void des(unsigned char *data_p,char type1,unsigned char* key_p,char type2,int type)
{
unsigned char tempbuf[12];
unsigned char key[7];
unsigned char i;
unsigned char j;
unsigned char count;
void (*f)(unsigned char* data_p,char type);
selectbits(data_p,type1,(unsigned char *)DesIp,PGROM,tempbuf,SRAM,64);/*????*/
movram(tempbuf,SRAM,data_p,type1,8);
selectbits((unsigned char *)key_p,type2,(unsigned char *)DesPc_1,PGROM,(unsigned char *)key,SRAM,56);/*KEY?????*/
for(i = 0;i < 16;i ++)
{
selectbits((unsigned char *)data_p + 4,type1,(unsigned char *)DesE,PGROM,tempbuf,SRAM,48);/*????*/
if(type ==1) //jia mi
{
f = shlc;
count = i;
}
else
{
count = 16 - i;
f = shrc;
}
for(j = 0;j < pgm_read_byte(&DesRots[count]);j++)/*KEY ???*/
{
f(key,SRAM);
}
selectbits(key,SRAM,(unsigned char *)DesPc_2,PGROM,tempbuf + 6,SRAM,48);/*KEY ????*/
doxor(tempbuf,SRAM,tempbuf + 6,SRAM,6);
strans(tempbuf,SRAM,tempbuf + 6,SRAM);
selectbits(tempbuf + 6,SRAM,(unsigned char *)DesP,PGROM,tempbuf,SRAM,32);
doxor(tempbuf,SRAM,data_p,type1,4);
if(i < 15)
{
movram(data_p + 4,type1,data_p,type1,4);
movram(tempbuf,SRAM,data_p + 4,type1,4);
}
}
movram(tempbuf,SRAM,data_p,type1,4);
selectbits(data_p,type1,(unsigned char *)DesIp_1,PGROM,tempbuf,SRAM,64);
movram(tempbuf,SRAM,data_p,type1,8);
}
// Reset the Kalman Filter for a CARMA(p,q) process
void KalmanFilterp::Reset() {
// Initialize the matrix of Eigenvectors. We will work with the state vector
// in the space spanned by the Eigenvectors because in this space the state
// transition matrix is diagonal, so the calculation of the matrix exponential
// is fast.
arma::cx_mat EigenMat(p_,p_);
EigenMat.row(0) = arma::ones<arma::cx_rowvec>(p_);
EigenMat.row(1) = omega_.st();
for (int i=2; i<p_; i++) {
EigenMat.row(i) = strans(arma::pow(omega_, i));
}
// Input vector under original state space representation
arma::cx_vec Rvector = arma::zeros<arma::cx_vec>(p_);
Rvector(p_-1) = 1.0;
// Transform the input vector to the rotated state space representation.
// The notation R and J comes from Belcher et al. (1994).
arma::cx_vec Jvector(p_);
Jvector = arma::solve(EigenMat, Rvector);
// Transform the moving average coefficients to the space spanned by EigenMat.
rotated_ma_coefs_ = ma_coefs_ * EigenMat;
// Calculate the stationary covariance matrix of the state vector.
for (int i=0; i<p_; i++) {
for (int j=i; j<p_; j++) {
// Only fill in upper triangle of StateVar because of symmetry
StateVar_(i,j) = -sigsqr_ * Jvector(i) * std::conj(Jvector(j)) /
(omega_(i) + std::conj(omega_(j)));
}
}
StateVar_ = arma::symmatu(StateVar_); // StateVar is symmetric
PredictionVar_ = StateVar_; // One-step state prediction error
state_vector_.zeros(); // Initial state is set to zero
// Initialize the Kalman mean and variance. These are the forecasted value
// for the measured time series values and its variance, conditional on the
// previous measurements
mean(0) = 0.0;
var(0) = std::real( arma::as_scalar(rotated_ma_coefs_ * StateVar_ * rotated_ma_coefs_.t()) );
var(0) += yerr_(0) * yerr_(0); // Add in measurement error contribution
innovation_ = y_(0); // The innovation
current_index_ = 1;
}
inline
typename
enable_if2
<
(is_arma_type<T1>::value && is_complex_strict<typename T1::elem_type>::value),
Mat< std::complex<typename T1::pod_type> >
>::result
ifft2(const T1& A)
{
arma_extra_debug_sigprint();
// not exactly efficient, but "better-than-nothing" implementation
typedef typename T1::pod_type T;
Mat< std::complex<T> > B = ifft(A);
// for square matrices, strans() will work out that an inplace transpose can be done,
// hence we can potentially avoid creating a temporary matrix
B = strans(B);
return strans( ifft(B) );
}
/**
* Participation coefficient is a measure of diversity of intermodular
* connections of individual nodes.
* Inputs: W, binary/weighted, directed/undirected
* connection matrix
* Ci, community affiliation vector
* Output: P, participation coefficient
* Note: The output for directed graphs is the "out-neighbor"
* participation coefficient.
* Reference: Guimera R, Amaral L. Nature (2005) 433:895-900.
*/
rowvec Connectome::participationCoefficient(const mat &W, const urowvec &Ci)
{
uint n = W.n_rows;
mat binaryW = zeros(n,n);
binaryW(find(W!=0)).fill(1);
rowvec Ko = sum(binaryW,0);
// A^T = (B C)^T = C^T B^T = C B^T since C = C^T
mat Gc = diagmat(conv_to<rowvec>::from(Ci))*strans(abs(sign(W)));
rowvec Kc2 = zeros(1,n);
for (uint i=0;i<=Ci.max();++i) {
mat c = zeros(n,n);
c(find(Gc==i)).fill(1);
Kc2 += arma::pow(sum((W%c),0),2);
}
rowvec P = ones(1,n) - Kc2/arma::pow(Ko,2);
P(find(Ko == 0)).fill(0);
return P;
}
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");
}
}
inline
void
running_stat_vec_aux::update_stats(running_stat_vec< std::complex<T> >& x, const Mat< std::complex<T> >& sample)
{
arma_extra_debug_sigprint();
typedef typename std::complex<T> eT;
const T N = x.counter.value();
if(N > T(0))
{
arma_debug_assert_same_size(x.r_mean, sample, "running_stat_vec(): dimensionality mismatch");
const uword n_elem = sample.n_elem;
const eT* sample_mem = sample.memptr();
eT* r_mean_mem = x.r_mean.memptr();
T* r_var_mem = x.r_var.memptr();
eT* min_val_mem = x.min_val.memptr();
eT* max_val_mem = x.max_val.memptr();
T* min_val_norm_mem = x.min_val_norm.memptr();
T* max_val_norm_mem = x.max_val_norm.memptr();
const T N_plus_1 = x.counter.value_plus_1();
const T N_minus_1 = x.counter.value_minus_1();
if(x.calc_cov == true)
{
Mat<eT>& tmp1 = x.tmp1;
Mat<eT>& tmp2 = x.tmp2;
tmp1 = sample - x.r_mean;
if(sample.n_cols == 1)
{
tmp2 = arma::conj(tmp1)*strans(tmp1);
}
else
{
tmp2 = trans(tmp1)*tmp1; //tmp2 = strans(conj(tmp1))*tmp1;
}
x.r_cov *= (N_minus_1/N);
x.r_cov += tmp2 / N_plus_1;
}
for(uword i=0; i<n_elem; ++i)
{
const eT& val = sample_mem[i];
const T val_norm = std::norm(val);
if(val_norm < min_val_norm_mem[i])
{
min_val_norm_mem[i] = val_norm;
min_val_mem[i] = val;
}
if(val_norm > max_val_norm_mem[i])
{
max_val_norm_mem[i] = val_norm;
max_val_mem[i] = val;
}
const eT& r_mean_val = r_mean_mem[i];
r_var_mem[i] = N_minus_1/N * r_var_mem[i] + std::norm(val - r_mean_val)/N_plus_1;
r_mean_mem[i] = r_mean_val + (val - r_mean_val)/N_plus_1;
}
}
else
{
arma_debug_check( (sample.is_vec() == false), "running_stat_vec(): given sample is not a vector");
x.r_mean.set_size(sample.n_rows, sample.n_cols);
x.r_var.zeros(sample.n_rows, sample.n_cols);
if(x.calc_cov == true)
{
x.r_cov.zeros(sample.n_elem, sample.n_elem);
}
x.min_val.set_size(sample.n_rows, sample.n_cols);
x.max_val.set_size(sample.n_rows, sample.n_cols);
x.min_val_norm.set_size(sample.n_rows, sample.n_cols);
x.max_val_norm.set_size(sample.n_rows, sample.n_cols);
const uword n_elem = sample.n_elem;
const eT* sample_mem = sample.memptr();
eT* r_mean_mem = x.r_mean.memptr();
eT* min_val_mem = x.min_val.memptr();
eT* max_val_mem = x.max_val.memptr();
T* min_val_norm_mem = x.min_val_norm.memptr();
T* max_val_norm_mem = x.max_val_norm.memptr();
for(uword i=0; i<n_elem; ++i)
{
const eT& val = sample_mem[i];
const T val_norm = std::norm(val);
r_mean_mem[i] = val;
min_val_mem[i] = val;
max_val_mem[i] = val;
min_val_norm_mem[i] = val_norm;
max_val_norm_mem[i] = val_norm;
}
}
x.counter++;
}
