view decoders/timidity/playmidi.c @ 306:c97be6e1bd27

Added framework for Sound_Seek() support.
author Ryan C. Gordon <icculus@icculus.org>
date Sun, 21 Apr 2002 18:39:47 +0000
parents 2d887640d300
children 498240aa76f1
line wrap: on
line source

/*

    TiMidity -- Experimental MIDI to WAVE converter
    Copyright (C) 1995 Tuukka Toivonen <toivonen@clinet.fi>

    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.

    playmidi.c -- random stuff in need of rearrangement

*/

#if HAVE_CONFIG_H
#  include <config.h>
#endif

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "SDL_sound.h"

#define __SDL_SOUND_INTERNAL__
#include "SDL_sound_internal.h"

#include "timidity.h"
#include "options.h"
#include "instrum.h"
#include "playmidi.h"
#include "output.h"
#include "mix.h"
#include "tables.h"

static void adjust_amplification(MidiSong *song)
{ 
  song->master_volume = (float)(song->amplification) / (float)100.0;
}

static void reset_voices(MidiSong *song)
{
  int i;
  for (i=0; i<MAX_VOICES; i++)
    song->voice[i].status=VOICE_FREE;
}

/* Process the Reset All Controllers event */
static void reset_controllers(MidiSong *song, int c)
{
  song->channel[c].volume=90; /* Some standard says, although the SCC docs say 0. */
  song->channel[c].expression=127; /* SCC-1 does this. */
  song->channel[c].sustain=0;
  song->channel[c].pitchbend=0x2000;
  song->channel[c].pitchfactor=0; /* to be computed */
}

static void reset_midi(MidiSong *song)
{
  int i;
  for (i=0; i<16; i++)
    {
      reset_controllers(song, i);
      /* The rest of these are unaffected by the Reset All Controllers event */
      song->channel[i].program=song->default_program;
      song->channel[i].panning=NO_PANNING;
      song->channel[i].pitchsens=2;
      song->channel[i].bank=0; /* tone bank or drum set */
    }
  reset_voices(song);
}

static void select_sample(MidiSong *song, int v, Instrument *ip)
{
  Sint32 f, cdiff, diff;
  int s,i;
  Sample *sp, *closest;

  s=ip->samples;
  sp=ip->sample;

  if (s==1)
    {
      song->voice[v].sample=sp;
      return;
    }

  f=song->voice[v].orig_frequency;
  for (i=0; i<s; i++)
    {
      if (sp->low_freq <= f && sp->high_freq >= f)
	{
	  song->voice[v].sample=sp;
	  return;
	}
      sp++;
    }

  /* 
     No suitable sample found! We'll select the sample whose root
     frequency is closest to the one we want. (Actually we should
     probably convert the low, high, and root frequencies to MIDI note
     values and compare those.) */

  cdiff=0x7FFFFFFF;
  closest=sp=ip->sample;
  for(i=0; i<s; i++)
    {
      diff=sp->root_freq - f;
      if (diff<0) diff=-diff;
      if (diff<cdiff)
	{
	  cdiff=diff;
	  closest=sp;
	}
      sp++;
    }
  song->voice[v].sample=closest;
  return;
}

static void recompute_freq(MidiSong *song, int v)
{
  int 
    sign=(song->voice[v].sample_increment < 0), /* for bidirectional loops */
    pb=song->channel[song->voice[v].channel].pitchbend;
  double a;
  
  if (!song->voice[v].sample->sample_rate)
    return;

  if (song->voice[v].vibrato_control_ratio)
    {
      /* This instrument has vibrato. Invalidate any precomputed
         sample_increments. */

      int i=VIBRATO_SAMPLE_INCREMENTS;
      while (i--)
	song->voice[v].vibrato_sample_increment[i]=0;
    }

  if (pb==0x2000 || pb<0 || pb>0x3FFF)
    song->voice[v].frequency = song->voice[v].orig_frequency;
  else
    {
      pb-=0x2000;
      if (!(song->channel[song->voice[v].channel].pitchfactor))
	{
	  /* Damn. Somebody bent the pitch. */
	  Sint32 i=pb*song->channel[song->voice[v].channel].pitchsens;
	  if (pb<0)
	    i=-i;
	  song->channel[song->voice[v].channel].pitchfactor=
	    (float)(bend_fine[(i>>5) & 0xFF] * bend_coarse[i>>13]);
	}
      if (pb>0)
	song->voice[v].frequency=
	  (Sint32)(song->channel[song->voice[v].channel].pitchfactor *
		  (double)(song->voice[v].orig_frequency));
      else
	song->voice[v].frequency=
	  (Sint32)((double)(song->voice[v].orig_frequency) /
		  song->channel[song->voice[v].channel].pitchfactor);
    }

  a = FSCALE(((double)(song->voice[v].sample->sample_rate) *
	      (double)(song->voice[v].frequency)) /
	     ((double)(song->voice[v].sample->root_freq) *
	      (double)(song->rate)),
	     FRACTION_BITS);

  if (sign) 
    a = -a; /* need to preserve the loop direction */

  song->voice[v].sample_increment = (Sint32)(a);
}

static void recompute_amp(MidiSong *song, int v)
{
  Sint32 tempamp;

  /* TODO: use fscale */

  tempamp= (song->voice[v].velocity *
	    song->channel[song->voice[v].channel].volume * 
	    song->channel[song->voice[v].channel].expression); /* 21 bits */

  if (!(song->encoding & PE_MONO))
    {
      if (song->voice[v].panning > 60 && song->voice[v].panning < 68)
	{
	  song->voice[v].panned=PANNED_CENTER;

	  song->voice[v].left_amp=
	    FSCALENEG((double)(tempamp) * song->voice[v].sample->volume * song->master_volume,
		      21);
	}
      else if (song->voice[v].panning<5)
	{
	  song->voice[v].panned = PANNED_LEFT;

	  song->voice[v].left_amp=
	    FSCALENEG((double)(tempamp) * song->voice[v].sample->volume * song->master_volume,
		      20);
	}
      else if (song->voice[v].panning>123)
	{
	  song->voice[v].panned = PANNED_RIGHT;

	  song->voice[v].left_amp= /* left_amp will be used */
	    FSCALENEG((double)(tempamp) * song->voice[v].sample->volume * song->master_volume,
		      20);
	}
      else
	{
	  song->voice[v].panned = PANNED_MYSTERY;

	  song->voice[v].left_amp=
	    FSCALENEG((double)(tempamp) * song->voice[v].sample->volume * song->master_volume,
		      27);
	  song->voice[v].right_amp = song->voice[v].left_amp * (song->voice[v].panning);
	  song->voice[v].left_amp *= (float)(127 - song->voice[v].panning);
	}
    }
  else
    {
      song->voice[v].panned = PANNED_CENTER;

      song->voice[v].left_amp=
	FSCALENEG((double)(tempamp) * song->voice[v].sample->volume * song->master_volume,
		  21);
    }
}

static void start_note(MidiSong *song, MidiEvent *e, int i)
{
  Instrument *ip;
  int j;

  if (ISDRUMCHANNEL(song, e->channel))
    {
      if (!(ip=song->drumset[song->channel[e->channel].bank]->instrument[e->a]))
	{
	  if (!(ip=song->drumset[0]->instrument[e->a]))
	    return; /* No instrument? Then we can't play. */
	}
      if (ip->samples != 1)
	{
	  SNDDBG(("Strange: percussion instrument with %d samples!",
		  ip->samples));
	}

      if (ip->sample->note_to_use) /* Do we have a fixed pitch? */
	song->voice[i].orig_frequency = freq_table[(int)(ip->sample->note_to_use)];
      else
	song->voice[i].orig_frequency = freq_table[e->a & 0x7F];
      
      /* drums are supposed to have only one sample */
      song->voice[i].sample = ip->sample;
    }
  else
    {
      if (song->channel[e->channel].program == SPECIAL_PROGRAM)
	ip=song->default_instrument;
      else if (!(ip=song->tonebank[song->channel[e->channel].bank]->
		 instrument[song->channel[e->channel].program]))
	{
	  if (!(ip=song->tonebank[0]->instrument[song->channel[e->channel].program]))
	    return; /* No instrument? Then we can't play. */
	}

      if (ip->sample->note_to_use) /* Fixed-pitch instrument? */
	song->voice[i].orig_frequency = freq_table[(int)(ip->sample->note_to_use)];
      else
	song->voice[i].orig_frequency = freq_table[e->a & 0x7F];
      select_sample(song, i, ip);
    }

  song->voice[i].status = VOICE_ON;
  song->voice[i].channel = e->channel;
  song->voice[i].note = e->a;
  song->voice[i].velocity = e->b;
  song->voice[i].sample_offset = 0;
  song->voice[i].sample_increment = 0; /* make sure it isn't negative */

  song->voice[i].tremolo_phase = 0;
  song->voice[i].tremolo_phase_increment = song->voice[i].sample->tremolo_phase_increment;
  song->voice[i].tremolo_sweep = song->voice[i].sample->tremolo_sweep_increment;
  song->voice[i].tremolo_sweep_position = 0;

  song->voice[i].vibrato_sweep = song->voice[i].sample->vibrato_sweep_increment;
  song->voice[i].vibrato_sweep_position = 0;
  song->voice[i].vibrato_control_ratio = song->voice[i].sample->vibrato_control_ratio;
  song->voice[i].vibrato_control_counter = song->voice[i].vibrato_phase = 0;
  for (j=0; j<VIBRATO_SAMPLE_INCREMENTS; j++)
    song->voice[i].vibrato_sample_increment[j] = 0;

  if (song->channel[e->channel].panning != NO_PANNING)
    song->voice[i].panning = song->channel[e->channel].panning;
  else
    song->voice[i].panning = song->voice[i].sample->panning;

  recompute_freq(song, i);
  recompute_amp(song, i);
  if (song->voice[i].sample->modes & MODES_ENVELOPE)
    {
      /* Ramp up from 0 */
      song->voice[i].envelope_stage = 0;
      song->voice[i].envelope_volume = 0;
      song->voice[i].control_counter = 0;
      recompute_envelope(song, i);
      apply_envelope_to_amp(song, i);
    }
  else
    {
      song->voice[i].envelope_increment = 0;
      apply_envelope_to_amp(song, i);
    }
}

static void kill_note(MidiSong *song, int i)
{
  song->voice[i].status = VOICE_DIE;
}

/* Only one instance of a note can be playing on a single channel. */
static void note_on(MidiSong *song)
{
  int i = song->voices, lowest=-1; 
  Sint32 lv=0x7FFFFFFF, v;
  MidiEvent *e = song->current_event;

  while (i--)
    {
      if (song->voice[i].status == VOICE_FREE)
	lowest=i; /* Can't get a lower volume than silence */
      else if (song->voice[i].channel==e->channel && 
	       (song->voice[i].note==e->a || song->channel[song->voice[i].channel].mono))
	kill_note(song, i);
    }

  if (lowest != -1)
    {
      /* Found a free voice. */
      start_note(song,e,lowest);
      return;
    }
  
  /* Look for the decaying note with the lowest volume */
  i = song->voices;
  while (i--)
    {
      if ((song->voice[i].status != VOICE_ON) &&
	  (song->voice[i].status != VOICE_DIE))
	{
	  v = song->voice[i].left_mix;
	  if ((song->voice[i].panned == PANNED_MYSTERY)
	      && (song->voice[i].right_mix > v))
	    v = song->voice[i].right_mix;
	  if (v<lv)
	    {
	      lv=v;
	      lowest=i;
	    }
	}
    }

  if (lowest != -1)
    {
      /* This can still cause a click, but if we had a free voice to
	 spare for ramping down this note, we wouldn't need to kill it
	 in the first place... Still, this needs to be fixed. Perhaps
	 we could use a reserve of voices to play dying notes only. */
      
      song->cut_notes++;
      song->voice[lowest].status=VOICE_FREE;
      start_note(song,e,lowest);
    }
  else
    song->lost_notes++;
}

static void finish_note(MidiSong *song, int i)
{
  if (song->voice[i].sample->modes & MODES_ENVELOPE)
    {
      /* We need to get the envelope out of Sustain stage */
      song->voice[i].envelope_stage = 3;
      song->voice[i].status = VOICE_OFF;
      recompute_envelope(song, i);
      apply_envelope_to_amp(song, i);
    }
  else
    {
      /* Set status to OFF so resample_voice() will let this voice out
         of its loop, if any. In any case, this voice dies when it
         hits the end of its data (ofs>=data_length). */
      song->voice[i].status = VOICE_OFF;
    }
}

static void note_off(MidiSong *song)
{
  int i = song->voices;
  MidiEvent *e = song->current_event;

  while (i--)
    if (song->voice[i].status == VOICE_ON &&
	song->voice[i].channel == e->channel &&
	song->voice[i].note == e->a)
      {
	if (song->channel[e->channel].sustain)
	  {
	    song->voice[i].status = VOICE_SUSTAINED;
	  }
	else
	  finish_note(song, i);
	return;
      }
}

/* Process the All Notes Off event */
static void all_notes_off(MidiSong *song)
{
  int i = song->voices;
  int c = song->current_event->channel;

  SNDDBG(("All notes off on channel %d", c));
  while (i--)
    if (song->voice[i].status == VOICE_ON &&
	song->voice[i].channel == c)
      {
	if (song->channel[c].sustain) 
	  song->voice[i].status = VOICE_SUSTAINED;
	else
	  finish_note(song, i);
      }
}

/* Process the All Sounds Off event */
static void all_sounds_off(MidiSong *song)
{
  int i = song->voices;
  int c = song->current_event->channel;

  while (i--)
    if (song->voice[i].channel == c && 
	song->voice[i].status != VOICE_FREE &&
	song->voice[i].status != VOICE_DIE)
      {
	kill_note(song, i);
      }
}

static void adjust_pressure(MidiSong *song)
{
  MidiEvent *e = song->current_event;
  int i = song->voices;
  
  while (i--)
    if (song->voice[i].status == VOICE_ON &&
	song->voice[i].channel == e->channel &&
	song->voice[i].note == e->a)
      {
	song->voice[i].velocity = e->b;
	recompute_amp(song, i);
	apply_envelope_to_amp(song, i);
	return;
      }
}

static void drop_sustain(MidiSong *song)
{
  int i = song->voices;
  int c = song->current_event->channel;

  while (i--)
    if (song->voice[i].status == VOICE_SUSTAINED && song->voice[i].channel == c)
      finish_note(song, i);
}

static void adjust_pitchbend(MidiSong *song)
{
  int c = song->current_event->channel;
  int i = song->voices;
  
  while (i--)
    if (song->voice[i].status != VOICE_FREE && song->voice[i].channel == c)
      {
	recompute_freq(song, i);
      }
}

static void adjust_volume(MidiSong *song)
{
  int c = song->current_event->channel;
  int i = song->voices;

  while (i--)
    if (song->voice[i].channel == c &&
	(song->voice[i].status==VOICE_ON || song->voice[i].status==VOICE_SUSTAINED))
      {
	recompute_amp(song, i);
	apply_envelope_to_amp(song, i);
      }
}

static void seek_forward(MidiSong *song, Sint32 until_time)
{
  reset_voices(song);
  while (song->current_event->time < until_time)
    {
      switch(song->current_event->type)
	{
	  /* All notes stay off. Just handle the parameter changes. */

	case ME_PITCH_SENS:
	  song->channel[song->current_event->channel].pitchsens =
	    song->current_event->a;
	  song->channel[song->current_event->channel].pitchfactor = 0;
	  break;
	  
	case ME_PITCHWHEEL:
	  song->channel[song->current_event->channel].pitchbend =
	    song->current_event->a + song->current_event->b * 128;
	  song->channel[song->current_event->channel].pitchfactor = 0;
	  break;
	  
	case ME_MAINVOLUME:
	  song->channel[song->current_event->channel].volume =
	    song->current_event->a;
	  break;
	  
	case ME_PAN:
	  song->channel[song->current_event->channel].panning =
	    song->current_event->a;
	  break;
	      
	case ME_EXPRESSION:
	  song->channel[song->current_event->channel].expression =
	    song->current_event->a;
	  break;
	  
	case ME_PROGRAM:
	  if (ISDRUMCHANNEL(song, song->current_event->channel))
	    /* Change drum set */
	    song->channel[song->current_event->channel].bank =
	      song->current_event->a;
	  else
	    song->channel[song->current_event->channel].program =
	      song->current_event->a;
	  break;

	case ME_SUSTAIN:
	  song->channel[song->current_event->channel].sustain =
	    song->current_event->a;
	  break;

	case ME_RESET_CONTROLLERS:
	  reset_controllers(song, song->current_event->channel);
	  break;
	      
	case ME_TONE_BANK:
	  song->channel[song->current_event->channel].bank =
	    song->current_event->a;
	  break;
	  
	case ME_EOT:
	  song->current_sample = song->current_event->time;
	  return;
	}
      song->current_event++;
    }
  /*song->current_sample=song->current_event->time;*/
  if (song->current_event != song->events)
    song->current_event--;
  song->current_sample=until_time;
}

static void skip_to(MidiSong *song, Sint32 until_time)
{
  if (song->current_sample > until_time)
    song->current_sample = 0;

  reset_midi(song);
  song->buffered_count = 0;
  song->buffer_pointer = song->common_buffer;
  song->current_event = song->events;
  
  if (until_time)
    seek_forward(song, until_time);
}

static void do_compute_data(MidiSong *song, Sint32 count)
{
  int i;
  memset(song->buffer_pointer, 0, 
	 (song->encoding & PE_MONO) ? (count * 4) : (count * 8));
  for (i = 0; i < song->voices; i++)
    {
      if(song->voice[i].status != VOICE_FREE)
	mix_voice(song, song->buffer_pointer, i, count);
    }
  song->current_sample += count;
}

/* count=0 means flush remaining buffered data to output device, then
   flush the device itself */
static void compute_data(MidiSong *song, void *stream, Sint32 count)
{
  int channels;

  if ( song->encoding & PE_MONO )
    channels = 1;
  else
    channels = 2;

  if (!count)
    {
      if (song->buffered_count)
          song->write(stream, song->common_buffer, channels * song->buffered_count);
      song->buffer_pointer = song->common_buffer;
      song->buffered_count = 0;
      return;
    }

  while ((count + song->buffered_count) >= song->buffer_size)
    {
      do_compute_data(song, song->buffer_size - song->buffered_count);
      count -= song->buffer_size - song->buffered_count;
      song->write(stream, song->common_buffer, channels * song->buffer_size);
      song->buffer_pointer = song->common_buffer;
      song->buffered_count = 0;
    }
  if (count>0)
    {
      do_compute_data(song, count);
      song->buffered_count += count;
      song->buffer_pointer += (song->encoding & PE_MONO) ? count : count*2;
    }
}

void Timidity_Start(MidiSong *song)
{
  song->playing = 1;
  adjust_amplification(song);
  skip_to(song, 0);
}

int Timidity_PlaySome(MidiSong *song, void *stream, Sint32 len)
{
  Sint32 start_sample, end_sample, samples;
  int bytes_per_sample;

  if (!song->playing)
    return 0;
  
  bytes_per_sample =
        ((song->encoding & PE_MONO) ? 1 : 2)
      * ((song->encoding & PE_16BIT) ? 2 : 1);
  samples = len / bytes_per_sample;
  
  start_sample = song->current_sample;
  end_sample = song->current_sample+samples;
  while ( song->current_sample < end_sample ) {
    /* Handle all events that should happen at this time */
    while (song->current_event->time <= song->current_sample) {
      switch(song->current_event->type) {

        /* Effects affecting a single note */

        case ME_NOTEON:
          if (!(song->current_event->b)) /* Velocity 0? */
            note_off(song);
          else
            note_on(song);
          break;
  
        case ME_NOTEOFF:
          note_off(song);
          break;
  
        case ME_KEYPRESSURE:
          adjust_pressure(song);
          break;
  
          /* Effects affecting a single channel */
  
        case ME_PITCH_SENS:
          song->channel[song->current_event->channel].pitchsens =
	    song->current_event->a;
          song->channel[song->current_event->channel].pitchfactor = 0;
          break;
          
        case ME_PITCHWHEEL:
          song->channel[song->current_event->channel].pitchbend =
            song->current_event->a + song->current_event->b * 128;
          song->channel[song->current_event->channel].pitchfactor = 0;
          /* Adjust pitch for notes already playing */
          adjust_pitchbend(song);
          break;
          
        case ME_MAINVOLUME:
          song->channel[song->current_event->channel].volume =
	    song->current_event->a;
          adjust_volume(song);
          break;
          
        case ME_PAN:
          song->channel[song->current_event->channel].panning =
	    song->current_event->a;
          break;
          
        case ME_EXPRESSION:
          song->channel[song->current_event->channel].expression =
	    song->current_event->a;
          adjust_volume(song);
          break;
  
        case ME_PROGRAM:
          if (ISDRUMCHANNEL(song, song->current_event->channel)) {
            /* Change drum set */
            song->channel[song->current_event->channel].bank =
	      song->current_event->a;
          }
          else
            song->channel[song->current_event->channel].program =
	      song->current_event->a;
          break;
  
        case ME_SUSTAIN:
          song->channel[song->current_event->channel].sustain =
	    song->current_event->a;
          if (!song->current_event->a)
            drop_sustain(song);
          break;
          
        case ME_RESET_CONTROLLERS:
          reset_controllers(song, song->current_event->channel);
          break;
  
        case ME_ALL_NOTES_OFF:
          all_notes_off(song);
          break;
          
        case ME_ALL_SOUNDS_OFF:
          all_sounds_off(song);
          break;
          
        case ME_TONE_BANK:
          song->channel[song->current_event->channel].bank =
	    song->current_event->a;
          break;
  
        case ME_EOT:
          /* Give the last notes a couple of seconds to decay  */
          SNDDBG(("Playing time: ~%d seconds\n",
		  song->current_sample/song->rate+2));
          SNDDBG(("Notes cut: %d\n", song->cut_notes));
          SNDDBG(("Notes lost totally: %d\n", song->lost_notes));
	  song->playing = 0;
          return (song->current_sample - start_sample) * bytes_per_sample;
        }
      song->current_event++;
    }
    if (song->current_event->time > end_sample)
      compute_data(song, stream, end_sample-song->current_sample);
    else
      compute_data(song, stream, song->current_event->time-song->current_sample);
  }
  return samples * bytes_per_sample;
}

void Timidity_SetVolume(MidiSong *song, int volume)
{
  int i;
  if (volume > MAX_AMPLIFICATION)
    song->amplification = MAX_AMPLIFICATION;
  else
  if (volume < 0)
    song->amplification = 0;
  else
    song->amplification = volume;
  adjust_amplification(song);
  for (i = 0; i < song->voices; i++)
    if (song->voice[i].status != VOICE_FREE)
      {
        recompute_amp(song, i);
        apply_envelope_to_amp(song, i);
      }
}