int client(const char *hostname, int port, int nmin, int nmax, int nstep, int nrep){
int sock=connect_port(hostname, port, 0, 1);
if(sock<0 || stwriteint(sock, nstep)
|| stwriteint(sock, nmin)
|| stwriteint(sock, nmax)
|| stwriteint(sock, nrep)) {
warning("Unable to connecto to %s\n", hostname);
close(sock); return 1;
}
buf1=(char*)malloc(nmax*nstep);
for(int i=0;i<10;i++){//warm up
stwrite(sock, buf1, nmax);
stread(sock, buf1, 64);
usleep(500);
}
double tim1, tim2, tim3;
int nlen=(nmax-nmin+nstep)/nstep;
dmat *timing=dnew(nrep, nlen);
dmat *timing2=dnew(nrep, nlen);
int ilen=-1;
for(int len=nmin; len<=nmax; len+=nstep){
ilen++;
info("len=%d\n", len);
for(int irep=0; irep<nrep; irep++){
if(irep%800==0){
info("irep=%d of %d\n", irep, nrep);
}
usleep(500);
tim1=myclockd();
stwrite(sock, buf1, len);
tim2=myclockd();
stread(sock, buf1, 64);
tim3=myclockd();
timing->p[irep+ilen*nrep]=tim3-tim1;
timing2->p[irep+ilen*nrep]=tim2-tim1;
}
}
close(sock);
writebin(timing, "pix_timing_%s_%d_%d_%d", HOST, nmin, nmax, nstep);
writebin(timing2, "pix_timing2_%s_%d_%d_%d", HOST, nmin, nmax, nstep);
dbg("done\n");
return 0;
}
int server(int sock){
int nstep;
int nmin;
int nmax;
int nrep;
if(stread(sock, &nstep, sizeof(int))
|| stread(sock, &nmin, sizeof(int))
|| stread(sock, &nmax, sizeof(int))
|| stread(sock, &nrep, sizeof(int))){
return -1;
}
info("nstep=%d, nmin=%d, nmax=%d, nrep=%d\n", nstep,nmin, nmax, nrep);
buf2=(char*)malloc(nmax*nstep);
//warm up
for(int i=0; i<10; i++){
if(stread(sock, buf2, nmax) || stwrite(sock, buf2, 64)){
close(sock);
return -1;
}
}
for(int len=nmin; len<=nmax; len+=nstep){
for(int irep=0; irep<nrep; irep++){
if(stread(sock, buf2, len) || stwrite(sock, buf2, 64)){
close(sock);
return -1;
}
}
}
return -1;
}
/**
Read a string from socket
*/
int streadstr(int sfd, char **str){
int len;
int ans=streadint(sfd, &len);
if(!ans && len>0){
*str=calloc(1, sizeof(char)*len);
ans=stread(sfd, *str, len);
if(ans){
free(*str);
*str=NULL;
}
}else{
*str=NULL;
}
return ans;
}
int main(int argc, char *argv[]){
enum{
P_EXE,
P_FRAC,
P_NSTEP,
P_TOT,
};
if(argc!=P_TOT){
info2("Usage: \n\tenv MVM_CLIENT=hostname MVM_PORT=port MVM_SASTEP=sastep ./mvm_cpu fraction nstep\n");
_Exit(0);
}
int fraction=strtol(argv[P_FRAC], NULL, 10);
int nstep=strtol(argv[P_NSTEP], NULL, 10);
int nstep0=nstep>1?20:0;//warm up
dmat *d_saind=dread("NFIRAOS_saind");
const int nsa=(d_saind->nx-1)/fraction;
int *saind=mymalloc((1+nsa),int);
for(int i=0; i<nsa+1; i++){
saind[i]=(int)d_saind->p[i];
}
dfree(d_saind);
const int totpix=saind[nsa];
const int nact=6981;//active subapertures.
int ng=nsa*2;
float FSMdelta=-0.2;
smat *dm=snew(nact,1);
smat *mvm=snew(nact, ng);
smat *mtch=snew(totpix*2,1);
smat *grad=snew(ng,1);
smat *im0=snew(totpix,3);
short *pix=mymalloc(totpix,short);
short *pixbias=mymalloc(totpix,short);
{
rand_t rseed;
seed_rand(&rseed, 1);
srandu(mvm, 1e-7, &rseed);
srandu(mtch, 1, &rseed);
for(int i=0; i<totpix; i++){
pix[i]=(short)(randu(&rseed)*25565);
pixbias[i]=(short)(randu(&rseed)*1000);
}
}
smat *mvmt=strans(mvm);
int sastep=200;//how many subapertures each time
int nrep=1;
if(getenv("MVM_NREP")){
nrep=strtol(getenv("MVM_NREP"), NULL, 10);
}
if(getenv("MVM_SECT")){
sastep=nsa/strtol(getenv("MVM_SECT"), NULL, 10);
}
if(getenv("MVM_TRANS")){
use_trans=strtol(getenv("MVM_TRANS"), NULL, 10);
}
if(getenv("MVM_SASTEP")){
sastep=strtol(getenv("MVM_SASTEP"), NULL, 10);
}
info2("use_trans=%d, nrep=%d, sastep=%d\n", use_trans, nrep, sastep);
int sock=-1;
char* MVM_CLIENT=getenv("MVM_CLIENT");
if(MVM_CLIENT){
short port=(short)strtol(getenv("MVM_PORT"), NULL, 10);
sock=connect_port(MVM_CLIENT, port, 0 ,1);
if(sock!=-1) {
info2("Connected\n");
int cmd[7];
cmd[0]=nact;
cmd[1]=nsa;
cmd[2]=sastep;
cmd[3]=totpix;
cmd[4]=nstep;
cmd[5]=nstep0;
cmd[6]=2;
if(stwriteintarr(sock, cmd, 7)
|| stwriteintarr(sock, saind, nsa+1)
|| stwrite(sock, pix, sizeof(short)*totpix)){
close(sock); sock=-1;
warning("Failed: %s\n", strerror(errno));
}
}
}
int ready=0;
if(sock!=-1 && stwriteint(sock, ready)){
warning("error send ready signal: %s\n", strerror(errno));
close(sock); sock=-1;
}
smat *timing=snew(nstep, 1);
TIC;
float timtot=0, timmax=0, timmin=INFINITY;
set_realtime(-1, -20);
for(int jstep=-nstep0; jstep<nstep; jstep++){
int istep=jstep<0?0:jstep;
tic;
double theta=M_PI*0.5*istep+FSMdelta;
float cd=cos(theta);
float sd=cos(theta);
szero(dm);
for(int isa=0; isa<nsa; isa+=sastep){
int npixleft;
int nsaleft;
if(nsa<isa+sastep){//terminate
npixleft=totpix-saind[isa];
nsaleft=nsa-isa;
}else{
npixleft=saind[isa+sastep]-saind[isa];
nsaleft=sastep;
}
short *pcur=pix+saind[isa];
if(sock!=-1){
if(stread(sock, pcur, sizeof(short)*npixleft)){
warning("failed: %s\n", strerror(errno));
close(sock); sock=-1;
_Exit(1);
}
if(isa==0) tic;
}
//Matched filter
mtch_do(mtch->p, pix, pixbias,
grad->p+isa*2, im0->p, im0->p+totpix, im0->p+totpix*2,
saind+isa, nsaleft, cd, sd);
//MVM
for(int irep=0; irep<nrep; irep++){
if(use_trans){
mvmt_do(mvmt->p+isa*2, grad->p+isa*2,dm->p, nact, nsaleft*2, ng);
}else{
mvm_do(mvm->p+isa*2*nact, grad->p+isa*2, dm->p, nact, nsaleft*2);
}
}
}//for isa
if(sock!=-1){
if(stwrite(sock, dm->p, sizeof(float)*nact)){
warning("error write dmres: %s\n", strerror(errno));
close(sock); sock=-1;
_Exit(1);
}
if(streadint(sock, &ready)){//acknowledgement.
warning("error read ack failed: %s\n", strerror(errno));
close(sock), sock=-1;
_Exit(1);
}
timing->p[istep]=ready*1.e-6;
}else{
timing->p[istep]=toc3;//do not tic.
}
if(jstep==istep){
timtot+=timing->p[istep];
if(timmax<timing->p[istep]){
timmax=timing->p[istep];
}
if(timmin>timing->p[istep]){
timmin=timing->p[istep];
}
}
}//for istep
float timmean=timtot/nstep;
info2("Timing is mean %.3f, max %.3f min %.3f. BW is %.1f of 51.2GB/s\n",
timmean*1e3, timmax*1e3, timmin*1e3, nrep*(nact*ng+nact+ng)*sizeof(float)/timmean/(1024*1024*1024));
writebin(timing, "cpu_timing_%s", HOST);
if(nstep==1){
writearr("cpu_pix", 1, sizeof(short), M_INT16, NULL, pix, totpix, 1);
writearr("cpu_pixbias", 1, sizeof(short), M_INT16, NULL, pixbias, totpix, 1);
writebin(dm, "cpu_dm");
writebin(grad, "cpu_grad");
writebin(mvm, "cpu_mvm");
writebin(mtch, "cpu_mtch");
}
}
int
fsrestore(int argc, char *argv[])
{
int i, j, status;
int block, bytes;
int blk, nblks, size, mark;
struct partition *pp;
printf("Current SCSI device = ID %d\n", scsi_device);
getline("Is it sure ? (y/n) ", cons_buf);
if ((cons_buf[0] != 'y') && (cons_buf[0] != 'Y'))
return ST_ERROR;
st_rewind(rst0);
st_skip(rst0);
status = stread(rst0, index, LABEL_SIZE);
st_skip(rst0);
for (i = 0; i < MAXPARTITIONS; i++) {
pp = &(lp->d_partitions[i]);
if (pp->p_size > 0) {
printf("%c: ", i + 'A');
printf("size = %d(0x%s), ",
pp->p_size, hexstr(pp->p_size, 8));
printf("offset = %d(0x%s)\n",
pp->p_offset, hexstr(pp->p_offset, 8));
blk = pp->p_offset;
nblks = pp->p_size;
size = nblks << DEV_BSHIFT;
block = BUF_BLOCK;
bytes = BUF_BYTES;
mark = nblks / block;
if (nblks % block)
mark++;
for (j = 0; j < mark; j++)
printf("-");
for (j = 0; j < mark; j++)
printf("%c", '\x08');
while (nblks > 0) {
if (nblks < block) {
block = nblks;
bytes = nblks << DEV_BSHIFT;
}
if (stread(rst0, dump_buf, bytes) != bytes) {
printf("tape read failed !!!\n");
return ST_ERROR;
}
if (!scsi_write(blk, dump_buf, bytes)) {
printf("disk write failed !!!\n");
return ST_ERROR;
}
blk += block;
nblks -= block;
size -= bytes;
printf("#");
}
st_skip(rst0);
printf("\n\n");
}
}
return ST_NORMAL;
}
//respond to scheduler
static int respond(int sock){
int cmd[3];
//read fixed length header info.
if(streadintarr(sock, cmd, 3)){
return -1;//failed
}
int ihost=host_from_sock(sock);
if(ihost>=0){
htime[ihost]=myclockd();
}
int pid=cmd[2];
switch(cmd[0]){
case -1:{//server request shutdown
info("disconnect from %s\n", hosts[ihost]);
return -1;
}
break;
case MON_VERSION:
break;
case MON_STATUS:
{
if(ihost<0){
warning("Host not found\n");
return -1;
}
PROC_T *p=proc_get(ihost,pid);
if(!p){
p=proc_add(ihost,pid);
}
if(stread(sock, &p->status, sizeof(STATUS_T))){
return -1;
}
if(p->status.info==S_REMOVE){
proc_remove(ihost, pid);
}else{
if(cmd[1]!=ihost && cmd[1]!=cmd[2]){
/*A new mean to replace the ID of a job.*/
p->pid=cmd[1];
}
gdk_threads_add_idle((GSourceFunc)refresh, p);
}
}
break;
case MON_PATH:
{
if(ihost<0){
warning("Host not found\n");
return -1;
}
PROC_T *p=proc_get(ihost,pid);
if(!p){
p=proc_add(ihost,pid);
}
if(streadstr(sock, &p->path)){
return -1;
}
char *tmp=NULL;
while((tmp=strchr(p->path, '\n'))){
tmp[0]=' ';
}
}
break;
case MON_LOAD:
{
if(ihost<0){
warning("Host not found\n");
return -1;
}
usage_cpu[ihost]=(double)((pid>>16) & 0xFFFF)/100.;
usage_mem[ihost]=(double)(pid & 0xFFFF)/100.;
usage_cpu[ihost]=MAX(MIN(1,usage_cpu[ihost]),0);
usage_mem[ihost]=MAX(MIN(1,usage_mem[ihost]),0);
gdk_threads_add_idle((GSourceFunc)update_progress, GINT_TO_POINTER(ihost));
}
break;
case MON_ADDHOST:
if(cmd[1]>-1 && cmd[1]<nhost){
pthread_t tmp;
pthread_create(&tmp, NULL, (void*(*)(void*))add_host, GINT_TO_POINTER(cmd[1]));
}else if(cmd[1]==-2){
return -2;
}
break;
default:
warning_time("Invalid cmd %d\n",cmd[0]);
return -1;
}
return 0;
}
//server for mvmfull_real
int mvm_server(int sock){
int cmd[7];
if(streadintarr(sock, cmd, 7)){
return -1;
}
int nact=cmd[0];
int nsa=cmd[1];
int sastep=cmd[2];
int totpix=cmd[3];
int pixpsa=totpix;
int nstep=cmd[4];
int nstep0=cmd[5];
int type=cmd[6];
dbg("type=%d, nact=%d, nsa=%d, sastep=%d, %s=%d, nstep=%d\n",type,
nact, nsa, sastep, type==1?"pixpsa":"totpix", totpix, nstep);
int *saind=NULL;
if(type==1){//mvmfull_iwfs
totpix=pixpsa*nsa;
}else{//mvmfull_real
saind=mymalloc((nsa+1),int);
if(streadintarr(sock, saind, nsa+1)){
return -1;
}
}
short *pix=mymalloc(totpix,short);
if(type==1){
rand_t rseed;
seed_rand(&rseed, 1);
for(int i=0; i<totpix; i++){
pix[i]=(short)randu(&rseed);
}
}else{
if(stread(sock, pix, totpix*sizeof(short))){
return -1;
}
}
smat *dmres=snew(nact, 1);
int ready;
streadint(sock, &ready); //wait for client to be ready.
#if __linux__
struct timespec ct;
clock_gettime(CLOCK_MONOTONIC, &ct);
int readtime_ns=500000;//500 micro-second read out time.
int frametime_ns=1250000;//frame time.
int nsend=((nsa+sastep-1)/sastep);//number of segments sent along read out
int int1_ns=readtime_ns/nsend;//interval between segment sending
int int2_ns=frametime_ns-readtime_ns;//interval after last segment.
#endif
TIC;tic;
for(int istep=-nstep0; istep<nstep; istep++){
//info("\rSend trigger ");
#if __linux__
//scheduled start time of the frame.
double tk0=(double)ct.tv_sec+(double)ct.tv_nsec*1.e-9;
#else
tic;
#endif
for(int isa=0; isa<nsa; isa+=sastep){
#if __linux__
if(clock_nanosleep(CLOCK_MONOTONIC, TIMER_ABSTIME, &ct, NULL)){
warning("clock_nanosleep is interrupted\n");
}
if(isa==0){
tic;
}
#endif
int nleft;
if(type==1){
nleft=((nsa-isa)<sastep?(nsa-isa):sastep)*pixpsa;
}else{
if(nsa<isa+sastep){//terminate
nleft=totpix-saind[isa];
}else{
nleft=saind[isa+sastep]-saind[isa];
}
}
if(stwrite(sock, pix+(type==1?pixpsa*isa:saind[isa]), 2*nleft)){//2 byte data.
warning("failed: %s\n", strerror(errno));
return -1;
}
#if __linux__
ct.tv_nsec+=int1_ns;
while(ct.tv_nsec>=1000000000){
ct.tv_nsec-=1000000000;
ct.tv_sec++;
}
#endif
}
if(stread(sock, dmres->p, sizeof(float)*nact)){
warning("read dmres failed: %s\n", strerror(errno));
return -1;
}
ready=(int)(toc3*1e6);//mvm is finished.
#if __linux__
if(nstep<100){
dbg("tk=%.6f tic=%.6f, toc=%.6f, ready=%.6f\n", tk0, tk, myclockd(), ready*1e-6);
}
#endif
if(stwriteint(sock, ready)){
warning("write ready failed: %s\n", strerror(errno));
return -1;
}
//set next frame start time.
#if __linux__
ct.tv_nsec+=int2_ns;
while(ct.tv_nsec>=1000000000){
ct.tv_nsec-=1000000000;
ct.tv_sec++;
}
#endif
if((istep & 0xFF) == 0xFF){
info("%d %d us.\n", istep, ready);
}
}
info("\n");
return -1;
}
