double do_time_adolc(double *x, double **seed, int d, int nloop,
int tag, const F& f)
{
Teuchos::Time timer("F", false);
int n = F::n;
trace_on(tag);
adouble *x_ad = new adouble[n];
for (int i=0; i<n; i++)
x_ad[i] <<= x[i];
adouble y_ad = 0.0;
f(x_ad, y_ad);
double y;
y_ad >>= y;
delete [] x_ad;
trace_off();
double **jac = new double*[1];
jac[0] = new double[d];
timer.start(true);
for (int j=0; j<nloop; j++)
forward(tag, 1, n, d, x, seed, &y, jac);
timer.stop();
delete [] jac[0];
delete [] jac;
return timer.totalElapsedTime() / nloop;
}
/* The interface function */
double tapingScalarFunction( int tag, double* indeps ) {
int i;
trace_on(tag);
adouble* activeIndeps = new adouble[indepDim];
adouble* aIP = activeIndeps;
double* iP = indeps;
for (i=0; i<indepDim; i++)
*aIP++ <<= *iP++;
adouble ares = activeRosenbrock(indepDim, activeIndeps);
double res = 0;
ares >>= res;
trace_off();
return res;
}
/* MAIN PROGRAM */
int main() {
int i, vc;
int tag = 1;
double z, t, tmin, tmax, tdist, dz;
/*--------------------------------------------------------------------------*/
/* Preparation */
fprintf(stdout,"CUBIC LIGHTHOUSE Using CARDAN (ADOL-C Example)\n\n");
tmin = 0.15;
tmax = 0.24;
fprintf(stdout,"How many values = ? \n");
scanf("%d",&vc);
/*--------------------------------------------------------------------------*/
t = 0.1;
adouble az,at;
trace_on(tag);
at <<= t;
az = activeCubicLighthouse(at);
az >>= z;
trace_off();
/*--------------------------------------------------------------------------*/
size_t tape_stats[STAT_SIZE];
tapestats(tag,tape_stats);
fprintf(stdout,"\n independents %zu\n",tape_stats[NUM_INDEPENDENTS]);
fprintf(stdout," dependents %zu\n",tape_stats[NUM_DEPENDENTS]);
fprintf(stdout," operations %zu\n",tape_stats[NUM_OPERATIONS]);
fprintf(stdout," operations buffer size %zu\n",tape_stats[OP_BUFFER_SIZE]);
fprintf(stdout," locations buffer size %zu\n",tape_stats[LOC_BUFFER_SIZE]);
fprintf(stdout," constants buffer size %zu\n",tape_stats[VAL_BUFFER_SIZE]);
fprintf(stdout," maxlive %zu\n",tape_stats[NUM_MAX_LIVES]);
fprintf(stdout," valstack size %zu\n\n",tape_stats[TAY_STACK_SIZE]);
/*--------------------------------------------------------------------------*/
tdist = (tmax-tmin)/((double) (vc-1));
t = tmin;
for (i=0; i<vc; i++) {
function(tag,1,1,&t,&z);
gradient(tag,1,&t,&dz);
fprintf(stdout,"%e %e %e\n",t,z,dz);
t += tdist;
}
/*--------------------------------------------------------------------------*/
return 1;
}
static void ileaf_dump(struct btree *btree, void *vleaf)
{
if (!tux3_trace)
return;
struct ileaf_attr_ops *attr_ops = btree->ops->private_ops;
struct ileaf *leaf = vleaf;
inum_t inum = ibase(leaf);
__be16 *dict = ileaf_dict(btree, leaf);
unsigned offset = 0;
trace_on("inode table block 0x%Lx/%i (%x bytes free)",
ibase(leaf), icount(leaf), ileaf_free(btree, leaf));
for (int i = 0; i < icount(leaf); i++, inum++) {
int limit = __atdict(dict, i + 1), size = limit - offset;
if (!size)
continue;
if (size < 0)
trace_on(" 0x%Lx: <corrupt>\n", inum);
else if (!size)
trace_on(" 0x%Lx: <empty>\n", inum);
else if (attr_ops == &iattr_ops) {
/* FIXME: this doesn't work in kernel */
struct tux3_inode tuxnode = {};
struct inode *inode = &tuxnode.vfs_inode;
void *attrs = leaf->table + offset;
inode->i_sb = vfs_sb(btree->sb),
attr_ops->decode(btree, inode, attrs, size);
free_xcache(inode);
}
offset = limit;
}
}
/* The interface function */
void tapingVectorFunction( int tag, double* indEPS_, double* dEPS_ ) {
int i;
trace_on(tag);
adouble* activeIndEPS_ = new adouble[indepDim];
adouble* activeDEPS_ = new adouble[depDim];
adouble* aIP = activeIndEPS_;
double* iP = indEPS_;
for (i=0; i<indepDim; i++)
*aIP++ <<= *iP++;
activeEutroph(activeIndEPS_,activeDEPS_);
aIP = activeDEPS_;
iP = dEPS_;
for (i=0; i<depDim; i++)
*aIP++ >>= *iP++;
trace_off();
}
/* The interface function */
void tapingVectorFunction( int tag, double* indeps, double* deps ) {
int i;
trace_on(tag);
adouble* activeIndeps = new adouble[indepDim];
adouble* activeDeps = new adouble[depDim];
adouble* aIP = activeIndeps;
double* iP = indeps;
for (i=0; i<indepDim; i++)
*aIP++ <<= *iP++;
activeGearFunction(activeIndeps,activeDeps);
aIP = activeDeps;
iP = deps;
for (i=0; i<depDim; i++)
*aIP++ >>= *iP++;
trace_off();
}
int main(int argc, char* argv[]) {
vector<double> ind;
int n = 0;
int m = 0;
set_up(argc, argv, ind, n, m);
std::cout << "num_dep = " << m << ", num_ind = " << n << std::endl;
adouble* xad = new adouble[n];
adouble* yad = new adouble[m];
double* y = new double[m];
double* x = new double[n];
for (int i = 0; i < n; i++) {
x[i] = ind[i];
}
trace_on(TAG);
for (int i = 0; i < n; i++) {
xad[i] <<= x[i];
}
func_eval<adouble>(n, xad, m, yad);
#ifdef DUMMY_SCALAR
adouble zad = 0;
double z;
for (int i = 0; i < m; i++) {
zad = zad + yad[i];
}
zad >>= z;
std::cout << "func_eval = " << z << std::endl;
m = 1;
#else
for (int i = 0; i < m; i++) {
yad[i] >>= y[i];
//std::cout << "y["<<i<<"] = " << y[i] << std::endl;
}
#endif
trace_off();
tear_down();
// Evaluate derivatives;
int options[2] = {ORDER, METHOD};
double t;
std::cout << "m = " << m << " n = " << n << std::endl;
t = evaluate_derivatives(n, m, x, options);
delete[] xad;
delete[] yad;
}
int main() {
double xyz[3], f; /* variables */
adouble ax, ay, az, af; /* active varaibles */
xyz[0] = 1.2;
xyz[1] = 2.6; /* initialize any values */
xyz[2] = 0.03;
/* TRACING THE EVALUATION TAPE */
trace_on(1); /* start tracing of an evaluation
tape with the identifier 1 */
ax <<= xyz[0]; /* marking independent variables */
ay <<= xyz[1];
az <<= xyz[2];
af = myf(ax,ay,az); /* calling the 'active' version of
the function to be differentiated
to generate a tape of the evaluation
process;
NOTE: Instead of calling a
C function the whole evaluation code
can be placed here (see example file
DEX/powerexam.C) */
af >>= f; /* marking the only one dependent
variable */
trace_off(1); /* stop tracing */
/* NOTE: trace_off(..) is called with the value 1 (for the optional
! argument). This forces ADOL-C to save the generated tapes
! on harddisc. In particular these are the files
!
! _adol-op_tape.1 (operations = opcodes)
! _adol-in_tape.1 (integers = locations)
! _adol-rl_tape.1 (real values = doubles)
!
! The appendix '1' is determined by the used tape
! identifier, which was passed to trace_on(..).
*/
}
void adolc_init_fill(bool retape,
int tag,
const ElemData& e,
unsigned int neqn,
const std::vector<double>& x,
const std::vector< std::vector<double> >& w,
std::vector<double>& x_local,
std::vector<adouble>& x_ad,
double** w_local) {
if (retape)
trace_on(tag, 1);
for (unsigned int node=0; node<e.nnode; node++)
for (unsigned int eqn=0; eqn<neqn; eqn++) {
x_local[node*neqn+eqn] = x[e.gid[node]*neqn+eqn];
if (retape)
x_ad[node*neqn+eqn] <<= x_local[node*neqn+eqn];
}
for (unsigned int col=0; col<w.size(); col++)
for (unsigned int node=0; node<e.nnode; node++)
for (unsigned int eqn=0; eqn<neqn; eqn++)
w_local[col][node*neqn+eqn] = w[col][e.gid[node]*neqn+eqn];
}
bool fwd::run(SEXP xTrgt, SEXP xAryTrgt, SEXP xCtrl, SEXP xAryCtrl, SEXP xYrs)
{
if (!Trgt.Init(xTrgt,xAryTrgt,niters())) return false;
if (!Ctrl.Init(xCtrl,xAryTrgt,niters())) return false;
MinProjYear = (int)REAL(xYrs)[0];
MaxProjYear = (int)REAL(xYrs)[0];
int i;
for (i=1; i<LENGTH(xYrs); i++)
{
if (MinProjYear>REAL(xYrs)[i]) MinProjYear=(int)REAL(xYrs)[i];
if (MaxProjYear<REAL(xYrs)[i]) MaxProjYear=(int)REAL(xYrs)[i];
}
//ADol-C stuff
int n=1;
double *depen, *indep, *r, **jac;
adouble *depen_ad, *indep_ad;
int iter = 0;
//get N at start of year
double x=1.0;
for (iter=1; iter<=niters(); iter++)
project(&x, MinProjYear-1 ,iter, TRUE, TRUE);
int tag = 0;
for (int iYr=MinProjYear; iYr<=MaxProjYear; iYr++)
{
int _tag = 0; //(tag % 6 + 1);
n = (int)Trgt.n(iYr);
//n of independent variables MUST = n of equations
if (Ctrl.n(iYr) == n)
{
depen = new double[n];
indep = new double[n];
r = new double[n];
jac = new double*[n];
depen_ad = new adouble[n];
indep_ad = new adouble[n];
for (i=0; i<n; i++)
jac[i] = new double[n];
for (iter=1; iter<=niters(); iter++)
{
// set independent variables to estimaate
int j=0;
for (int iFleet=1; iFleet<=nfleet(); iFleet++)
for (int iMetier=1; iMetier<=nmetier(); iMetier++)
if (Ctrl.fit(iYr, iFleet, iMetier))
indep[j++] = 1.0;
// Taping the computation of the jacobian
trace_on(_tag);
// marking independent variables
for (i=0; i<n; i++)
indep_ad[i] <<= indep[i];
project(indep_ad,depen_ad,iYr,iter);
// marking dependent variables
for (i=0; i<n; i++)
depen_ad[i] >>= depen[i];
trace_off(_tag);
//jacobian(tag,m,n,indep,jac);
r[0]=1.0;
function(_tag,n,n,indep,r);
int NIters=0;
while (norm(r,n) > 1e-10 && NIters++<50)
{
jac_solv(_tag,n,indep,r,0,2);
for (i=0; i<n; i++)
indep[i] -= r[i];
function(_tag,n,n,indep,r);
}
project(indep, iYr, iter);
}
delete[] depen;
delete[] indep;
delete[] r;
delete[] depen_ad;
delete[] indep_ad;
for (i=0; i<n; i++)
delete[] jac[i];
delete[] jac;
}
tag++;
}
int main(int argc, char *argv[]) {
int n=get_num_ind();
int i,j;
struct timeval tv1,tv2;
adouble *xad;
adouble fad;
double f;
double *x;
x=new double[n];
xad=new adouble[n];
get_initial_value(x);
printf("evaluating the function...");
trace_on(tag);
for(i=0;i<n;i++)
{
xad[i] <<= x[i];
}
fad=func_eval(xad);
fad >>= f;
trace_off();
printf("done!\n");
// printf("function value =<%10.20f>\n",f);
// function(tag,1,n,x,&f);
// printf("adolc func value=<%10.20f>\n",f);
//tape_doc(tag,1,n,x,&f);
#ifdef _compare_with_full
double **H;
H = myalloc2(n,n);
printf("computing full hessain....");
gettimeofday(&tv1,NULL);
hessian(tag,n,x,H);
printf("done\n");
gettimeofday(&tv2,NULL);
printf("Computing the full hessian cost %10.6f seconds\n",(tv2.tv_sec-tv1.tv_sec)+(double)(tv2.tv_usec-tv1.tv_usec)/1000000);
#ifdef _PRINTOUT
for(i=0;i<n;i++){
for(j=0;j<n;j++){
printf("H[%d][%d]=<%10.10f>",i,j,H[i][j]);
}
printf("\n");
}
printf("\n");
#endif
#endif
#ifdef edge_pushing
unsigned int *rind = NULL;
unsigned int *cind = NULL;
double *values = NULL;
int nnz;
int options[2];
options[0]=PRE_ACC;
options[1]=COMPUT_GRAPH;
gettimeofday(&tv1,NULL);
// edge_hess(tag, 1, n, x, &nnz, &rind, &cind, &values, options);
sparse_hess(tag,n,0,x, &nnz, &rind, &cind, &values, options);
gettimeofday(&tv2,NULL);
printf("Sparse Hessian: edge pushing cost %10.6f seconds\n",(tv2.tv_sec-tv1.tv_sec)+(double)(tv2.tv_usec-tv1.tv_usec)/1000000);
#ifdef _PRINTOUT
for(i=0;i<nnz;i++){
printf("<%d,%d>:<%10.10f>\n",cind[i],rind[i],values[i]);
// printf("%d %d \n", rind[i], cind[i]);
}
#endif
#endif
#ifdef _compare_with_full
#ifdef edge_pushing
compare_matrix(n,H,nnz,cind,rind,values);
#endif
myfree2(H);
#endif
#ifdef edge_pushing
printf("nnz=%d\n", nnz);
free(rind); rind=NULL;
free(cind); cind=NULL;
free(values); values=NULL;
#endif
delete[] x;
delete[] xad;
return 0;
}
bool link_poly(
size_t size ,
size_t repeat ,
CppAD::vector<double> &a , // coefficients of polynomial
CppAD::vector<double> &z , // polynomial argument value
CppAD::vector<double> &ddp ) // second derivative w.r.t z
{
// -----------------------------------------------------
// setup
int tag = 0; // tape identifier
int keep = 1; // keep forward mode results in buffer
int m = 1; // number of dependent variables
int n = 1; // number of independent variables
int d = 2; // order of the derivative
double f; // function value
int i; // temporary index
// choose a vector of polynomial coefficients
CppAD::uniform_01(size, a);
// AD copy of the polynomial coefficients
std::vector<adouble> A(size);
for(i = 0; i < int(size); i++)
A[i] = a[i];
// domain and range space AD values
adouble Z, P;
// allocate arguments to hos_forward
double *x0 = 0;
x0 = CPPAD_TRACK_NEW_VEC(n, x0);
double *y0 = 0;
y0 = CPPAD_TRACK_NEW_VEC(m, y0);
double **x = 0;
x = CPPAD_TRACK_NEW_VEC(n, x);
double **y = 0;
y = CPPAD_TRACK_NEW_VEC(m, y);
for(i = 0; i < n; i++)
{ x[i] = 0;
x[i] = CPPAD_TRACK_NEW_VEC(d, x[i]);
}
for(i = 0; i < m; i++)
{ y[i] = 0;
y[i] = CPPAD_TRACK_NEW_VEC(d, y[i]);
}
// Taylor coefficient for argument
x[0][0] = 1.; // first order
x[0][1] = 0.; // second order
extern bool global_retape;
if( global_retape ) while(repeat--)
{ // choose an argument value
CppAD::uniform_01(1, z);
// declare independent variables
trace_on(tag, keep);
Z <<= z[0];
// AD computation of the function value
P = CppAD::Poly(0, A, Z);
// create function object f : Z -> P
P >>= f;
trace_off();
// get the next argument value
CppAD::uniform_01(1, z);
x0[0] = z[0];
// evaluate the polynomial at the new argument value
hos_forward(tag, m, n, d, keep, x0, x, y0, y);
// second derivative is twice second order Taylor coef
ddp[0] = 2. * y[0][1];
}
else
{
bool link_sparse_hessian(
size_t repeat ,
CppAD::vector<double> &x_arg ,
CppAD::vector<size_t> &i ,
CppAD::vector<size_t> &j ,
CppAD::vector<double> &h )
{
// -----------------------------------------------------
// setup
size_t k, m;
size_t order = 0; // derivative order corresponding to function
size_t tag = 0; // tape identifier
size_t keep = 1; // keep forward mode results in buffer
size_t n = x_arg.size(); // number of independent variables
size_t ell = i.size(); // number of indices in i and j
double f; // function value
typedef CppAD::vector<double> DblVector;
typedef CppAD::vector<adouble> ADVector;
typedef CppAD::vector<size_t> SizeVector;
ADVector X(n); // AD domain space vector
double *x; // double domain space vector
double **H; // Hessian
ADVector Y(1); // AD range space value
DblVector tmp(2 * ell); // double temporary vector
x = 0;
x = CPPAD_TRACK_NEW_VEC(n, x);
H = 0;
H = CPPAD_TRACK_NEW_VEC(n, H);
for(k = 0; k < n; k++)
{ H[k] = 0;
H[k] = CPPAD_TRACK_NEW_VEC(n, H[k]);
}
// choose a value for x
CppAD::uniform_01(n, x);
for(k = 0; k < n; k++)
x_arg[k] = x[k];
// ------------------------------------------------------
while(repeat--)
{
// get the next set of indices
CppAD::uniform_01(2 * ell, tmp);
for(k = 0; k < ell; k++)
{ i[k] = size_t( n * tmp[k] );
i[k] = std::min(n-1, i[k]);
//
j[k] = size_t( n * tmp[k + ell] );
j[k] = std::min(n-1, j[k]);
}
// declare independent variables
trace_on(tag, keep);
for(k = 0; k < n; k++)
X[k] <<= x[k];
// AD computation of f(x)
CppAD::sparse_evaluate(X, i, j, order, Y);
// create function object f : X -> Y
Y[0] >>= f;
trace_off();
// evaluate and return the hessian of f
hessian(int(tag), int(n), x, H);
}
for(k = 0; k < n; k++)
{ for(m = 0; m <= k; m++)
{ h[ k * n + m] = H[k][m];
h[ m * n + k] = H[k][m];
}
CPPAD_TRACK_DEL_VEC(H[k]);
}
CPPAD_TRACK_DEL_VEC(H);
CPPAD_TRACK_DEL_VEC(x);
return true;
}
/* MAIN PROGRAM */
int main() { /*------------------------------------------------------------------------*/
/* variables */
const int tag = 1; // tape tag
const int size = 5; // system size
const int indep = size*size+size; // # of indeps
const int depen = size; // # of deps
double A[size][size], a1[size], a2[size], // passive variables
b[size], x[size];
adouble **AA, *AAp, *Abx; // active variables
double *args = myalloc1(indep); // arguments
double **jac = myalloc2(depen,indep); // the Jacobian
double *laghessvec = myalloc1(indep); // Hessian-vector product
int i,j;
/*------------------------------------------------------------------------*/
/* Info */
fprintf(stdout,"LINEAR SYSTEM SOLVING by "
"LU-DECOMPOSITION (ADOL-C Example)\n\n");
/*------------------------------------------------------------------------*/
/* Allocation und initialization of the system matrix */
AA = new adouble*[size];
AAp = new adouble[size*size];
for (i=0; i<size; i++) {
AA[i] = AAp;
AAp += size;
}
Abx = new adouble[size];
for(i=0; i<size; i++) {
a1[i] = i*0.25;
a2[i] = i*0.33;
}
for(i=0; i<size; i++) {
for(j=0; j<size; j++)
A[i][j] = a1[i]*a2[j];
A[i][i] += i+1;
b[i] = -i-1;
}
/*------------------------------------------------------------------------*/
/* Taping the computation of the determinant */
trace_on(tag);
/* marking indeps */
for(i=0; i<size; i++)
for(j=0; j<size; j++)
AA[i][j] <<= (args[i*size+j] = A[i][j]);
for(i=0; i<size; i++)
Abx[i] <<= (args[size*size+i] = b[i]);
/* LU-factorization and computation of solution */
LUfact(size,AA);
LUsolve(size,AA,Abx);
/* marking deps */
for (i=0; i<size; i++)
Abx[i] >>= x[i];
trace_off();
fprintf(stdout," x[0] (original): %16.4le\n",x[0]);
/*------------------------------------------------------------------------*/
/* Recomputation */
function(tag,depen,indep,args,x);
fprintf(stdout," x[0] (from tape): %16.4le\n",x[0]);
/*------------------------------------------------------------------------*/
/* Computation of Jacobian */
jacobian(tag,depen,indep,args,jac);
fprintf(stdout," Jacobian:\n");
for (i=0; i<depen; i++) {
for (j=0; j<indep; j++)
fprintf(stdout," %14.6le",jac[i][j]);
fprintf(stdout,"\n");
}
/*------------------------------------------------------------------------*/
/* Computation of Lagrange-Hessian-vector product */
lagra_hess_vec(tag,depen,indep,args,args,x,laghessvec);
fprintf(stdout," Part of Lagrange-Hessian-vector product:\n");
for (i=0; i<size; i++) {
for (j=0; j<size; j++)
fprintf(stdout," %14.6le",laghessvec[i*size+j]);
fprintf(stdout,"\n");
}
/*------------------------------------------------------------------------*/
/* Tape-documentation */
tape_doc(tag,depen,indep,args,x);
/*------------------------------------------------------------------------*/
/* Tape statistics */
int tape_stats[STAT_SIZE];
tapestats(tag,tape_stats);
fprintf(stdout,"\n independents %d\n",tape_stats[NUM_INDEPENDENTS]);
fprintf(stdout," dependents %d\n",tape_stats[NUM_DEPENDENTS]);
fprintf(stdout," operations %d\n",tape_stats[NUM_OPERATIONS]);
fprintf(stdout," operations buffer size %d\n",tape_stats[OP_BUFFER_SIZE]);
fprintf(stdout," locations buffer size %d\n",tape_stats[LOC_BUFFER_SIZE]);
fprintf(stdout," constants buffer size %d\n",tape_stats[VAL_BUFFER_SIZE]);
fprintf(stdout," maxlive %d\n",tape_stats[NUM_MAX_LIVES]);
fprintf(stdout," valstack size %d\n\n",tape_stats[TAY_STACK_SIZE]);
/*------------------------------------------------------------------------*/
/* That's it */
return 1;
}
void trace_pulse(void)
{
trace_on();
_delay_us(1);
trace_off();
}
// alternative implementation calling high-level _sys->continuous
//--------------------------------------------------------------------------
void Omu_IntODE::syseq_forward(double t, const VECP y, const VECP u,
VECP f)
{
#ifdef OMU_WITH_ADOLC
int i, j;
Omu_DependentVec &Ft = *_Ft_ptr;
//
// form a vector of independent variables
//
for (i = 0; i < _nd; i++) {
_x[i] = u[i];
}
for (i = 0; i < _n; i++) {
_x[_nd + i] = y[i];
_x[_nd + _n + i] = 0.0; // yprime[i]
}
for (i = 0; i < _nu; i++) {
_x[_nd + 2 * _n + i] = u[_nd + i];
}
//
// evaluate residual
//
// adoublev ax(_nd + _n);
static adoublev ax; ax.alloc(_nd + _n);
// adoublev adx(_nd + _n);
static adoublev adx; adx.alloc(_nd + _n);
// adoublev au(_nu);
static adoublev au; au.alloc(_nu);
// adoublev aF(_nd + _n);
static adoublev aF; aF.alloc(_nd + _n);
for (i = 0; i < _nd; i++)
adx[i] = 0.0;
for (i = _nd; i < _nxt; i++)
aF[i] = 0.0;
if (_sa)
trace_on(3); // tape 3
ax <<= _x->ve;
for (i = 0; i < _n; i++)
adx[_nd + i] <<= _x->ve[_nd + _n + i];
au <<= _x->ve + _nd + 2 * _n;
_sys->continuous(_kk, t, ax, au, adx, aF);
for (i = _nd; i < _nxt; i++) {
aF[i] >>= f[i - _nd];
f[i - _nd] /= -Ft.Jdx[i][i];
}
if (_sa) {
trace_off();
int nindep = _nd + 2 * _n + _nu;
int npar = _nx + _nu;
m_zero(_X2);
for (i = 0; i < _nd; i++) {
_X2[i][i] = 1.0;
}
for (i = 0; i < _n; i++) {
for (j = 0; j < npar; j++) {
_X2[_nd + i][j] = y[(1 + j) * _n + i];
_X2[_nd + _n + i][j] = 0.0; // yprime[(1 + j) * _n + i];
}
}
for (i = 0; i < _nu; i++) {
_X2[_nd + 2 * _n + i][_nd + _n + i] = 1.0;
}
forward(3, _n, nindep, npar, _x->ve, _X2->me, f->ve, _Y2->me);
for (i = _nd; i < _nxt; i++) {
f[i - _nd] /= -Ft.Jdx[i][i];
for (j = 0; j < npar; j++) {
f[(1 + j) * _n + i - _nd] = _Y2[i - _nd][j] / -Ft.Jdx[i][i];
}
}
}
_res_evals++;
if (_sa)
_sen_evals++;
#else
m_error(E_NULL, "Omu_IntODE::syseq_forward: was compiled without ADOL-C");
#endif
}
int main(int argc, char *argv[]) {
int n=NUM_IND;
int i,j;
struct timeval tv1,tv2;
adouble *xad;
adouble fad;
double f;
double *x;
x=new double[n];
xad=new adouble[n];
get_initials(x, n);
// printf("evaluating the function...");
trace_on(tag);
for(i=0;i<n;i++)
{
xad[i] <<= x[i];
}
fad=eval_func<adouble>(xad, n);
fad >>= f;
trace_off();
// printf("done!\n");
std::cout << "y = " << f << std::endl;
#ifdef COMPARE_WITH_FULL_HESS
double **H;
H = myalloc2(n,n);
printf("computing full hessain....");
gettimeofday(&tv1,NULL);
hessian(tag,n,x,H);
printf("done\n");
gettimeofday(&tv2,NULL);
printf("Computing the full hessian cost %10.6f seconds\n",(tv2.tv_sec-tv1.tv_sec)+(double)(tv2.tv_usec-tv1.tv_usec)/1000000);
#ifdef PRINT_RESULTS
for(i=0;i<n;i++){
for(j=0;j<n;j++){
printf("H[%d][%d]=<%10.10f>",i,j,H[i][j]);
}
printf("\n");
}
printf("\n");
#endif
#endif
unsigned int *rind = NULL;
unsigned int *cind = NULL;
double *values = NULL;
int nnz;
int options[2];
#ifdef LIVARH
options[0]=0;
options[1]=1;
gettimeofday(&tv1,NULL);
edge_hess(tag, 1, n, x, &nnz, &rind, &cind, &values, options);
gettimeofday(&tv2,NULL);
printf("Sparse Hessian: LivarH cost %10.6f seconds\n",(tv2.tv_sec-tv1.tv_sec)+(double)(tv2.tv_usec-tv1.tv_usec)/1000000);
#endif
#ifdef LIVARHACC
options[0]=1;
options[1]=1;
gettimeofday(&tv1,NULL);
edge_hess(tag, 1, n, x, &nnz, &rind, &cind, &values, options);
gettimeofday(&tv2,NULL);
printf("Sparse Hessian: LivarHACC cost %10.6f seconds\n",(tv2.tv_sec-tv1.tv_sec)+(double)(tv2.tv_usec-tv1.tv_usec)/1000000);
#endif
// Sparse ADOL-C drivers report the upper matrix
#ifdef DIRECT
options[0]=0;
options[1]=1;
gettimeofday(&tv1,NULL);
sparse_hess(tag, n, 0, x, &nnz, &cind, &rind, &values, options);
gettimeofday(&tv2,NULL);
printf("Sparse Hessian: direct recovery cost %10.6f seconds\n",(tv2.tv_sec-tv1.tv_sec)+(double)(tv2.tv_usec-tv1.tv_usec)/1000000);
#endif
#ifdef INDIRECT
options[0]=0;
options[1]=0;
gettimeofday(&tv1,NULL);
sparse_hess(tag, n, 0, x, &nnz, &cind, &rind, &values, options);
gettimeofday(&tv2,NULL);
printf("Sparse Hessian: indirect recovery cost %10.6f seconds\n",(tv2.tv_sec-tv1.tv_sec)+(double)(tv2.tv_usec-tv1.tv_usec)/1000000);
#endif
#ifdef PRINT_RESULTS
for(i=0;i<nnz;i++){
printf("<%d,%d>:<%10.10f>\n",rind[i],cind[i],values[i]);
}
#endif
#ifdef COMPARE_WITH_FULL_HESS
compare_matrix(n,H,nnz,rind,cind,values);
myfree2(H);
#endif
free(rind); rind=NULL;
free(cind); cind=NULL;
free(values); values=NULL;
delete[] x;
delete[] xad;
return 0;
}
/* The interface function */
void tapingVectorFunction( int tag, double* indeps, double* deps ) {
int i;
trace_on(tag);
adouble* activeIndeps = new adouble[indepDim];
adouble* activeDeps = new adouble[depDim];
adouble* aIP = activeIndeps;
double* iP = indeps;
for (i=0; i<indepDim; i++)
*aIP++ <<= *iP++;
i = 3;
adouble * activeRadMotCoeff = activeIndeps+i;
i += radMotDegree;
adouble * activeVerMotCoeff = activeIndeps+i;
i += verMotDegree;
adouble * activeHorMotCoeff = activeIndeps+i;
i += horMotDegree;
adouble * activeHelMotCoeff = activeIndeps+i;
i += helMotDegree;
adouble * activeAngMotCoeff = activeIndeps+i;
i += angMotDegree;
adouble * activeModRolCoeff = activeIndeps+i;
activeGearFunction(
activeIndeps,
activeDeps,
// jetzt kommen die ganzen Parameter
xmk, // Messerversatz
ymk, // MK-Versatz
kopspw, // Kopfspanwinkel
flaspw, // Flankenspanwinkel
meschw, // Messerschwenkwinkel
flkrrd, // Flugkreisradius
e, // Exzentrizitaet
exzenw, // Exzentrizitaetswinkel
thetas, // Messerkopfschwenkung
thetan, // Messerkopfneigung
xmw, // MK-x
ymw, // MK-y
zmw, // MK-z
thetaw, // Wiegenwinkel
m, // Achsversatz
zwr, // Verschiebung Werkradachse
delta, // Teilkegeloeffnungswinkel
omega, //
c,
r, // Kopfradius
rs, // Sphaerikradius
ys, // Sphaerik-Mitte-Y
zs, // Sphaerik-Mitte-Z
// jetzt die Zusatzbewegungen
radMotDegree,
activeRadMotCoeff,
verMotDegree,
activeVerMotCoeff,
horMotDegree,
activeHorMotCoeff,
helMotDegree,
activeHelMotCoeff,
angMotDegree,
activeAngMotCoeff,
modRolDegree,
activeModRolCoeff
);
aIP = activeDeps;
iP = deps;
for (i=0; i<depDim; i++)
*aIP++ >>= *iP++;
trace_off();
delete [] activeDeps;
delete [] activeIndeps;
}
void run_client(const char *host, int port, int tun_fd, const char *username, const char *password, int use_ssl) {
int flags = fcntl(tun_fd, F_GETFL, 0);
assert(fcntl(tun_fd, F_SETFL, flags | O_NONBLOCK) == 0);
char buff[BUFF_SZ];
char url[URL_SZ];
int read_len;
CURL *curl;
CURLcode res;
double speed_upload, total_time;
struct curl_httppost *post;
struct curl_httppost *last_post;
int backoff = 0;
if (MAX_HOST_LEN < strlen(host)) {
log_crit("client", "host-name too long");
return;
}
make_pkt_post_url(host, port, use_ssl, url, URL_SZ);
curl = curl_easy_init();
curl_easy_setopt(curl, CURLOPT_URL, url);
if (trace_on()) curl_easy_setopt(curl, CURLOPT_VERBOSE, 1L);
curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, write_callback);
curl_easy_setopt(curl, CURLOPT_WRITEDATA, (void *)&tun_fd);
curl_easy_setopt(curl, CURLOPT_USERNAME, username);
curl_easy_setopt(curl, CURLOPT_PASSWORD, password);
if (use_ssl) {
curl_easy_setopt(curl, CURLOPT_SSL_VERIFYPEER, 0L);
curl_easy_setopt(curl, CURLOPT_SSL_VERIFYHOST, 0L);
}
assert(curl != NULL);
post = last_post = NULL;
while(! do_stop) {
read_len = read(tun_fd, buff, BUFF_SZ);
if (read_len == -1 && errno != EAGAIN)
log_warn("client", "Failed to read from tun");
if (read_len > 0) {
if (post != NULL) {
curl_formfree(post);
post = last_post = NULL;
}
curl_formadd(&post, &last_post,
CURLFORM_COPYNAME, "pkt",
CURLFORM_BUFFER, "pkt",
CURLFORM_BUFFERPTR, buff,
CURLFORM_BUFFERLENGTH, read_len,
CURLFORM_END);
curl_easy_setopt(curl, CURLOPT_HTTPPOST, post);
backoff = MIN_BACKOFF_MICRO_SEC;
log_debug("client", "Sending %d bytes of data in current request", read_len);
} else {
increase_backoff(&backoff);
curl_easy_setopt(curl, CURLOPT_HTTPGET, 1);
log_debug("client", "Sending NO-data in current request");
}
res = curl_easy_perform(curl);
if(res != CURLE_OK) {
log_warnx("client", "curl_easy_perform() failed: %s", curl_easy_strerror(res));
} else {
curl_easy_getinfo(curl, CURLINFO_SPEED_UPLOAD, &speed_upload);
curl_easy_getinfo(curl, CURLINFO_TOTAL_TIME, &total_time);
log_debug("client", "Speed: %.3f bytes/sec during %.3f seconds\n", speed_upload, total_time);
}
do_backoff(backoff, tun_fd);
}
if (post != NULL) curl_formfree(post);
curl_easy_cleanup(curl);
}
/* MAIN PROGRAM */
int main() {
int n,i,it;
size_t tape_stats[STAT_SIZE];
/*--------------------------------------------------------------------------*/
/* Input */
fprintf(stdout,"SPEELPENNINGS PRODUCT Type 1 (ADOL-C Example)\n\n");
fprintf(stdout,"number of independent variables = ? \n");
scanf("%d",&n);
int itu;
fprintf(stdout,"number of evaluations = ? \n");
scanf("%d",&itu);
/*--------------------------------------------------------------------------*/
double yp=0.0; /* 0. time (undifferentiated double code) */
double *xp = new double[n];
/* Init */
for (i=0;i<n;i++)
xp[i] = (i+1.0)/(2.0+i);
double t00 = myclock(1);
for (it=0; it<itu; it++) {
yp = 1.0;
for (i=0; i<n; i++)
yp *= xp[i];
}
double t01 = myclock();
/*--------------------------------------------------------------------------*/
double yout=0; /* 1. time (tracing ! no keep) */
double t10 = myclock();
trace_on(TAG);
adouble* x;
x = new adouble[n];
adouble y;
y = 1;
for (i=0; i<n; i++) {
x[i] <<= xp[i];
y *= x[i];
}
y >>= yout;
delete [] x;
trace_off();
double t11 = myclock();
fprintf(stdout,"%E =? %E function values should be the same \n",yout,yp);
/*--------------------------------------------------------------------------*/
tapestats(TAG,tape_stats);
fprintf(stdout,"\n independents %zu\n",tape_stats[NUM_INDEPENDENTS]);
fprintf(stdout," dependents %zu\n",tape_stats[NUM_DEPENDENTS]);
fprintf(stdout," operations %zu\n",tape_stats[NUM_OPERATIONS]);
fprintf(stdout," operations buffer size %zu\n",tape_stats[OP_BUFFER_SIZE]);
fprintf(stdout," locations buffer size %zu\n",tape_stats[LOC_BUFFER_SIZE]);
fprintf(stdout," constants buffer size %zu\n",tape_stats[VAL_BUFFER_SIZE]);
fprintf(stdout," maxlive %zu\n",tape_stats[NUM_MAX_LIVES]);
fprintf(stdout," valstack size %zu\n\n",tape_stats[TAY_STACK_SIZE]);
/*--------------------------------------------------------------------------*/
double **r = new double*[1];
r[0] = new double[1];
r[0][0] = yp;
double err;
double *z = new double[n];
double *g = new double[n];
double* h = new double[n];
double *ind = new double[n];
/*--------------------------------------------------------------------------*/
double t60 = myclock(); /* 6. time (forward no keep) */
for (it=0; it<itu; it++)
forward(TAG,1,n,0,xp,*r);
double t61 = myclock();
/*--------------------------------------------------------------------------*/
double t20 = myclock(); /* 2. time (forward+keep) */
for (it=0; it<itu; it++)
forward(TAG,1,n,1,xp,*r);
double t21 = myclock();
/*--------------------------------------------------------------------------*/
double t30 = myclock(); /* 3. time (reverse) */
for (it=0; it<itu; it++)
reverse(TAG,1,n,0,1.0,g);
double t31 = myclock();
err=0;
for (i=0; i<n; i++) // Compare with deleted product
{ err = maxabs(err,xp[i]*g[i]/r[0][0] - 1.0);
ind[i] = xp[i];
}
fprintf(stdout,"%E = maximum relative errors in gradient (fw+rv)\n",err);
/*--------------------------------------------------------------------------*/
double t40 = myclock(); /* 4. time (gradient) */
for (it=0; it<itu; it++)
gradient(TAG,n,ind,z); //last argument lagrange is ommitted
double t41 = myclock();
err = 0;
for (i=0; i<n; i++) // Compare with previous numerical result
err = maxabs(err,g[i]/z[i] - 1.0);
fprintf(stdout,"%E = gradient error should be exactly zero \n",err);
/*--------------------------------------------------------------------------*/
double *tan = new double[n]; /* 5. time (first row of Hessian) */
for (i=1; i<n; i++)
tan[i] = 0.0 ;
tan[0]=1.0;
double t50 = myclock();
for (it=0; it<itu; it++)
hess_vec(TAG,n,ind,tan,h); // Computes Hessian times direction tan.
double t51 = myclock();
err = abs(h[0]);
for (i=1; i<n; i++) //Compare with doubly deleted product
err = maxabs(err,xp[0]*h[i]/g[i]-1.0);
fprintf(stdout,"%E = maximum relative error in Hessian column \n",err);
/*--------------------------------------------------------------------------*/
double h1n = h[n-1]; /* Check for symmetry */
tan[0]=0;
tan[n-1]=1;
hess_vec(TAG,n,ind,tan,h); // Computes Hessian times direction tan.
fprintf(stdout,
"%E = %E (1,n) and (n,1) entry should be the same\n",h1n,h[0]);
/*--------------------------------------------------------------------------*/
/* output of results */
if (t01-t00) {
double rtu = 1.0/(t01-t00);
fprintf(stdout,"\n\n times for ");
fprintf(stdout,"\n unitime : \t%E seconds",(t01-t00)/itu);
fprintf(stdout,"\n tracing : \t%E",(t11-t10)*rtu*itu);
fprintf(stdout," units \t%E seconds",(t11-t10));
fprintf(stdout,
"\n----------------------------------------------------------");
fprintf(stdout,"\n forward (no keep): \t%E",(t61-t60)*rtu);
fprintf(stdout," units \t%E seconds",(t61-t60)/itu);
fprintf(stdout,"\n forward + keep : \t%E",(t21-t20)*rtu);
fprintf(stdout," units \t%E seconds",(t21-t20)/itu);
fprintf(stdout,"\n reverse : \t%E",(t31-t30)*rtu);
fprintf(stdout," units \t%E seconds",(t31-t30)/itu);
fprintf(stdout,
"\n----------------------------------------------------------");
fprintf(stdout,"\n gradient : \t%E",(t41-t40)*rtu);
fprintf(stdout," units \t%E seconds",(t41-t40)/itu);
fprintf(stdout,"\n hess*vec : \t%E",(t51-t50)*rtu);
fprintf(stdout," units \t%E seconds\n",(t51-t50)/itu);
} else
fprintf(stdout,"\n-> zero timing due to small problem dimension \n");
return 1;
}
