void Vertsplit::evalAdj(const std::vector<pv_MX>& adjSeed, const std::vector<pv_MX>& adjSens) {
int nadj = adjSeed.size();
int nx = offset_.size()-1;
// Get row offsets
vector<int> row_offset;
row_offset.reserve(offset_.size());
row_offset.push_back(0);
for (std::vector<Sparsity>::const_iterator it=output_sparsity_.begin();
it!=output_sparsity_.end();
++it) {
row_offset.push_back(row_offset.back() + it->size1());
}
for (int d=0; d<nadj; ++d) {
if (adjSens[d][0]!=0) {
vector<MX> v;
for (int i=0; i<nx; ++i) {
MX* x_i = adjSeed[d][i];
if (x_i!=0) {
v.push_back(*x_i);
*x_i = MX();
} else {
v.push_back(MX(output_sparsity_[i].shape()));
}
}
adjSens[d][0]->addToSum(vertcat(v));
}
}
}
int write_traj(FILE *log,t_commrec *cr,
char *traj,t_nsborder *nsb,
int step,real t,real lambda,t_nrnb nrnb[],
int natoms,rvec *xx,rvec *vv,rvec *ff,matrix box)
{
static int fp=-1;
if ((fp == -1) && MASTER(cr)) {
#ifdef DEBUG
fprintf(log,"Going to open trajectory file: %s\n",traj);
#endif
fp = open_trn(traj,"w");
}
#define MX(xvf) moveit(log,cr->left,cr->right,#xvf,xvf,nsb)
if (cr->nnodes > 1) {
MX(xx);
MX(vv);
MX(ff);
}
if ((xx || vv || ff) && MASTER(cr)) {
fwrite_trn(fp,step,t,lambda,box,natoms,xx,vv,ff);
gmx_fio_flush(fp);
}
return fp;
}
void DrawWindowContent() {
int w = nWWidth;
int h = nWHeight;
WSetColor(DARKGRAY);
WFillRectangle(0, 0, w, h);
WSetColor(BLACK);
// axis
WDrawLine(0, h / 2, w, h / 2);
WDrawLine(w / 2, 0, w / 2, h);
// arrows
WDrawLine(w / 2, 0, w / 2 - 5, 5);
WDrawLine(w / 2, 0, w / 2 + 5, 5);
WDrawLine(w, h / 2, w - 5, h / 2 - 5);
WDrawLine(w, h / 2, w - 5, h / 2 + 5);
WDrawString("Press Q to quit, F1...F4 to change function", 10, 20);
WSetColor(GREEN);
for (int i = 0; i < N; i++)
WDrawLine(MX(x[i]), MY(f(x[i])), MX(x[i + 1]), MY(f(x[i + 1])));
WSetColor(BLUE);
double stp = 0.0001;
for (double x = a; x < b; x += stp)
WDrawLine(MX(x), MY(f(x)), MX(x + stp), MY(f(x + stp)));
// WSetColor (RED);
// for (double x = a; x < b; x+=stp)
// WDrawLine(MX(x), MY(F(x)), MX(x+stp), MY(F(x+stp)));
}
void apply_translations(void)
{
unsigned int i;
#define MX(a) (bit_count (conversions_mask & (a)))
if ((MX (C_ASCII | C_EBCDIC | C_IBM) > 1)
|| (MX (C_BLOCK | C_UNBLOCK) > 1)
|| (MX (C_LCASE | C_UCASE) > 1)
|| (MX (C_UNBLOCK | C_SYNC) > 1))
{
log_info("\
only one conv in {ascii,ebcdic,ibm}, {lcase,ucase}, {block,unblock}, {unblock,sync}");
void Inverse::evaluateMX(const MXPtrV& input, MXPtrV& output, const MXPtrVV& fwdSeed,
MXPtrVV& fwdSens, const MXPtrVV& adjSeed, MXPtrVV& adjSens,
bool output_given) {
const MX& X = *input[0];
MX& inv_X = *output[0];
if (!output_given) {
inv_X = inv(X);
}
// Forward sensitivities
int nfwd = fwdSens.size();
for (int d=0; d<nfwd; ++d) {
*fwdSens[d][0] = -mul(inv_X, mul(*fwdSeed[d][0], inv_X));
}
// Adjoint sensitivities
int nadj = adjSeed.size();
if (nadj>0) {
MX trans_inv_X = inv_X.T();
for (int d=0; d<nadj; ++d) {
adjSens[d][0]->addToSum(-mul(trans_inv_X, mul(*adjSeed[d][0], trans_inv_X)));
*adjSeed[d][0] = MX();
}
}
}
void UnaryMX::evaluateMX(const MXPtrV& input, MXPtrV& output, const MXPtrVV& fwdSeed, MXPtrVV& fwdSens, const MXPtrVV& adjSeed, MXPtrVV& adjSens, bool output_given){
// Evaluate function
MX f, dummy; // Function value, dummy second argument
if(output_given){
f = *output[0];
} else {
casadi_math<MX>::fun(op_,*input[0],dummy,f);
}
// Number of forward directions
int nfwd = fwdSens.size();
int nadj = adjSeed.size();
if(nfwd>0 || nadj>0){
// Get partial derivatives
MX pd[2];
casadi_math<MX>::der(op_,*input[0],dummy,f,pd);
// Propagate forward seeds
for(int d=0; d<nfwd; ++d){
*fwdSens[d][0] = pd[0]*(*fwdSeed[d][0]);
}
// Propagate adjoint seeds
for(int d=0; d<nadj; ++d){
MX s = *adjSeed[d][0];
*adjSeed[d][0] = MX();
*adjSens[d][0] += pd[0]*s;
}
}
// Perform the assignment (which may be inplace, hence delayed)
if(!output_given){
*output[0] = f;
}
}
void Determinant::evaluateMX(const MXPtrV& input, MXPtrV& output, const MXPtrVV& fwdSeed,
MXPtrVV& fwdSens, const MXPtrVV& adjSeed, MXPtrVV& adjSens,
bool output_given) {
int nfwd = fwdSens.size();
int nadj = adjSeed.size();
// Non-differentiated output
const MX& X = *input[0];
MX& det_X = *output[0];
if (!output_given) {
det_X = det(X);
}
// Quick return
if (nfwd==0 && nadj==0) return;
// Create only once
MX trans_inv_X = inv(X).T();
// Forward sensitivities
for (int d=0; d<nfwd; ++d) {
*fwdSens[d][0] = det_X * inner_prod(trans_inv_X, *fwdSeed[d][0]);
}
// Adjoint sensitivities
for (int d=0; d<nadj; ++d) {
adjSens[d][0]->addToSum((*adjSeed[d][0]*det_X) * trans_inv_X);
*adjSeed[d][0] = MX();
}
}
AnyScalar pow(const AnyScalar&x, int i) {
if (x.is_double()) return pow(x.as_double(), i);
if (x.is_SX()) return pow(x.as_SX(), SX(i));
if (x.is_MX()) return pow(x.as_MX(), MX(i));
tensor_assert(false);
return 0;
}
void Reshape::evaluateMX(const MXPtrV& input, MXPtrV& output, const MXPtrVV& fwdSeed,
MXPtrVV& fwdSens, const MXPtrVV& adjSeed, MXPtrVV& adjSens,
bool output_given) {
// Quick return if inplace
if (input[0]==output[0]) return;
if (!output_given) {
*output[0] = reshape(*input[0], shape());
}
// Forward sensitivities
int nfwd = fwdSens.size();
for (int d = 0; d<nfwd; ++d) {
*fwdSens[d][0] = reshape(*fwdSeed[d][0], shape());
}
// Adjoint sensitivities
int nadj = adjSeed.size();
for (int d=0; d<nadj; ++d) {
MX& aseed = *adjSeed[d][0];
MX& asens = *adjSens[d][0];
asens.addToSum(reshape(aseed, dep().shape()));
aseed = MX();
}
}
Function Switch
::get_reverse(casadi_int nadj, const std::string& name,
const std::vector<std::string>& inames,
const std::vector<std::string>& onames,
const Dict& opts) const {
// Derivative of each case
vector<Function> der(f_.size());
for (casadi_int k=0; k<f_.size(); ++k) {
if (!f_[k].is_null()) der[k] = f_[k].reverse(nadj);
}
// Default case
Function der_def;
if (!f_def_.is_null()) der_def = f_def_.reverse(nadj);
// New Switch for derivatives
Function sw = Function::conditional("switch_" + name, der, der_def);
// Get expressions for the derivative switch
vector<MX> arg = sw.mx_in();
vector<MX> res = sw(arg);
// No derivatives with respect to index
res.insert(res.begin(), MX(1, nadj));
// Create wrapper
return Function(name, arg, res, inames, onames, opts);
}
/// Convert scalar to matrix
inline static MX toMatrix(const MX& x, const Sparsity& sp) {
if (x.size()==sp.size()) {
return x;
} else {
return MX(sp, x);
}
}
Function Switch
::get_forward(casadi_int nfwd, const std::string& name,
const std::vector<std::string>& inames,
const std::vector<std::string>& onames,
const Dict& opts) const {
// Derivative of each case
vector<Function> der(f_.size());
for (casadi_int k=0; k<f_.size(); ++k) {
if (!f_[k].is_null()) der[k] = f_[k].forward(nfwd);
}
// Default case
Function der_def;
if (!f_def_.is_null()) der_def = f_def_.forward(nfwd);
// New Switch for derivatives
Function sw = Function::conditional("switch_" + name, der, der_def);
// Get expressions for the derivative switch
vector<MX> arg = sw.mx_in();
vector<MX> res = sw(arg);
// Ignore seed for ind
arg.insert(arg.begin() + n_in_ + n_out_, MX(1, nfwd));
// Create wrapper
return Function(name, arg, res, inames, onames, opts);
}
// calculate per cell scores
static void mkscores(void) {int i,j,k,l,m; int*p; float f;
for(i=0;i<nc;i++) for(j=0;j<nl;j++) cells[i].score[j]=0.0;
for(i=0;i<nw;i++) {
m=words[i].length;
p=words[i].flist;
l=words[i].flistlen;
// if(p==NULL) {p=dwds[words[i].length];l=dcount[words[i].length];} // default feasible word list
if(words[i].fe) continue;
if(afunique&&words[i].commit>=0) { // avoid zero score if we've committed
if(l==1) for(k=0;k<m;k++) words[i].c[k]->score[chartol[(int)lts[p[0]].s[k]]]+=1.0;
else assert(l==0);
}
else {
for(j=0;j<l;j++) if(!(afunique&&isused(p[j]))) { // for each remaining feasible word
f=ansp[lts[p[j]].ans]->score;
for(k=0;k<m;k++) words[i].c[k]->score[chartol[(int)lts[p[j]].s[k]]]+=f; // add in its score to this cell's score
}
}
}
for(i=0;i<ne;i++) for(j=0;j<nl;j++) entries[i].score[j]=1.0;
for(i=0;i<nc;i++) {
// f=(float)lcount[cells[i].w->length];
// if(f!=0.0) f=1.0/f;
f=1.0;
for(j=0;j<nl;j++) cells[i].e->score[j]*=f*cells[i].score[j]; // copy scores to entries, scaled by total word count at this length
}
for(i=0;i<ne;i++) {
f=-BIGF; for(j=0;j<nl;j++) f=MX(f,entries[i].score[j]);
entries[i].crux=f; // crux at an entry is the greatest score over all possible letters
}
}
/**
\fn pre-open
*/
bool ADM_ffMpeg2Encoder::configureContext(void)
{
switch(Settings.params.mode)
{
case COMPRESS_2PASS:
case COMPRESS_2PASS_BITRATE:
if(false==setupPass())
{
printf("[ffmpeg] Multipass setup failed\n");
return false;
}
break;
case COMPRESS_SAME:
case COMPRESS_CQ:
_context->flags |= CODEC_FLAG_QSCALE;
_context->bit_rate = 0;
break;
case COMPRESS_CBR:
_context->bit_rate=Settings.params.bitrate*1000; // kb->b;
break;
default:
return false;
}
presetContext(&Settings);
// Override some parameters specific to this codec
// Set matrix if any...
#define MX(a,b,c) case a: _context->intra_matrix=b,_context->inter_matrix=c;break;
switch(Mp2Settings.matrix)
{
MX(MPEG2_MATRIX_DEFAULT,NULL,NULL);
MX(MPEG2_MATRIX_TMPGENC,tmpgenc_intra,tmpgenc_inter);
MX(MPEG2_MATRIX_ANIME,anime_intra,anime_inter);
MX(MPEG2_MATRIX_KVCD,kvcd_intra,kvcd_inter);
default:
ADM_error("unknown matrix type : %d\n",(int)Mp2Settings.matrix);
ADM_assert(0);
break;
}
_context->rc_buffer_size=Mp2Settings.lavcSettings.bufferSize*8*1024;
_context->rc_buffer_size_header=Mp2Settings.lavcSettings.bufferSize*8*1024;
_context->rc_initial_buffer_occupancy=_context->rc_buffer_size;
_context->rc_max_rate=Mp2Settings.lavcSettings.maxBitrate*1000;
_context->rc_max_rate_header=Mp2Settings.lavcSettings.maxBitrate*1000;
// /Override some parameters specific to this codec
return true;
}
void Inverse::evalAdj(const std::vector<pv_MX>& adjSeed, const std::vector<pv_MX>& adjSens) {
MX inv_X = shared_from_this<MX>();
MX trans_inv_X = inv_X.T();
for (int d=0; d<adjSeed.size(); ++d) {
adjSens[d][0]->addToSum(-mul(trans_inv_X, mul(*adjSeed[d][0], trans_inv_X)));
*adjSeed[d][0] = MX();
}
}
void mark_node_id(DI *di, int node_index, int deep, size_t begin, size_t len)
{
(*di).all_node[node_index].id = node_index;
(*di).all_node[node_index].deep = deep;
(*di).all_node[node_index].begin = begin;
(*di).all_node[node_index].len = len;
(*di).the_deep = MX((*di).the_deep, deep+1);
return;
}
void Determinant::evalAdj(const std::vector<pv_MX>& adjSeed, const std::vector<pv_MX>& adjSens) {
const MX& X = dep();
MX det_X = shared_from_this<MX>();
MX trans_inv_X = inv(X).T();
for (int d=0; d<adjSeed.size(); ++d) {
adjSens[d][0]->addToSum((*adjSeed[d][0]*det_X) * trans_inv_X);
*adjSeed[d][0] = MX();
}
}
MX SymbolicMX::join_primitives(std::vector<MX>::const_iterator& it) const {
MX ret = *it++;
if (ret.size()==size()) {
return ret;
} else {
casadi_assert(ret.is_empty(true));
return MX(size());
}
}
//calculate M and B
static inline void calcMB()
{
//todo code of EZ(i, j+1) and EZ(i,j) may be reverse ?
for(int i=1; i<N_PX-1; i++){
for(int j=1; j<N_PY-1; j++){
double complex nowMx = MX(i,j);
Mx[ind(i,j)] = CMX(i,j)*MX(i,j) - CMXEZ(i,j)*(EZ(i,j+1) - EZ(i,j));
Bx[ind(i,j)] = CBX(i,j)*BX(i,j) + CBXMX1(i,j)*MX(i,j) - CBXMX0(i,j)*nowMx;
}
}
//todo code of EZ(i+1, j) and EZ(i,j)
for(int i=1; i<N_PX-1; i++){
for(int j=1; j<N_PY-1; j++){
double complex nowMy = MY(i,j);
My[ind(i,j)] = CMY(i,j)*MY(i,j) - CMYEZ(i,j)*(-EZ(i+1,j) + EZ(i,j));
By[ind(i,j)] = CBY(i,j)*BY(i,j) + CBYMY1(i,j)*MY(i,j) - CBYMY0(i,j)*nowMy;
}
}
}
Function SwitchInternal
::getDerReverse(const std::string& name, int nadj, Dict& opts) {
// Derivative of each case
vector<Function> der(f_.size());
for (int k=0; k<f_.size(); ++k) {
if (!f_[k].isNull()) der[k] = f_[k].derReverse(nadj);
}
// Default case
Function der_def;
if (!f_def_.isNull()) der_def = f_def_.derReverse(nadj);
// New Switch for derivatives
stringstream ss;
ss << "adj" << nadj << "_" << name_;
Switch sw(ss.str(), der, der_def);
// Construct wrapper inputs and arguments for calling sw
vector<MX> arg = symbolicInput();
vector<MX> res = symbolicOutput();
vector<vector<MX> > seed = symbolicAdjSeed(nadj, res);
vector<MX> w_in = arg;
w_in.insert(w_in.end(), res.begin(), res.end());
// Arguments for calling sw being constructed
vector<MX> v;
v.push_back(arg.at(0)); // index
for (int d=0; d<nadj; ++d) {
// Add to wrapper input
w_in.insert(w_in.end(), seed[d].begin(), seed[d].end());
// Add to sw argument vector
v.insert(v.end(), arg.begin()+1, arg.end());
v.insert(v.end(), res.begin(), res.end());
v.insert(v.end(), seed[d].begin(), seed[d].end());
}
// Construct wrapper outputs
casadi_assert(v.size()==sw.nIn());
v = sw(v);
vector<MX> w_out;
MX ind_sens = MX(1, 1); // no dependency on index
vector<MX>::const_iterator v_it = v.begin(), v_it_next;
for (int d=0; d<nadj; ++d) {
w_out.push_back(ind_sens);
v_it_next = v_it + (nIn()-1);
w_out.insert(w_out.end(), v_it, v_it_next);
v_it = v_it_next;
}
// Create wrapper
return MXFunction(name, w_in, w_out, opts);
}
void ClothoidPath::SetOffset( const CarModel& cm, double k, double t, PathPt* l3, const PathPt* l2, const PathPt* l4 )
{
double marg = cm.WIDTH / 2 + 0.02;//1.0;//1.1
double wl = -MN(m_maxL, l3->Wl()) + marg;
double wr = MN(m_maxR, l3->Wr()) - marg;
double buf = MN(1.5, 100 * fabs(k)); // a = v*v/r;
if( k >= 0 )// 0.00001 )
{
if( t < wl )
t = wl;
else if( t > wr - l3->rBuf - buf )
{
if( l3->offs > wr - l3->rBuf - buf )
t = MN(t, l3->offs);
else
t = wr - l3->rBuf - buf;
t = MN(t, wr);
}
}
else //if( k < -0.00001 )
{
if( t > wr )
t = wr;
else if( t < wl + l3->lBuf + buf )
{
if( l3->offs < wl + l3->lBuf + buf )
t = MX(t, l3->offs);
else
t = wl + l3->lBuf + buf;
t = MX(t, wl);
}
}
l3->offs = t;
l3->pt = l3->CalcPt();
l3->k = Utils::CalcCurvatureXY(l2->pt, l3->pt, l4->pt);
}
void write_xtc_traj(FILE *log,t_commrec *cr,
char *xtc_traj,t_nsborder *nsb,t_mdatoms *md,
int step,real t,rvec *xx,matrix box,real prec)
{
static bool bFirst=TRUE;
static rvec *x_sel;
static int natoms;
int i,j;
if ((bFirst) && MASTER(cr)) {
#ifdef DEBUG
fprintf(log,"Going to open compressed trajectory file: %s\n",xtc_traj);
#endif
xd = open_xtc(xtc_traj,"w");
/* Count the number of atoms in the selection */
natoms=0;
for(i=0; (i<md->nr); i++)
if (md->cXTC[i] == 0)
natoms++;
if(log)
fprintf(log,"There are %d atoms in your xtc output selection\n",natoms);
if (natoms != md->nr)
snew(x_sel,natoms);
bFirst=FALSE;
}
if (cr->nnodes > 1) {
MX(xx);
}
if ((xx) && MASTER(cr)) {
if (natoms == md->nr)
x_sel = xx;
else {
/* We need to copy everything into a temp array */
for(i=j=0; (i<md->nr); i++) {
if (md->cXTC[i] == 0) {
copy_rvec(xx[i],x_sel[j]);
j++;
}
}
}
if (write_xtc(xd,natoms,step,t,box,x_sel,prec) == 0)
fatal_error(0,"XTC error");
}
}
void DrawWindowContent() {
int w = nWWidth;
int h = nWHeight;
WSetColor(DARKGRAY);
WFillRectangle(0, 0, w, h);
WSetColor(BLACK);
// axis
WDrawLine(0, h / 2, w, h / 2);
WDrawLine(w / 2, 0, w / 2, h);
// arrows
WDrawLine(w / 2, 0, w / 2 - 5, 5);
WDrawLine(w / 2, 0, w / 2 + 5, 5);
WDrawLine(w, h / 2, w - 5, h / 2 - 5);
WDrawLine(w, h / 2, w - 5, h / 2 + 5);
WSetColor(RED);
WDrawString("Q=quit, F1..F4 -- change scale, F5/F6 -- change node count", 10,
20);
double diff = 0;
for (int i = 1; i <= N - 1; i++) {
int k = i;
if (i == 1)
k = 2;
if (i == N - 1)
k = N - 2; // edge conditions
int ss = int((X(i) - a) / (b - a) * w);
int se = int((X(i + 1) - a) / (b - a) * w);
for (int j = ss; j <= se; j++) {
double t1 = (double(j) / double(w)) * (b - a) + a;
double t2 = (double(j + 1) / double(w)) * (b - a) + a;
WSetColor(GREEN);
WDrawLine(MX(t1), MY(Par(k, t1)), MX(t2), MY(Par(k, t2)));
WSetColor(BLUE);
WDrawLine(MX(t1), MY(f(t1)), MX(t2), MY(f(t2)));
if (diff < fabs(f(t1) - Par(k, t1)))
diff = fabs(f(t1) - Par(k, t1));
}
}
WSetColor(RED);
for (int i = 1; i <= N; i++)
WDrawLine(MX(X(i)), h / 2 - 3, MX(X(i)), h / 2 + 3);
char str[256];
WSetColor(RED);
sprintf(str, "Difference: %1.20lf", diff);
WDrawString(str, 10, 40);
}
void InnerProd::evaluateMX(const MXPtrV& input, MXPtrV& output, const MXPtrVV& fwdSeed, MXPtrVV& fwdSens, const MXPtrVV& adjSeed, MXPtrVV& adjSens, bool output_given){
if(!output_given){
*output[0] = (*input[0])->getInnerProd(*input[1]);
}
// Forward sensitivities
int nfwd = fwdSens.size();
for(int d=0; d<nfwd; ++d){
*fwdSens[d][0] = (*input[0])->getInnerProd(*fwdSeed[d][1]) + (*fwdSeed[d][0])->getInnerProd(*input[1]);
}
// Adjoint sensitivities
int nadj = adjSeed.size();
for(int d=0; d<nadj; ++d){
adjSens[d][0]->addToSum(*adjSeed[d][0] * *input[1]);
adjSens[d][1]->addToSum(*adjSeed[d][0] * *input[0]);
*adjSeed[d][0] = MX();
}
}
void Multiplication<TrX,TrY>::evaluateMX(const MXPtrV& input, MXPtrV& output, const MXPtrVV& fwdSeed, MXPtrVV& fwdSens, const MXPtrVV& adjSeed, MXPtrVV& adjSens, bool output_given){
if(!output_given)
*output[0] = *input[0] + mul(tr<TrX>(*input[1]),tr<TrY>(*input[2]),(*input[0]).sparsity());
// Forward sensitivities
int nfwd = fwdSens.size();
for(int d=0; d<nfwd; ++d){
*fwdSens[d][0] = *fwdSeed[d][0] + mul(tr<TrX>(*input[1]),tr<TrY>(*fwdSeed[d][2]),(*input[0]).sparsity()) + mul(tr<TrX>(*fwdSeed[d][1]),tr<TrY>(*input[2]),(*input[0]).sparsity());
}
// Adjoint sensitivities
int nadj = adjSeed.size();
for(int d=0; d<nadj; ++d){
adjSens[d][1]->addToSum(tr<TrX>(mul(*adjSeed[d][0],tr<!TrY>(*input[2]),tr<TrX>(*input[1]).sparsity())));
adjSens[d][2]->addToSum(tr<TrY>(mul(tr<!TrX>(*input[1]),*adjSeed[d][0],tr<TrY>(*input[2]).sparsity())));
if(adjSeed[d][0]!=adjSens[d][0]){
adjSens[d][0]->addToSum(*adjSeed[d][0]);
*adjSeed[d][0] = MX();
}
}
}
void Diagcat::evaluateMX(const MXPtrV& input, MXPtrV& output, const MXPtrVV& fwdSeed,
MXPtrVV& fwdSens, const MXPtrVV& adjSeed, MXPtrVV& adjSens,
bool output_given) {
int nfwd = fwdSens.size();
int nadj = adjSeed.size();
// Non-differentiated output
if (!output_given) {
*output[0] = diagcat(getVector(input));
}
// Forward sensitivities
for (int d = 0; d<nfwd; ++d) {
*fwdSens[d][0] = diagcat(getVector(fwdSeed[d]));
}
// Quick return?
if (nadj==0) return;
// Get offsets for each row and column
vector<int> offset1(ndep()+1, 0);
vector<int> offset2(ndep()+1, 0);
for (int i=0; i<ndep(); ++i) {
int ncol = dep(i).sparsity().size2();
int nrow = dep(i).sparsity().size1();
offset2[i+1] = offset2[i] + ncol;
offset1[i+1] = offset1[i] + nrow;
}
// Adjoint sensitivities
for (int d=0; d<nadj; ++d) {
MX& aseed = *adjSeed[d][0];
vector<MX> s = diagsplit(aseed, offset1, offset2);
aseed = MX();
for (int i=0; i<ndep(); ++i) {
adjSens[d][i]->addToSum(s[i]);
}
}
}
void OrthBoundingBox::AllDirectionDerivative(const Variable *x, Vector **dgfs, integer UpperBound, double threshold, integer &idxgf)
{
integer *maxidx = new integer[d * 2];
integer *minidx = maxidx + d;
double *maxv = new double[d * 2];
double *minv = maxv + d;
integer idx = 0;
bool findmax = true;
double volumn = 1;
const double *xptr = x->ObtainReadData();
double *XE = new double[d * n];
Matrix MX(xptr, d, d), ME(E, d, n), MXE(XE, d, n);
Matrix::DGEMM(1, MX, false, ME, false, 0, MXE);
//ForDebug::Print("XE:", XE, d, n);//---
RecursiveDirDeri(x, maxidx, minidx, maxv, minv, idx, volumn, findmax, idxgf, dgfs, UpperBound, XE, threshold);
delete[] XE;
delete[] maxidx;
delete[] maxv;
};
MX GenericCall::projectArg(const MX& x, const Sparsity& sp, int i) {
if (x.size()==sp.size()) {
// Insert sparsity projection nodes if needed
return project(x, sp);
} else {
// Different dimensions
if (x.is_empty() || sp.is_empty()) { // NOTE: To permissive?
// Replace nulls with zeros of the right dimension
return MX::zeros(sp);
} else if (x.is_scalar()) {
// Scalar argument means set all
return MX(sp, x);
} else if (x.size1()==sp.size2() && x.size2()==sp.size1() && sp.is_vector()) {
// Transposed vector
return projectArg(x.T(), sp, i);
} else {
// Mismatching dimensions
casadi_error("Cannot create function call node: Dimension mismatch for argument "
<< i << ". Argument has shape " << x.size()
<< " but function input has shape " << sp.size());
}
}
}
void SimpleIndefDleInternal::init() {
DleInternal::init();
casadi_assert_message(!pos_def_,
"pos_def option set to True: Solver only handles the indefinite case.");
n_ = A_.size1();
MX As = MX::sym("A", A_);
MX Vs = MX::sym("V", V_);
MX Vss = (Vs+Vs.T())/2;
MX A_total = DMatrix::eye(n_*n_) - kron(As, As);
MX Pf = solve(A_total, vec(Vss), getOption("linear_solver"));
MX P = reshape(Pf, n_, n_);
f_ = MXFunction(dleIn("a", As, "v", Vs),
dleOut("p", MX(P(output().sparsity()))));
f_.init();
casadi_assert(getNumOutputs()==f_.getNumOutputs());
for (int i=0;i<getNumInputs();++i) {
casadi_assert_message(input(i).sparsity()==f_.input(i).sparsity(),
"Sparsity mismatch for input " << i << ":" <<
input(i).dimString() << " <-> " << f_.input(i).dimString() << ".");
}
for (int i=0;i<getNumOutputs();++i) {
casadi_assert_message(output(i).sparsity()==f_.output(i).sparsity(),
"Sparsity mismatch for output " << i << ":" <<
output(i).dimString() << " <-> " << f_.output(i).dimString() << ".");
}
}
MXFunction vec (const FX &a_) {
FX a = a_;
// Pass null if input is null
if (a.isNull()) return MXFunction();
// Get the MX inputs, only used for shape
const std::vector<MX> &symbolicInputMX = a.symbolicInput();
// Have a vector with MX that have the shape of vec(symbolicInputMX )
std::vector<MX> symbolicInputMX_vec(a.getNumInputs());
// Make vector valued MX's out of them
std::vector<MX> symbolicInputMX_vec_reshape(a.getNumInputs());
// Apply the vec-transformation to the inputs
for (int i=0;i<symbolicInputMX.size();++i) {
std::stringstream s;
s << "X_flat_" << i;
symbolicInputMX_vec[i] = MX(s.str(),vec(symbolicInputMX[i].sparsity()));
symbolicInputMX_vec_reshape[i] = trans(reshape(symbolicInputMX_vec[i],trans(symbolicInputMX[i].sparsity())));
}
// Call the original function with the vecced inputs
std::vector<MX> symbolicOutputMX = a.call(symbolicInputMX_vec_reshape);
// Apply the vec-transformation to the outputs
for (int i=0;i<symbolicOutputMX.size();++i)
symbolicOutputMX[i] = vec(symbolicOutputMX[i]);
// Make a new function with the vecced input/outputs
MXFunction ret(symbolicInputMX_vec,symbolicOutputMX);
// Initialize it if a was
if (a.isInit()) ret.init();
return ret;
}