plugin-pin management.. nearly there
[ardour.git] / libs / ardour / interpolation.cc
index c3a45a0401735797551ca1cf949bbeff07a55b55..286030c26db487cb8d825d0bd712119d3be9116a 100644 (file)
+/*
+    Copyright (C) 2012 Paul Davis
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program; if not, write to the Free Software
+    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+*/
+
 #include <stdint.h>
 #include <cstdio>
 
 #include "ardour/interpolation.h"
+#include "ardour/midi_buffer.h"
 
 using namespace ARDOUR;
 
-nframes_t
-FixedPointLinearInterpolation::interpolate (int channel, nframes_t nframes, Sample *input, Sample *output)
+
+framecnt_t
+LinearInterpolation::interpolate (int channel, framecnt_t nframes, Sample *input, Sample *output)
 {
-       // the idea behind phase is that when the speed is not 1.0, we have to 
-       // interpolate between samples and then we have to store where we thought we were. 
-       // rather than being at sample N or N+1, we were at N+0.8792922
-       // so the "phase" element, if you want to think about this way, 
-       // varies from 0 to 1, representing the "offset" between samples
-       uint64_t        the_phase = last_phase[channel];
-       
-       // acceleration
-       int64_t  phi_delta;
-
-       // phi = fixed point speed
-       if (phi != target_phi) {
-               phi_delta = ((int64_t)(target_phi - phi)) / nframes;
-       } else {
-               phi_delta = 0;
-       }
-       
        // index in the input buffers
-       nframes_t   i = 0;
+       framecnt_t i = 0;
+
+       double acceleration = 0;
+
+       if (_speed != _target_speed) {
+               acceleration = _target_speed - _speed;
+       }
+
+       for (framecnt_t outsample = 0; outsample < nframes; ++outsample) {
+               double const d = phase[channel] + outsample * (_speed + acceleration);
+               i = floor(d);
+               Sample fractional_phase_part = d - i;
+               if (fractional_phase_part >= 1.0) {
+                       fractional_phase_part -= 1.0;
+                       i++;
+               }
 
-       for (nframes_t outsample = 0; outsample < nframes; ++outsample) {
-               i = the_phase >> 24;
-               Sample fractional_phase_part = (the_phase & fractional_part_mask) / binary_scaling_factor;
-               
                if (input && output) {
                        // Linearly interpolate into the output buffer
-                       output[outsample] = 
+                       output[outsample] =
                                input[i] * (1.0f - fractional_phase_part) +
                                input[i+1] * fractional_phase_part;
                }
-               
-               the_phase += phi + phi_delta;
        }
 
-       last_phase[channel] = (the_phase & fractional_part_mask);
-       
-       // playback distance
+       double const distance = phase[channel] + nframes * (_speed + acceleration);
+       i = floor(distance);
+       phase[channel] = distance - i;
        return i;
 }
 
-void 
-FixedPointLinearInterpolation::add_channel_to (int /*input_buffer_size*/, int /*output_buffer_size*/)
-{
-       last_phase.push_back (0);
-}
-
-void 
-FixedPointLinearInterpolation::remove_channel_from ()
-{
-       last_phase.pop_back ();
-}
-
-void
-FixedPointLinearInterpolation::reset() 
-{
-       for (size_t i = 0; i <= last_phase.size(); i++) {
-               last_phase[i] = 0;
-       }
-}
-
-
-nframes_t
-LinearInterpolation::interpolate (int channel, nframes_t nframes, Sample *input, Sample *output)
+framecnt_t
+CubicInterpolation::interpolate (int channel, framecnt_t nframes, Sample *input, Sample *output)
 {
        // index in the input buffers
-       nframes_t   i = 0;
-       
+       framecnt_t   i = 0;
+
        double acceleration;
        double distance = 0.0;
-       
+
        if (_speed != _target_speed) {
                acceleration = _target_speed - _speed;
        } else {
                acceleration = 0.0;
        }
-       
+
        distance = phase[channel];
-       for (nframes_t outsample = 0; outsample < nframes; ++outsample) {
-               i = floor(distance);
-               Sample fractional_phase_part = distance - i;
-               if (fractional_phase_part >= 1.0) {
-                       fractional_phase_part -= 1.0;
-                       i++;
-               }
-               
+
+       if (nframes < 3) {
+               /* no interpolation possible */
+
                if (input && output) {
-               // Linearly interpolate into the output buffer
-                       output[outsample] = 
-                               input[i] * (1.0f - fractional_phase_part) +
-                               input[i+1] * fractional_phase_part;
+                       for (i = 0; i < nframes; ++i) {
+                               output[i] = input[i];
+                       }
                }
-               distance += _speed + acceleration;
+
+               return nframes;
        }
-       
-       i = floor(distance);
-       phase[channel] = distance - floor(distance);
-       
-       return i;
-}
 
-nframes_t
-CubicInterpolation::interpolate (int channel, nframes_t nframes, Sample *input, Sample *output)
-{
-    // index in the input buffers
-    nframes_t   i = 0;
-    
-    double acceleration;
-    double distance = 0.0;
-    
-    if (_speed != _target_speed) {
-        acceleration = _target_speed - _speed;
-    } else {
-        acceleration = 0.0;
-    }
-    
-    distance = phase[channel];
-    for (nframes_t outsample = 0; outsample < nframes; ++outsample) {
-        i = floor(distance);
-        Sample fractional_phase_part = distance - i;
-        if (fractional_phase_part >= 1.0) {
-            fractional_phase_part -= 1.0;
-            i++;
-        }
-        
-        if (input && output) {
-            // Cubically interpolate into the output buffer
-            output[outsample] = cube_interp(fractional_phase_part, input[i-1], input[i], input[i+1], input[i+2]);
-        }
-        distance += _speed + acceleration;
-    }
-    
-    i = floor(distance);
-    phase[channel] = distance - floor(distance);
-    
-    return i;
-}
+       /* keep this condition out of the inner loop */
 
-SplineInterpolation::SplineInterpolation()
-{
-    reset ();
-}
+       if (input && output) {
 
-void SplineInterpolation::reset()
-{
-    Interpolation::reset();
-    M[0] = 0.0;
-    M[1] = 0.0;
-    M[2] = 0.0;
-}
+               Sample inm1;
 
-nframes_t
-SplineInterpolation::interpolate (int channel, nframes_t nframes, Sample *input, Sample *output)
-{
-    
-    // now interpolate
-    // index in the input buffers
-    nframes_t   i = 0, delta_i = 0;
-    
-    double acceleration;
-    double distance = 0.0;
-    
-    if (_speed != _target_speed) {
-        acceleration = _target_speed - _speed;
-    } else {
-        acceleration = 0.0;
-    }
-    
-    distance = phase[channel];
-    assert(distance >= 0.0 && distance < 1.0);
-    
-    for (nframes_t outsample = 0; outsample < nframes; outsample++) {
-        i = floor(distance);
-        
-        double x = double(distance) - double(i);
-        
-        // if distance is something like 0.999999999999
-        // it will get rounded to 1 in the conversion to float above
-        while (x >= 1.0) {
-            x -= 1.0;
-            i++;
-        } 
-        
-        assert(x >= 0.0 && x < 1.0);
-        
-        if (input && output) {
-            // if i changed, recalculate coefficients
-            if (delta_i == 1) {
-                // if i changed, rotate the M's
-                M[0] = M[1];
-                M[1] = M[2];
-                M[2] = 6.0 * (input[i] - 2.0*input[i+1] + input[i+2]) - 4.0*M[1] - M[0];
-                printf("\ny[%d] = %lf\n", i, input[i]);
-                printf("y[%d] = %lf\n", i+1, input[i+1]);
-                printf("y[%d] = %lf\n\n", i+2, input[i+2]);
-                printf("M[2] = %lf  M[1] = %lf  M[0] = %lf y-term: %lf M-term: %lf\n", 
-                        M[2], M[1], M[0],  6.0 * (input[i] - 2.0*input[i+1] + input[i+2]),
-                        - 4.0*M[1] - M[0]);
-            }
-            double a3 = (M[1] - M[0]) / 6.0;
-            double a2 = M[0] / 2.0;
-            double a1 = input[i+1] - input[i] - (M[1] + 2.0*M[0]) / 6.0;
-            double a0 = input[i];
-            // interpolate into the output buffer
-            output[outsample] = ((a3*x + a2)*x + a1)*x + a0;
-            //printf( "input[%d/%d] = %lf/%lf  distance: %lf output[%d] = %lf\n", i, i+1, input[i], input[i+1], distance, outsample, output[outsample]);
-            
-        }
-        distance += _speed + acceleration;
-
-        delta_i = floor(distance) - i;
-    }
-    
-    i = floor(distance);
-    phase[channel] = distance - floor(distance);
-    assert (phase[channel] >= 0.0 && phase[channel] < 1.0);
-    
-    return i;
-}
+               if (floor (distance) == 0.0) {
+                       /* best guess for the fake point we have to add to be able to interpolate at i == 0:
+                          .... maintain slope of first actual segment ...
+                          */
+                       inm1 = input[i] - (input[i+1] - input[i]);
+               } else {
+                       inm1 = input[i-1];
+               }
 
-LibSamplerateInterpolation::LibSamplerateInterpolation() : state (0)
-{
-       _speed = 1.0;
-}
+               for (framecnt_t outsample = 0; outsample < nframes; ++outsample) {
 
-LibSamplerateInterpolation::~LibSamplerateInterpolation() 
-{
-       for (size_t i = 0; i < state.size(); i++) {
-               state[i] = src_delete (state[i]);
-       }
-}
+                       float f = floor (distance);
+                       float fractional_phase_part = distance - f;
 
-void
-LibSamplerateInterpolation::set_speed (double new_speed)
-{ 
-       _speed = new_speed; 
-       for (size_t i = 0; i < state.size(); i++) {
-               src_set_ratio (state[i], 1.0/_speed);
-       }
-}
+                       /* get the index into the input we should start with */
 
-void
-LibSamplerateInterpolation::reset_state ()
-{
-       printf("INTERPOLATION: reset_state()\n");
-       for (size_t i = 0; i < state.size(); i++) {
-               if (state[i]) {
-                       src_reset (state[i]);
-               } else {
-                       state[i] = src_new (SRC_SINC_FASTEST, 1, &error);
+                       i = lrintf (f);
+
+                       /* fractional_phase_part only reaches 1.0 thanks to float imprecision. In theory
+                          it should always be < 1.0. If it ever >= 1.0, then bump the index we use
+                          and back it off. This is the point where we "skip" an entire sample in the
+                          input, because the phase part has accumulated so much error that we should
+                          really be closer to the next sample. or something like that ...
+                          */
+
+                       if (fractional_phase_part >= 1.0) {
+                               fractional_phase_part -= 1.0;
+                               ++i;
+                       }
+
+                       // Cubically interpolate into the output buffer: keep this inlined for speed and rely on compiler
+                       // optimization to take care of the rest
+                       // shamelessly ripped from Steve Harris' swh-plugins (ladspa-util.h)
+
+                       output[outsample] = input[i] + 0.5f * fractional_phase_part * (input[i+1] - inm1 +
+                                       fractional_phase_part * (4.0f * input[i+1] + 2.0f * inm1 - 5.0f * input[i] - input[i+2] +
+                                               fractional_phase_part * (3.0f * (input[i] - input[i+1]) - inm1 + input[i+2])));
+
+                       distance += _speed + acceleration;
+                       inm1 = input[i];
                }
+
+               i = floor(distance);
+               phase[channel] = distance - floor(distance);
+
+       } else {
+               /* used to calculate play-distance with acceleration (silent roll)
+                * (use same algorithm as real playback for identical rounding/floor'ing)
+                */
+               for (framecnt_t outsample = 0; outsample < nframes; ++outsample) {
+                       distance += _speed + acceleration;
+               }
+               i = floor(distance);
        }
-}
 
-void
-LibSamplerateInterpolation::add_channel_to (int input_buffer_size, int output_buffer_size) 
-{
-       SRC_DATA* newdata = new SRC_DATA;
-       
-       /* Set up sample rate converter info. */
-       newdata->end_of_input = 0 ; 
-
-       newdata->input_frames  = input_buffer_size;
-       newdata->output_frames = output_buffer_size;
-
-       newdata->input_frames_used = 0 ;
-       newdata->output_frames_gen = 0 ;
-
-       newdata->src_ratio = 1.0/_speed;
-       
-       data.push_back (newdata);
-       state.push_back (0);
-       
-       reset_state ();
+       return i;
 }
 
-void
-LibSamplerateInterpolation::remove_channel_from () 
+framecnt_t
+CubicMidiInterpolation::distance (framecnt_t nframes, bool roll)
 {
-       SRC_DATA* d = data.back ();
-       delete d;
-       data.pop_back ();
-       if (state.back ()) {
-               src_delete (state.back ());
+       assert(phase.size() == 1);
+
+       framecnt_t i = 0;
+
+       double acceleration;
+       double distance = 0.0;
+
+       if (nframes < 3) {
+               return nframes;
        }
-       state.pop_back ();
-       reset_state ();
-}
 
-nframes_t
-LibSamplerateInterpolation::interpolate (int channel, nframes_t nframes, Sample *input, Sample *output)
-{      
-       if (!data.size ()) {
-               printf ("ERROR: trying to interpolate with no channels\n");
-               return 0;
+       if (_speed != _target_speed) {
+               acceleration = _target_speed - _speed;
+       } else {
+               acceleration = 0.0;
+       }
+
+       distance = phase[0];
+
+       for (framecnt_t outsample = 0; outsample < nframes; ++outsample) {
+               distance += _speed + acceleration;
        }
-       
-       data[channel]->data_in     = input;
-       data[channel]->data_out   = output;
-       
-       data[channel]->input_frames  = nframes * _speed;
-       data[channel]->output_frames = nframes;
-       data[channel]->src_ratio         = 1.0/_speed; 
-
-       if ((error = src_process (state[channel], data[channel]))) {    
-               printf ("\nError : %s\n\n", src_strerror (error));
-               exit (1);
+
+       if (roll) {
+               phase[0] = distance - floor(distance);
        }
-       
-       //printf("INTERPOLATION: channel %d input_frames_used: %d\n", channel, data[channel]->input_frames_used);
-       
-       return data[channel]->input_frames_used;
+
+       i = floor(distance);
+
+       return i;
 }