X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=libs%2Fardour%2Finterpolation.cc;h=3ba9253dee7016045c92108f2c9f1454fe2a331f;hb=9bf40bde3aed831791108bfccc4b1e10b071afdc;hp=bccaa45553484bb0eceb9e82eeb6b578a3b174a4;hpb=0013feddbf260f0b57ee74786c316739806ea47a;p=ardour.git diff --git a/libs/ardour/interpolation.cc b/libs/ardour/interpolation.cc index bccaa45553..3ba9253dee 100644 --- a/libs/ardour/interpolation.cc +++ b/libs/ardour/interpolation.cc @@ -1,5 +1,5 @@ /* - Copyright (C) 2012 Paul Davis + 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 @@ -17,136 +17,211 @@ */ -#include +#include #include +#include + #include "ardour/interpolation.h" +#include "ardour/midi_buffer.h" using namespace ARDOUR; +using std::cerr; +using std::endl; - -framecnt_t -LinearInterpolation::interpolate (int channel, framecnt_t nframes, Sample *input, Sample *output) +CubicInterpolation::CubicInterpolation () + : valid_z_bits (0) { - // index in the input buffers - framecnt_t i = 0; +} - double acceleration = 0; +samplecnt_t +CubicInterpolation::interpolate (int channel, samplecnt_t input_samples, Sample *input, samplecnt_t & output_samples, Sample *output) +{ + assert (input_samples > 0); + assert (output_samples > 0); + assert (input); + assert (output); + assert (phase.size () > channel); + + _speed = fabs (_speed); + + if (invalid (0)) { + + /* z[0] not set. Two possibilities + * + * 1) we have just been constructed or ::reset() + * + * 2) we were only given 1 sample after construction or + * ::reset, and stored it in z[1] + */ + + if (invalid (1)) { + + /* first call after construction or after ::reset */ + + switch (input_samples) { + case 1: + /* store one sample for use next time. We don't + * have enough points to interpolate or even + * compute the first z[0] value, but keep z[1] + * around. + */ + z[1] = input[0]; validate (1); + output_samples = 0; + return 0; + case 2: + /* store two samples for use next time, and + * compute a value for z[0] that will maintain + * the slope of the first actual segment. We + * still don't have enough samples to interpolate. + */ + z[0] = input[0] - (input[1] - input[0]); validate (0); + z[1] = input[0]; validate (1); + z[2] = input[1]; validate (2); + output_samples = 0; + return 0; + default: + /* We have enough samples to interpolate this time, + * but don't have a valid z[0] value because this is the + * first call after construction or ::reset. + * + * First point is based on a requirement to maintain + * the slope of the first actual segment + */ + z[0] = input[0] - (input[1] - input[0]); validate (0); + break; + } + } else { + + /* at least one call since construction or + * after::reset, since we have z[1] set + * + * we can now compute z[0] as required + */ + + z[0] = z[1] - (input[0] - z[1]); validate (0); + + /* we'll check the number of samples we've been given + in the next switch() statement below, and either + just save some more samples or actual interpolate + */ + } - if (_speed != _target_speed) { - acceleration = _target_speed - _speed; + assert (is_valid (0)); } - 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++; + switch (input_samples) { + case 1: + /* one more sample of input. find the right vX to store + it in, and decide if we're ready to interpolate + */ + if (invalid (1)) { + z[1] = input[0]; validate (1); + /* still not ready to interpolate */ + output_samples = 0; + return 0; + } else if (invalid (2)) { + /* still not ready to interpolate */ + z[2] = input[0]; validate (2); + output_samples = 0; + return 0; + } else if (invalid (3)) { + z[3] = input[0]; validate (3); + /* ready to interpolate */ } - - if (input && output) { - // Linearly interpolate into the output buffer - output[outsample] = - input[i] * (1.0f - fractional_phase_part) + - input[i+1] * fractional_phase_part; + break; + case 2: + /* two more samples of input. find the right vX to store + them in, and decide if we're ready to interpolate + */ + if (invalid (1)) { + z[1] = input[0]; validate (1); + z[2] = input[1]; validate (2); + /* still not ready to interpolate */ + output_samples = 0; + return 0; + } else if (invalid (2)) { + z[2] = input[0]; validate (2); + z[3] = input[1]; validate (3); + /* ready to interpolate */ + } else if (invalid (3)) { + z[3] = input[0]; validate (3); + /* ready to interpolate */ } + break; + + default: + /* caller has given us at least enough samples to interpolate a + single value. + */ + z[1] = input[0]; validate (1); + z[2] = input[1]; validate (2); + z[3] = input[2]; validate (3); } - double const distance = phase[channel] + nframes * (_speed + acceleration); - i = floor(distance); - phase[channel] = distance - i; - return i; -} - -framecnt_t -CubicInterpolation::interpolate (int channel, framecnt_t nframes, Sample *input, Sample *output) -{ - // index in the input buffers - framecnt_t i = 0; - - double acceleration; - double distance = 0.0; + /* ready to interpolate using z[0], z[1], z[2] and z[3] */ - if (_speed != _target_speed) { - acceleration = _target_speed - _speed; - } else { - acceleration = 0.0; - } + assert (is_valid (0)); + assert (is_valid (1)); + assert (is_valid (2)); + assert (is_valid (3)); - distance = phase[channel]; + /* we can use up to (input_samples - 2) of the input, so compute the + * maximum number of output samples that represents. + * + * Remember that the expected common case here is to be given + * input_samples that is substantially larger than output_samples, + * thus allowing us to always compute output_samples in one call. + */ - if (nframes < 3) { - /* no interpolation possible */ + const samplecnt_t output_from_input = floor ((input_samples - 2) / _speed); - for (i = 0; i < nframes; ++i) { - output[i] = input[i]; - } + /* limit output to either the caller's requested number or the number + * determined by the input size. + */ - return nframes; - } + const samplecnt_t limit = std::min (output_samples, output_from_input); - /* keep this condition out of the inner loop */ + samplecnt_t outsample = 0; + double distance = phase[channel]; + samplecnt_t used = floor (distance); + samplecnt_t i = 0; - if (input && output) { + while (outsample < limit) { - Sample inm1; + i = floor (distance); - 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]; - } + /* this call may stop the loop from being vectorized */ + float fractional_phase_part = fmod (distance, 1.0); - for (framecnt_t outsample = 0; outsample < nframes; ++outsample) { + /* Cubically interpolate into the output buffer */ + output[outsample++] = z[1] + 0.5f * fractional_phase_part * + (z[2] - z[0] + fractional_phase_part * (4.0f * z[2] + 2.0f * z[0] - 5.0f * z[1] - z[3] + + fractional_phase_part * (3.0f * (z[1] - z[2]) - z[0] + z[3]))); - float f = floor (distance); - float fractional_phase_part = distance - f; + distance += _speed; - /* get the index into the input we should start with */ - - 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]; - } + z[0] = z[1]; + z[1] = input[i]; + z[2] = input[i+1]; + z[3] = input[i+2]; + } - i = floor(distance); - phase[channel] = distance - floor(distance); + output_samples = outsample; + phase[channel] = fmod (distance, 1.0); + return i - used; +} - } 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 +CubicInterpolation::reset () +{ + Interpolation::reset (); + valid_z_bits = 0; +} - return i; +samplecnt_t +CubicInterpolation::distance (samplecnt_t nsamples) +{ + assert (phase.size () > 0); + return floor (floor (phase[0]) + (_speed * nsamples)); }