Paul Davis
d116af22db
git-svn-id: svn://localhost/ardour2/branches/3.0@8590 d708f5d6-7413-0410-9779-e7cbd77b26cf
503 lines
14 KiB
C++
503 lines
14 KiB
C++
/*
|
|
Copyright (C) 2004-2011 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 <inttypes.h>
|
|
|
|
#include <cmath>
|
|
#include <cerrno>
|
|
#include <fstream>
|
|
#include <cstdlib>
|
|
#include <string>
|
|
#include <cstdio>
|
|
#include <locale.h>
|
|
#include <unistd.h>
|
|
#include <float.h>
|
|
#include <iomanip>
|
|
|
|
#include <glibmm.h>
|
|
|
|
#include "pbd/cartesian.h"
|
|
#include "pbd/convert.h"
|
|
#include "pbd/error.h"
|
|
#include "pbd/failed_constructor.h"
|
|
#include "pbd/xml++.h"
|
|
#include "pbd/enumwriter.h"
|
|
|
|
#include "evoral/Curve.hpp"
|
|
|
|
#include "ardour/audio_buffer.h"
|
|
#include "ardour/audio_buffer.h"
|
|
#include "ardour/buffer_set.h"
|
|
#include "ardour/pan_controllable.h"
|
|
#include "ardour/pannable.h"
|
|
#include "ardour/runtime_functions.h"
|
|
#include "ardour/session.h"
|
|
#include "ardour/utils.h"
|
|
|
|
#include "panner_2in2out.h"
|
|
|
|
#include "i18n.h"
|
|
|
|
#include "pbd/mathfix.h"
|
|
|
|
using namespace std;
|
|
using namespace ARDOUR;
|
|
using namespace PBD;
|
|
|
|
static PanPluginDescriptor _descriptor = {
|
|
"Equal Power Stereo",
|
|
2, 2,
|
|
Panner2in2out::factory
|
|
};
|
|
|
|
extern "C" { PanPluginDescriptor* panner_descriptor () { return &_descriptor; } }
|
|
|
|
Panner2in2out::Panner2in2out (boost::shared_ptr<Pannable> p)
|
|
: Panner (p)
|
|
{
|
|
if (!_pannable->has_state()) {
|
|
_pannable->pan_azimuth_control->set_value (0.5);
|
|
_pannable->pan_width_control->set_value (1.0);
|
|
}
|
|
|
|
update ();
|
|
|
|
/* LEFT SIGNAL */
|
|
left_interp[0] = left[0] = desired_left[0];
|
|
right_interp[0] = right[0] = desired_right[0];
|
|
|
|
/* RIGHT SIGNAL */
|
|
left_interp[1] = left[1] = desired_left[1];
|
|
right_interp[1] = right[1] = desired_right[1];
|
|
|
|
_pannable->pan_azimuth_control->Changed.connect_same_thread (*this, boost::bind (&Panner2in2out::update, this));
|
|
_pannable->pan_width_control->Changed.connect_same_thread (*this, boost::bind (&Panner2in2out::update, this));
|
|
}
|
|
|
|
Panner2in2out::~Panner2in2out ()
|
|
{
|
|
}
|
|
|
|
double
|
|
Panner2in2out::position () const
|
|
{
|
|
return _pannable->pan_azimuth_control->get_value();
|
|
}
|
|
|
|
double
|
|
Panner2in2out::width () const
|
|
{
|
|
return _pannable->pan_width_control->get_value();
|
|
}
|
|
|
|
void
|
|
Panner2in2out::set_position (double p)
|
|
{
|
|
if (clamp_position (p)) {
|
|
_pannable->pan_azimuth_control->set_value (p);
|
|
}
|
|
}
|
|
|
|
void
|
|
Panner2in2out::set_width (double p)
|
|
{
|
|
if (clamp_width (p)) {
|
|
_pannable->pan_width_control->set_value (p);
|
|
}
|
|
}
|
|
|
|
void
|
|
Panner2in2out::update ()
|
|
{
|
|
/* it would be very nice to split this out into a virtual function
|
|
that can be accessed from BaseStereoPanner and used in do_distribute_automated().
|
|
|
|
but the place where its used in do_distribute_automated() is a tight inner loop,
|
|
and making "nframes" virtual function calls to compute values is an absurd
|
|
overhead.
|
|
*/
|
|
|
|
/* x == 0 => hard left = 180.0 degrees
|
|
x == 1 => hard right = 0.0 degrees
|
|
*/
|
|
|
|
float pos[2];
|
|
const double width = _pannable->pan_width_control->get_value();
|
|
const double direction_as_lr_fract = _pannable->pan_azimuth_control->get_value();
|
|
|
|
if (width < 0.0) {
|
|
pos[0] = direction_as_lr_fract + (width/2.0); // left signal lr_fract
|
|
pos[1] = direction_as_lr_fract - (width/2.0); // right signal lr_fract
|
|
} else {
|
|
pos[1] = direction_as_lr_fract + (width/2.0); // right signal lr_fract
|
|
pos[0] = direction_as_lr_fract - (width/2.0); // left signal lr_fract
|
|
}
|
|
|
|
/* compute target gain coefficients for both input signals */
|
|
|
|
float const pan_law_attenuation = -3.0f;
|
|
float const scale = 2.0f - 4.0f * powf (10.0f,pan_law_attenuation/20.0f);
|
|
float panR;
|
|
float panL;
|
|
|
|
/* left signal */
|
|
|
|
panR = pos[0];
|
|
panL = 1 - panR;
|
|
desired_left[0] = panL * (scale * panL + 1.0f - scale);
|
|
desired_right[0] = panR * (scale * panR + 1.0f - scale);
|
|
|
|
/* right signal */
|
|
|
|
panR = pos[1];
|
|
panL = 1 - panR;
|
|
desired_left[1] = panL * (scale * panL + 1.0f - scale);
|
|
desired_right[1] = panR * (scale * panR + 1.0f - scale);
|
|
}
|
|
|
|
bool
|
|
Panner2in2out::clamp_position (double& p)
|
|
{
|
|
double w = _pannable->pan_width_control->get_value();
|
|
return clamp_stereo_pan (p, w);
|
|
}
|
|
|
|
bool
|
|
Panner2in2out::clamp_width (double& w)
|
|
{
|
|
double p = _pannable->pan_azimuth_control->get_value();
|
|
return clamp_stereo_pan (p, w);
|
|
}
|
|
|
|
bool
|
|
Panner2in2out::clamp_stereo_pan (double& direction_as_lr_fract, double& width)
|
|
{
|
|
double r_pos;
|
|
double l_pos;
|
|
|
|
width = max (min (width, 1.0), -1.0);
|
|
direction_as_lr_fract = max (min (direction_as_lr_fract, 1.0), 0.0);
|
|
|
|
r_pos = direction_as_lr_fract + (width/2.0);
|
|
l_pos = direction_as_lr_fract - (width/2.0);
|
|
|
|
if (width < 0.0) {
|
|
swap (r_pos, l_pos);
|
|
}
|
|
|
|
/* if the new left position is less than or equal to zero (hard left) and the left panner
|
|
is already there, we're not moving the left signal.
|
|
*/
|
|
|
|
if (l_pos < 0.0) {
|
|
return false;
|
|
}
|
|
|
|
/* if the new right position is less than or equal to 1.0 (hard right) and the right panner
|
|
is already there, we're not moving the right signal.
|
|
*/
|
|
|
|
if (r_pos > 1.0) {
|
|
return false;
|
|
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
void
|
|
Panner2in2out::distribute_one (AudioBuffer& srcbuf, BufferSet& obufs, gain_t gain_coeff, pframes_t nframes, uint32_t which)
|
|
{
|
|
assert (obufs.count().n_audio() == 2);
|
|
|
|
pan_t delta;
|
|
Sample* dst;
|
|
pan_t pan;
|
|
|
|
Sample* const src = srcbuf.data();
|
|
|
|
/* LEFT OUTPUT */
|
|
|
|
dst = obufs.get_audio(0).data();
|
|
|
|
if (fabsf ((delta = (left[which] - desired_left[which]))) > 0.002) { // about 1 degree of arc
|
|
|
|
/* we've moving the pan by an appreciable amount, so we must
|
|
interpolate over 64 frames or nframes, whichever is smaller */
|
|
|
|
pframes_t const limit = min ((pframes_t) 64, nframes);
|
|
pframes_t n;
|
|
|
|
delta = -(delta / (float) (limit));
|
|
|
|
for (n = 0; n < limit; n++) {
|
|
left_interp[which] = left_interp[which] + delta;
|
|
left[which] = left_interp[which] + 0.9 * (left[which] - left_interp[which]);
|
|
dst[n] += src[n] * left[which] * gain_coeff;
|
|
}
|
|
|
|
/* then pan the rest of the buffer; no need for interpolation for this bit */
|
|
|
|
pan = left[which] * gain_coeff;
|
|
|
|
mix_buffers_with_gain (dst+n,src+n,nframes-n,pan);
|
|
|
|
} else {
|
|
|
|
left[which] = desired_left[which];
|
|
left_interp[which] = left[which];
|
|
|
|
if ((pan = (left[which] * gain_coeff)) != 1.0f) {
|
|
|
|
if (pan != 0.0f) {
|
|
|
|
/* pan is 1 but also not 0, so we must do it "properly" */
|
|
|
|
mix_buffers_with_gain(dst,src,nframes,pan);
|
|
|
|
/* mark that we wrote into the buffer */
|
|
|
|
// obufs[0] = 0;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* pan is 1 so we can just copy the input samples straight in */
|
|
|
|
mix_buffers_no_gain(dst,src,nframes);
|
|
|
|
/* XXX it would be nice to mark that we wrote into the buffer */
|
|
}
|
|
}
|
|
|
|
/* RIGHT OUTPUT */
|
|
|
|
dst = obufs.get_audio(1).data();
|
|
|
|
if (fabsf ((delta = (right[which] - desired_right[which]))) > 0.002) { // about 1 degree of arc
|
|
|
|
/* we're moving the pan by an appreciable amount, so we must
|
|
interpolate over 64 frames or nframes, whichever is smaller */
|
|
|
|
pframes_t const limit = min ((pframes_t) 64, nframes);
|
|
pframes_t n;
|
|
|
|
delta = -(delta / (float) (limit));
|
|
|
|
for (n = 0; n < limit; n++) {
|
|
right_interp[which] = right_interp[which] + delta;
|
|
right[which] = right_interp[which] + 0.9 * (right[which] - right_interp[which]);
|
|
dst[n] += src[n] * right[which] * gain_coeff;
|
|
}
|
|
|
|
/* then pan the rest of the buffer, no need for interpolation for this bit */
|
|
|
|
pan = right[which] * gain_coeff;
|
|
|
|
mix_buffers_with_gain(dst+n,src+n,nframes-n,pan);
|
|
|
|
/* XXX it would be nice to mark the buffer as written to */
|
|
|
|
} else {
|
|
|
|
right[which] = desired_right[which];
|
|
right_interp[which] = right[which];
|
|
|
|
if ((pan = (right[which] * gain_coeff)) != 1.0f) {
|
|
|
|
if (pan != 0.0f) {
|
|
|
|
/* pan is not 1 but also not 0, so we must do it "properly" */
|
|
|
|
mix_buffers_with_gain(dst,src,nframes,pan);
|
|
|
|
/* XXX it would be nice to mark the buffer as written to */
|
|
}
|
|
|
|
} else {
|
|
|
|
/* pan is 1 so we can just copy the input samples straight in */
|
|
|
|
mix_buffers_no_gain(dst,src,nframes);
|
|
|
|
/* XXX it would be nice to mark the buffer as written to */
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
Panner2in2out::distribute_one_automated (AudioBuffer& srcbuf, BufferSet& obufs,
|
|
framepos_t start, framepos_t end, pframes_t nframes,
|
|
pan_t** buffers, uint32_t which)
|
|
{
|
|
assert (obufs.count().n_audio() == 2);
|
|
|
|
Sample* dst;
|
|
pan_t* pbuf;
|
|
Sample* const src = srcbuf.data();
|
|
pan_t* const position = buffers[0];
|
|
pan_t* const width = buffers[1];
|
|
|
|
/* fetch positional data */
|
|
|
|
if (!_pannable->pan_azimuth_control->list()->curve().rt_safe_get_vector (start, end, position, nframes)) {
|
|
/* fallback */
|
|
distribute_one (srcbuf, obufs, 1.0, nframes, which);
|
|
return;
|
|
}
|
|
|
|
if (!_pannable->pan_width_control->list()->curve().rt_safe_get_vector (start, end, width, nframes)) {
|
|
/* fallback */
|
|
distribute_one (srcbuf, obufs, 1.0, nframes, which);
|
|
return;
|
|
}
|
|
|
|
/* apply pan law to convert positional data into pan coefficients for
|
|
each buffer (output)
|
|
*/
|
|
|
|
const float pan_law_attenuation = -3.0f;
|
|
const float scale = 2.0f - 4.0f * powf (10.0f,pan_law_attenuation/20.0f);
|
|
|
|
for (pframes_t n = 0; n < nframes; ++n) {
|
|
|
|
float panR;
|
|
|
|
if (which == 0) {
|
|
// panning left signal
|
|
panR = position[n] - (width[n]/2.0f); // center - width/2
|
|
} else {
|
|
// panning right signal
|
|
panR = position[n] + (width[n]/2.0f); // center - width/2
|
|
}
|
|
|
|
const float panL = 1 - panR;
|
|
|
|
/* note that are overwriting buffers, but its OK
|
|
because we're finished with their old contents
|
|
(position/width automation data) and are
|
|
replacing it with panning/gain coefficients
|
|
that we need to actually process the data.
|
|
*/
|
|
|
|
buffers[0][n] = panL * (scale * panL + 1.0f - scale);
|
|
buffers[1][n] = panR * (scale * panR + 1.0f - scale);
|
|
}
|
|
|
|
/* LEFT OUTPUT */
|
|
|
|
dst = obufs.get_audio(0).data();
|
|
pbuf = buffers[0];
|
|
|
|
for (pframes_t n = 0; n < nframes; ++n) {
|
|
dst[n] += src[n] * pbuf[n];
|
|
}
|
|
|
|
/* XXX it would be nice to mark the buffer as written to */
|
|
|
|
/* RIGHT OUTPUT */
|
|
|
|
dst = obufs.get_audio(1).data();
|
|
pbuf = buffers[1];
|
|
|
|
for (pframes_t n = 0; n < nframes; ++n) {
|
|
dst[n] += src[n] * pbuf[n];
|
|
}
|
|
|
|
/* XXX it would be nice to mark the buffer as written to */
|
|
}
|
|
|
|
Panner*
|
|
Panner2in2out::factory (boost::shared_ptr<Pannable> p, Speakers& /* ignored */)
|
|
{
|
|
return new Panner2in2out (p);
|
|
}
|
|
|
|
XMLNode&
|
|
Panner2in2out::get_state (void)
|
|
{
|
|
return state (true);
|
|
}
|
|
|
|
XMLNode&
|
|
Panner2in2out::state (bool /*full_state*/)
|
|
{
|
|
XMLNode& root (Panner::get_state ());
|
|
root.add_property (X_("type"), _descriptor.name);
|
|
return root;
|
|
}
|
|
|
|
int
|
|
Panner2in2out::set_state (const XMLNode& node, int version)
|
|
{
|
|
LocaleGuard lg (X_("POSIX"));
|
|
Panner::set_state (node, version);
|
|
return 0;
|
|
}
|
|
|
|
std::set<Evoral::Parameter>
|
|
Panner2in2out::what_can_be_automated() const
|
|
{
|
|
set<Evoral::Parameter> s;
|
|
s.insert (Evoral::Parameter (PanAzimuthAutomation));
|
|
s.insert (Evoral::Parameter (PanWidthAutomation));
|
|
return s;
|
|
}
|
|
|
|
string
|
|
Panner2in2out::describe_parameter (Evoral::Parameter p)
|
|
{
|
|
switch (p.type()) {
|
|
case PanAzimuthAutomation:
|
|
return _("L/R");
|
|
case PanWidthAutomation:
|
|
return _("Width");
|
|
default:
|
|
return _pannable->describe_parameter (p);
|
|
}
|
|
}
|
|
|
|
string
|
|
Panner2in2out::value_as_string (boost::shared_ptr<AutomationControl> ac) const
|
|
{
|
|
/* DO NOT USE LocaleGuard HERE */
|
|
double val = ac->get_value();
|
|
|
|
switch (ac->parameter().type()) {
|
|
case PanAzimuthAutomation:
|
|
/* We show the position of the center of the image relative to the left & right.
|
|
This is expressed as a pair of percentage values that ranges from (100,0)
|
|
(hard left) through (50,50) (hard center) to (0,100) (hard right).
|
|
|
|
This is pretty wierd, but its the way audio engineers expect it. Just remember that
|
|
the center of the USA isn't Kansas, its (50LA, 50NY) and it will all make sense.
|
|
*/
|
|
|
|
return string_compose (_("L:%1 R:%2"), (int) rint (100.0 * (1.0 - val)),
|
|
(int) rint (100.0 * val));
|
|
|
|
case PanWidthAutomation:
|
|
return string_compose (_("Width: %1%%"), (int) floor (100.0 * val));
|
|
|
|
default:
|
|
return _pannable->value_as_string (ac);
|
|
}
|
|
}
|