCCESS != magma_zmalloc_cpu( &work, ldwork*nb ) ) {
magma_free( dAP );
if ( dA != dAT )
magma_free( dAT );
*info = MAGMA_ERR_HOST_ALLOC;
return *info;
}
*/
cl_mem work_mapped = clCreateBuffer( gContext, CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR, ldwork*nb * sizeof(magmaDoubleComplex), NULL, NULL );
work = (magmaDoubleComplex*) clEnqueueMapBuffer( queues[0], work_mapped, CL_TRUE, CL_MAP_READ | CL_MAP_WRITE, 0, ldwork*nb * sizeof(magmaDoubleComplex), 0, NULL, NULL, NULL );
for( j=0; j < s; j++ ) {
// download j-th panel
magmablas_ztranspose( nb, m-j*nb, dAT(j,j), lddat, dAP(0,0), maxm, queues[0] );
clFlush( queues[0] );
magma_queue_sync( queues[0] );
magma_zgetmatrix_async( m-j*nb, nb, dAP(0,0), maxm, work(0), ldwork, queues[1], NULL );
clFlush( queues[1] );
if ( j > 0 ) {
magma_ztrsm( MagmaRight, MagmaUpper, MagmaNoTrans, MagmaUnit,
n - (j+1)*nb, nb,
c_one, dAT(j-1,j-1), lddat,
dAT(j-1,j+1), lddat, queues[0] );
magma_zgemm( MagmaNoTrans, MagmaNoTrans,
n-(j+1)*nb, m-j*nb, nb,
c_neg_one, dAT(j-1,j+1), lddat,
dAT(j, j-1), lddat,
c_one, dAT(j, j+1), lddat, queues[0] );
}
magma_queue_sync( queues[1] );
// do the cpu part
rows = m - j*nb;
lapackf77_zgetrf( &rows, &nb, work, &ldwork, ipiv+j*nb, &iinfo );
if ( *info == 0 && iinfo > 0 )
*info = iinfo + j*nb;
for( i=j*nb; i < j*nb + nb; ++i ) {
ipiv[i] += j*nb;
}
magmablas_zlaswp( n, dAT(0,0), lddat, j*nb + 1, j*nb + nb, ipiv, 1, queues[0] );
clFlush( queues[0] );
// upload j-th panel
magma_zsetmatrix_async( m-j*nb, nb, work(0), ldwork, dAP(0,0), maxm, queues[1], NULL );
magma_queue_sync( queues[1] );
magmablas_ztranspose( m-j*nb, nb, dAP(0,0), maxm, dAT(j,j), lddat, queues[0] );
clFlush( queues[0] );
// do the small non-parallel computations (next panel update)
if ( s > (j+1) ) {
magma_ztrsm( MagmaRight, MagmaUpper, MagmaNoTrans, MagmaUnit,
nb, nb,
c_one, dAT(j, j ), lddat,
dAT(j, j+1), lddat, queues[0] );
magma_zgemm( MagmaNoTrans, MagmaNoTrans,
nb, m-(j+1)*nb, nb,
c_neg_one, dAT(j, j+1), lddat,
dAT(j+1, j ), lddat,
c_one, dAT(j+1, j+1), lddat, queues[0] );
}
else {
magma_ztrsm( MagmaRight, MagmaUpper, MagmaNoTrans, MagmaUnit,
n-s*nb, nb,
c_one, dAT(j, j ), lddat,
dAT(j, j+1), lddat, queues[0] );
magma_zgemm( MagmaNoTrans, MagmaNoTrans,
n-(j+1)*nb, m-(j+1)*nb, nb,
c_neg_one, dAT(j, j+1), lddat,
dAT(j+1, j ), lddat,
c_one, dAT(j+1, j+1), lddat, queues[0] );
}
}
magma_int_t nb0 = min( m - s*nb, n - s*nb );
if ( nb0 > 0 ) {
rows = m - s*nb;
magmablas_ztranspose( nb0, rows, dAT(s,s), lddat, dAP(0,0), maxm, queues[0] );
clFlush( queues[0] );
magma_queue_sync( queues[0] );
magma_zgetmatrix_async( rows, nb0, dAP(0,0), maxm, work(0), ldwork, queues[1], NULL );
magma_queue_sync( queues[1] );
// do the cpu part
lapackf77_zgetrf( &rows, &nb0, work, &ldwork, ipiv+s*nb, &iinfo );
if ( (*info == 0) && (iinfo > 0) )
*info = iinfo + s*nb;
for( i=s*nb; i < s*nb + nb0; ++i ) {
ipiv[i] += s*nb;
}
magmablas_zlaswp( n, dAT(0,0), lddat, s*nb + 1, s*nb + nb0, ipiv, 1, queues[0] );
clFlush( queues[0] );
// upload j-th panel
magma_zsetmatrix_async( rows, nb0, work(0), ldwork, dAP(0,0), maxm, queues[1], NULL );
magma_queue_sync( queues[1] );
magmablas_ztranspose( rows, nb0, dAP(0,0), maxm, dAT(s,s), lddat, queues[0] );
clFlush( queues[0] );
magma_ztrsm( MagmaRight, MagmaUpper, MagmaNoTrans, MagmaUnit,
n-s*nb-nb0, nb0,
c_one, dAT(s,s), lddat,
dAT(s,s)+nb0, lddat, queues[0] );
}
// undo transpose
if ( dA == dAT ) {
magmablas_ztranspose_inplace( m, dAT(0,0), lddat, queues[0] );
}
else {
magmablas_ztranspose( n, m, dAT(0,0), lddat, dA(0,0), ldda, queues[0] );
magma_free( dAT );
}
magma_queue_sync( queues[0] );
magma_queue_sync( queues[1] );
magma_free( dAP );
// magma_free_cpu( work );
clEnqueueUnmapMemObject( queues[0], work_mapped, work, 0, NULL, NULL );
clReleaseMemObject( work_mapped );
}
return *info;
} /* magma_zgetrf_gpu */
