void listdisk(int nlist)
{
int i;
real r, phi, vcir, omega, Adisk, kappa, sigma1, sigma2,
mu_eff, sig_r, sig_p, sig_z, vrad, vorb2, vorb;
printf("#%5s %6s %6s %6s %6s %6s %6s %6s %6s %6s %6s\n",
"r", "vcir", "omega", "kappa", "sig_z", "sig_r", "sig_p", "vorb",
"Q", "rhomid", "fmax");
for (i = 1; i <= nlist; i++) { /* loop over radii */
r = (i * rcut) / ((real) nlist);
vcir = seval(r, &rdtab[0], &vctab[0], &vctab[NTAB], NTAB);
omega = vcir / r;
Adisk = (omega - spldif(r, &rdtab[0], &vctab[0], &vctab[NTAB], NTAB)) / 2;
if (omega - Adisk < 0.0)
error("%s: kappa undefined (omega - Adisk < 0)\n"
" r, omega, Adisk = %f %f %f\n", getargv0(), r, omega, Adisk);
kappa = 2 * rsqrt(rsqr(omega) - Adisk * omega);
sigma1 = rsqr(alpha1) * mdisk1 * rexp(- alpha1 * r) / TWO_PI;
sigma2 = rsqr(alpha2) * mdisk2 * rexp(- alpha2 * r) / TWO_PI;
mu_eff = (r_mu>0 ? 1 + (mu - 1) * (r / (r + r_mu)) : mu);
sig_z = rsqrt(PI * (sigma1 + sigma2) * zdisk);
sig_r = mu_eff * sig_z;
sig_p = (0.5 * kappa / omega) * sig_r;
vorb2 = rsqr(vcir) + rsqr(sig_r) * (1 - 2 * alpha1 * r) - rsqr(sig_p) +
(r_mu>0 ? rsqr(sig_z) * r * mu_eff*(2*mu-2)*r_mu/rsqr(r+r_mu) : 0);
vorb = rsqrt(MAX(vorb2, 0.0));
printf("%6.4f %6.4f %6.2f %6.2f %6.4f %6.4f %6.4f %6.4f "
"%6.3f %6.1f %6.1f\n",
r, vcir, omega, kappa, sig_z, sig_r, sig_p, vorb,
kappa * sig_r / (3.358*(sigma1+sigma2)), sigma1/(2*zdisk),
sigma1 / (2*zdisk * rsqrt(rqbe(2*PI)) * sig_z*sig_r*sig_p));
}
}
local void testdata(void) {
real rsc, vsc, r, v, x, y;
bodyptr p;
float scale = 1.0f;
float vscale = 1.0f;
float mscale = 1.0f;
if (nbody < 1) // check for silly values
error("%s: absurd value for nbody\n", getargv0());
bodytab = (bodyptr) allocate(nbody * sizeof(body));
// alloc space for bodies
rsc = (3 * PI) / 16; // set length scale factor
vsc = rsqrt(1.0 / rsc); // find speed scale factor
for (p = bodytab; p < bodytab+nbody; p++) { // loop over particles
Type(p) = BODY; // tag as a body
//Mass(p) = 1.0 / nbody; // set masses equal
// Set mass randomly, like in brute
Mass(p) = (rand() / (float) RAND_MAX) * mscale;
x = xrandom(0.0, MFRAC); // pick enclosed mass
r = 1.0 / rsqrt(rpow(x, -2.0/3.0) - 1); // find enclosing radius
pickshell(Pos(p), NDIM, rsc * r); // pick position vector
do { // select from fn g(x)
x = xrandom(0.0, 1.0); // for x in range 0:1
y = xrandom(0.0, 0.1); // max of g(x) is 0.092
} while (y > x*x * rpow(1 - x*x, 3.5)); // using von Neumann tech
v = x * rsqrt(2.0 / rsqrt(1 + r*r)); // find resulting speed
pickshell(Vel(p), NDIM, vsc * v); // pick velocity vector
}
tnow = 0.0; // set elapsed model time
}
int main()
{
int i, j, k, h, t;
real g, gg;
//freopen("input.txt", "rt", stdin);
scanf("%d%d", &m, &n);
for (i = 0; i < m; i++) scanf(real_in, &point[i]);
for (k = 0; k < n; k++) for (i = 0; i < m; i++)
{
scanf(real_in, &vec[k][i]);
}
sdist = calc_vec_slen(point);
for (i = 0; i < n; i++) koeff[i] = 0;
h = 1;
for (i = 0; i < n; i++) h *= 3;
for (j = 0; j < h; j++)
{
for (t = 0; t < m; t++) p[t] = -point[t];
k = j; w = 0;
for (i = 0; i < n; i++)
{
cur_koeff[i] = 0;
way[i] = k % 3; k /= 3;
if (way[i] == 2)
{
for (t = 0; t < m; t++) a[w][t] = vec[i][t];
where_koeff[w] = i;
w++;
}
else if (way[i] == 1)
{
for (t = 0; t < m; t++) p[t] -= vec[i][t];
cur_koeff[i] = 1;
}
}
find_dist();
}
printf(real_out"\n", rsqrt(sdist));
for (i = 0; i < n; i++)
{
printf(real_out"%c", koeff[i], (i==n-1) ? '\n' : ' ');
}
gg = 0;
for (i = 0; i < m; i++)
{
g = point[i];
for (j = 0; j < n; j++) g += vec[j][i] * koeff[j];
gg += g*g;
}
printf(real_out"\n", rsqrt(gg));
return 0;
}
void gspmodel(void)
{
real beta_a, mcut, vcut, vfac;
static real *sig2 = NULL;
real r, vmax2, sigr, sig, x, y, vr, v1, v2;
bodyptr bp;
vector rhat, vec1, vec2, vtmp;
beta_a = getdparam("beta_a");
assert(beta_a <= 1.0);
nbody = getiparam("nbody");
assert(nbody > 0);
mcut = getdparam("mcut");
assert(0.0 < mcut && mcut <= 1.0);
vcut = getdparam("vcut");
assert(vcut > 0.0);
if (sig2 == NULL)
sig2 = calc_sig2_gsp(gsp, ggsp, beta_a);
if (btab == NULL)
btab = (bodyptr) allocate(nbody * SizeofBody);
vfac = rsqrt(1 - beta_a);
for (bp = btab; bp < NthBody(btab, nbody); bp = NextBody(bp)) {
Mass(bp) = gsp->mtot / nbody;
r = r_mass_gsp(gsp, xrandom(0.0, mcut * gsp->mtot));
vmax2 = -2 * rsqr(vcut) * phi_gsp(ggsp, r);
if (vfac > 1.0)
vmax2 = vmax2 / rsqr(vfac);
sigr = rsqrt(sig2_gsp(gsp, ggsp, beta_a, sig2, r));
sig = fixsigma(sigr, rsqrt(vmax2));
do {
vr = grandom(0.0, sig);
v1 = grandom(0.0, sig);
v2 = grandom(0.0, sig);
} while (vr*vr + v1*v1 + v2*v2 > vmax2);
picktriad(rhat, vec1, vec2);
MULVS(Pos(bp), rhat, r);
MULVS(Vel(bp), rhat, vr);
MULVS(vtmp, vec1, v1 * vfac);
ADDV(Vel(bp), Vel(bp), vtmp);
MULVS(vtmp, vec2, v2 * vfac);
ADDV(Vel(bp), Vel(bp), vtmp);
Phi(bp) = phi_gsp(ggsp, r);
Aux(bp) = Phi(bp) + 0.5 * dotvp(Vel(bp), Vel(bp));
}
if (getbparam("besort"))
qsort(btab, nbody, SizeofBody, berank);
if (getbparam("zerocm"))
snapcenter(btab, nbody, MassField.offset);
if (! strnull(getparam("auxvar")))
setauxvar(btab, nbody);
}
local void sumcell(cellptr start, cellptr finish, vector pos0,
real *phi0, vector acc0)
{
real eps2, eps2thrd, dr2, dr2i, dr1i, mdr1i, mdr3i, qdr2, dr5i, phi_q;
vector dr, qdr;
eps2 = eps * eps; // premultiply softening
eps2thrd = eps2 / 3.0; // predivide for soft corr
for (cellptr p = start; p < finish; p++) { // loop over node list
DOTPSUBV(dr2, dr, Pos(p), pos0); // do mono part of force
dr2i = ((real) 1.0) / (dr2 + eps2); // perform only division
dr1i = rsqrt(dr2i); // set inverse soft distance
mdr1i = Mass(p) * dr1i; // form mono potential
mdr3i = mdr1i * dr2i; // get scale factor for dr
DOTPMULMV(qdr2, qdr, Quad(p), dr); // do quad part of force
#if defined(SOFTCORR)
qdr2 -= eps2thrd * Trace(p); // apply Keigo's correction
#endif
dr5i = ((real) 3.0) * dr2i * dr2i * dr1i; // factor 3 saves a multiply
phi_q = ((real) 0.5) * dr5i * qdr2; // form quad potential
mdr3i += ((real) 5.0) * phi_q * dr2i; // adjust radial term
*phi0 -= mdr1i + phi_q; // sum mono and quad pot
ADDMULVS2(acc0, dr, mdr3i, qdr, - dr5i); // sum mono and quad acc
}
}
real pickdist(real eta, real sigma)
{
static real eta0 = -1.0, sigcorr;
int niter;
real x, y, q;
if (eta != eta0) {
sigcorr =
rsqrt(8 * eta /
(bessel_K(0.75, 1/(32*eta)) / bessel_K(0.25, 1/(32*eta)) - 1));
eprintf("[%s: sigma correction factor = %f]\n", getargv0(), sigcorr);
eta0 = eta;
}
niter = 0;
do {
x = xrandom(-1.0, 1.0);
y = xrandom(0.0, YMAX);
q = rexp(- 0.5 * rsqr(fmap(x)) - eta * rsqr(rsqr(fmap(x)))) *
(1 + x*x) / rsqr(1 - x*x);
if (q > YMAX) /* should not ever happen */
error("%s: guess out of bounds\n x = %f q = %f > %f\n",
getargv0(), x, q, YMAX);
niter++;
if (niter > 1000)
error("%s: 1000 iterations without success\n", getargv0());
} while (y > q || x*x == 1); /* 2nd test prevents infty */
return (sigcorr * sigma * fmap(x));
}
local void sum1force(bodyptr btab, int nbody, int i0, real eps2)
{
bodyptr bp0, bp;
double phi0, acc0[NDIM];
vector pos0, dr;
real dr2, drab, mri, mr3i;
int j;
bp0 = NthBody(btab, i0);
phi0 = 0.0;
CLRV(acc0);
SETV(pos0, Pos(bp0));
for (j = 0; j < nbody; j++)
if (j != i0) {
bp = NthBody(btab, j);
DOTPSUBV(dr2, dr, Pos(bp), pos0);
dr2 += eps2;
drab = rsqrt(dr2);
mri = Mass(bp) / drab;
phi0 -= mri;
mr3i = mri / dr2;
ADDMULVS(acc0, dr, mr3i);
}
Phi(bp0) = phi0;
SETV(Acc(bp0), acc0);
}
const Quaternion NormalizeEst(const Quaternion& quaternion)
{
float lenSqr, lenInv;
lenSqr = Norm(quaternion);
lenInv = rsqrt(lenSqr);
return quaternion * lenInv;
}
local void gravsub(nodeptr q)
{
real drab, phii, mor3;
vector ai, quaddr;
real dr5inv, phiquad, drquaddr;
if (q != (long) qmem) { /* cant use memorized data? */
SUBV(dr, Pos(q), pos0); /* then compute sep. */
DOTVP(drsq, dr, dr); /* and sep. squared */
}
drsq += eps*eps; /* use standard softening */
drab = rsqrt(drsq);
phii = Mass(q) / drab;
mor3 = phii / drsq;
MULVS(ai, dr, mor3);
phi0 -= phii; /* add to total grav. pot. */
ADDV(acc0, acc0, ai); /* ... and to total accel. */
if (usequad && Type(q) == CELL) { /* if cell, add quad term */
cellptr SAFE c= TC(q);
dr5inv = 1.0/(drsq * drsq * drab); /* form dr^-5 */
MULMV(quaddr, Quad(c), dr); /* form Q * dr */
DOTVP(drquaddr, dr, quaddr); /* form dr * Q * dr */
phiquad = -0.5 * dr5inv * drquaddr; /* get quad. part of phi */
phi0 = phi0 + phiquad; /* increment potential */
phiquad = 5.0 * phiquad / drsq; /* save for acceleration */
MULVS(ai, dr, phiquad); /* components of acc. */
SUBV(acc0, acc0, ai); /* increment */
MULVS(quaddr, quaddr, dr5inv);
SUBV(acc0, acc0, quaddr); /* acceleration */
}
}
void setauxvar(bodyptr btab, int nbody)
{
bodyptr bp;
vector jvec;
real jtot, etot, r0, r1, r;
if (streq(getparam("auxvar"), "mass"))
for (bp = btab; bp < NthBody(btab, nbody); bp = NextBody(bp))
Aux(bp) = mass_gsp(ggsp, absv(Pos(bp)));
else if (streq(getparam("auxvar"), "rperi"))
for (bp = btab; bp < NthBody(btab, nbody); bp = NextBody(bp)) {
CROSSVP(jvec, Pos(bp), Vel(bp));
jtot = absv(jvec);
etot = 0.5 * dotvp(Vel(bp), Vel(bp)) + Phi(bp);
r0 = 0.0;
r1 = absv(Pos(bp));
r = 0.5 * (r0 + r1);
while ((r1 - r0) > TOL * r) {
if (rsqrt(2 * (etot - phi_gsp(ggsp, r))) > jtot/r)
r1 = r;
else
r0 = r;
r = 0.5 * (r0 + r1);
}
Aux(bp) = r;
}
else
error("%s: unknown auxvar option %s\n",
getargv0(), getparam("auxvar"));
}
void polyscale(void)
{
double r_henon, m_henon, p_henon, scl_r, scl_m, scl_v;
int i;
r_henon = (4*mpol + 6) / (3*mpol - npol + 5);
m_henon = 1.0;
p_henon = m_henon / r_henon;
scl_r = r_henon / rad1;
scl_m = m_henon / mtot;
scl_v = rsqrt(scl_m / scl_r);
for (i = 0; i < nstep; i++) {
rtab[i] = scl_r * rtab[i];
mtab[i] = scl_m * mtab[i];
ptab[i] = rsqr(scl_v) * ptab[i] - p_henon;
if (i > 0 && mtab[i-1] >= mtab[i])
error("%s: mass not monotonic! mtab[%d:%d] = %f,%f\n", getprog(),
i-1, i, mtab[i-1], mtab[i]);
}
rad1 = scl_r * rad1;
mtot = scl_m * mtot;
phi1 = rsqr(scl_v) * phi1 - p_henon; // == - p_henon by def
Kprime = Kprime * scl_m / rqbe(scl_r * scl_v);
eprintf("[%s.polyscale: Kprime = %f]\n", getprog(), Kprime);
}
Vector3 Cone::randomDirectionInCone(Random& rng) const {
const float cosThresh = cos(angle);
float cosAngle;
float normalizer;
Vector3 v;
do {
float vlenSquared;
// Sample uniformly on a sphere by rejection sampling and then normalizing
do {
v.x = rng.uniform(-1, 1);
v.y = rng.uniform(-1, 1);
v.z = rng.uniform(-1, 1);
// Sample uniformly on a cube
vlenSquared = v.squaredLength();
} while (vlenSquared > 1);
const float temp = v.dot(direction);
// Compute 1 / ||v||, but
// if the vector is in the wrong hemisphere, flip the sign
normalizer = rsqrt(vlenSquared) * sign(temp);
// Cosine of the angle between v and the light's negative-z axis
cosAngle = temp * normalizer;
} while (cosAngle < cosThresh);
// v was within the cone. Normalize it and maybe flip the hemisphere.
return v * normalizer;
}
/**
* Step one of the algorithm. All pairs run in parallel. This
* function calculates intermediate results for a pair and
* stores it into shared array.
*
* @arg ij Integer number refering to a pair
*/
GENERIC void calc_pair(int ij)const{
int i = first<nbod>( ij );
int j = second<nbod>( ij );
if(i != j){
// Relative vector from planet i to planet j
double dx[3] = { sys[j][0].pos()- sys[i][0].pos(),sys[j][1].pos()- sys[i][1].pos(), sys[j][2].pos()- sys[i][2].pos() };
// Relative velocity between the planets
double dv[3] = { sys[j][0].vel()- sys[i][0].vel(),sys[j][1].vel()- sys[i][1].vel(), sys[j][2].vel()- sys[i][2].vel() };
// Distance between the planets
double r2 = dx[0]*dx[0]+dx[1]*dx[1]+dx[2]*dx[2];
double jerk_mag = inner_product(dx,dv) * 3.0 / r2;
double acc_mag = rsqrt(r2) / r2;
#pragma unroll
for(int c = 0; c < 3; c ++)
{
shared[ij][c].acc() = dx[c]* acc_mag;
shared[ij][c].jerk() = (dv[c] - dx[c] * jerk_mag ) * acc_mag;
}
}
}
local void sumforces(bodyptr btab, int nbody, bodyptr gtab, int ngrav,
real eps2)
{
bodyptr bp, gp;
double phi0, acc0[NDIM];
vector pos0, dr;
real dr2, dr2i, dr1i, mdr1i, mdr3i;
for (bp = btab; bp < NthBody(btab, nbody); bp = NextBody(bp)) {
phi0 = 0.0;
CLRV(acc0);
SETV(pos0, Pos(bp));
for (gp = gtab; gp < NthBody(gtab, ngrav); gp = NextBody(gp)) {
DOTPSUBV(dr2, dr, Pos(gp), pos0);
dr2i = ((real) 1.0) / (dr2 + eps2);
dr1i = rsqrt(dr2i);
mdr1i = Mass(gp) * dr1i;
mdr3i = mdr1i * dr2i;
phi0 -= mdr1i;
ADDMULVS(acc0, dr, mdr3i);
}
Phi(bp) = phi0;
SETV(Acc(bp), acc0);
}
}
void sphreport(stream ostr, smxptr sm, string options)
{
bodyptr *bptr = sm->kd->bptr;
int i, nupd, nlev[MAXLEV+1], *rprof = sm->rprof;
real rbsum, rbmin, rbmax, flev[MAXLEV+1], flev4, xi[NRPROF-1];
nupd = 0;
for (i = 0; i <= MAXLEV; i++)
nlev[i] = 0;
rbsum = 0.0;
for (i = 0; i < sm->kd->ngas; i++) {
if (Update(bptr[i])) {
nupd++;
nlev[NewLevel(bptr[i])]++;
rbsum += Smooth(bptr[i]);
rbmin = (nupd == 1 ? Smooth(bptr[i]) : MIN(rbmin, Smooth(bptr[i])));
rbmax = (nupd == 1 ? Smooth(bptr[i]) : MAX(rbmin, Smooth(bptr[i])));
}
}
flev4 = 0.0;
for (i = 0; i <= MAXLEV; i++) {
flev[i] = 100.0 * ((real) nlev[i]) / ((real) nupd);
if (i >= 4)
flev4 += flev[i];
}
fprintf(ostr, "\n%9s %9s %9s %9s %29s %9s\n",
"log hmin", "log havg", "log hmax", "freqavg",
"timestep level distribution", "CPUsph");
fprintf(ostr,
"%9.4f %9.4f %9.4f %9.2f %5.1f %5.1f %5.1f %5.1f %5.1f %9.3f\n",
rlog10(rbmin), rlog10(rbsum / nupd), rlog10(rbmax),
sm->freqsum / nupd, flev[0], flev[1], flev[2], flev[3], flev4,
cputime() - sm->cpustart);
if (scanopt(options, "corrfunc")) {
for (i = 0; i <= NRPROF - 2; i++)
xi[i] = -1 + rpow(8.0, i/2.0) * (rprof[i] - rprof[i+1]) /
((1 - rsqrt(0.125)) * rprof[0]);
fprintf(ostr, "\n ");
for (i = NRPROF - 2; i >= 0; i--)
fprintf(ostr, " xi%d", i);
fprintf(ostr, "\n ");
for (i = NRPROF - 2; i >= 0; i--)
fprintf(ostr, "%6.2f", xi[i]);
fprintf(ostr, "\n");
}
if (scanopt(options, "levelhist")) {
fprintf(ostr, "\n ");
for (i = 0; i <= MAXLEV; i++)
if (nlev[i] != 0)
fprintf(ostr, i<10 ? " lev%d" : " lev%d", i);
fprintf(ostr, "\n ");
for (i = 0; i <= MAXLEV; i++)
if (nlev[i] != 0)
fprintf(ostr, "%6d", nlev[i]);
fprintf(ostr, "\n");
}
fflush(ostr);
}
/**
@brief ハウスホルダー・ベクトルへの変換.
@details h[0]=0; ...; h[k-1]=0; h[k]=-s*xi; h[k+1]=x[k+1]; ...; h[n-1]=x[n-1];
@param[in] h 初期化済みのベクトル.サイズはn.
@param[in] x 初期化済みのベクトル.サイズはn.
@param[in] n ベクトルのサイズ.
@param[in] k 第k要素が基準.
@param[out] h ハウスホルダー・ベクトル.
*/
void rhouseholder_vec(int n, int k, rmulti **h, rmulti *alpha, rmulti **x)
{
int p0,p1,prec;
rmulti *eta=NULL,*zeta=NULL,*xi=NULL,*axk=NULL;
// allocate
p0=rget_prec(alpha);
p1=rvec_get_prec_max(n,h);
prec=MAX2(p0,p1);
eta=rallocate_prec(prec);
zeta=rallocate_prec(prec);
xi=rallocate_prec(prec);
axk=rallocate_prec(prec);
//----------- norm
rvec_sum_pow2(xi,n-k-1,&x[k+1]); // xi=sum(abs(x((k+1):end)).^2);
rmul(axk,x[k],x[k]); // axk=|x[k]|^2
radd(eta,axk,xi); // eta=|x[k]|^2+...
rsqrt(axk,axk); // axk=|x[k]|
rsqrt(eta,eta); // eta=sqrt(|x[k]|^2+...)
if(req_d(eta,0)){rsub(xi,eta,axk);} // xi=eta-|x(k)|
else{ // xi=xi/(|x(k)|+eta)
radd(zeta,axk,eta);
rdiv(xi,xi,zeta);
}
//----------- h
rvec_set_zeros(k,h);
rvec_copy(n-k-1,&h[k+1],&x[k+1]); // h((k+1):end)=x((k+1):end);
if(ris_zero(x[k])){
rcopy(h[k],xi); rneg(h[k],h[k]); // h[k]=-xi
}else{
rdiv(zeta,xi,axk); rneg(zeta,zeta); // zeta=-xi/axk;
rmul(h[k],x[k],zeta); // h[k]=zeta*x[k];
}
//----------- alpha
if(req_d(xi,0) || req_d(eta,0)){
rset_d(alpha,0);
}else{
rmul(alpha,xi,eta); // alpha=1/(xi*eta)
rinv(alpha,alpha);
}
// free
eta=rfree(eta);
zeta=rfree(zeta);
xi=rfree(xi);
axk=rfree(axk);
}
void makedisk(void)
{
real m, r, phi, vcir, omega, Adisk, kappa, sigma1, sigma2,
mu_eff, sig_r, sig_p, sig_z, vrad, vorb2, vorb, vphi;
double Trr = 0.0, Tpp = 0.0, Tzz = 0.0;
int i;
bodyptr dp;
for (i = 0; i < ndisk; i++) { /* loop initializing bodies */
m = mdtab[NTAB-1] * ((real) i + 0.5) / ndisk;
r = seval(m, &mdtab[0], &rdtab[0], &rdtab[NTAB], NTAB);
vcir = seval(r, &rdtab[0], &vctab[0], &vctab[NTAB], NTAB);
omega = vcir / r;
Adisk = (omega - spldif(r, &rdtab[0], &vctab[0], &vctab[NTAB], NTAB)) / 2;
if (omega - Adisk < 0.0)
error("%s: kappa undefined (omega - Adisk < 0)\n"
" r, omega, Adisk = %f %f %f\n", getargv0(), r, omega, Adisk);
kappa = 2 * rsqrt(rsqr(omega) - Adisk * omega);
sigma1 = rsqr(alpha1) * mdisk1 * rexp(- alpha1 * r) / TWO_PI;
sigma2 = rsqr(alpha2) * mdisk2 * rexp(- alpha2 * r) / TWO_PI;
mu_eff = (r_mu>0 ? 1 + (mu - 1) * (r / (r + r_mu)) : mu);
sig_z = rsqrt(PI * (sigma1 + sigma2) * zdisk);
sig_r = mu_eff * sig_z;
sig_p = (0.5 * kappa / omega) * sig_r;
vorb2 = rsqr(vcir) + rsqr(sig_r) * (1 - 2 * alpha1 * r) - rsqr(sig_p) +
(r_mu>0 ? rsqr(sig_z) * r * mu_eff*(2*mu-2)*r_mu/rsqr(r+r_mu) : 0);
vorb = rsqrt(MAX(vorb2, 0.0));
dp = NthBody(disk, i); /* set up ptr to disk body */
Mass(dp) = mdisk1 / ndisk;
phi = xrandom(0.0, TWO_PI);
Pos(dp)[0] = r * rsin(phi);
Pos(dp)[1] = r * rcos(phi);
Pos(dp)[2] = zdisk * ratanh(xrandom(-1.0, 1.0));
vrad = (eta > 0 ? pickdist(eta, sig_r) : grandom(0.0, sig_r));
vphi = (eta > 0 ? pickdist(eta, sig_p) : grandom(0.0, sig_p)) + vorb;
Vel(dp)[0] = vrad * rsin(phi) + vphi * rcos(phi);
Vel(dp)[1] = vrad * rcos(phi) - vphi * rsin(phi);
Vel(dp)[2] = grandom(0.0, sig_z);
Trr += Mass(dp) * rsqr(sig_r) / 2;
Tpp += Mass(dp) * (rsqr(vorb) + rsqr(sig_p)) / 2;
Tzz += Mass(dp) * rsqr(sig_z) / 2;
}
eprintf("[%s: Trr = %f Tpp = %f Tzz = %f]\n", getargv0(), Trr, Tpp, Tzz);
}
real sigeff(real sig, real vmax)
{
real x, y, z, Q;
x = vmax / sig;
y = (SQRTPI / SQRT2) * erf(x / SQRT2);
z = rexp(rsqr(x) / 2);
Q = ((-3 * x - rqbe(x)) / z + 3 * y) / (- x / z + y);
return (sig * rsqrt(Q / 3));
}
int main(int argc, string argv[])
{
stream istr;
string bodytags[] = { PosTag, NULL }, intags[MaxBodyFields];
bodyptr btab = NULL, bp;
int nbody, nshell, n;
real tnow, vals[3];
matrix tmpm, qmat;
vector v1, v2, v3;
initparam(argv, defv);
istr = stropen(getparam("in"), "r");
get_history(istr);
layout_body(bodytags, Precision, NDIM);
printf("#%11s %3s %11s %11s %11s\n",
"time", "n", "r_rms", "c/a", "b/a");
while (get_snap(istr, &btab, &nbody, &tnow, intags, FALSE)) {
if (! set_member(intags, PosTag))
error("%s: %s data missing\n", getargv0(), PosTag);
if (nbody % getiparam("nbin") != 0)
error("%s: nbin does not divide number of bodies\n", getargv0());
nshell = nbody / getiparam("nbin");
for (n = 0; n < nbody; n += nshell) {
CLRM(qmat);
for (bp = NthBody(btab, n); bp < NthBody(btab, n + nshell);
bp = NextBody(bp)) {
OUTVP(tmpm, Pos(bp), Pos(bp));
ADDM(qmat, qmat, tmpm);
}
eigensolve(v1, v2, v3, vals, qmat);
printf(" %11.6f %3d %11.6f %11.6f %11.6f\n",
tnow, n / nshell, rsqrt(tracem(qmat) / nshell),
rsqrt(vals[2] / vals[0]), rsqrt(vals[1] / vals[0]));
if (getbparam("listvec")) {
printf("#\t\t\t\t\t\t\t%8.5f %8.5f %8.5f\n", v1[0], v1[1], v1[2]);
printf("#\t\t\t\t\t\t\t%8.5f %8.5f %8.5f\n", v2[0], v2[1], v2[2]);
printf("#\t\t\t\t\t\t\t%8.5f %8.5f %8.5f\n", v3[0], v3[1], v3[2]);
}
}
}
return (0);
}
void testdisk(void)
{
int ndisk, i;
real rmin2, rmax2, eps2, sigma, r_i, theta_i, m_i, v_i;
bodyptr gp, sp;
ndisk = getiparam("ndisk");
ngalaxy = ndisk + (getbparam("nosphr") ? 0 : nspheroid);
galaxy = (bodyptr) allocate(ngalaxy * SizeofBody);
rmin2 = rsqr(getdparam("rmin"));
rmax2 = rsqr(getdparam("rmax"));
eps2 = rsqr(getdparam("eps"));
sigma = getdparam("sigma");
init_random(getiparam("seed"));
for (i = 0; i < ndisk; i++) { /* build disk */
gp = NthBody(galaxy, i);
Mass(gp) = 0.0; /* set mass to zero */
r_i = rsqrt(rmin2 + i * (rmax2 - rmin2) / (ndisk - 1.0));
theta_i = xrandom(0.0, TWO_PI);
Pos(gp)[0] = r_i * rsin(theta_i); /* set positions */
Pos(gp)[1] = r_i * rcos(theta_i);
Pos(gp)[2] = 0.0;
if (r_i < rsph[NTAB-1])
m_i = seval(r_i, &rsph[0], &msph[0], &msph[NTAB], NTAB);
else
m_i = msph[NTAB-1];
v_i = rsqrt(MAX(m_i, 0.0) * r_i*r_i / rpow(r_i*r_i + eps2, 1.5));
/* set velocities */
Vel(gp)[0] = grandom( v_i * rcos(theta_i), sigma);
Vel(gp)[1] = grandom(- v_i * rsin(theta_i), sigma);
Vel(gp)[2] = grandom( 0.0, sigma);
}
if (! getbparam("nosphr"))
for (i = 0; i < nspheroid; i++) { /* append spheroid */
sp = NthBody(spheroid, i);
gp = NthBody(galaxy, ndisk + i);
memcpy(gp, sp, SizeofBody);
}
if (getbparam("zerocm"))
snapcenter(galaxy, ngalaxy, MassField.offset);
}
__device__ mmm1dgpu_real dev_K0(mmm1dgpu_real x)
{
mmm1dgpu_real c = evaluateAsChebychevSeriesAt(
x<=2? bk0_data :x<=8? ak0_data : ak02_data,
x<=2? bk0_size :x<=8? ak0_size : ak02_size,
x<=2? x*x/2-1.0f :x<=8? (16/x-5.0f)/3.0f : (16/x-1.0f)
);
if (x <= 2) {
mmm1dgpu_real I0 = evaluateAsChebychevSeriesAt(bi0_data, bi0_size, x*x/4.5f-1.0f);
return (-log(x) + M_LN2f)*I0 + c;
}
return exp(-x)*c*rsqrt(x);
}
__device__ mmm1dgpu_real dev_K1(mmm1dgpu_real x)
{
mmm1dgpu_real c = evaluateAsChebychevSeriesAt(
x<=2? bk1_data :x<=8? ak1_data : ak12_data,
x<=2? bk1_size :x<=8? ak1_size : ak12_size,
x<=2? x*x/2-1.0f :x<=8? (16/x-5.0f)/3.0f : (16/x-1.0f)
);
if(x <= 2) {
mmm1dgpu_real I1 = x * evaluateAsChebychevSeriesAt(bi1_data, bi1_size, x*x/4.5f-1.0f);
return (log(x) - M_LN2f)*I1 + c/x;
}
return exp(-x)*c*rsqrt(x);
}
local void hqmforces(bodyptr btab, int nbody, real M, real a, real b,
real tol)
{
bodyptr bp;
double r, mr3i, params[4], phi0, aR0, az0, abserr[3];
static gsl_integration_workspace *wksp = NULL;
gsl_function FPhi, F_aR, F_az;
static double maxerr = 0.0;
int stat[3];
if (a == b) { // spherical case is easy!
for (bp = btab; bp < NthBody(btab, nbody); bp = NextBody(bp)) {
r = absv(Pos(bp));
Phi(bp) = - M / (a + r);
mr3i = M * rsqr(r / (a + r)) / rqbe(r);
MULVS(Acc(bp), Pos(bp), - mr3i);
}
} else { // flattened case is harder
if (wksp == NULL) { // on first call, initialze
wksp = gsl_integration_workspace_alloc(1000);
gsl_set_error_handler_off(); // handle errors below
}
FPhi.function = &intPhi;
F_aR.function = &int_aR;
F_az.function = &int_az;
FPhi.params = F_aR.params = F_az.params = params;
a2(params) = rsqr(a);
b2(params) = rsqr(b);
for (bp = btab; bp < NthBody(btab, nbody); bp = NextBody(bp)) {
R(params) = rsqrt(rsqr(Pos(bp)[0]) + rsqr(Pos(bp)[1]));
z(params) = Pos(bp)[2];
stat[0] = gsl_integration_qagiu(&FPhi, 0.0, tol, 0.0,
1000, wksp, &phi0, &abserr[0]);
stat[1] = gsl_integration_qagiu(&F_aR, 0.0, tol, 0.0,
1000, wksp, &aR0, &abserr[1]);
stat[2] = gsl_integration_qagiu(&F_az, 0.0, tol, 0.0,
1000, wksp, &az0, &abserr[2]);
if (stat[0] || stat[1] || stat[2]) // any errors reported?
for (int i = 0; i < 3; i++)
if (stat[i] != 0 && abserr[i] > maxerr) {
eprintf("[%s.hqmforces: warning: %s abserr[%d] = %g]\n",
getprog(), gsl_strerror(stat[i]), i+1, abserr[i]);
maxerr = abserr[i]; // adjust reporting threshold
}
Phi(bp) = - M * phi0;
Acc(bp)[0] = - M * (Pos(bp)[0] / R(params)) * aR0;
Acc(bp)[1] = - M * (Pos(bp)[1] / R(params)) * aR0;
Acc(bp)[2] = - M * az0;
}
}
}
CUDA_DEVICE static inline void
calc_vxi(particle_real_t vxi[3], particle_t *prt)
{
particle_real_t *px = &particle_px(prt);
#ifdef __CUDACC__
particle_real_t root = rsqrt(1.f + sqr(px[0]) + sqr(px[1]) + sqr(px[2]));
#else
particle_real_t root = 1.f
/ particle_real_sqrt(1.f + sqr(px[0]) + sqr(px[1]) + sqr(px[2]));
#endif
vxi[0] = px[0] * root;
vxi[1] = px[1] * root;
vxi[2] = px[2] * root;
}
void plummodel(real mfrac)
{
real rsc, vsc, r, v, x, y;
bodyptr p;
if (mfrac < 0 || mfrac > 1) // check for silly values
error("%s: absurd value for mfrac\n", getprog());
rsc = (3 * PI) / 16; // set length scale factor
vsc = rsqrt(1.0 / rsc); // and speed scale factor
for (p = btab; p < NthBody(btab,nbody); p = NextBody(p)) {
// loop over particles
Mass(p) = 1.0 / nbody; // set masses equal
x = xrandom(0.0, mfrac); // pick enclosed mass
r = 1.0 / rsqrt(rpow(x, -2.0/3.0) - 1); // find enclosing radius
pickshell(Pos(p), NDIM, rsc * r); // pick position vector
do { // use von Neumann rejection
x = xrandom(0.0, 1.0); // for x in range 0:1
y = xrandom(0.0, 0.1); // max of g(x) is 0.092
} while (y > x*x * rpow(1 - x*x, 3.5)); // select from g(x)
v = x * rsqrt(2.0 / rsqrt(1 + r*r)); // find resulting speed
pickshell(Vel(p), NDIM, vsc * v); // pick velocity vector
Phi(p) = -1.0 / (rsc * rsqrt(rsqr(r) + 1)); // compute model potential
}
}
/*! Samples the distribution. \param s is the sample location
* provided by the caller. */
inline Vec3fa AnisotropicBlinn__sample(const AnisotropicBlinn* This, const float sx, const float sy)
{
const float phi =float(two_pi)*sx;
const float sinPhi0 = sqrtf(This->nx+1)*sinf(phi);
const float cosPhi0 = sqrtf(This->ny+1)*cosf(phi);
const float norm = rsqrt(sqr(sinPhi0)+sqr(cosPhi0));
const float sinPhi = sinPhi0*norm;
const float cosPhi = cosPhi0*norm;
const float n = This->nx*sqr(cosPhi)+This->ny*sqr(sinPhi);
const float cosTheta = powf(sy,rcp(n+1));
const float sinTheta = cos2sin(cosTheta);
const float pdf = This->norm1*powf(cosTheta,n);
const Vec3fa h = Vec3fa(cosPhi * sinTheta, sinPhi * sinTheta, cosTheta);
const Vec3fa wh = h.x*This->dx + h.y*This->dy + h.z*This->dz;
return Vec3fa(wh,pdf);
}
void polymodel1(void)
{
bodyptr p;
real r, x, v;
for (p = btab; p < NthBody(btab, nbody); p = NextBody(p)) {
Mass(p) = 1.0 / nbody;
r = xrandom(0.0, 2.0);
pickshell(Pos(p), 3, r);
x = SQRT2 * rcos(PI * (2.0 - xrandom(0.0, 1.0)) / 4.0);
v = (xrandom(-1.0, 1.0) < 0.0 ? -1.0 : 1.0) *
(1 - x*x) * rsqrt(rlog(2 / r));
MULVS(Vel(p), Pos(p), v/r);
Phi(p) = 0.5 * rlog(r / 2.0) - 0.5;
}
bodyfields[4] = NULL; // don't output Aux data
}
static
void
test_fixed_err()
{
bool err_ok=true;
for (fixed x=fixed(0.0001); x< fixed(100000.0); x*= fixed(1.5)) {
fixed y0 = fixed(1)/sqrt(x);
fixed y1 = rsqrt(x);
fixed err = fabs(y1 / y0 - fixed(1));
bool this_ok = err< fixed(0.0001);
err_ok &= this_ok;
if (!this_ok) {
printf("%g %g %g %g\n", (double)x, (double)y0, (double)y1, (double)err);
}
}
ok(err_ok, "fixed rsqrt error", 0);
}
Light_SampleRes PointLight_sample(const Light* super,
const DifferentialGeometry& dg,
const Vec2f& s)
{
const PointLight* self = (PointLight*)super;
Light_SampleRes res;
// extant light vector from the hit point
const Vec3fa dir = self->position - dg.P;
const float dist2 = dot(dir, dir);
const float invdist = rsqrt(dist2);
// normalized light vector
res.dir = dir * invdist;
res.dist = dist2 * invdist;
res.pdf = inf; // per default we always take this res
// convert from power to radiance by attenuating by distance^2
res.weight = self->power * sqr(invdist);
const float sinTheta = self->radius * invdist;
if ((self->radius > 0.f) & (sinTheta > 0.005f)) {
// res surface of sphere as seen by hit point -> cone of directions
// for very small cones treat as point light, because float precision is not good enough
if (sinTheta < 1.f) {
const float cosTheta = sqrt(1.f - sinTheta * sinTheta);
const Vec3fa localDir = uniformSampleCone(cosTheta, s);
res.dir = frame(res.dir) * localDir;
res.pdf = uniformSampleConePDF(cosTheta);
const float c = localDir.z;
res.dist = c*res.dist - sqrt(sqr(self->radius) - (1.f - c*c) * dist2);
// TODO scale radiance by actual distance
} else { // inside sphere
const Vec3fa localDir = cosineSampleHemisphere(s);
res.dir = frame(dg.Ns) * localDir;
res.pdf = cosineSampleHemispherePDF(localDir);
// TODO:
res.weight = self->power * rcp(sqr(self->radius));
res.dist = self->radius;
}
}
return res;
}
int init_tunnel_offsets(struct fb *tunnel)
{
int x, y;
int u, v;
float angle, radius;
int *p = tunnel->pixbuf;
for(y = -tunnel->height/2; y < tunnel->height/2; y++) {
for(x = -tunnel->width/2; x < tunnel->width/2; x++) {
radius = rsqrt(x*x + y*y);
if(radius < 1) radius = 1;
angle = ratan2(y, x) + PI;
v = 100000.0f/radius;
u = angle*128/PI;
*p++ = (u & 0xff) + ((v & 0xff)<<8);
}
}
return 0;
}