diff options
author | IOhannes m zmölnig <zmoelnig@users.sourceforge.net> | 2015-08-28 17:55:26 +0000 |
---|---|---|
committer | IOhannes m zmölnig <zmoelnig@users.sourceforge.net> | 2015-08-28 17:55:26 +0000 |
commit | 739a8290a8b05645a2d055aed8e5eb76ba2e24e6 (patch) | |
tree | 13713efb9e01a6f2912ce4fe0a12af4869a7593b /src/doublepole~.c | |
parent | 863ac0a263247d43d48c06ed942ddd470f811507 (diff) |
one-true indentation
svn path=/trunk/externals/zexy/; revision=17537
Diffstat (limited to 'src/doublepole~.c')
-rw-r--r-- | src/doublepole~.c | 163 |
1 files changed, 84 insertions, 79 deletions
diff --git a/src/doublepole~.c b/src/doublepole~.c index 93a2a74..a06e861 100644 --- a/src/doublepole~.c +++ b/src/doublepole~.c @@ -2,121 +2,126 @@ * For information on usage and redistribution, and for a DISCLAIMER OF ALL * WARRANTIES, see the file, "LICENSE.txt," in this distribution. */ -/* "filters", both linear and nonlinear. +/* "filters", both linear and nonlinear. */ #include "m_pd.h" #include <math.h> //* ---------------- doublepole~ - raw doublepole filter ----------------- */ -typedef struct doublepolectl -{ - t_sample c_x1; - t_sample c_x2; - t_sample c_fb1; - t_sample c_fb2; +typedef struct doublepolectl { + t_sample c_x1; + t_sample c_x2; + t_sample c_fb1; + t_sample c_fb2; } t_doublepolectl; -typedef struct sigdoublepole -{ - t_object x_obj; - t_float x_f; - t_doublepolectl x_cspace; - t_doublepolectl *x_ctl; +typedef struct sigdoublepole { + t_object x_obj; + t_float x_f; + t_doublepolectl x_cspace; + t_doublepolectl *x_ctl; } t_sigdoublepole; static t_class *sigdoublepole_class = NULL; -static void sigdoublepole_list(t_sigdoublepole *x, t_symbol *s, int argc, t_atom *argv); +static void sigdoublepole_list(t_sigdoublepole *x, t_symbol *s, int argc, + t_atom *argv); static void *sigdoublepole_new(t_symbol *s, int argc, t_atom *argv) { - t_sigdoublepole *x = (t_sigdoublepole *)pd_new(sigdoublepole_class); - outlet_new(&x->x_obj, &s_signal); - x->x_ctl = &x->x_cspace; - x->x_cspace.c_x1 = x->x_cspace.c_x2 = 0; - sigdoublepole_list(x, s, argc, argv); - x->x_f = 0; - return (x); + t_sigdoublepole *x = (t_sigdoublepole *)pd_new(sigdoublepole_class); + outlet_new(&x->x_obj, &s_signal); + x->x_ctl = &x->x_cspace; + x->x_cspace.c_x1 = x->x_cspace.c_x2 = 0; + sigdoublepole_list(x, s, argc, argv); + x->x_f = 0; + return (x); } static t_int *sigdoublepole_perform(t_int *w) { - t_sample *in = (t_sample *)(w[1]); - t_sample *out = (t_sample *)(w[2]); - t_doublepolectl *c = (t_doublepolectl *)(w[3]); - int n = (t_int)(w[4]); - int i; - t_sample last = c->c_x1; - t_sample prev = c->c_x2; - t_sample fb1 = c->c_fb1; - t_sample fb2 = c->c_fb2; - for (i = 0; i < n; i++) - { - t_sample output = *in++ + fb1 * last + fb2 * prev; - if (PD_BIGORSMALL(output)) - output = 0; - *out++ = output; - prev = last; - last = output; + t_sample *in = (t_sample *)(w[1]); + t_sample *out = (t_sample *)(w[2]); + t_doublepolectl *c = (t_doublepolectl *)(w[3]); + int n = (t_int)(w[4]); + int i; + t_sample last = c->c_x1; + t_sample prev = c->c_x2; + t_sample fb1 = c->c_fb1; + t_sample fb2 = c->c_fb2; + for (i = 0; i < n; i++) { + t_sample output = *in++ + fb1 * last + fb2 * prev; + if (PD_BIGORSMALL(output)) { + output = 0; } - c->c_x1 = last; - c->c_x2 = prev; - return (w+5); + *out++ = output; + prev = last; + last = output; + } + c->c_x1 = last; + c->c_x2 = prev; + return (w+5); } -static void sigdoublepole_list(t_sigdoublepole *x, t_symbol *s, int argc, t_atom *argv) +static void sigdoublepole_list(t_sigdoublepole *x, t_symbol *s, int argc, + t_atom *argv) { - t_float fb1 = atom_getfloatarg(0, argc, argv); - t_float fb2 = atom_getfloatarg(1, argc, argv); - t_float discriminant = fb1 * fb1 + 4 * fb2; - t_doublepolectl *c = x->x_ctl; - if (discriminant < 0) /* imaginary roots -- resonant filter */ - { - /* they're conjugates so we just check that the product - is less than one */ - if (fb2 >= -1.0f) goto stable; + t_float fb1 = atom_getfloatarg(0, argc, argv); + t_float fb2 = atom_getfloatarg(1, argc, argv); + t_float discriminant = fb1 * fb1 + 4 * fb2; + t_doublepolectl *c = x->x_ctl; + if (discriminant < 0) { /* imaginary roots -- resonant filter */ + /* they're conjugates so we just check that the product + is less than one */ + if (fb2 >= -1.0f) { + goto stable; } - else /* real roots */ - { - /* check that the parabola 1 - fb1 x - fb2 x^2 has a - vertex between -1 and 1, and that it's nonnegative - at both ends, which implies both roots are in [1-,1]. */ - if (fb1 <= 2.0f && fb1 >= -2.0f && - 1.0f - fb1 -fb2 >= 0 && 1.0f + fb1 - fb2 >= 0) - goto stable; + } else { /* real roots */ + /* check that the parabola 1 - fb1 x - fb2 x^2 has a + vertex between -1 and 1, and that it's nonnegative + at both ends, which implies both roots are in [1-,1]. */ + if (fb1 <= 2.0f && fb1 >= -2.0f && + 1.0f - fb1 -fb2 >= 0 && 1.0f + fb1 - fb2 >= 0) { + goto stable; } - /* if unstable, just bash to zero */ - fb1 = fb2 = 0; + } + /* if unstable, just bash to zero */ + fb1 = fb2 = 0; stable: - c->c_fb1 = fb1; - c->c_fb2 = fb2; + c->c_fb1 = fb1; + c->c_fb2 = fb2; } -static void sigdoublepole_set(t_sigdoublepole *x, t_symbol *s, int argc, t_atom *argv) +static void sigdoublepole_set(t_sigdoublepole *x, t_symbol *s, int argc, + t_atom *argv) { - t_doublepolectl *c = x->x_ctl; - c->c_x1 = atom_getfloatarg(0, argc, argv); - c->c_x2 = atom_getfloatarg(1, argc, argv); + t_doublepolectl *c = x->x_ctl; + c->c_x1 = atom_getfloatarg(0, argc, argv); + c->c_x2 = atom_getfloatarg(1, argc, argv); } static void sigdoublepole_dsp(t_sigdoublepole *x, t_signal **sp) { - dsp_add(sigdoublepole_perform, 4, - sp[0]->s_vec, sp[1]->s_vec, - x->x_ctl, sp[0]->s_n); + dsp_add(sigdoublepole_perform, 4, + sp[0]->s_vec, sp[1]->s_vec, + x->x_ctl, sp[0]->s_n); } void doublepole_tilde_setup(void) { - sigdoublepole_class = class_new(gensym("doublepole~"), (t_newmethod)sigdoublepole_new, - 0, sizeof(t_sigdoublepole), 0, A_GIMME, 0); - CLASS_MAINSIGNALIN(sigdoublepole_class, t_sigdoublepole, x_f); - class_addmethod(sigdoublepole_class, (t_method)sigdoublepole_dsp, gensym("dsp"), 0); - class_addlist(sigdoublepole_class, sigdoublepole_list); - class_addmethod(sigdoublepole_class, (t_method)sigdoublepole_set, gensym("set"), - A_GIMME, 0); - class_addmethod(sigdoublepole_class, (t_method)sigdoublepole_set, gensym("clear"), - A_GIMME, 0); + sigdoublepole_class = class_new(gensym("doublepole~"), + (t_newmethod)sigdoublepole_new, + 0, sizeof(t_sigdoublepole), 0, A_GIMME, 0); + CLASS_MAINSIGNALIN(sigdoublepole_class, t_sigdoublepole, x_f); + class_addmethod(sigdoublepole_class, (t_method)sigdoublepole_dsp, + gensym("dsp"), 0); + class_addlist(sigdoublepole_class, sigdoublepole_list); + class_addmethod(sigdoublepole_class, (t_method)sigdoublepole_set, + gensym("set"), + A_GIMME, 0); + class_addmethod(sigdoublepole_class, (t_method)sigdoublepole_set, + gensym("clear"), + A_GIMME, 0); } |