[PD-cvs] pd/src d_filter.c,1.3.4.3,1.3.4.4

[Tim Blechmann]

Update of /cvsroot/pure-data/pd/src
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv6391

Modified Files:
      Tag: devel_0_38
	d_filter.c 
Log Message:
unrolled biquad and relaxed denormal bashing


Index: d_filter.c
===================================================================
RCS file: /cvsroot/pure-data/pd/src/d_filter.c,v
retrieving revision 1.3.4.3
retrieving revision 1.3.4.4
diff -C2 -d -r1.3.4.3 -r1.3.4.4
*** d_filter.c	21 May 2005 13:08:13 -0000	1.3.4.3
--- d_filter.c	22 Jun 2005 18:47:52 -0000	1.3.4.4
***************
*** 399,402 ****
--- 399,472 ----
  }
  
+ /* tb: some loop unrolling & do some relaxed denormal bashing */
+ #if 1
+ static t_int *sigbiquad_perf8(t_int *w)
+ {
+     float *in = (float *)(w[1]);
+     float *out = (float *)(w[2]);
+     t_biquadctl *c = (t_biquadctl *)(w[3]);
+     int n = (t_int)(w[4])>>3;
+     int i;
+     float last = c->c_x1;
+     float prev = c->c_x2;
+     float fb1 = c->c_fb1;
+     float fb2 = c->c_fb2;
+     float ff1 = c->c_ff1;
+     float ff2 = c->c_ff2;
+     float ff3 = c->c_ff3;
+     for (i = 0; i < n; i++)
+     {
+         float output =  *in++ + fb1 * last + fb2 * prev;
+ #ifdef __i386__
+         output += 1e-10; /* quantizing should be faster than PD_BIGORSMALL */
+         output -= 1e-10; /* only doing this every 8th sample should be ok */
+ #endif
+ 		*out++ = ff1 * output + ff2 * last + ff3 * prev;
+         prev = last;
+         last = output;
+ 
+         output =  *in++ + fb1 * last + fb2 * prev;
+         *out++ = ff1 * output + ff2 * last + ff3 * prev;
+         prev = last;
+         last = output;
+ 
+         output =  *in++ + fb1 * last + fb2 * prev;
+         *out++ = ff1 * output + ff2 * last + ff3 * prev;
+         prev = last;
+         last = output;
+ 
+         output =  *in++ + fb1 * last + fb2 * prev;
+         *out++ = ff1 * output + ff2 * last + ff3 * prev;
+         prev = last;
+         last = output;
+ 
+         output =  *in++ + fb1 * last + fb2 * prev;
+ 		output += 1e-10;
+ 		output -= 1e-10;
+ 		*out++ = ff1 * output + ff2 * last + ff3 * prev;
+         prev = last;
+         last = output;
+ 
+         output =  *in++ + fb1 * last + fb2 * prev;
+         *out++ = ff1 * output + ff2 * last + ff3 * prev;
+         prev = last;
+         last = output;
+ 
+         output =  *in++ + fb1 * last + fb2 * prev;
+         *out++ = ff1 * output + ff2 * last + ff3 * prev;
+         prev = last;
+         last = output;
+ 
+         output =  *in++ + fb1 * last + fb2 * prev;
+         *out++ = ff1 * output + ff2 * last + ff3 * prev;
+         prev = last;
+         last = output;
+     }
+     c->c_x1 = last;
+     c->c_x2 = prev;
+     return (w+5);
+ }
+ #endif
+ 
  static void sigbiquad_list(t_sigbiquad *x, t_symbol *s, int argc, t_atom *argv)
  {
***************
*** 442,449 ****
  static void sigbiquad_dsp(t_sigbiquad *x, t_signal **sp)
  {
      dsp_add(sigbiquad_perform, 4,
          sp[0]->s_vec, sp[1]->s_vec, 
              x->x_ctl, sp[0]->s_n);
! 
  }
  
--- 512,530 ----
  static void sigbiquad_dsp(t_sigbiquad *x, t_signal **sp)
  {
+ #if 1
+ 	const int n = sp[0]->s_n;
+ 	if (n&7)
+ 		dsp_add(sigbiquad_perform, 4,
+ 			sp[0]->s_vec, sp[1]->s_vec, 
+             x->x_ctl, sp[0]->s_n);
+ 	else
+ 		dsp_add(sigbiquad_perf8, 4,
+ 			sp[0]->s_vec, sp[1]->s_vec, 
+             x->x_ctl, sp[0]->s_n);
+ #else
      dsp_add(sigbiquad_perform, 4,
          sp[0]->s_vec, sp[1]->s_vec, 
              x->x_ctl, sp[0]->s_n);
! #endif
  }