/*========================================*/
int initCAN(int bus)
{
NCTYPE_ATTRID AttrIdList[8];
NCTYPE_UINT32 AttrValueList[8];
//NCTYPE_UINT32 Baudrate = 125000; // BAUD_125K
NCTYPE_UINT32 Baudrate = NC_BAUD_1000K;
char Interface[15];
NCTYPE_STATUS Status = 0;
NCTYPE_OBJH TxHandle;
sprintf_s(Interface, "CAN%d", bus);
// Configure the CAN Network Interface Object
AttrIdList[0] = NC_ATTR_BAUD_RATE;
AttrValueList[0] = Baudrate;
AttrIdList[1] = NC_ATTR_START_ON_OPEN;
AttrValueList[1] = NC_TRUE;
AttrIdList[2] = NC_ATTR_READ_Q_LEN;
AttrValueList[2] = 100;
AttrIdList[3] = NC_ATTR_WRITE_Q_LEN;
AttrValueList[3] = 10;
AttrIdList[4] = NC_ATTR_CAN_COMP_STD;
AttrValueList[4] = 0;//0xCFFFFFFF;
AttrIdList[5] = NC_ATTR_CAN_MASK_STD;
AttrValueList[5] = NC_CAN_MASK_STD_DONTCARE;
AttrIdList[6] = NC_ATTR_CAN_COMP_XTD;
AttrValueList[6] = 0;//0xCFFFFFFF;
AttrIdList[7] = NC_ATTR_CAN_MASK_XTD;
AttrValueList[7] = NC_CAN_MASK_XTD_DONTCARE;
printf("<< CAN: Config\n");
Status = ncConfig(Interface, 2, AttrIdList, AttrValueList);
if (Status < 0)
{
PrintStat(bus, Status, "ncConfig");
return Status;
}
printf(" - Done\n");
// open the CAN Network Interface Object
printf("<< CAN: Open Channel\n");
Status = ncOpenObject (Interface, &TxHandle);
if (Status < 0)
{
PrintStat(bus, Status, "ncOpenObject");
return Status;
}
canDev[bus] = TxHandle;
printf(" - Done\n");
return 0;
}
void RemoveLowLocalAndOutputChanged(SuccinctDBG &dbg, FILE *contigs_file, FILE *multi_file, FILE *final_contig_file,
FILE *addi_contig_file, FILE *addi_multi_file,
double min_depth, int min_len, double local_ratio, int min_final_contig_len) {
xtimer_t timer;
timer.reset();
timer.start();
UnitigGraph unitig_graph(&dbg);
unitig_graph.InitFromSdBG();
timer.stop();
printf("Simple path graph size: %u, time for building: %lf\n", unitig_graph.size(), timer.elapsed());
timer.reset();
timer.start();
histogram.clear();
if (final_contig_file == NULL) {
unitig_graph.OutputInitUnitigs(contigs_file, multi_file, histogram);
} else {
unitig_graph.OutputInitUnitigs(contigs_file, multi_file, final_contig_file, histogram, min_final_contig_len);
}
PrintStat();
timer.stop();
printf("Time to output: %lf\n", timer.elapsed());
const double kMaxDepth = 65535;
const int kLocalWidth = 1000;
int64_t num_removed = 0;
timer.reset();
timer.start();
while (min_depth < kMaxDepth) {
// xtimer_t local_timer;
// local_timer.reset();
// local_timer.start();
if (!unitig_graph.RemoveLocalLowDepth(min_depth, min_len, kLocalWidth, local_ratio, num_removed)) {
break;
}
min_depth *= 1.1;
// local_timer.stop();
// printf("depth: %lf, num: %ld, time: %lf\n", min_depth, num_removed, local_timer.elapsed());
}
timer.stop();
printf("Number of unitigs removed: %ld, time: %lf\n", num_removed, timer.elapsed());
histogram.clear();
unitig_graph.OutputChangedUnitigs(addi_contig_file, addi_multi_file, histogram);
PrintStat();
}
int canSendMsg(int bus, int id, char len, unsigned char *data, int blocking)
{
if (!canDev[bus])
return -1;
NCTYPE_STATUS Status = 0;
NCTYPE_STATE State = 0;
NCTYPE_CAN_FRAME TxFrame;
TxFrame.IsRemote = NC_FRMTYPE_DATA;
TxFrame.ArbitrationId = id;
TxFrame.DataLength = len;
for (int i=0; i<len; i++)
TxFrame.Data[i] = ((NCTYPE_UINT8_P)data)[i];
Status = ncWrite(canDev[bus], sizeof(NCTYPE_CAN_FRAME), &TxFrame);
if (Status < 0)
{
PrintStat(bus, Status, "ncWrite");
return Status;
}
if (blocking)
{
Status = ncWaitForState(canDev[bus], (NC_ST_WRITE_SUCCESS | NC_ST_WARNING | NC_ST_ERROR), NC_DURATION_INFINITE, &State);
if (Status < 0 || !(NC_ST_WRITE_SUCCESS & State))
return -1; // failed
}
return Status;
}
int main (int argc, char *argv[]) {
t_fmsg *f;
int i;
/* Creer la file */
f = CreerFile();
// Créer le thread consommateur
CreateProc((function_t)Consommateur,(void *)f, 0);
// Créer les threads producteurs
for (i = 0; i < NBPROD; i++)
CreateProc((function_t)Producteur,(void *)f, 0);
// Lancer l'ordonnanceur en mode non "verbeux"
sched(1);
// Imprimer les statistiques
PrintStat();
return EXIT_SUCCESS;
}
//=======================
void plotSignalAndBGofMC(Char_t *fileNameIn = "histos_jPsi_2360GeV_STARTUP", //don't put ".root" at the end!
Char_t *enLabel = "2360 GeV",
Bool_t takeBest = kTRUE, //uses the histograms after all cuts
Bool_t normalize = kTRUE, //normalizes the histos to 1 nb-1
Double_t intLumi = 261.//jPsi: DESIGN: 2.36 TeV: 266.; 900 GeV: 263.
//jPsi: STARTUP: 2.36 TeV: 261.; 900 GeV: 384.
) {
LoadHistos(fileNameIn, takeBest);
Normalize(normalize, intLumi);
LoadBGHistos("histos_ppMuXLoose_2360GeV_STARTUP.root", takeBest, 5.5);
PlotHistos(enLabel, fileNameIn);
PrintStat();
}
int main (int argc, char *argv[]) {
// Créer les processus long
CreateProc((function_t)Thread1,(void *)NULL, 0);
// Lancer l'ordonnanceur en mode "verbeux"
sched(0);
// Imprimer les statistiques
PrintStat();
return EXIT_SUCCESS;
}
void AssembleFinalFromUnitigGraph(SuccinctDBG &dbg, FILE *final_contig_file, int min_final_contig_len) {
xtimer_t timer;
timer.reset();
timer.start();
UnitigGraph unitig_graph(&dbg);
unitig_graph.InitFromSdBG();
timer.stop();
printf("unitig graph size: %u, time for building: %lf\n", unitig_graph.size(), timer.elapsed());
timer.reset();
timer.start();
histogram.clear();
unitig_graph.OutputFinalUnitigs(final_contig_file, histogram, min_final_contig_len);
PrintStat();
timer.stop();
printf("Time to output: %lf\n", timer.elapsed());
}
void main(int argc, char *argv[])
{
char *path;
if (argc == 2)
{
g_simulate = false;
path = argv[1];
}
else if (argc == 3 && !strcmp(argv[1], "-s"))
{
g_simulate = true;
g_verbose = false;
path = argv[2];
}
else if (argc == 3 && !strcmp(argv[1], "-v"))
{
g_simulate = false;
g_verbose = true;
path = argv[2];
}
else if (argc == 4 &&
((!strcmp(argv[1], "-s") && !strcmp(argv[2], "-v")) ||
(!strcmp(argv[1], "-v") && !strcmp(argv[2], "-s"))))
{
g_simulate = true;
g_verbose = true;
path = argv[3];
}
else
{
printf(
"detfs 1.4\n"
"\n"
"Usage: detfs [-s] [-v] <path>\n"
"\n"
"-s Simulate\n"
"-v Verbose logging\n");
return;
}
DelinkTFS(path);
PrintStat();
}
int freeCAN(int bus)
{
if (!canDev[bus])
return -1;
NCTYPE_STATUS Status = 0;
printf("<< CAN: Close\n");
Status = ncCloseObject(canDev[bus]);
if (Status < 0)
{
PrintStat(bus, Status, "ncCloseObject");
return Status;
}
canDev[bus] = 0;
printf(" - Done\n");
return 0;
}
void RemoveLowLocalAndOutputFinal(SuccinctDBG &dbg, FILE *final_contig_file,
double min_depth, int min_len, double local_ratio, int min_final_contig_len) {
UnitigGraph unitig_graph(&dbg);
unitig_graph.InitFromSdBG();
printf("Simple path graph size: %u\n", unitig_graph.size());
const double kMaxDepth = 65535;
const int kLocalWidth = 1000;
int64_t num_removed = 0;
while (min_depth < kMaxDepth &&
unitig_graph.RemoveLocalLowDepth(min_depth, min_len, kLocalWidth, local_ratio, num_removed)) {
min_depth *= 1.1;
}
printf("Number of unitigs removed: %ld\n", num_removed);
histogram.clear();
unitig_graph.OutputFinalUnitigs(final_contig_file, histogram, min_final_contig_len);
PrintStat();
}
int main (int argc, char *argv[]) {
int i;
int *j;
// Créer les processus long
for (i = 0; i < 2; i++) {
j = (int *) malloc(sizeof(int));
*j= i;
CreateProc((function_t)ProcLong,(void *)j, 80);
}
// Créer les processus court
for (i = 0; i < 2; i++) {
j = (int *) malloc(sizeof(int));
*j= i;
CreateProc((function_t)ProcCourt,(void *)j, 10);
}
// Definir une nouvelle primitive d'election avec un quantum de 1 seconde
// SchedParam(NEW, 1, RandomElect);
// Definir une nouvelle primitive d'election sans quantum (batch)
//SchedParam(NEW, 0, SJFElect);
// Redefinir le quantum par defaut
//SchedParam(PREMPT, 2, NULL);
// Passer en mode batch
//SchedParam(BATCH, 0, NULL);
// Lancer l'ordonnanceur en mode non "verbeux"
sched(0);
// Imprimer les statistiques
PrintStat();
return EXIT_SUCCESS;
}
int canReadMsg(int bus, int *id, int *len, unsigned char *data, int blocking)
{
if (!canDev[bus])
return -1;
NCTYPE_STATUS Status = 0;
NCTYPE_STATE State = 0;
NCTYPE_CAN_STRUCT RxFrame;
if (blocking)
{
Status = ncWaitForState(canDev[bus], NC_ST_READ_AVAIL, NC_DURATION_INFINITE, &State);
if (Status < 0 || !(NC_ST_READ_AVAIL & State))
return -1; // failed
}
memset(&RxFrame, 0, sizeof(NCTYPE_CAN_FRAME));
//RxFrame.FrameType = NC_FRMTYPE_DATA;
RxFrame.DataLength = *len;
Status = ncRead(canDev[bus], sizeof(NCTYPE_CAN_STRUCT), &RxFrame);
if (Status < 0 || NC_FRMTYPE_DATA != RxFrame.FrameType)
{
PrintStat(bus, Status, "ncRead");
return Status;
}
else if (Status == CanWarnOldData)
{
(*len) = 0;
return Status;
}
(*id) = RxFrame.ArbitrationId;
//(*time) = (FILETIME)(RxFrame.Timestamp);
(*len) = RxFrame.DataLength;
for (int i=0; i<RxFrame.DataLength; i++)
((NCTYPE_UINT8_P)data)[i] = RxFrame.Data[i];
#ifdef _DUMP_RXFRAME
PrintRxFrame(RxFrame);
#endif
return Status;
}
main(int argc, char *argv[])
{
char buf[80];
Point3 **v;
Point3 *BaseV;
int *usedpoint;
List T=NULL_LIST,
Q=NULL_LIST;
int n,i=1;
FILE *fp=stdout;
double sec;
SetProgramName("DeWall");
if((argc<2) ||
(strcmp(argv[i],"/?")==0)||
(strcmp(argv[i],"-?")==0)||
(strcmp(argv[i],"/h")==0)||
(strcmp(argv[i],"-h")==0) ) Error(USAGE_MESSAGE, EXIT);
while(*argv[i]=='-')
{
switch(argv[i][1])
{
case 'p' : UpdateFlag=ON; break;
case 'c' : CheckFlag=ON; break;
case 't' : SafeTetraFlag=ON; break;
case 's' : StatFlag=ON;
if(argv[i][2]=='1') NumStatFlag=ON;
if(argv[i][2]=='2')
{
NumStatTitleFlag=ON;
NumStatFlag=ON;
}
break;
case 'u' : UGScaleFlag=ON;
if(argv[i][2]==0 && isdigit(argv[i+1][0]))
UGScale=atof(argv[++i]);
else UGScale=atof(argv[i]+2);
break;
default : sprintf(buf,"Unknown options '%s'\n",argv[i]);
Error(buf,NO_EXIT);
}
i++;
}
BaseV=ReadPoints(argv[i++],&n);
if(argc>i) fp=fopen(argv[i],"w");
SI.Point=n;
v=(Point3 **)malloc(n*sizeof(Point3 *));
usedpoint=(int *)malloc(n*sizeof(int));
if(!v || !usedpoint) Error("Unable to allocate memory for Points\n",EXIT);
for(i=0;i<n;i++) v[i]=&(BaseV[i]);
for(i=0;i<n;i++) usedpoint[i] = -1;
Q=NewList(FIFO,sizeof(Face)); /* Initialize First Face */
ChangeEqualObjectList(EqualFace,Q); /* List Q. */
if(n>40) HashList(n/4,HashFace,Q);
T=NewList(LIFO,sizeof(ShortTetra)); /* Initialize Built Tetra-*/
ChangeEqualObjectList(EqualTetra,T); /* hedra List T. */
if(SafeTetraFlag && n>40) HashList(n/4,HashTetra,T);
ResetChronos(USER_CHRONOS);
StartChronos(USER_CHRONOS);
DeWall(v,BaseV,usedpoint,n,Q,T,XAxis);
StopChronos(USER_CHRONOS);
sec=ReadChronos(USER_CHRONOS);
SI.Tetra=CountList(T);
SI.Secs=sec;
if(StatFlag && !NumStatFlag) PrintStat();
if(StatFlag && NumStatFlag && NumStatTitleFlag) PrintNumStatTitle();
if(StatFlag && NumStatFlag) PrintNumStat();
if(!StatFlag)
printf("Points:%7i Secs:%6.2f Tetras:%7i\n",SI.Point,SI.Secs,SI.Tetra);
WriteTetraList(T,fp);
return 0;
}
void *arrivalThread(void *id)
{
My402ListElem *elem;
Packet *pkt;
int i;
long pkt_sleep, pkt_service;
int pkt_token;
long sleep_time;
long prev_pkt_time=0, instant_time;
long avg_ia=0;
for (i=0;i<num_packets;i++) {
getStatistics(&pkt_sleep,&pkt_service,&pkt_token,i);
if(pkt_sleep > 10000000)
pkt_sleep = 10000000;
if(pkt_service > 1000000)
pkt_service = 10000000;
taend=getinstanttime();
sleep_time = pkt_sleep - (taend-tastart);
usleep(sleep_time);
// Creating the Packet
pkt = (Packet *)malloc(sizeof(struct tagPacket));
pkt->pkt_id = i;
pkt->num_tokens = pkt_token;
pkt->service_time = pkt_service;
pkt->sys_enter = tastart = getinstanttime();
if(pkt->num_tokens > B) {
// Drop the packet
pkts_to_arrive--;
PrintStat(getinstanttime());
fprintf(stdout, "p%d arrives, Invalid token requirement, p%d Dropped!!\n",
pkt->pkt_id, pkt->pkt_id);
continue;
}
else {
instant_time = getinstanttime();
if(prev_pkt_time==0) {
prev_pkt_time = pkt->inter_arrival_time = (instant_time - temulation_start);
prev_pkt_time = instant_time;
}
else {
pkt->inter_arrival_time = instant_time - prev_pkt_time;
prev_pkt_time = instant_time;
}
PrintStat(instant_time);
avg_ia += pkt->inter_arrival_time;
fprintf(stdout, "p%d arrives, needs %d tokens, inter-arrival time = %.3fms\n",
pkt->pkt_id,pkt->num_tokens,(double)(pkt->inter_arrival_time)/1000);
}
pthread_mutex_lock(&my_mutex);
if(My402ListEmpty(&queue1)){
My402ListAppend(&queue1,pkt);
pkt->q1_enter = getinstanttime();
PrintStat(getinstanttime());
fprintf(stdout,"p%d enters Q1\n", pkt->pkt_id);
if(!My402ListEmpty(&queue2)) {
pthread_cond_signal(&queue2_cond);
}
else {
if(token_bucket >= pkt->num_tokens) {
elem = My402ListFirst(&queue1);
pkt = (Packet *)elem->obj;
My402ListUnlink(&queue1,elem);
pkt->q1_exit = getinstanttime();
token_bucket-=pkt->num_tokens;
PrintStat(getinstanttime());
fprintf(stdout, "p%d leaves Q1, time in Q1 = %.3fms, token bucket now has %d tokens\n",
pkt->pkt_id, (double)(pkt->q1_exit - pkt->q1_enter)/1000, token_bucket);
avg_pkt_q1 += (pkt->q1_exit - pkt->q1_enter);
pkts_left_q1++;
My402ListAppend(&queue2,pkt);
pkt->q1_enter = 0;
pkt->q1_exit = 0;
pkt->q2_enter = getinstanttime();
PrintStat(getinstanttime());
fprintf(stdout,"p%d enters Q2\n", pkt->pkt_id);
pthread_cond_signal(&queue2_cond);
}
}
}
else {
My402ListAppend(&queue1,pkt);
pkt->q1_enter = getinstanttime();
PrintStat(getinstanttime());
fprintf(stdout,"p%d enters Q1\n", pkt->pkt_id);
}
pthread_mutex_unlock(&my_mutex);
}
pthread_mutex_lock(&my_mutex);
pthread_cond_signal(&queue2_cond); // This can also be a false signal just to wake up server, if
pthread_mutex_unlock(&my_mutex); // say last packet is dropped and there is no pkt to be queued to q2.
if(i==-1) {
avg_inter_arrival = 0;
pkt_drop_prob = 0;
}
else {
avg_inter_arrival = (double)((double)avg_ia/(double)((i+1)*1000000));
pkt_drop_prob = (double)(num_packets - pkts_to_arrive)/(i+1);
}
pthread_exit(NULL);
}
int main(int argc, char **argv)
{
long t1=1, t2=2, t3=3;
pthread_t arrThread, serThread, tokThread;
long total_emulation_time;
// Setting to default values
lambda = 0.5;
mu = 0.35;
r = 1.5;
P = 3;
B = 10;
num_packets = 20;
token_bucket = 0;
server_die = 0;
token_die = 0;
avg_pkt_q1 = avg_pkt_q2 = avg_pkt_s = 0;
// Done
// Initialize Queues
My402ListInit(&queue1);
My402ListInit(&queue2);
pthread_mutex_init(&my_mutex,NULL);
pthread_cond_init(&queue2_cond,NULL);
parseCommandline(argc,argv);
tokenarrival = (long)((1/r)*1000000);
pkts_to_arrive = num_packets;
pthread_create(&arrThread, NULL, arrivalThread,(void *)t1);
pthread_create(&tokThread, NULL, tokenThread,(void *)t2);
pthread_create(&serThread, NULL, serverThread, (void *)t3);
temulation_start = tastart = ttstart = getinstanttime();
PrintStat(getinstanttime());
fprintf(stdout, "Emulation Begins\n");
pthread_join(arrThread, NULL);
pthread_join(tokThread, NULL);
pthread_join(serThread, NULL);
temulation_end = getinstanttime();
PrintStat(getinstanttime());
fprintf(stdout, "Emulation Ends \n");
total_emulation_time = temulation_end - temulation_start;
fprintf(stdout,"Statistics: \n\n");
fprintf(stdout,"\taverage packet inter arrival time = %.08g sec\n",avg_inter_arrival);
fprintf(stdout,"\taverage packet service time = %.08g sec\n\n",avg_serv_time);
fprintf(stdout,"\taverage number of packets in Q1 = %.08g\n",(double)avg_pkt_q1/total_emulation_time);
fprintf(stdout,"\taverage number of packets in Q2 = %.08g\n",(double)avg_pkt_q2/total_emulation_time);
fprintf(stdout,"\taverage number of packets at S = %.08g\n\n",(double)avg_pkt_s/total_emulation_time);
fprintf(stdout,"\taverage time a packet spent in the system = %.08g sec\n",avg_pkt_sys_time);
fprintf(stdout,"\tstandard deviation for time spent in system = %.08g sec \n\n",std_deviation);
fprintf(stdout,"\ttoken drop probability = %.08g\n",token_drop_prob);
fprintf(stdout,"\tpacket drop probability = %.08g\n",pkt_drop_prob);
pthread_mutex_destroy(&my_mutex);
pthread_cond_destroy(&queue2_cond);
pthread_exit(NULL);
}
void CDynamoRIOView::OnEditCopystats()
{
if (m_ProcessList.GetCurSel() == 0) {
MessageBox(_T("No instance selected"), _T("Error"), MB_OK | MYMBFLAGS);
return;
}
if (!OpenClipboard()) {
MessageBox(_T("Error opening clipboard"), _T("Error"),
MB_OK | MYMBFLAGS);
return;
}
EmptyClipboard();
#define CLIPBOARD_BUFSZ USHRT_MAX
TCHAR buf[CLIPBOARD_BUFSZ];
TCHAR *pos = buf;
if (m_selected_pid > 0) {
if (m_stats != NULL) {
pos += _stprintf(pos, _T("Process id = %d\r\n"),
m_stats->process_id);
pos += _stprintf(pos, _T("Process name = %") ASCII_PRINTF
_T("\r\n"), m_stats->process_name);
pos += _stprintf(pos, _T("Status = %s\r\n"),
m_Exited.GetBuffer(0));
#ifndef DRSTATS_DEMO
pos += _stprintf(pos, _T("Log mask = %s\r\n"),
m_LogMask.GetBuffer(0));
pos += _stprintf(pos, _T("Log level = %s\r\n"),
m_LogLevel.GetBuffer(0));
pos += _stprintf(pos, _T("Log file = %s\r\n"),
m_LogDir.GetBuffer(0));
#endif
pos += _stprintf(pos, _T("\r\nSTATS\r\n"));
uint i;
for (i = 0; i < m_stats->num_stats; i++) {
if (pos >= &buf[STATS_BUFSZ-STAT_NAME_MAX_LEN*2])
break;
// FIXME: Filter here too
pos += PrintStat(pos, i, TRUE/*filter*/);
assert(pos < &buf[STATS_BUFSZ-1]);
}
} else {
CString pname;
m_ProcessList.GetLBText(m_ProcessList.GetCurSel(), pname);
_stprintf(pos, _T("%s"), pname);
pos += _tcslen(pos);
}
}
if (m_clientStats != NULL) {
pos += PrintClientStats(pos, &buf[STATS_BUFSZ-1]);
}
size_t len = _tcslen(buf);
// Allocate a global memory object for the text.
HGLOBAL hglbCopy = GlobalAlloc(GMEM_DDESHARE,
(len + 1) * sizeof(TCHAR));
if (hglbCopy == NULL) {
CloseClipboard();
return;
}
// Lock the handle and copy the text to the buffer.
LPTSTR lptstrCopy = (LPTSTR) GlobalLock(hglbCopy);
memcpy(lptstrCopy, buf, (len + 1) * sizeof(TCHAR));
lptstrCopy[len] = (TCHAR) 0; // null TCHARacter
GlobalUnlock(hglbCopy);
// Place the handle on the clipboard.
SetClipboardData(
#ifdef UNICODE
CF_UNICODETEXT,
#else
CF_TEXT,
#endif
hglbCopy);
CloseClipboard();
}
void
pzgssv(int nprocs, SuperMatrix *A, int *perm_c, int *perm_r,
SuperMatrix *L, SuperMatrix *U, SuperMatrix *B, int *info )
{
/*
* -- SuperLU MT routine (version 2.0) --
* Lawrence Berkeley National Lab, Univ. of California Berkeley,
* and Xerox Palo Alto Research Center.
* September 10, 2007
*
* Purpose
* =======
*
* PZGSSV solves the system of linear equations A*X=B, using the parallel
* LU factorization routine PZGSTRF. It performs the following steps:
*
* 1. If A is stored column-wise (A->Stype = NC):
*
* 1.1. Permute the columns of A, forming A*Pc, where Pc is a
* permutation matrix.
* For more details of this step, see sp_preorder.c.
*
* 1.2. Factor A as Pr*A*Pc=L*U with the permutation Pr determined
* by Gaussian elimination with partial pivoting.
* L is unit lower triangular with offdiagonal entries
* bounded by 1 in magnitude, and U is upper triangular.
*
* 1.3. Solve the system of equations A*X=B using the factored
* form of A.
*
* 2. If A is stored row-wise (A->Stype = NR), apply the above algorithm
* to the tranpose of A:
*
* 2.1. Permute columns of tranpose(A) (rows of A),
* forming transpose(A)*Pc, where Pc is a permutation matrix.
* For more details of this step, see sp_preorder.c.
*
* 2.2. Factor A as Pr*transpose(A)*Pc=L*U with the permutation Pr
* determined by Gaussian elimination with partial pivoting.
* L is unit lower triangular with offdiagonal entries
* bounded by 1 in magnitude, and U is upper triangular.
*
* 2.3. Solve the system of equations A*X=B using the factored
* form of A.
*
* See supermatrix.h for the definition of "SuperMatrix" structure.
*
*
* Arguments
* =========
*
* nprocs (input) int
* Number of processes (or threads) to be spawned and used to perform
* the LU factorization by pzgstrf(). There is a single thread of
* control to call pzgstrf(), and all threads spawned by pzgstrf()
* are terminated before returning from pzgstrf().
*
* A (input) SuperMatrix*
* Matrix A in A*X=B, of dimension (A->nrow, A->ncol), where
* A->nrow = A->ncol. Currently, the type of A can be:
* Stype = NC or NR; Dtype = _D; Mtype = GE. In the future,
* more general A will be handled.
*
* perm_c (input/output) int*
* If A->Stype=NC, column permutation vector of size A->ncol,
* which defines the permutation matrix Pc; perm_c[i] = j means
* column i of A is in position j in A*Pc.
* On exit, perm_c may be overwritten by the product of the input
* perm_c and a permutation that postorders the elimination tree
* of Pc'*A'*A*Pc; perm_c is not changed if the elimination tree
* is already in postorder.
*
* If A->Stype=NR, column permutation vector of size A->nrow
* which describes permutation of columns of tranpose(A)
* (rows of A) as described above.
*
* perm_r (output) int*,
* If A->Stype=NR, row permutation vector of size A->nrow,
* which defines the permutation matrix Pr, and is determined
* by partial pivoting. perm_r[i] = j means row i of A is in
* position j in Pr*A.
*
* If A->Stype=NR, permutation vector of size A->ncol, which
* determines permutation of rows of transpose(A)
* (columns of A) as described above.
*
* L (output) SuperMatrix*
* The factor L from the factorization
* Pr*A*Pc=L*U (if A->Stype=NC) or
* Pr*transpose(A)*Pc=L*U (if A->Stype=NR).
* Uses compressed row subscripts storage for supernodes, i.e.,
* L has types: Stype = SCP, Dtype = _D, Mtype = TRLU.
*
* U (output) SuperMatrix*
* The factor U from the factorization
* Pr*A*Pc=L*U (if A->Stype=NC) or
* Pr*transpose(A)*Pc=L*U (if A->Stype=NR).
* Use column-wise storage scheme, i.e., U has types:
* Stype = NCP, Dtype = _D, Mtype = TRU.
*
* B (input/output) SuperMatrix*
* B has types: Stype = DN, Dtype = _D, Mtype = GE.
* On entry, the right hand side matrix.
* On exit, the solution matrix if info = 0;
*
* info (output) int*
* = 0: successful exit
* > 0: if info = i, and i is
* <= A->ncol: U(i,i) is exactly zero. The factorization has
* been completed, but the factor U is exactly singular,
* so the solution could not be computed.
* > A->ncol: number of bytes allocated when memory allocation
* failure occurred, plus A->ncol.
*
*/
trans_t trans;
NCformat *Astore;
DNformat *Bstore;
SuperMatrix *AA; /* A in NC format used by the factorization routine.*/
SuperMatrix AC; /* Matrix postmultiplied by Pc */
int i, n, panel_size, relax;
fact_t fact;
yes_no_t refact, usepr;
double diag_pivot_thresh, drop_tol;
void *work;
int lwork;
superlumt_options_t superlumt_options;
Gstat_t Gstat;
double t; /* Temporary time */
double *utime;
flops_t *ops, flopcnt;
/* ------------------------------------------------------------
Test the input parameters.
------------------------------------------------------------*/
Astore = A->Store;
Bstore = B->Store;
*info = 0;
if ( nprocs <= 0 ) *info = -1;
else if ( A->nrow != A->ncol || A->nrow < 0 ||
(A->Stype != SLU_NC && A->Stype != SLU_NR) ||
A->Dtype != SLU_Z || A->Mtype != SLU_GE )
*info = -2;
else if ( B->ncol < 0 || Bstore->lda < SUPERLU_MAX(1, A->nrow) )*info = -7;
if ( *info != 0 ) {
i = -(*info);
xerbla_("pzgssv", &i);
return;
}
#if 0
/* Use the best sequential code.
if this part is commented out, we will use the parallel code
run on one processor. */
if ( nprocs == 1 ) {
return;
}
#endif
fact = EQUILIBRATE;
refact = NO;
trans = NOTRANS;
panel_size = sp_ienv(1);
relax = sp_ienv(2);
diag_pivot_thresh = 1.0;
usepr = NO;
drop_tol = 0.0;
work = NULL;
lwork = 0;
/* ------------------------------------------------------------
Allocate storage and initialize statistics variables.
------------------------------------------------------------*/
n = A->ncol;
StatAlloc(n, nprocs, panel_size, relax, &Gstat);
StatInit(n, nprocs, &Gstat);
utime = Gstat.utime;
ops = Gstat.ops;
/* ------------------------------------------------------------
Convert A to NC format when necessary.
------------------------------------------------------------*/
if ( A->Stype == SLU_NR ) {
NRformat *Astore = A->Store;
AA = (SuperMatrix *) SUPERLU_MALLOC( sizeof(SuperMatrix) );
zCreate_CompCol_Matrix(AA, A->ncol, A->nrow, Astore->nnz,
Astore->nzval, Astore->colind, Astore->rowptr,
SLU_NC, A->Dtype, A->Mtype);
trans = TRANS;
} else if ( A->Stype == SLU_NC ) AA = A;
/* ------------------------------------------------------------
Initialize the option structure superlumt_options using the
user-input parameters;
Apply perm_c to the columns of original A to form AC.
------------------------------------------------------------*/
pzgstrf_init(nprocs, fact, trans, refact, panel_size, relax,
diag_pivot_thresh, usepr, drop_tol, perm_c, perm_r,
work, lwork, AA, &AC, &superlumt_options, &Gstat);
/* ------------------------------------------------------------
Compute the LU factorization of A.
The following routine will create nprocs threads.
------------------------------------------------------------*/
pzgstrf(&superlumt_options, &AC, perm_r, L, U, &Gstat, info);
flopcnt = 0;
for (i = 0; i < nprocs; ++i) flopcnt += Gstat.procstat[i].fcops;
ops[FACT] = flopcnt;
#if ( PRNTlevel==1 )
printf("nprocs = %d, flops %e, Mflops %.2f\n",
nprocs, flopcnt, flopcnt/utime[FACT]*1e-6);
printf("Parameters: w %d, relax %d, maxsuper %d, rowblk %d, colblk %d\n",
sp_ienv(1), sp_ienv(2), sp_ienv(3), sp_ienv(4), sp_ienv(5));
fflush(stdout);
#endif
/* ------------------------------------------------------------
Solve the system A*X=B, overwriting B with X.
------------------------------------------------------------*/
if ( *info == 0 ) {
t = SuperLU_timer_();
zgstrs (trans, L, U, perm_r, perm_c, B, &Gstat, info);
utime[SOLVE] = SuperLU_timer_() - t;
ops[SOLVE] = ops[TRISOLVE];
}
/* ------------------------------------------------------------
Deallocate storage after factorization.
------------------------------------------------------------*/
pxgstrf_finalize(&superlumt_options, &AC);
if ( A->Stype == SLU_NR ) {
Destroy_SuperMatrix_Store(AA);
SUPERLU_FREE(AA);
}
/* ------------------------------------------------------------
Print timings, then deallocate statistic variables.
------------------------------------------------------------*/
#ifdef PROFILE
{
SCPformat *Lstore = (SCPformat *) L->Store;
ParallelProfile(n, Lstore->nsuper+1, Gstat.num_panels, nprocs, &Gstat);
}
#endif
PrintStat(&Gstat);
StatFree(&Gstat);
}
void
dgssvx(char *fact, char *trans, char *refact,
SuperMatrix *A, factor_param_t *factor_params, int *perm_c,
int *perm_r, int *etree, char *equed, double *R, double *C,
SuperMatrix *L, SuperMatrix *U, void *work, int lwork,
SuperMatrix *B, SuperMatrix *X, double *recip_pivot_growth,
double *rcond, double *ferr, double *berr,
mem_usage_t *mem_usage, int *info )
{
/*
* Purpose
* =======
*
* DGSSVX solves the system of linear equations A*X=B or A'*X=B, using
* the LU factorization from dgstrf(). Error bounds on the solution and
* a condition estimate are also provided. It performs the following steps:
*
* 1. If A is stored column-wise (A->Stype = NC):
*
* 1.1. If fact = 'E', scaling factors are computed to equilibrate the
* system:
* trans = 'N': diag(R)*A*diag(C) *inv(diag(C))*X = diag(R)*B
* trans = 'T': (diag(R)*A*diag(C))**T *inv(diag(R))*X = diag(C)*B
* trans = 'C': (diag(R)*A*diag(C))**H *inv(diag(R))*X = diag(C)*B
* Whether or not the system will be equilibrated depends on the
* scaling of the matrix A, but if equilibration is used, A is
* overwritten by diag(R)*A*diag(C) and B by diag(R)*B (if trans='N')
* or diag(C)*B (if trans = 'T' or 'C').
*
* 1.2. Permute columns of A, forming A*Pc, where Pc is a permutation
* matrix that usually preserves sparsity.
* For more details of this step, see sp_preorder.c.
*
* 1.3. If fact = 'N' or 'E', the LU decomposition is used to factor the
* matrix A (after equilibration if fact = 'E') as Pr*A*Pc = L*U,
* with Pr determined by partial pivoting.
*
* 1.4. Compute the reciprocal pivot growth factor.
*
* 1.5. If some U(i,i) = 0, so that U is exactly singular, then the
* routine returns with info = i. Otherwise, the factored form of
* A is used to estimate the condition number of the matrix A. If
* the reciprocal of the condition number is less than machine
* precision, info = A->ncol+1 is returned as a warning, but the
* routine still goes on to solve for X and computes error bounds
* as described below.
*
* 1.6. The system of equations is solved for X using the factored form
* of A.
*
* 1.7. Iterative refinement is applied to improve the computed solution
* matrix and calculate error bounds and backward error estimates
* for it.
*
* 1.8. If equilibration was used, the matrix X is premultiplied by
* diag(C) (if trans = 'N') or diag(R) (if trans = 'T' or 'C') so
* that it solves the original system before equilibration.
*
* 2. If A is stored row-wise (A->Stype = NR), apply the above algorithm
* to the transpose of A:
*
* 2.1. If fact = 'E', scaling factors are computed to equilibrate the
* system:
* trans = 'N': diag(R)*A'*diag(C) *inv(diag(C))*X = diag(R)*B
* trans = 'T': (diag(R)*A'*diag(C))**T *inv(diag(R))*X = diag(C)*B
* trans = 'C': (diag(R)*A'*diag(C))**H *inv(diag(R))*X = diag(C)*B
* Whether or not the system will be equilibrated depends on the
* scaling of the matrix A, but if equilibration is used, A' is
* overwritten by diag(R)*A'*diag(C) and B by diag(R)*B
* (if trans='N') or diag(C)*B (if trans = 'T' or 'C').
*
* 2.2. Permute columns of transpose(A) (rows of A),
* forming transpose(A)*Pc, where Pc is a permutation matrix that
* usually preserves sparsity.
* For more details of this step, see sp_preorder.c.
*
* 2.3. If fact = 'N' or 'E', the LU decomposition is used to factor the
* transpose(A) (after equilibration if fact = 'E') as
* Pr*transpose(A)*Pc = L*U with the permutation Pr determined by
* partial pivoting.
*
* 2.4. Compute the reciprocal pivot growth factor.
*
* 2.5. If some U(i,i) = 0, so that U is exactly singular, then the
* routine returns with info = i. Otherwise, the factored form
* of transpose(A) is used to estimate the condition number of the
* matrix A. If the reciprocal of the condition number
* is less than machine precision, info = A->nrow+1 is returned as
* a warning, but the routine still goes on to solve for X and
* computes error bounds as described below.
*
* 2.6. The system of equations is solved for X using the factored form
* of transpose(A).
*
* 2.7. Iterative refinement is applied to improve the computed solution
* matrix and calculate error bounds and backward error estimates
* for it.
*
* 2.8. If equilibration was used, the matrix X is premultiplied by
* diag(C) (if trans = 'N') or diag(R) (if trans = 'T' or 'C') so
* that it solves the original system before equilibration.
*
* See supermatrix.h for the definition of 'SuperMatrix' structure.
*
* Arguments
* =========
*
* fact (input) char*
* Specifies whether or not the factored form of the matrix
* A is supplied on entry, and if not, whether the matrix A should
* be equilibrated before it is factored.
* = 'F': On entry, L, U, perm_r and perm_c contain the factored
* form of A. If equed is not 'N', the matrix A has been
* equilibrated with scaling factors R and C.
* A, L, U, perm_r are not modified.
* = 'N': The matrix A will be factored, and the factors will be
* stored in L and U.
* = 'E': The matrix A will be equilibrated if necessary, then
* factored into L and U.
*
* trans (input) char*
* Specifies the form of the system of equations:
* = 'N': A * X = B (No transpose)
* = 'T': A**T * X = B (Transpose)
* = 'C': A**H * X = B (Transpose)
*
* refact (input) char*
* Specifies whether we want to re-factor the matrix.
* = 'N': Factor the matrix A.
* = 'Y': Matrix A was factored before, now we want to re-factor
* matrix A with perm_r and etree as inputs. Use
* the same storage for the L\U factors previously allocated,
* expand it if necessary. User should insure to use the same
* memory model. In this case, perm_r may be modified due to
* different pivoting determined by diagonal threshold.
* If fact = 'F', then refact is not accessed.
*
* A (input/output) SuperMatrix*
* Matrix A in A*X=B, of dimension (A->nrow, A->ncol). The number
* of the linear equations is A->nrow. Currently, the type of A can be:
* Stype = NC or NR, Dtype = _D, Mtype = GE. In the future,
* more general A can be handled.
*
* On entry, If fact = 'F' and equed is not 'N', then A must have
* been equilibrated by the scaling factors in R and/or C.
* A is not modified if fact = 'F' or 'N', or if fact = 'E' and
* equed = 'N' on exit.
*
* On exit, if fact = 'E' and equed is not 'N', A is scaled as follows:
* If A->Stype = NC:
* equed = 'R': A := diag(R) * A
* equed = 'C': A := A * diag(C)
* equed = 'B': A := diag(R) * A * diag(C).
* If A->Stype = NR:
* equed = 'R': transpose(A) := diag(R) * transpose(A)
* equed = 'C': transpose(A) := transpose(A) * diag(C)
* equed = 'B': transpose(A) := diag(R) * transpose(A) * diag(C).
*
* factor_params (input) factor_param_t*
* The structure defines the input scalar parameters, consisting of
* the following fields. If factor_params = NULL, the default
* values are used for all the fields; otherwise, the values
* are given by the user.
* - panel_size (int): Panel size. A panel consists of at most
* panel_size consecutive columns. If panel_size = -1, use
* default value 8.
* - relax (int): To control degree of relaxing supernodes. If the
* number of nodes (columns) in a subtree of the elimination
* tree is less than relax, this subtree is considered as one
* supernode, regardless of the row structures of those columns.
* If relax = -1, use default value 8.
* - diag_pivot_thresh (double): Diagonal pivoting threshold.
* At step j of the Gaussian elimination, if
* abs(A_jj) >= diag_pivot_thresh * (max_(i>=j) abs(A_ij)),
* then use A_jj as pivot. 0 <= diag_pivot_thresh <= 1.
* If diag_pivot_thresh = -1, use default value 1.0,
* which corresponds to standard partial pivoting.
* - drop_tol (double): Drop tolerance threshold. (NOT IMPLEMENTED)
* At step j of the Gaussian elimination, if
* abs(A_ij)/(max_i abs(A_ij)) < drop_tol,
* then drop entry A_ij. 0 <= drop_tol <= 1.
* If drop_tol = -1, use default value 0.0, which corresponds to
* standard Gaussian elimination.
*
* perm_c (input/output) int*
* If A->Stype = NC, Column permutation vector of size A->ncol,
* which defines the permutation matrix Pc; perm_c[i] = j means
* column i of A is in position j in A*Pc.
* On exit, perm_c may be overwritten by the product of the input
* perm_c and a permutation that postorders the elimination tree
* of Pc'*A'*A*Pc; perm_c is not changed if the elimination tree
* is already in postorder.
*
* If A->Stype = NR, column permutation vector of size A->nrow,
* which describes permutation of columns of transpose(A)
* (rows of A) as described above.
*
* perm_r (input/output) int*
* If A->Stype = NC, row permutation vector of size A->nrow,
* which defines the permutation matrix Pr, and is determined
* by partial pivoting. perm_r[i] = j means row i of A is in
* position j in Pr*A.
*
* If A->Stype = NR, permutation vector of size A->ncol, which
* determines permutation of rows of transpose(A)
* (columns of A) as described above.
*
* If refact is not 'Y', perm_r is output argument;
* If refact = 'Y', the pivoting routine will try to use the input
* perm_r, unless a certain threshold criterion is violated.
* In that case, perm_r is overwritten by a new permutation
* determined by partial pivoting or diagonal threshold pivoting.
*
* etree (input/output) int*, dimension (A->ncol)
* Elimination tree of Pc'*A'*A*Pc.
* If fact is not 'F' and refact = 'Y', etree is an input argument,
* otherwise it is an output argument.
* Note: etree is a vector of parent pointers for a forest whose
* vertices are the integers 0 to A->ncol-1; etree[root]==A->ncol.
*
* equed (input/output) char*
* Specifies the form of equilibration that was done.
* = 'N': No equilibration.
* = 'R': Row equilibration, i.e., A was premultiplied by diag(R).
* = 'C': Column equilibration, i.e., A was postmultiplied by diag(C).
* = 'B': Both row and column equilibration, i.e., A was replaced
* by diag(R)*A*diag(C).
* If fact = 'F', equed is an input argument, otherwise it is
* an output argument.
*
* R (input/output) double*, dimension (A->nrow)
* The row scale factors for A or transpose(A).
* If equed = 'R' or 'B', A (if A->Stype = NC) or transpose(A) (if
* A->Stype = NR) is multiplied on the left by diag(R).
* If equed = 'N' or 'C', R is not accessed.
* If fact = 'F', R is an input argument; otherwise, R is output.
* If fact = 'F' and equed = 'R' or 'B', each element of R must
* be positive.
*
* C (input/output) double*, dimension (A->ncol)
* The column scale factors for A or transpose(A).
* If equed = 'C' or 'B', A (if A->Stype = NC) or transpose(A) (if
* A->Stype = NR) is multiplied on the right by diag(C).
* If equed = 'N' or 'R', C is not accessed.
* If fact = 'F', C is an input argument; otherwise, C is output.
* If fact = 'F' and equed = 'C' or 'B', each element of C must
* be positive.
*
* L (output) SuperMatrix*
* The factor L from the factorization
* Pr*A*Pc=L*U (if A->Stype = NC) or
* Pr*transpose(A)*Pc=L*U (if A->Stype = NR).
* Uses compressed row subscripts storage for supernodes, i.e.,
* L has types: Stype = SC, Dtype = _D, Mtype = TRLU.
*
* U (output) SuperMatrix*
* The factor U from the factorization
* Pr*A*Pc=L*U (if A->Stype = NC) or
* Pr*transpose(A)*Pc=L*U (if A->Stype = NR).
* Uses column-wise storage scheme, i.e., U has types:
* Stype = NC, Dtype = _D, Mtype = TRU.
*
* work (workspace/output) void*, size (lwork) (in bytes)
* User supplied workspace, should be large enough
* to hold data structures for factors L and U.
* On exit, if fact is not 'F', L and U point to this array.
*
* lwork (input) int
* Specifies the size of work array in bytes.
* = 0: allocate space internally by system malloc;
* > 0: use user-supplied work array of length lwork in bytes,
* returns error if space runs out.
* = -1: the routine guesses the amount of space needed without
* performing the factorization, and returns it in
* mem_usage->total_needed; no other side effects.
*
* See argument 'mem_usage' for memory usage statistics.
*
* B (input/output) SuperMatrix*
* B has types: Stype = DN, Dtype = _D, Mtype = GE.
* On entry, the right hand side matrix.
* On exit,
* if equed = 'N', B is not modified; otherwise
* if A->Stype = NC:
* if trans = 'N' and equed = 'R' or 'B', B is overwritten by
* diag(R)*B;
* if trans = 'T' or 'C' and equed = 'C' of 'B', B is
* overwritten by diag(C)*B;
* if A->Stype = NR:
* if trans = 'N' and equed = 'C' or 'B', B is overwritten by
* diag(C)*B;
* if trans = 'T' or 'C' and equed = 'R' of 'B', B is
* overwritten by diag(R)*B.
*
* X (output) SuperMatrix*
* X has types: Stype = DN, Dtype = _D, Mtype = GE.
* If info = 0 or info = A->ncol+1, X contains the solution matrix
* to the original system of equations. Note that A and B are modified
* on exit if equed is not 'N', and the solution to the equilibrated
* system is inv(diag(C))*X if trans = 'N' and equed = 'C' or 'B',
* or inv(diag(R))*X if trans = 'T' or 'C' and equed = 'R' or 'B'.
*
* recip_pivot_growth (output) double*
* The reciprocal pivot growth factor max_j( norm(A_j)/norm(U_j) ).
* The infinity norm is used. If recip_pivot_growth is much less
* than 1, the stability of the LU factorization could be poor.
*
* rcond (output) double*
* The estimate of the reciprocal condition number of the matrix A
* after equilibration (if done). If rcond is less than the machine
* precision (in particular, if rcond = 0), the matrix is singular
* to working precision. This condition is indicated by a return
* code of info > 0.
*
* FERR (output) double*, dimension (B->ncol)
* The estimated forward error bound for each solution vector
* X(j) (the j-th column of the solution matrix X).
* If XTRUE is the true solution corresponding to X(j), FERR(j)
* is an estimated upper bound for the magnitude of the largest
* element in (X(j) - XTRUE) divided by the magnitude of the
* largest element in X(j). The estimate is as reliable as
* the estimate for RCOND, and is almost always a slight
* overestimate of the true error.
*
* BERR (output) double*, dimension (B->ncol)
* The componentwise relative backward error of each solution
* vector X(j) (i.e., the smallest relative change in
* any element of A or B that makes X(j) an exact solution).
*
* mem_usage (output) mem_usage_t*
* Record the memory usage statistics, consisting of following fields:
* - for_lu (float)
* The amount of space used in bytes for L\U data structures.
* - total_needed (float)
* The amount of space needed in bytes to perform factorization.
* - expansions (int)
* The number of memory expansions during the LU factorization.
*
* info (output) int*
* = 0: successful exit
* < 0: if info = -i, the i-th argument had an illegal value
* > 0: if info = i, and i is
* <= A->ncol: U(i,i) is exactly zero. The factorization has
* been completed, but the factor U is exactly
* singular, so the solution and error bounds
* could not be computed.
* = A->ncol+1: U is nonsingular, but RCOND is less than machine
* precision, meaning that the matrix is singular to
* working precision. Nevertheless, the solution and
* error bounds are computed because there are a number
* of situations where the computed solution can be more
* accurate than the value of RCOND would suggest.
* > A->ncol+1: number of bytes allocated when memory allocation
* failure occurred, plus A->ncol.
*
*/
DNformat *Bstore, *Xstore;
double *Bmat, *Xmat;
int ldb, ldx, nrhs;
SuperMatrix *AA; /* A in NC format used by the factorization routine.*/
SuperMatrix AC; /* Matrix postmultiplied by Pc */
int colequ, equil, nofact, notran, rowequ;
char trant[1], norm[1];
int i, j, info1;
double amax, anorm, bignum, smlnum, colcnd, rowcnd, rcmax, rcmin;
int relax, panel_size;
double diag_pivot_thresh, drop_tol;
double t0; /* temporary time */
double *utime;
extern SuperLUStat_t SuperLUStat;
/* External functions */
extern double dlangs(char *, SuperMatrix *);
extern double dlamch_(char *);
Bstore = B->Store;
Xstore = X->Store;
Bmat = Bstore->nzval;
Xmat = Xstore->nzval;
ldb = Bstore->lda;
ldx = Xstore->lda;
nrhs = B->ncol;
#if 0
printf("dgssvx: fact=%c, trans=%c, refact=%c, equed=%c\n",
*fact, *trans, *refact, *equed);
#endif
*info = 0;
nofact = lsame_(fact, "N");
equil = lsame_(fact, "E");
notran = lsame_(trans, "N");
if (nofact || equil) {
*(unsigned char *)equed = 'N';
rowequ = FALSE;
colequ = FALSE;
} else {
rowequ = lsame_(equed, "R") || lsame_(equed, "B");
colequ = lsame_(equed, "C") || lsame_(equed, "B");
smlnum = dlamch_("Safe minimum");
bignum = 1. / smlnum;
}
/* Test the input parameters */
if (!nofact && !equil && !lsame_(fact, "F")) *info = -1;
else if (!notran && !lsame_(trans, "T") && !lsame_(trans, "C")) *info = -2;
else if ( !(lsame_(refact,"Y") || lsame_(refact, "N")) ) *info = -3;
else if ( A->nrow != A->ncol || A->nrow < 0 ||
(A->Stype != NC && A->Stype != NR) ||
A->Dtype != _D || A->Mtype != GE )
*info = -4;
else if (lsame_(fact, "F") && !(rowequ || colequ || lsame_(equed, "N")))
*info = -9;
else {
if (rowequ) {
rcmin = bignum;
rcmax = 0.;
for (j = 0; j < A->nrow; ++j) {
rcmin = MIN(rcmin, R[j]);
rcmax = MAX(rcmax, R[j]);
}
if (rcmin <= 0.) *info = -10;
else if ( A->nrow > 0)
rowcnd = MAX(rcmin,smlnum) / MIN(rcmax,bignum);
else rowcnd = 1.;
}
if (colequ && *info == 0) {
rcmin = bignum;
rcmax = 0.;
for (j = 0; j < A->nrow; ++j) {
rcmin = MIN(rcmin, C[j]);
rcmax = MAX(rcmax, C[j]);
}
if (rcmin <= 0.) *info = -11;
else if (A->nrow > 0)
colcnd = MAX(rcmin,smlnum) / MIN(rcmax,bignum);
else colcnd = 1.;
}
if (*info == 0) {
if ( lwork < -1 ) *info = -15;
else if ( B->ncol < 0 || Bstore->lda < MAX(0, A->nrow) ||
B->Stype != DN || B->Dtype != _D ||
B->Mtype != GE )
*info = -16;
else if ( X->ncol < 0 || Xstore->lda < MAX(0, A->nrow) ||
B->ncol != X->ncol || X->Stype != DN ||
X->Dtype != _D || X->Mtype != GE )
*info = -17;
}
}
if (*info != 0) {
i = -(*info);
xerbla_("dgssvx", &i);
return;
}
/* Default values for factor_params */
panel_size = sp_ienv(1);
relax = sp_ienv(2);
diag_pivot_thresh = 1.0;
drop_tol = 0.0;
if ( factor_params != NULL ) {
if ( factor_params->panel_size != -1 )
panel_size = factor_params->panel_size;
if ( factor_params->relax != -1 ) relax = factor_params->relax;
if ( factor_params->diag_pivot_thresh != -1 )
diag_pivot_thresh = factor_params->diag_pivot_thresh;
if ( factor_params->drop_tol != -1 )
drop_tol = factor_params->drop_tol;
}
StatInit(panel_size, relax);
utime = SuperLUStat.utime;
/* Convert A to NC format when necessary. */
if ( A->Stype == NR ) {
NRformat *Astore = A->Store;
AA = (SuperMatrix *) SUPERLU_MALLOC( sizeof(SuperMatrix) );
dCreate_CompCol_Matrix(AA, A->ncol, A->nrow, Astore->nnz,
Astore->nzval, Astore->colind, Astore->rowptr,
NC, A->Dtype, A->Mtype);
if ( notran ) { /* Reverse the transpose argument. */
*trant = 'T';
notran = 0;
} else {
*trant = 'N';
notran = 1;
}
} else { /* A->Stype == NC */
*trant = *trans;
AA = A;
}
if ( equil ) {
t0 = SuperLU_timer_();
/* Compute row and column scalings to equilibrate the matrix A. */
dgsequ(AA, R, C, &rowcnd, &colcnd, &amax, &info1);
if ( info1 == 0 ) {
/* Equilibrate matrix A. */
dlaqgs(AA, R, C, rowcnd, colcnd, amax, equed);
rowequ = lsame_(equed, "R") || lsame_(equed, "B");
colequ = lsame_(equed, "C") || lsame_(equed, "B");
}
utime[EQUIL] = SuperLU_timer_() - t0;
}
/* Scale the right hand side if equilibration was performed. */
if ( notran ) {
if ( rowequ ) {
for (j = 0; j < nrhs; ++j)
for (i = 0; i < A->nrow; ++i) {
Bmat[i + j*ldb] *= R[i];
}
}
} else if ( colequ ) {
for (j = 0; j < nrhs; ++j)
for (i = 0; i < A->nrow; ++i) {
Bmat[i + j*ldb] *= C[i];
}
}
if ( nofact || equil ) {
t0 = SuperLU_timer_();
sp_preorder(refact, AA, perm_c, etree, &AC);
utime[ETREE] = SuperLU_timer_() - t0;
/* printf("Factor PA = LU ... relax %d\tw %d\tmaxsuper %d\trowblk %d\n",
relax, panel_size, sp_ienv(3), sp_ienv(4));
fflush(stdout); */
/* Compute the LU factorization of A*Pc. */
t0 = SuperLU_timer_();
dgstrf(refact, &AC, diag_pivot_thresh, drop_tol, relax, panel_size,
etree, work, lwork, perm_r, perm_c, L, U, info);
utime[FACT] = SuperLU_timer_() - t0;
if ( lwork == -1 ) {
mem_usage->total_needed = *info - A->ncol;
return;
}
}
if ( *info > 0 ) {
if ( *info <= A->ncol ) {
/* Compute the reciprocal pivot growth factor of the leading
rank-deficient *info columns of A. */
*recip_pivot_growth = dPivotGrowth(*info, AA, perm_c, L, U);
}
return;
}
/* Compute the reciprocal pivot growth factor *recip_pivot_growth. */
*recip_pivot_growth = dPivotGrowth(A->ncol, AA, perm_c, L, U);
/* Estimate the reciprocal of the condition number of A. */
t0 = SuperLU_timer_();
if ( notran ) {
*(unsigned char *)norm = '1';
} else {
*(unsigned char *)norm = 'I';
}
anorm = dlangs(norm, AA);
dgscon(norm, L, U, anorm, rcond, info);
utime[RCOND] = SuperLU_timer_() - t0;
/* Compute the solution matrix X. */
for (j = 0; j < nrhs; j++) /* Save a copy of the right hand sides */
for (i = 0; i < B->nrow; i++)
Xmat[i + j*ldx] = Bmat[i + j*ldb];
t0 = SuperLU_timer_();
dgstrs (trant, L, U, perm_r, perm_c, X, info);
utime[SOLVE] = SuperLU_timer_() - t0;
/* Use iterative refinement to improve the computed solution and compute
error bounds and backward error estimates for it. */
t0 = SuperLU_timer_();
dgsrfs(trant, AA, L, U, perm_r, perm_c, equed, R, C, B,
X, ferr, berr, info);
utime[REFINE] = SuperLU_timer_() - t0;
/* Transform the solution matrix X to a solution of the original system. */
if ( notran ) {
if ( colequ ) {
for (j = 0; j < nrhs; ++j)
for (i = 0; i < A->nrow; ++i) {
Xmat[i + j*ldx] *= C[i];
}
}
} else if ( rowequ ) {
for (j = 0; j < nrhs; ++j)
for (i = 0; i < A->nrow; ++i) {
Xmat[i + j*ldx] *= R[i];
}
}
/* Set INFO = A->ncol+1 if the matrix is singular to working precision. */
if ( *rcond < dlamch_("E") ) *info = A->ncol + 1;
dQuerySpace(L, U, panel_size, mem_usage);
if ( nofact || equil ) Destroy_CompCol_Permuted(&AC);
if ( A->Stype == NR ) {
Destroy_SuperMatrix_Store(AA);
SUPERLU_FREE(AA);
}
PrintStat( &SuperLUStat );
StatFree();
}
void MapMenu()
{
if(Level == 0)
{
Pic[sPlayer]->redraw = true;
Menu->redraw = true;
Hand->redraw = true;
Tile2->redraw = true;
Map->redraw = true;
if(!uButton && P1[UP].v)
{
MoveUp();
mTick = GetTickCount();
}
else if(uButton && P1[UP].v && (GetTickCount() - mTick > mDelay))
{
MoveUp();
mTick = GetTickCount();
}
else if(uButton && !P1[UP].v )
uButton = false;
else if(!dButton && P1[DOWN].v)
{
MoveDown();
mTick = GetTickCount();
}
else if(dButton && P1[DOWN].v && (GetTickCount() - mTick > mDelay))
{
MoveDown();
mTick = GetTickCount();
}
else if(dButton && !P1[DOWN].v )
dButton = false;
else if( !dButton && !uButton && !bButton && P1[B1].v )
{
if(mNum == 0)
{
Level = 1;
bButton = true;
}
else if(mNum == 1)
{
/*
Map->moveable = true;
Tile1->moveable = false;
Hand->Hide();
Camp->Hide();
Tile1->redraw = true;
Menu->redraw = true;
Level = 3;
*/
WalkOnMap = true;
Gray->visable = true;
for(i = 0; i < nGoo; i++)
{
Goo[i]->visable = true;
Goo[i]->moveable = true;
Goo[i]->animate = true;
Goo[i]->movement_done = true;
}
Super->visable = true;
Super->moveable = true;
Super->animate = true;
Super->movement_done = true;
Hand->visable = false;
bButton = true;
}
else if(mNum == 2)
{
gv.ScreenNum = gData->LevelNum;
gv.SubScreenNum = gData->SubLevelNum;
gv.h = 456;
gv.m = 12;
gv.s = 56;
gData->size = sizeof(gv);
gData->data = &gv;
Val->Main->LoadLevel = false;
gData->SubLevelNum = 1;
gData->LevelNum = 1;
bButton = true;
}
else if(mNum == 3)
{
PostMessage(Val->Main->Get_MainHwnd(), WM_DESTROY, 0, 0);
bButton = true;
}
}
else if(!P1[B1].v)
bButton = false;
}
else if(Level == 1)
{
Tile1->moveable = false;
Menu->Hide();
Tile1->redraw = true;
Camp->animate = false;
Tile2->redraw = true;
Map->redraw = true;
Stat[sPlayer]->Show();
Stat[sPlayer]->redraw = true;
ShowStatus(true);
RedrawStat();
PrintStat(sPlayer);
Hand->SetPos(25, 227);
Level = 2;
}
else if(Level == 2)
{
if( !bButton && P1[B1].v )
{
Level = 0;
Menu->Show();
Pic[sPlayer]->Show();
Pic[sPlayer]->redraw = true;
Pic[!sPlayer]->Hide();
Tile1->moveable = true;
Hand->SetPos(Pos[mNum].x, Pos[mNum].y );
Hand->redraw = true;
Map->redraw = true;
Tile2->redraw = true;
Camp->animate = true;
ShowStatus(false);
Stat[0]->Hide();
Stat[1]->Hide();
bButton = true;
}
else if(!P1[B1].v)
bButton = false;
}
/*
else if( (Level == 3) && (Map->movement_done))
{
Tile1->moveable = true;
Hand->Show();
Hand->SetPos(Pos[mNum].x, Pos[mNum].y );
Menu->redraw = true;
Level = 0;
}
*/
if(!lButton && P1[LEFT].v)
{
if(Level == 0)
{
Pic[sPlayer]->Hide();
sPlayer = sPlayer ? 0 : 1;
Pic[sPlayer]->Show();
Pic[sPlayer]->redraw = true;
PrintStat(sPlayer);
RedrawStat();
}
else if(Level == 2)
{
Hand->redraw = true;
Stat[sPlayer]->Hide();
sPlayer = sPlayer ? 0 : 1;
Stat[sPlayer]->Show();
Stat[sPlayer]->redraw = true;
PrintStat(sPlayer);
RedrawStat();
}
lButton = true;
}
else if(lButton && !P1[LEFT].v )
lButton = false;
else if(!rButton && P1[RIGHT].v)
{
if(Level == 0)
{
Pic[sPlayer]->Hide();
sPlayer = sPlayer ? 0 : 1;
Pic[sPlayer]->Show();
Pic[sPlayer]->redraw = true;
PrintStat(sPlayer);
RedrawStat();
}
else if(Level == 2)
{
Hand->redraw = true;
Stat[sPlayer]->Hide();
sPlayer = sPlayer ? 0 : 1;
Stat[sPlayer]->Show();
Stat[sPlayer]->redraw = true;
PrintStat(sPlayer);
RedrawStat();
}
rButton = true;
}
else if(rButton && !P1[RIGHT].v )
rButton = false;
}
void sigusr1_handler(int signal)
{
fprintf(stderr, "ArchC: Received signal %d. Printing statistics\n", signal);
PrintStat();
fprintf(stderr, "ArchC: -------------------- Continuing Simulation ------------------\n");
}
void sigsegv_handler(int signal)
{
fprintf(stderr, "ArchC Error: Segmentation fault.\n");
PrintStat();
exit(EXIT_FAILURE);
}
void sigint_handler(int signal)
{
fprintf(stderr, "ArchC: INTERUPTED BY THE SIGNAL %d\n", signal);
PrintStat();
exit(EXIT_FAILURE);
}
void
psgssvx(int nprocs, superlumt_options_t *superlumt_options, SuperMatrix *A,
int *perm_c, int *perm_r, equed_t *equed, float *R, float *C,
SuperMatrix *L, SuperMatrix *U,
SuperMatrix *B, SuperMatrix *X, float *recip_pivot_growth,
float *rcond, float *ferr, float *berr,
superlu_memusage_t *superlu_memusage, int *info)
{
/*
* -- SuperLU MT routine (version 2.0) --
* Lawrence Berkeley National Lab, Univ. of California Berkeley,
* and Xerox Palo Alto Research Center.
* September 10, 2007
*
* Purpose
* =======
*
* psgssvx() solves the system of linear equations A*X=B or A'*X=B, using
* the LU factorization from sgstrf(). Error bounds on the solution and
* a condition estimate are also provided. It performs the following steps:
*
* 1. If A is stored column-wise (A->Stype = NC):
*
* 1.1. If fact = EQUILIBRATE, scaling factors are computed to equilibrate
* the system:
* trans = NOTRANS: diag(R)*A*diag(C)*inv(diag(C))*X = diag(R)*B
* trans = TRANS: (diag(R)*A*diag(C))**T *inv(diag(R))*X = diag(C)*B
* trans = CONJ: (diag(R)*A*diag(C))**H *inv(diag(R))*X = diag(C)*B
* Whether or not the system will be equilibrated depends on the
* scaling of the matrix A, but if equilibration is used, A is
* overwritten by diag(R)*A*diag(C) and B by diag(R)*B
* (if trans = NOTRANS) or diag(C)*B (if trans = TRANS or CONJ).
*
* 1.2. Permute columns of A, forming A*Pc, where Pc is a permutation matrix
* that usually preserves sparsity.
* For more details of this step, see ssp_colorder.c.
*
* 1.3. If fact = DOFACT or EQUILIBRATE, the LU decomposition is used to
* factor the matrix A (after equilibration if fact = EQUILIBRATE) as
* Pr*A*Pc = L*U, with Pr determined by partial pivoting.
*
* 1.4. Compute the reciprocal pivot growth factor.
*
* 1.5. If some U(i,i) = 0, so that U is exactly singular, then the routine
* returns with info = i. Otherwise, the factored form of A is used to
* estimate the condition number of the matrix A. If the reciprocal of
* the condition number is less than machine precision,
* info = A->ncol+1 is returned as a warning, but the routine still
* goes on to solve for X and computes error bounds as described below.
*
* 1.6. The system of equations is solved for X using the factored form
* of A.
*
* 1.7. Iterative refinement is applied to improve the computed solution
* matrix and calculate error bounds and backward error estimates
* for it.
*
* 1.8. If equilibration was used, the matrix X is premultiplied by
* diag(C) (if trans = NOTRANS) or diag(R) (if trans = TRANS or CONJ)
* so that it solves the original system before equilibration.
*
* 2. If A is stored row-wise (A->Stype = NR), apply the above algorithm
* to the tranpose of A:
*
* 2.1. If fact = EQUILIBRATE, scaling factors are computed to equilibrate
* the system:
* trans = NOTRANS:diag(R)*A'*diag(C)*inv(diag(C))*X = diag(R)*B
* trans = TRANS: (diag(R)*A'*diag(C))**T *inv(diag(R))*X = diag(C)*B
* trans = CONJ: (diag(R)*A'*diag(C))**H *inv(diag(R))*X = diag(C)*B
* Whether or not the system will be equilibrated depends on the
* scaling of the matrix A, but if equilibration is used, A' is
* overwritten by diag(R)*A'*diag(C) and B by diag(R)*B
* (if trans = NOTRANS) or diag(C)*B (if trans = TRANS or CONJ).
*
* 2.2. Permute columns of transpose(A) (rows of A),
* forming transpose(A)*Pc, where Pc is a permutation matrix that
* usually preserves sparsity.
* For more details of this step, see ssp_colorder.c.
*
* 2.3. If fact = DOFACT or EQUILIBRATE, the LU decomposition is used to
* factor the matrix A (after equilibration if fact = EQUILIBRATE) as
* Pr*transpose(A)*Pc = L*U, with the permutation Pr determined by
* partial pivoting.
*
* 2.4. Compute the reciprocal pivot growth factor.
*
* 2.5. If some U(i,i) = 0, so that U is exactly singular, then the routine
* returns with info = i. Otherwise, the factored form of transpose(A)
* is used to estimate the condition number of the matrix A.
* If the reciprocal of the condition number is less than machine
* precision, info = A->nrow+1 is returned as a warning, but the
* routine still goes on to solve for X and computes error bounds
* as described below.
*
* 2.6. The system of equations is solved for X using the factored form
* of transpose(A).
*
* 2.7. Iterative refinement is applied to improve the computed solution
* matrix and calculate error bounds and backward error estimates
* for it.
*
* 2.8. If equilibration was used, the matrix X is premultiplied by
* diag(C) (if trans = NOTRANS) or diag(R) (if trans = TRANS or CONJ)
* so that it solves the original system before equilibration.
*
* See supermatrix.h for the definition of 'SuperMatrix' structure.
*
* Arguments
* =========
*
* nprocs (input) int
* Number of processes (or threads) to be spawned and used to perform
* the LU factorization by psgstrf(). There is a single thread of
* control to call psgstrf(), and all threads spawned by psgstrf()
* are terminated before returning from psgstrf().
*
* superlumt_options (input) superlumt_options_t*
* The structure defines the input parameters and data structure
* to control how the LU factorization will be performed.
* The following fields should be defined for this structure:
*
* o fact (fact_t)
* Specifies whether or not the factored form of the matrix
* A is supplied on entry, and if not, whether the matrix A should
* be equilibrated before it is factored.
* = FACTORED: On entry, L, U, perm_r and perm_c contain the
* factored form of A. If equed is not NOEQUIL, the matrix A has
* been equilibrated with scaling factors R and C.
* A, L, U, perm_r are not modified.
* = DOFACT: The matrix A will be factored, and the factors will be
* stored in L and U.
* = EQUILIBRATE: The matrix A will be equilibrated if necessary,
* then factored into L and U.
*
* o trans (trans_t)
* Specifies the form of the system of equations:
* = NOTRANS: A * X = B (No transpose)
* = TRANS: A**T * X = B (Transpose)
* = CONJ: A**H * X = B (Transpose)
*
* o refact (yes_no_t)
* Specifies whether this is first time or subsequent factorization.
* = NO: this factorization is treated as the first one;
* = YES: it means that a factorization was performed prior to this
* one. Therefore, this factorization will re-use some
* existing data structures, such as L and U storage, column
* elimination tree, and the symbolic information of the
* Householder matrix.
*
* o panel_size (int)
* A panel consists of at most panel_size consecutive columns.
*
* o relax (int)
* To control degree of relaxing supernodes. If the number
* of nodes (columns) in a subtree of the elimination tree is less
* than relax, this subtree is considered as one supernode,
* regardless of the row structures of those columns.
*
* o diag_pivot_thresh (float)
* Diagonal pivoting threshold. At step j of the Gaussian
* elimination, if
* abs(A_jj) >= diag_pivot_thresh * (max_(i>=j) abs(A_ij)),
* use A_jj as pivot, else use A_ij with maximum magnitude.
* 0 <= diag_pivot_thresh <= 1. The default value is 1,
* corresponding to partial pivoting.
*
* o usepr (yes_no_t)
* Whether the pivoting will use perm_r specified by the user.
* = YES: use perm_r; perm_r is input, unchanged on exit.
* = NO: perm_r is determined by partial pivoting, and is output.
*
* o drop_tol (double) (NOT IMPLEMENTED)
* Drop tolerance parameter. At step j of the Gaussian elimination,
* if abs(A_ij)/(max_i abs(A_ij)) < drop_tol, drop entry A_ij.
* 0 <= drop_tol <= 1. The default value of drop_tol is 0,
* corresponding to not dropping any entry.
*
* o work (void*) of size lwork
* User-supplied work space and space for the output data structures.
* Not referenced if lwork = 0;
*
* o lwork (int)
* Specifies the length of work array.
* = 0: allocate space internally by system malloc;
* > 0: use user-supplied work array of length lwork in bytes,
* returns error if space runs out.
* = -1: the routine guesses the amount of space needed without
* performing the factorization, and returns it in
* superlu_memusage->total_needed; no other side effects.
*
* A (input/output) SuperMatrix*
* Matrix A in A*X=B, of dimension (A->nrow, A->ncol), where
* A->nrow = A->ncol. Currently, the type of A can be:
* Stype = NC or NR, Dtype = _D, Mtype = GE. In the future,
* more general A will be handled.
*
* On entry, If superlumt_options->fact = FACTORED and equed is not
* NOEQUIL, then A must have been equilibrated by the scaling factors
* in R and/or C. On exit, A is not modified
* if superlumt_options->fact = FACTORED or DOFACT, or
* if superlumt_options->fact = EQUILIBRATE and equed = NOEQUIL.
*
* On exit, if superlumt_options->fact = EQUILIBRATE and equed is not
* NOEQUIL, A is scaled as follows:
* If A->Stype = NC:
* equed = ROW: A := diag(R) * A
* equed = COL: A := A * diag(C)
* equed = BOTH: A := diag(R) * A * diag(C).
* If A->Stype = NR:
* equed = ROW: transpose(A) := diag(R) * transpose(A)
* equed = COL: transpose(A) := transpose(A) * diag(C)
* equed = BOTH: transpose(A) := diag(R) * transpose(A) * diag(C).
*
* perm_c (input/output) int*
* If A->Stype = NC, Column permutation vector of size A->ncol,
* which defines the permutation matrix Pc; perm_c[i] = j means
* column i of A is in position j in A*Pc.
* On exit, perm_c may be overwritten by the product of the input
* perm_c and a permutation that postorders the elimination tree
* of Pc'*A'*A*Pc; perm_c is not changed if the elimination tree
* is already in postorder.
*
* If A->Stype = NR, column permutation vector of size A->nrow,
* which describes permutation of columns of tranpose(A)
* (rows of A) as described above.
*
* perm_r (input/output) int*
* If A->Stype = NC, row permutation vector of size A->nrow,
* which defines the permutation matrix Pr, and is determined
* by partial pivoting. perm_r[i] = j means row i of A is in
* position j in Pr*A.
*
* If A->Stype = NR, permutation vector of size A->ncol, which
* determines permutation of rows of transpose(A)
* (columns of A) as described above.
*
* If superlumt_options->usepr = NO, perm_r is output argument;
* If superlumt_options->usepr = YES, the pivoting routine will try
* to use the input perm_r, unless a certain threshold criterion
* is violated. In that case, perm_r is overwritten by a new
* permutation determined by partial pivoting or diagonal
* threshold pivoting.
*
* equed (input/output) equed_t*
* Specifies the form of equilibration that was done.
* = NOEQUIL: No equilibration.
* = ROW: Row equilibration, i.e., A was premultiplied by diag(R).
* = COL: Column equilibration, i.e., A was postmultiplied by diag(C).
* = BOTH: Both row and column equilibration, i.e., A was replaced
* by diag(R)*A*diag(C).
* If superlumt_options->fact = FACTORED, equed is an input argument,
* otherwise it is an output argument.
*
* R (input/output) double*, dimension (A->nrow)
* The row scale factors for A or transpose(A).
* If equed = ROW or BOTH, A (if A->Stype = NC) or transpose(A)
* (if A->Stype = NR) is multiplied on the left by diag(R).
* If equed = NOEQUIL or COL, R is not accessed.
* If fact = FACTORED, R is an input argument; otherwise, R is output.
* If fact = FACTORED and equed = ROW or BOTH, each element of R must
* be positive.
*
* C (input/output) double*, dimension (A->ncol)
* The column scale factors for A or transpose(A).
* If equed = COL or BOTH, A (if A->Stype = NC) or trnspose(A)
* (if A->Stype = NR) is multiplied on the right by diag(C).
* If equed = NOEQUIL or ROW, C is not accessed.
* If fact = FACTORED, C is an input argument; otherwise, C is output.
* If fact = FACTORED and equed = COL or BOTH, each element of C must
* be positive.
*
* L (output) SuperMatrix*
* The factor L from the factorization
* Pr*A*Pc=L*U (if A->Stype = NC) or
* Pr*transpose(A)*Pc=L*U (if A->Stype = NR).
* Uses compressed row subscripts storage for supernodes, i.e.,
* L has types: Stype = SCP, Dtype = _D, Mtype = TRLU.
*
* U (output) SuperMatrix*
* The factor U from the factorization
* Pr*A*Pc=L*U (if A->Stype = NC) or
* Pr*transpose(A)*Pc=L*U (if A->Stype = NR).
* Uses column-wise storage scheme, i.e., U has types:
* Stype = NCP, Dtype = _D, Mtype = TRU.
*
* B (input/output) SuperMatrix*
* B has types: Stype = DN, Dtype = _D, Mtype = GE.
* On entry, the right hand side matrix.
* On exit,
* if equed = NOEQUIL, B is not modified; otherwise
* if A->Stype = NC:
* if trans = NOTRANS and equed = ROW or BOTH, B is overwritten
* by diag(R)*B;
* if trans = TRANS or CONJ and equed = COL of BOTH, B is
* overwritten by diag(C)*B;
* if A->Stype = NR:
* if trans = NOTRANS and equed = COL or BOTH, B is overwritten
* by diag(C)*B;
* if trans = TRANS or CONJ and equed = ROW of BOTH, B is
* overwritten by diag(R)*B.
*
* X (output) SuperMatrix*
* X has types: Stype = DN, Dtype = _D, Mtype = GE.
* If info = 0 or info = A->ncol+1, X contains the solution matrix
* to the original system of equations. Note that A and B are modified
* on exit if equed is not NOEQUIL, and the solution to the
* equilibrated system is inv(diag(C))*X if trans = NOTRANS and
* equed = COL or BOTH, or inv(diag(R))*X if trans = TRANS or CONJ
* and equed = ROW or BOTH.
*
* recip_pivot_growth (output) float*
* The reciprocal pivot growth factor computed as
* max_j ( max_i(abs(A_ij)) / max_i(abs(U_ij)) ).
* If recip_pivot_growth is much less than 1, the stability of the
* LU factorization could be poor.
*
* rcond (output) float*
* The estimate of the reciprocal condition number of the matrix A
* after equilibration (if done). If rcond is less than the machine
* precision (in particular, if rcond = 0), the matrix is singular
* to working precision. This condition is indicated by a return
* code of info > 0.
*
* ferr (output) float*, dimension (B->ncol)
* The estimated forward error bound for each solution vector
* X(j) (the j-th column of the solution matrix X).
* If XTRUE is the true solution corresponding to X(j), FERR(j)
* is an estimated upper bound for the magnitude of the largest
* element in (X(j) - XTRUE) divided by the magnitude of the
* largest element in X(j). The estimate is as reliable as
* the estimate for RCOND, and is almost always a slight
* overestimate of the true error.
*
* berr (output) float*, dimension (B->ncol)
* The componentwise relative backward error of each solution
* vector X(j) (i.e., the smallest relative change in
* any element of A or B that makes X(j) an exact solution).
*
* superlu_memusage (output) superlu_memusage_t*
* Record the memory usage statistics, consisting of following fields:
* - for_lu (float)
* The amount of space used in bytes for L\U data structures.
* - total_needed (float)
* The amount of space needed in bytes to perform factorization.
* - expansions (int)
* The number of memory expansions during the LU factorization.
*
* info (output) int*
* = 0: successful exit
* < 0: if info = -i, the i-th argument had an illegal value
* > 0: if info = i, and i is
* <= A->ncol: U(i,i) is exactly zero. The factorization has
* been completed, but the factor U is exactly
* singular, so the solution and error bounds
* could not be computed.
* = A->ncol+1: U is nonsingular, but RCOND is less than machine
* precision, meaning that the matrix is singular to
* working precision. Nevertheless, the solution and
* error bounds are computed because there are a number
* of situations where the computed solution can be more
* accurate than the value of RCOND would suggest.
* > A->ncol+1: number of bytes allocated when memory allocation
* failure occurred, plus A->ncol.
*
*/
NCformat *Astore;
DNformat *Bstore, *Xstore;
float *Bmat, *Xmat;
int ldb, ldx, nrhs;
SuperMatrix *AA; /* A in NC format used by the factorization routine.*/
SuperMatrix AC; /* Matrix postmultiplied by Pc */
int colequ, equil, dofact, notran, rowequ;
char norm[1];
trans_t trant;
int i, j, info1;
float amax, anorm, bignum, smlnum, colcnd, rowcnd, rcmax, rcmin;
int n, relax, panel_size;
Gstat_t Gstat;
double t0; /* temporary time */
double *utime;
flops_t *ops, flopcnt;
/* External functions */
extern float slangs(char *, SuperMatrix *);
extern double slamch_(char *);
Astore = A->Store;
Bstore = B->Store;
Xstore = X->Store;
Bmat = Bstore->nzval;
Xmat = Xstore->nzval;
n = A->ncol;
ldb = Bstore->lda;
ldx = Xstore->lda;
nrhs = B->ncol;
superlumt_options->perm_c = perm_c;
superlumt_options->perm_r = perm_r;
*info = 0;
dofact = (superlumt_options->fact == DOFACT);
equil = (superlumt_options->fact == EQUILIBRATE);
notran = (superlumt_options->trans == NOTRANS);
if (dofact || equil) {
*equed = NOEQUIL;
rowequ = FALSE;
colequ = FALSE;
} else {
rowequ = (*equed == ROW) || (*equed == BOTH);
colequ = (*equed == COL) || (*equed == BOTH);
smlnum = slamch_("Safe minimum");
bignum = 1. / smlnum;
}
/* ------------------------------------------------------------
Test the input parameters.
------------------------------------------------------------*/
if ( nprocs <= 0 ) *info = -1;
else if ( (!dofact && !equil && (superlumt_options->fact != FACTORED))
|| (!notran && (superlumt_options->trans != TRANS) &&
(superlumt_options->trans != CONJ))
|| (superlumt_options->refact != YES &&
superlumt_options->refact != NO)
|| (superlumt_options->usepr != YES &&
superlumt_options->usepr != NO)
|| superlumt_options->lwork < -1 )
*info = -2;
else if ( A->nrow != A->ncol || A->nrow < 0 ||
(A->Stype != SLU_NC && A->Stype != SLU_NR) ||
A->Dtype != SLU_S || A->Mtype != SLU_GE )
*info = -3;
else if ((superlumt_options->fact == FACTORED) &&
!(rowequ || colequ || (*equed == NOEQUIL))) *info = -6;
else {
if (rowequ) {
rcmin = bignum;
rcmax = 0.;
for (j = 0; j < A->nrow; ++j) {
rcmin = SUPERLU_MIN(rcmin, R[j]);
rcmax = SUPERLU_MAX(rcmax, R[j]);
}
if (rcmin <= 0.) *info = -7;
else if ( A->nrow > 0)
rowcnd = SUPERLU_MAX(rcmin,smlnum) / SUPERLU_MIN(rcmax,bignum);
else rowcnd = 1.;
}
if (colequ && *info == 0) {
rcmin = bignum;
rcmax = 0.;
for (j = 0; j < A->nrow; ++j) {
rcmin = SUPERLU_MIN(rcmin, C[j]);
rcmax = SUPERLU_MAX(rcmax, C[j]);
}
if (rcmin <= 0.) *info = -8;
else if (A->nrow > 0)
colcnd = SUPERLU_MAX(rcmin,smlnum) / SUPERLU_MIN(rcmax,bignum);
else colcnd = 1.;
}
if (*info == 0) {
if ( B->ncol < 0 || Bstore->lda < SUPERLU_MAX(0, A->nrow) ||
B->Stype != SLU_DN || B->Dtype != SLU_S ||
B->Mtype != SLU_GE )
*info = -11;
else if ( X->ncol < 0 || Xstore->lda < SUPERLU_MAX(0, A->nrow) ||
B->ncol != X->ncol || X->Stype != SLU_DN ||
X->Dtype != SLU_S || X->Mtype != SLU_GE )
*info = -12;
}
}
if (*info != 0) {
i = -(*info);
xerbla_("psgssvx", &i);
return;
}
/* ------------------------------------------------------------
Allocate storage and initialize statistics variables.
------------------------------------------------------------*/
panel_size = superlumt_options->panel_size;
relax = superlumt_options->relax;
StatAlloc(n, nprocs, panel_size, relax, &Gstat);
StatInit(n, nprocs, &Gstat);
utime = Gstat.utime;
ops = Gstat.ops;
/* ------------------------------------------------------------
Convert A to NC format when necessary.
------------------------------------------------------------*/
if ( A->Stype == SLU_NR ) {
NRformat *Astore = A->Store;
AA = (SuperMatrix *) SUPERLU_MALLOC( sizeof(SuperMatrix) );
sCreate_CompCol_Matrix(AA, A->ncol, A->nrow, Astore->nnz,
Astore->nzval, Astore->colind, Astore->rowptr,
SLU_NC, A->Dtype, A->Mtype);
if ( notran ) { /* Reverse the transpose argument. */
trant = TRANS;
notran = 0;
} else {
trant = NOTRANS;
notran = 1;
}
} else { /* A->Stype == NC */
trant = superlumt_options->trans;
AA = A;
}
/* ------------------------------------------------------------
Diagonal scaling to equilibrate the matrix.
------------------------------------------------------------*/
if ( equil ) {
t0 = SuperLU_timer_();
/* Compute row and column scalings to equilibrate the matrix A. */
sgsequ(AA, R, C, &rowcnd, &colcnd, &amax, &info1);
if ( info1 == 0 ) {
/* Equilibrate matrix A. */
slaqgs(AA, R, C, rowcnd, colcnd, amax, equed);
rowequ = (*equed == ROW) || (*equed == BOTH);
colequ = (*equed == COL) || (*equed == BOTH);
}
utime[EQUIL] = SuperLU_timer_() - t0;
}
/* ------------------------------------------------------------
Scale the right hand side.
------------------------------------------------------------*/
if ( notran ) {
if ( rowequ ) {
for (j = 0; j < nrhs; ++j)
for (i = 0; i < A->nrow; ++i) {
Bmat[i + j*ldb] *= R[i];
}
}
} else if ( colequ ) {
for (j = 0; j < nrhs; ++j)
for (i = 0; i < A->nrow; ++i) {
Bmat[i + j*ldb] *= C[i];
}
}
/* ------------------------------------------------------------
Perform the LU factorization.
------------------------------------------------------------*/
if ( dofact || equil ) {
/* Obtain column etree, the column count (colcnt_h) and supernode
partition (part_super_h) for the Householder matrix. */
t0 = SuperLU_timer_();
sp_colorder(AA, perm_c, superlumt_options, &AC);
utime[ETREE] = SuperLU_timer_() - t0;
#if ( PRNTlevel >= 2 )
printf("Factor PA = LU ... relax %d\tw %d\tmaxsuper %d\trowblk %d\n",
relax, panel_size, sp_ienv(3), sp_ienv(4));
fflush(stdout);
#endif
/* Compute the LU factorization of A*Pc. */
t0 = SuperLU_timer_();
psgstrf(superlumt_options, &AC, perm_r, L, U, &Gstat, info);
utime[FACT] = SuperLU_timer_() - t0;
flopcnt = 0;
for (i = 0; i < nprocs; ++i) flopcnt += Gstat.procstat[i].fcops;
ops[FACT] = flopcnt;
if ( superlumt_options->lwork == -1 ) {
superlu_memusage->total_needed = *info - A->ncol;
return;
}
}
if ( *info > 0 ) {
if ( *info <= A->ncol ) {
/* Compute the reciprocal pivot growth factor of the leading
rank-deficient *info columns of A. */
*recip_pivot_growth = sPivotGrowth(*info, AA, perm_c, L, U);
}
} else {
/* ------------------------------------------------------------
Compute the reciprocal pivot growth factor *recip_pivot_growth.
------------------------------------------------------------*/
*recip_pivot_growth = sPivotGrowth(A->ncol, AA, perm_c, L, U);
/* ------------------------------------------------------------
Estimate the reciprocal of the condition number of A.
------------------------------------------------------------*/
t0 = SuperLU_timer_();
if ( notran ) {
*(unsigned char *)norm = '1';
} else {
*(unsigned char *)norm = 'I';
}
anorm = slangs(norm, AA);
sgscon(norm, L, U, anorm, rcond, info);
utime[RCOND] = SuperLU_timer_() - t0;
/* ------------------------------------------------------------
Compute the solution matrix X.
------------------------------------------------------------*/
for (j = 0; j < nrhs; j++) /* Save a copy of the right hand sides */
for (i = 0; i < B->nrow; i++)
Xmat[i + j*ldx] = Bmat[i + j*ldb];
t0 = SuperLU_timer_();
sgstrs(trant, L, U, perm_r, perm_c, X, &Gstat, info);
utime[SOLVE] = SuperLU_timer_() - t0;
ops[SOLVE] = ops[TRISOLVE];
/* ------------------------------------------------------------
Use iterative refinement to improve the computed solution and
compute error bounds and backward error estimates for it.
------------------------------------------------------------*/
t0 = SuperLU_timer_();
sgsrfs(trant, AA, L, U, perm_r, perm_c, *equed,
R, C, B, X, ferr, berr, &Gstat, info);
utime[REFINE] = SuperLU_timer_() - t0;
/* ------------------------------------------------------------
Transform the solution matrix X to a solution of the original
system.
------------------------------------------------------------*/
if ( notran ) {
if ( colequ ) {
for (j = 0; j < nrhs; ++j)
for (i = 0; i < A->nrow; ++i) {
Xmat[i + j*ldx] *= C[i];
}
}
} else if ( rowequ ) {
for (j = 0; j < nrhs; ++j)
for (i = 0; i < A->nrow; ++i) {
Xmat[i + j*ldx] *= R[i];
}
}
/* Set INFO = A->ncol+1 if the matrix is singular to
working precision.*/
if ( *rcond < slamch_("E") ) *info = A->ncol + 1;
}
superlu_sQuerySpace(nprocs, L, U, panel_size, superlu_memusage);
/* ------------------------------------------------------------
Deallocate storage after factorization.
------------------------------------------------------------*/
if ( superlumt_options->refact == NO ) {
SUPERLU_FREE(superlumt_options->etree);
SUPERLU_FREE(superlumt_options->colcnt_h);
SUPERLU_FREE(superlumt_options->part_super_h);
}
if ( dofact || equil ) {
Destroy_CompCol_Permuted(&AC);
}
if ( A->Stype == SLU_NR ) {
Destroy_SuperMatrix_Store(AA);
SUPERLU_FREE(AA);
}
/* ------------------------------------------------------------
Print timings, then deallocate statistic variables.
------------------------------------------------------------*/
#ifdef PROFILE
{
SCPformat *Lstore = (SCPformat *) L->Store;
ParallelProfile(n, Lstore->nsuper+1, Gstat.num_panels, nprocs, &Gstat);
}
#endif
PrintStat(&Gstat);
StatFree(&Gstat);
}
void
dgssv(SuperMatrix *A, int *perm_c, int *perm_r, SuperMatrix *L,
SuperMatrix *U, SuperMatrix *B, int *info )
{
/*
* Purpose
* =======
*
* DGSSV solves the system of linear equations A*X=B, using the
* LU factorization from DGSTRF. It performs the following steps:
*
* 1. If A is stored column-wise (A->Stype = NC):
*
* 1.1. Permute the columns of A, forming A*Pc, where Pc
* is a permutation matrix. For more details of this step,
* see sp_preorder.c.
*
* 1.2. Factor A as Pr*A*Pc=L*U with the permutation Pr determined
* by Gaussian elimination with partial pivoting.
* L is unit lower triangular with offdiagonal entries
* bounded by 1 in magnitude, and U is upper triangular.
*
* 1.3. Solve the system of equations A*X=B using the factored
* form of A.
*
* 2. If A is stored row-wise (A->Stype = NR), apply the
* above algorithm to the transpose of A:
*
* 2.1. Permute columns of transpose(A) (rows of A),
* forming transpose(A)*Pc, where Pc is a permutation matrix.
* For more details of this step, see sp_preorder.c.
*
* 2.2. Factor A as Pr*transpose(A)*Pc=L*U with the permutation Pr
* determined by Gaussian elimination with partial pivoting.
* L is unit lower triangular with offdiagonal entries
* bounded by 1 in magnitude, and U is upper triangular.
*
* 2.3. Solve the system of equations A*X=B using the factored
* form of A.
*
* See supermatrix.h for the definition of 'SuperMatrix' structure.
*
* Arguments
* =========
*
* A (input) SuperMatrix*
* Matrix A in A*X=B, of dimension (A->nrow, A->ncol). The number
* of linear equations is A->nrow. Currently, the type of A can be:
* Stype = NC or NR; Dtype = _D; Mtype = GE. In the future, more
* general A will be handled.
*
* perm_c (input/output) int*
* If A->Stype = NC, column permutation vector of size A->ncol
* which defines the permutation matrix Pc; perm_c[i] = j means
* column i of A is in position j in A*Pc.
* On exit, perm_c may be overwritten by the product of the input
* perm_c and a permutation that postorders the elimination tree
* of Pc'*A'*A*Pc; perm_c is not changed if the elimination tree
* is already in postorder.
*
* If A->Stype = NR, column permutation vector of size A->nrow
* which describes permutation of columns of transpose(A)
* (rows of A) as described above.
*
* perm_r (output) int*
* If A->Stype = NC, row permutation vector of size A->nrow,
* which defines the permutation matrix Pr, and is determined
* by partial pivoting. perm_r[i] = j means row i of A is in
* position j in Pr*A.
*
* If A->Stype = NR, permutation vector of size A->ncol, which
* determines permutation of rows of transpose(A)
* (columns of A) as described above.
*
* L (output) SuperMatrix*
* The factor L from the factorization
* Pr*A*Pc=L*U (if A->Stype = NC) or
* Pr*transpose(A)*Pc=L*U (if A->Stype = NR).
* Uses compressed row subscripts storage for supernodes, i.e.,
* L has types: Stype = SC, Dtype = _D, Mtype = TRLU.
*
* U (output) SuperMatrix*
* The factor U from the factorization
* Pr*A*Pc=L*U (if A->Stype = NC) or
* Pr*transpose(A)*Pc=L*U (if A->Stype = NR).
* Uses column-wise storage scheme, i.e., U has types:
* Stype = NC, Dtype = _D, Mtype = TRU.
*
* B (input/output) SuperMatrix*
* B has types: Stype = DN, Dtype = _D, Mtype = GE.
* On entry, the right hand side matrix.
* On exit, the solution matrix if info = 0;
*
* info (output) int*
* = 0: successful exit
* > 0: if info = i, and i is
* <= A->ncol: U(i,i) is exactly zero. The factorization has
* been completed, but the factor U is exactly singular,
* so the solution could not be computed.
* > A->ncol: number of bytes allocated when memory allocation
* failure occurred, plus A->ncol.
*
*/
double t1; /* Temporary time */
char refact[1], trans[1];
DNformat *Bstore;
SuperMatrix *AA; /* A in NC format used by the factorization routine.*/
SuperMatrix AC; /* Matrix postmultiplied by Pc */
int lwork = 0, *etree, i;
/* Set default values for some parameters */
double diag_pivot_thresh = 1.0;
double drop_tol = 0;
int panel_size; /* panel size */
int relax; /* no of columns in a relaxed snodes */
double *utime;
extern SuperLUStat_t SuperLUStat;
/* Test the input parameters ... */
*info = 0;
Bstore = B->Store;
if ( A->nrow != A->ncol || A->nrow < 0 ||
(A->Stype != NC && A->Stype != NR) ||
A->Dtype != _D || A->Mtype != GE )
*info = -1;
else if ( B->ncol < 0 || Bstore->lda < MAX(0, A->nrow) ||
B->Stype != DN || B->Dtype != _D || B->Mtype != GE )
*info = -6;
if ( *info != 0 ) {
i = -(*info);
xerbla_("dgssv", &i);
return;
}
*refact = 'N';
*trans = 'N';
panel_size = sp_ienv(1);
relax = sp_ienv(2);
StatInit(panel_size, relax);
utime = SuperLUStat.utime;
/* Convert A to NC format when necessary. */
if ( A->Stype == NR ) {
NRformat *Astore = A->Store;
AA = (SuperMatrix *) SUPERLU_MALLOC( sizeof(SuperMatrix) );
dCreate_CompCol_Matrix(AA, A->ncol, A->nrow, Astore->nnz,
Astore->nzval, Astore->colind, Astore->rowptr,
NC, A->Dtype, A->Mtype);
*trans = 'T';
} else if ( A->Stype == NC ) AA = A;
etree = intMalloc(A->ncol);
t1 = SuperLU_timer_();
sp_preorder(refact, AA, perm_c, etree, &AC);
utime[ETREE] = SuperLU_timer_() - t1;
/*printf("Factor PA = LU ... relax %d\tw %d\tmaxsuper %d\trowblk %d\n",
relax, panel_size, sp_ienv(3), sp_ienv(4));*/
t1 = SuperLU_timer_();
/* Compute the LU factorization of A. */
dgstrf(refact, &AC, diag_pivot_thresh, drop_tol, relax, panel_size,
etree, NULL, lwork, perm_r, perm_c, L, U, info);
utime[FACT] = SuperLU_timer_() - t1;
t1 = SuperLU_timer_();
if ( *info == 0 ) {
/* Solve the system A*X=B, overwriting B with X. */
dgstrs (trans, L, U, perm_r, perm_c, B, info);
}
utime[SOLVE] = SuperLU_timer_() - t1;
SUPERLU_FREE (etree);
Destroy_CompCol_Permuted(&AC);
if ( A->Stype == NR ) {
Destroy_SuperMatrix_Store(AA);
SUPERLU_FREE(AA);
}
PrintStat( &SuperLUStat );
StatFree();
}
BOOL CDynamoRIOView::Refresh()
{
if (m_selected_pid <= 0) {
ZeroStrings();
return TRUE;
}
// We have to grab new stats every refresh
dr_statistics_t *new_stats = get_dynamorio_stats(m_selected_pid);
if (new_stats == NULL) {
if (m_stats != NULL) {
// Leave the stats for an exited process in place, for viewing
// and copying
} else
return TRUE;
} else {
if (m_stats != NULL)
free_dynamorio_stats(m_stats);
m_stats = new_stats;
}
uint i;
#define MAX_VISIBLE_STATS 75
#define STATS_BUFSZ (MAX_VISIBLE_STATS*(sizeof(single_stat_t)*2/*cover %10u, etc.*/))
TCHAR buf[STATS_BUFSZ];
TCHAR *c = buf;
buf[0] = _T('\0');
// If I just use a CString hooked up via DDX, re-setting the
// text causes the top of the visible box to show the top of
// the stats string; plus the scrollbar goes to the top,
// unless held in place; probably any reset of the string
// does an auto-reset of the display and scrollbar position.
// To have all the text in there do this:
// int scroll_pos = m_StatsCtl.GetScrollPos(SB_VERT);
// m_StatsCtl.SetWindowText(buf);
// m_StatsCtl.LineScroll(scroll_pos);
// But then there's a lot of flickering: too much.
// We only put visible text lines in the CEdit box, to reduce the
// flicker. It's too hard to do it w/ the CEdit's own scrollbar,
// which sets itself based on the actual text, so we have a
// separate one.
// HACK: I don't know how to find out how many lines will
// fit w/o putting actual text in there
if (m_StatsViewLines == 0) {
for (i = 0; i < m_stats->num_stats &&
i < MAX_VISIBLE_STATS /* can't be more than this */; i++) {
if (c >= &buf[STATS_BUFSZ-STAT_NAME_MAX_LEN*2])
break;
c += PrintStat(c, i, FALSE/*no filter*/);
assert(c < &buf[STATS_BUFSZ-1]);
}
m_StatsCtl.SetWindowText(buf);
UpdateData(FALSE); // write to screen
// I tried having only one screenful of string there, and
// setting the scrollbar range to be larger, but it doesn't
// seem to support that.
RECT rect;
m_StatsCtl.GetRect(&rect);
CPoint pos(rect.right, rect.bottom);
m_StatsViewLines = HIWORD(m_StatsCtl.CharFromPos(pos));
assert(m_StatsViewLines > 0);
m_StatsSB.SetScrollRange(0, m_stats->num_stats-1, TRUE/*redraw*/);
c = buf;
SCROLLINFO info;
info.cbSize = sizeof(SCROLLINFO);
info.fMask = SIF_PAGE;
info.nPage = m_StatsViewLines;
m_StatsSB.SetScrollInfo(&info);
}
int scroll_pos = m_StatsSB.GetScrollPos();
DWORD shown = 0;
uint printed;
for (i = scroll_pos;
i < m_stats->num_stats && shown < m_StatsViewLines; i++) {
if (c >= &buf[STATS_BUFSZ-STAT_NAME_MAX_LEN*2])
break;
printed = PrintStat(c, i, TRUE/*filter*/);
c += printed;
assert(c < &buf[STATS_BUFSZ-1]);
if (printed > 0)
shown++;
}
m_StatsCtl.SetWindowText(buf);
// num_stats could have changed so update
m_StatsSB.SetScrollRange(0, m_stats->num_stats-1, TRUE/*redraw*/);
if (new_stats == NULL)
m_Exited = _T(" Exited"); // line right end up with Running
else
m_Exited = _T("Running");
#ifndef DRSTATS_DEMO
m_LogLevel.Format(_T("%d"), m_stats->loglevel);
m_LogMask.Format(_T("0x%05X"), m_stats->logmask);
m_LogDir.Format(_T("%") ASCII_PRINTF, m_stats->logdir);
#endif
if (m_clientStats != NULL) {
#define CLIENTSTATS_BUFSZ USHRT_MAX
TCHAR buf[CLIENTSTATS_BUFSZ];
PrintClientStats(buf, &buf[CLIENTSTATS_BUFSZ-1]);
m_ClientStats.Format(_T("%s"), buf);
} else
m_ClientStats.Format(_T(""));
UpdateData(FALSE); // write to screen
return TRUE;
}
void *tokenThread(void *id)
{
long sleep_time;
int queue2_empty;
int token_drop=0;
Packet *pkt;
My402ListElem *elem;
while(1) {
if(token_die==1) {
token_drop_prob = (double)token_drop/tokencount;
pthread_exit(NULL);
}
ttend=getinstanttime();
sleep_time = tokenarrival - (ttend-ttstart);
usleep(sleep_time);
ttstart = getinstanttime();
pthread_mutex_lock(&my_mutex);
if(token_bucket<B) {
token_bucket++;
tokencount++;
PrintStat(getinstanttime());
fprintf(stdout,"token t%d arrives, token bucket now has %d tokens\n",tokencount,token_bucket);
}
else {
PrintStat(getinstanttime());
tokencount++;
token_drop++;
fprintf(stdout,"token t%d arrives, token bucket full, token dropped!!\n",tokencount);
}
if(!My402ListEmpty(&queue1)) {
elem = My402ListFirst(&queue1);
pkt = (Packet *)elem->obj;
if(token_bucket >= pkt->num_tokens) {
token_bucket-=pkt->num_tokens;
My402ListUnlink(&queue1,elem);
pkt->q1_exit = getinstanttime();
PrintStat(getinstanttime());
fprintf(stdout, "p%d leaves Q1, time in Q1 = %.3fms, token bucket now has %d tokens\n",
pkt->pkt_id, (double)(pkt->q1_exit - pkt->q1_enter)/1000, token_bucket);
avg_pkt_q1 += (pkt->q1_exit - pkt->q1_enter);
pkts_left_q1++;
queue2_empty = My402ListEmpty(&queue2);
My402ListAppend(&queue2,pkt);
pkt->q2_enter = getinstanttime();
PrintStat(getinstanttime());
fprintf(stdout,"p%d enters Q2\n", pkt->pkt_id);
if(queue2_empty==1)
pthread_cond_signal(&queue2_cond);
}
}
pthread_mutex_unlock(&my_mutex);
if(pkts_left_q1 == pkts_to_arrive) {
token_drop_prob = (double)token_drop/tokencount;
pthread_exit(NULL);
}
}
}
int main( int argc, char **argv ) {
struct stat stat_buff;
stat_record_t sum_stat;
printer_t print_header, print_record;
nfprof_t profile_data;
char *rfile, *Rfile, *Mdirs, *wfile, *ffile, *filter, *tstring, *stat_type;
char *byte_limit_string, *packet_limit_string, *print_format, *record_header;
char *print_order, *query_file, *UnCompress_file, *nameserver, *aggr_fmt;
int c, ffd, ret, element_stat, fdump;
int i, user_format, quiet, flow_stat, topN, aggregate, aggregate_mask, bidir;
int print_stat, syntax_only, date_sorted, do_tag, compress, do_xstat;
int plain_numbers, GuessDir, pipe_output, csv_output;
time_t t_start, t_end;
uint32_t limitflows;
char Ident[IDENTLEN];
rfile = Rfile = Mdirs = wfile = ffile = filter = tstring = stat_type = NULL;
byte_limit_string = packet_limit_string = NULL;
fdump = aggregate = 0;
aggregate_mask = 0;
bidir = 0;
t_start = t_end = 0;
syntax_only = 0;
topN = -1;
flow_stat = 0;
print_stat = 0;
element_stat = 0;
do_xstat = 0;
limitflows = 0;
date_sorted = 0;
total_bytes = 0;
total_flows = 0;
skipped_blocks = 0;
do_tag = 0;
quiet = 0;
user_format = 0;
compress = 0;
plain_numbers = 0;
pipe_output = 0;
csv_output = 0;
is_anonymized = 0;
GuessDir = 0;
nameserver = NULL;
print_format = NULL;
print_header = NULL;
print_record = NULL;
print_order = NULL;
query_file = NULL;
UnCompress_file = NULL;
aggr_fmt = NULL;
record_header = NULL;
Ident[0] = '\0';
while ((c = getopt(argc, argv, "6aA:Bbc:D:E:s:hHn:i:j:f:qzr:v:w:K:M:NImO:R:XZt:TVv:x:l:L:o:")) != EOF) {
switch (c) {
case 'h':
usage(argv[0]);
exit(0);
break;
case 'a':
aggregate = 1;
break;
case 'A':
if ( !ParseAggregateMask(optarg, &aggr_fmt ) ) {
exit(255);
}
aggregate_mask = 1;
break;
case 'B':
GuessDir = 1;
case 'b':
if ( !SetBidirAggregation() ) {
exit(255);
}
bidir = 1;
// implies
aggregate = 1;
break;
case 'D':
nameserver = optarg;
if ( !set_nameserver(nameserver) ) {
exit(255);
}
break;
case 'E':
query_file = optarg;
if ( !InitExporterList() ) {
exit(255);
}
PrintExporters(query_file);
exit(0);
break;
case 'X':
fdump = 1;
break;
case 'Z':
syntax_only = 1;
break;
case 'q':
quiet = 1;
break;
case 'z':
compress = 1;
break;
case 'c':
limitflows = atoi(optarg);
if ( !limitflows ) {
LogError("Option -c needs a number > 0\n");
exit(255);
}
break;
case 's':
stat_type = optarg;
if ( !SetStat(stat_type, &element_stat, &flow_stat) ) {
exit(255);
}
break;
case 'V': {
char *e1, *e2;
e1 = "";
e2 = "";
#ifdef NSEL
e1 = "NSEL-NEL";
#endif
printf("%s: Version: %s%s%s\n",argv[0], e1, e2, nfdump_version);
exit(0);
} break;
case 'l':
packet_limit_string = optarg;
break;
case 'K':
LogError("*** Anonymisation moved! Use nfanon to anonymise flows!\n");
exit(255);
break;
case 'H':
do_xstat = 1;
break;
case 'L':
byte_limit_string = optarg;
break;
case 'N':
plain_numbers = 1;
break;
case 'f':
ffile = optarg;
break;
case 't':
tstring = optarg;
break;
case 'r':
rfile = optarg;
if ( strcmp(rfile, "-") == 0 )
rfile = NULL;
break;
case 'm':
print_order = "tstart";
Parse_PrintOrder(print_order);
date_sorted = 1;
LogError("Option -m depricated. Use '-O tstart' instead\n");
break;
case 'M':
Mdirs = optarg;
break;
case 'I':
print_stat++;
break;
case 'o': // output mode
print_format = optarg;
break;
case 'O': { // stat order by
int ret;
print_order = optarg;
ret = Parse_PrintOrder(print_order);
if ( ret < 0 ) {
LogError("Unknown print order '%s'\n", print_order);
exit(255);
}
date_sorted = ret == 6; // index into order_mode
} break;
case 'R':
Rfile = optarg;
break;
case 'w':
wfile = optarg;
break;
case 'n':
topN = atoi(optarg);
if ( topN < 0 ) {
LogError("TopnN number %i out of range\n", topN);
exit(255);
}
break;
case 'T':
do_tag = 1;
break;
case 'i':
strncpy(Ident, optarg, IDENT_SIZE);
Ident[IDENT_SIZE - 1] = 0;
if ( strchr(Ident, ' ') ) {
LogError("Ident must not contain spaces\n");
exit(255);
}
break;
case 'j':
UnCompress_file = optarg;
UnCompressFile(UnCompress_file);
exit(0);
break;
case 'x':
query_file = optarg;
InitExtensionMaps(NO_EXTENSION_LIST);
DumpExMaps(query_file);
exit(0);
break;
case 'v':
query_file = optarg;
QueryFile(query_file);
exit(0);
break;
case '6': // print long IPv6 addr
Setv6Mode(1);
break;
default:
usage(argv[0]);
exit(0);
}
}
if (argc - optind > 1) {
usage(argv[0]);
exit(255);
} else {
/* user specified a pcap filter */
filter = argv[optind];
FilterFilename = NULL;
}
// Change Ident only
if ( rfile && strlen(Ident) > 0 ) {
ChangeIdent(rfile, Ident);
exit(0);
}
if ( (element_stat || flow_stat) && (topN == -1) )
topN = 10;
if ( topN < 0 )
topN = 0;
if ( (element_stat && !flow_stat) && aggregate_mask ) {
LogError("Warning: Aggregation ignored for element statistics\n");
aggregate_mask = 0;
}
if ( !flow_stat && aggregate_mask ) {
aggregate = 1;
}
if ( rfile && Rfile ) {
LogError("-r and -R are mutually exclusive. Plase specify either -r or -R\n");
exit(255);
}
if ( Mdirs && !(rfile || Rfile) ) {
LogError("-M needs either -r or -R to specify the file or file list. Add '-R .' for all files in the directories.\n");
exit(255);
}
extension_map_list = InitExtensionMaps(NEEDS_EXTENSION_LIST);
if ( !InitExporterList() ) {
exit(255);
}
SetupInputFileSequence(Mdirs, rfile, Rfile);
if ( print_stat ) {
nffile_t *nffile;
if ( !rfile && !Rfile && !Mdirs) {
LogError("Expect data file(s).\n");
exit(255);
}
memset((void *)&sum_stat, 0, sizeof(stat_record_t));
sum_stat.first_seen = 0x7fffffff;
sum_stat.msec_first = 999;
nffile = GetNextFile(NULL, 0, 0);
if ( !nffile ) {
LogError("Error open file: %s\n", strerror(errno));
exit(250);
}
while ( nffile && nffile != EMPTY_LIST ) {
SumStatRecords(&sum_stat, nffile->stat_record);
nffile = GetNextFile(nffile, 0, 0);
}
PrintStat(&sum_stat);
exit(0);
}
// handle print mode
if ( !print_format ) {
// automatically select an appropriate output format for custom aggregation
// aggr_fmt is compiled by ParseAggregateMask
if ( aggr_fmt ) {
int len = strlen(AggrPrependFmt) + strlen(aggr_fmt) + strlen(AggrAppendFmt) + 7; // +7 for 'fmt:', 2 spaces and '\0'
print_format = malloc(len);
if ( !print_format ) {
LogError("malloc() error in %s line %d: %s\n", __FILE__, __LINE__, strerror(errno) );
exit(255);
}
snprintf(print_format, len, "fmt:%s %s %s",AggrPrependFmt, aggr_fmt, AggrAppendFmt );
print_format[len-1] = '\0';
} else if ( bidir ) {
print_format = "biline";
} else
print_format = DefaultMode;
}
if ( strncasecmp(print_format, "fmt:", 4) == 0 ) {
// special user defined output format
char *format = &print_format[4];
if ( strlen(format) ) {
if ( !ParseOutputFormat(format, plain_numbers, printmap) )
exit(255);
print_record = format_special;
record_header = get_record_header();
user_format = 1;
} else {
LogError("Missing format description for user defined output format!\n");
exit(255);
}
} else {
// predefined output format
// Check for long_v6 mode
i = strlen(print_format);
if ( i > 2 ) {
if ( print_format[i-1] == '6' ) {
Setv6Mode(1);
print_format[i-1] = '\0';
} else
Setv6Mode(0);
}
i = 0;
while ( printmap[i].printmode ) {
if ( strncasecmp(print_format, printmap[i].printmode, MAXMODELEN) == 0 ) {
if ( printmap[i].Format ) {
if ( !ParseOutputFormat(printmap[i].Format, plain_numbers, printmap) )
exit(255);
// predefined custom format
print_record = printmap[i].func;
record_header = get_record_header();
user_format = 1;
} else {
// To support the pipe output format for element stats - check for pipe, and remember this
if ( strncasecmp(print_format, "pipe", MAXMODELEN) == 0 ) {
pipe_output = 1;
}
if ( strncasecmp(print_format, "csv", MAXMODELEN) == 0 ) {
csv_output = 1;
set_record_header();
record_header = get_record_header();
}
// predefined static format
print_record = printmap[i].func;
user_format = 0;
}
break;
}
i++;
}
}
if ( !print_record ) {
LogError("Unknown output mode '%s'\n", print_format);
exit(255);
}
// this is the only case, where headers are printed.
if ( strncasecmp(print_format, "raw", 16) == 0 )
print_header = format_file_block_header;
if ( aggregate && (flow_stat || element_stat) ) {
aggregate = 0;
LogError("Command line switch -s overwrites -a\n");
}
if ( !filter && ffile ) {
if ( stat(ffile, &stat_buff) ) {
LogError("Can't stat filter file '%s': %s\n", ffile, strerror(errno));
exit(255);
}
filter = (char *)malloc(stat_buff.st_size+1);
if ( !filter ) {
LogError("malloc() error in %s line %d: %s\n", __FILE__, __LINE__, strerror(errno) );
exit(255);
}
ffd = open(ffile, O_RDONLY);
if ( ffd < 0 ) {
LogError("Can't open filter file '%s': %s\n", ffile, strerror(errno));
exit(255);
}
ret = read(ffd, (void *)filter, stat_buff.st_size);
if ( ret < 0 ) {
perror("Error reading filter file");
close(ffd);
exit(255);
}
total_bytes += ret;
filter[stat_buff.st_size] = 0;
close(ffd);
FilterFilename = ffile;
}
// if no filter is given, set the default ip filter which passes through every flow
if ( !filter || strlen(filter) == 0 )
filter = "any";
Engine = CompileFilter(filter);
if ( !Engine )
exit(254);
if ( fdump ) {
printf("StartNode: %i Engine: %s\n", Engine->StartNode, Engine->Extended ? "Extended" : "Fast");
DumpList(Engine);
exit(0);
}
if ( syntax_only )
exit(0);
if ( print_order && flow_stat ) {
printf("-s record and -O (-m) are mutually exclusive options\n");
exit(255);
}
if ((aggregate || flow_stat || print_order) && !Init_FlowTable() )
exit(250);
if (element_stat && !Init_StatTable(HashBits, NumPrealloc) )
exit(250);
SetLimits(element_stat || aggregate || flow_stat, packet_limit_string, byte_limit_string);
if ( tstring ) {
if ( !ScanTimeFrame(tstring, &t_start, &t_end) )
exit(255);
}
if ( !(flow_stat || element_stat || wfile || quiet ) && record_header ) {
if ( user_format ) {
printf("%s\n", record_header);
} else {
// static format - no static format with header any more, but keep code anyway
if ( Getv6Mode() ) {
printf("%s\n", record_header);
} else
printf("%s\n", record_header);
}
}
nfprof_start(&profile_data);
sum_stat = process_data(wfile, element_stat, aggregate || flow_stat, print_order != NULL,
print_header, print_record, t_start, t_end,
limitflows, do_tag, compress, do_xstat);
nfprof_end(&profile_data, total_flows);
if ( total_bytes == 0 ) {
printf("No matched flows\n");
exit(0);
}
if (aggregate || print_order) {
if ( wfile ) {
nffile_t *nffile = OpenNewFile(wfile, NULL, compress, is_anonymized, NULL);
if ( !nffile )
exit(255);
if ( ExportFlowTable(nffile, aggregate, bidir, date_sorted, extension_map_list) ) {
CloseUpdateFile(nffile, Ident );
} else {
CloseFile(nffile);
unlink(wfile);
}
DisposeFile(nffile);
} else {
PrintFlowTable(print_record, topN, do_tag, GuessDir, extension_map_list);
}
}
if (flow_stat) {
PrintFlowStat(record_header, print_record, topN, do_tag, quiet, csv_output, extension_map_list);
#ifdef DEVEL
printf("Loopcnt: %u\n", loopcnt);
#endif
}
if (element_stat) {
PrintElementStat(&sum_stat, plain_numbers, record_header, print_record, topN, do_tag, quiet, pipe_output, csv_output);
}
if ( !quiet ) {
if ( csv_output ) {
PrintSummary(&sum_stat, plain_numbers, csv_output);
} else if ( !wfile ) {
if (is_anonymized)
printf("IP addresses anonymised\n");
PrintSummary(&sum_stat, plain_numbers, csv_output);
if ( t_last_flow == 0 ) {
// in case of a pre 1.6.6 collected and empty flow file
printf("Time window: <unknown>\n");
} else {
printf("Time window: %s\n", TimeString(t_first_flow, t_last_flow));
}
printf("Total flows processed: %u, Blocks skipped: %u, Bytes read: %llu\n",
total_flows, skipped_blocks, (unsigned long long)total_bytes);
nfprof_print(&profile_data, stdout);
}
}
Dispose_FlowTable();
Dispose_StatTable();
FreeExtensionMaps(extension_map_list);
#ifdef DEVEL
if ( hash_hit || hash_miss )
printf("Hash hit: %i, miss: %i, skip: %i, ratio: %5.3f\n", hash_hit, hash_miss, hash_skip, (float)hash_hit/((float)(hash_hit+hash_miss)));
#endif
return 0;
}
void *serverThread(void *id)
{
Packet *pkt;
My402ListElem *elem;
int i,pkt_processed = 0;
long serv_start, serv_end, avg_st = 0, avg_pkst = 0,sys_time;
double sys_avg = 0, sys_avg_sq = 0;
double var_sys, var_sys_sq;
for(i=0;i<pkts_to_arrive;i++) {
pthread_mutex_lock(&my_mutex);
while(My402ListEmpty(&queue2) && server_die==0) {
if(i==pkts_to_arrive) // This is the case when the last packet is dropped. And the
break; // server thread is waiting on q2 and no pkt is there to be queued to q2.
pthread_cond_wait(&queue2_cond, &my_mutex);
}
if(i==pkts_to_arrive)
server_die=1;
if(server_die!=0) {
My402ListUnlinkAll(&queue1);
My402ListUnlinkAll(&queue2);
pthread_mutex_unlock(&my_mutex);
break;
}
elem=My402ListFirst(&queue2);
pkt = (Packet *)elem->obj;
My402ListUnlink(&queue2,elem);
pkt->q2_exit = getinstanttime();
PrintStat(getinstanttime());
fprintf(stdout,"p%d begins service at S, time in Q2 = %.3fms\n", pkt->pkt_id,
(double)(pkt->q2_exit - pkt->q2_enter)/1000);
pthread_mutex_unlock(&my_mutex);
avg_pkt_q2 += (pkt->q2_exit - pkt->q2_enter);
serv_start = getinstanttime();
usleep(pkt->service_time);
pkt->sys_exit = serv_end = getinstanttime();
pkt->serv_time = serv_end - serv_start;
avg_st += pkt->serv_time;
avg_pkst += (pkt->sys_exit - pkt->sys_enter);
pkt_processed++;
PrintStat(getinstanttime());
sys_time = pkt->sys_exit - pkt->sys_enter;
fprintf(stdout,"p%d departs from S, service time = %.3fms, time in system = %.3fms\n", pkt->pkt_id,
(double)(pkt->serv_time)/1000,
(double)(sys_time)/1000);
avg_pkt_s += pkt->serv_time;
sys_avg += (double)(sys_time)/1000000;
sys_avg_sq += ((double)(sys_time)/1000000) * ((double)(sys_time)/1000000);
}
// Informing token thread to die.
if(pkt_processed!=0) {
avg_serv_time = (double)((double)avg_st/(double)(pkt_processed*1000000));
avg_pkt_sys_time = (double)((double)avg_pkst/(double)(pkt_processed*1000000));
var_sys = sys_avg/pkt_processed;
var_sys_sq = sys_avg_sq/pkt_processed;
std_deviation = sqrt(var_sys_sq - var_sys*var_sys);
}
else {
avg_serv_time = 0;
avg_pkt_sys_time = 0;
var_sys = 0;
var_sys_sq = 0;
std_deviation = 0;
}
token_die=1;
pthread_exit(NULL);
}
