|
@@ -86,22 +86,22 @@ extract_to_matrix(FLOATNAME(LMatrix4) &m) const {
|
|
|
// (from Real-time Rendering, p.49)
|
|
// (from Real-time Rendering, p.49)
|
|
|
////////////////////////////////////////////////////////////////////
|
|
////////////////////////////////////////////////////////////////////
|
|
|
void FLOATNAME(LQuaternion)::
|
|
void FLOATNAME(LQuaternion)::
|
|
|
-set_hpr(const FLOATNAME(LVecBase3) &hpr) {
|
|
|
|
|
|
|
+set_hpr(const FLOATNAME(LVecBase3) &hpr, CoordinateSystem cs) {
|
|
|
FLOATNAME(LQuaternion) quat_h, quat_p, quat_r;
|
|
FLOATNAME(LQuaternion) quat_h, quat_p, quat_r;
|
|
|
|
|
|
|
|
FLOATNAME(LVector3) v;
|
|
FLOATNAME(LVector3) v;
|
|
|
FLOATTYPE a, s, c;
|
|
FLOATTYPE a, s, c;
|
|
|
|
|
|
|
|
- v = FLOATNAME(LVector3)::up();
|
|
|
|
|
|
|
+ v = FLOATNAME(LVector3)::up(cs);
|
|
|
a = deg_2_rad(hpr[0] * 0.5f);
|
|
a = deg_2_rad(hpr[0] * 0.5f);
|
|
|
csincos(a, &s, &c);
|
|
csincos(a, &s, &c);
|
|
|
quat_h.set(c, v[0] * s, v[1] * s, v[2] * s);
|
|
quat_h.set(c, v[0] * s, v[1] * s, v[2] * s);
|
|
|
- v = FLOATNAME(LVector3)::right();
|
|
|
|
|
|
|
+ v = FLOATNAME(LVector3)::right(cs);
|
|
|
a = deg_2_rad(hpr[1] * 0.5f);
|
|
a = deg_2_rad(hpr[1] * 0.5f);
|
|
|
csincos(a, &s, &c);
|
|
csincos(a, &s, &c);
|
|
|
s = csin(a);
|
|
s = csin(a);
|
|
|
quat_p.set(c, v[0] * s, v[1] * s, v[2] * s);
|
|
quat_p.set(c, v[0] * s, v[1] * s, v[2] * s);
|
|
|
- v = FLOATNAME(LVector3)::forward();
|
|
|
|
|
|
|
+ v = FLOATNAME(LVector3)::forward(cs);
|
|
|
a = deg_2_rad(hpr[2] * 0.5f);
|
|
a = deg_2_rad(hpr[2] * 0.5f);
|
|
|
csincos(a, &s, &c);
|
|
csincos(a, &s, &c);
|
|
|
quat_r.set(c, v[0] * s, v[1] * s, v[2] * s);
|
|
quat_r.set(c, v[0] * s, v[1] * s, v[2] * s);
|
|
@@ -136,7 +136,7 @@ set_hpr(const FLOATNAME(LVecBase3) &hpr) {
|
|
|
// quaternion.
|
|
// quaternion.
|
|
|
////////////////////////////////////////////////////////////////////
|
|
////////////////////////////////////////////////////////////////////
|
|
|
FLOATNAME(LVecBase3) FLOATNAME(LQuaternion)::
|
|
FLOATNAME(LVecBase3) FLOATNAME(LQuaternion)::
|
|
|
-get_hpr() const {
|
|
|
|
|
|
|
+get_hpr(CoordinateSystem cs) const {
|
|
|
if (!temp_hpr_fix) {
|
|
if (!temp_hpr_fix) {
|
|
|
// With the old, broken hpr code in place, just go through the
|
|
// With the old, broken hpr code in place, just go through the
|
|
|
// existing matrix decomposition code. Not particularly speedy,
|
|
// existing matrix decomposition code. Not particularly speedy,
|
|
@@ -146,49 +146,67 @@ get_hpr() const {
|
|
|
FLOATNAME(LMatrix3) mat;
|
|
FLOATNAME(LMatrix3) mat;
|
|
|
extract_to_matrix(mat);
|
|
extract_to_matrix(mat);
|
|
|
FLOATNAME(LVecBase3) scale, hpr;
|
|
FLOATNAME(LVecBase3) scale, hpr;
|
|
|
- decompose_matrix(mat, scale, hpr);
|
|
|
|
|
|
|
+ decompose_matrix(mat, scale, hpr, cs);
|
|
|
return hpr;
|
|
return hpr;
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
|
|
+ if (cs == CS_default) {
|
|
|
|
|
+ cs = get_default_coordinate_system();
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
FLOATNAME(LVecBase3) hpr;
|
|
FLOATNAME(LVecBase3) hpr;
|
|
|
- FLOATTYPE N = (_v.data[0] * _v.data[0]) + (_v.data[1] * _v.data[1]) + (_v.data[2] * _v.data[2]) + (_v.data[3] * _v.data[3]);
|
|
|
|
|
- FLOATTYPE s = (N == 0.0f) ? 0.0f : (2.0f / N);
|
|
|
|
|
- FLOATTYPE xs, ys, zs, wx, wy, wz, xx, xy, xz, yy, yz, zz, c1, c2, c3, c4;
|
|
|
|
|
- FLOATTYPE cr, sr, cp, sp, ch, sh;
|
|
|
|
|
|
|
|
|
|
- xs = _v.data[1] * s; ys = _v.data[2] * s; zs = _v.data[3] * s;
|
|
|
|
|
- wx = _v.data[0] * xs; wy = _v.data[0] * ys; wz = _v.data[0] * zs;
|
|
|
|
|
- xx = _v.data[1] * xs; xy = _v.data[1] * ys; xz = _v.data[1] * zs;
|
|
|
|
|
- yy = _v.data[2] * ys; yz = _v.data[2] * zs; zz = _v.data[3] * zs;
|
|
|
|
|
- c1 = xz - wy;
|
|
|
|
|
- c2 = 1.0f - (xx + yy);
|
|
|
|
|
- c3 = 1.0f - (yy + zz);
|
|
|
|
|
- c4 = xy + wz;
|
|
|
|
|
-
|
|
|
|
|
- if (c1 == 0.0f) { // (roll = 0 or 180) or (pitch = +/- 90)
|
|
|
|
|
- if (c2 >= 0.0f) {
|
|
|
|
|
- hpr[2] = 0.0f;
|
|
|
|
|
- ch = c3;
|
|
|
|
|
- sh = c4;
|
|
|
|
|
- cp = c2;
|
|
|
|
|
|
|
+ if (cs == CS_zup_right) {
|
|
|
|
|
+ FLOATTYPE N = (_v.data[0] * _v.data[0]) + (_v.data[1] * _v.data[1]) + (_v.data[2] * _v.data[2]) + (_v.data[3] * _v.data[3]);
|
|
|
|
|
+ FLOATTYPE s = (N == 0.0f) ? 0.0f : (2.0f / N);
|
|
|
|
|
+ FLOATTYPE xs, ys, zs, wx, wy, wz, xx, xy, xz, yy, yz, zz, c1, c2, c3, c4;
|
|
|
|
|
+ FLOATTYPE cr, sr, cp, sp, ch, sh;
|
|
|
|
|
+
|
|
|
|
|
+ xs = _v.data[1] * s; ys = _v.data[2] * s; zs = _v.data[3] * s;
|
|
|
|
|
+ wx = _v.data[0] * xs; wy = _v.data[0] * ys; wz = _v.data[0] * zs;
|
|
|
|
|
+ xx = _v.data[1] * xs; xy = _v.data[1] * ys; xz = _v.data[1] * zs;
|
|
|
|
|
+ yy = _v.data[2] * ys; yz = _v.data[2] * zs; zz = _v.data[3] * zs;
|
|
|
|
|
+ c1 = xz - wy;
|
|
|
|
|
+ c2 = 1.0f - (xx + yy);
|
|
|
|
|
+ c3 = 1.0f - (yy + zz);
|
|
|
|
|
+ c4 = xy + wz;
|
|
|
|
|
+
|
|
|
|
|
+ if (c1 == 0.0f) { // (roll = 0 or 180) or (pitch = +/- 90)
|
|
|
|
|
+ if (c2 >= 0.0f) {
|
|
|
|
|
+ hpr[2] = 0.0f;
|
|
|
|
|
+ ch = c3;
|
|
|
|
|
+ sh = c4;
|
|
|
|
|
+ cp = c2;
|
|
|
|
|
+ } else {
|
|
|
|
|
+ hpr[2] = 180.0f;
|
|
|
|
|
+ ch = -c3;
|
|
|
|
|
+ sh = -c4;
|
|
|
|
|
+ cp = -c2;
|
|
|
|
|
+ }
|
|
|
} else {
|
|
} else {
|
|
|
- hpr[2] = 180.0f;
|
|
|
|
|
- ch = -c3;
|
|
|
|
|
- sh = -c4;
|
|
|
|
|
- cp = -c2;
|
|
|
|
|
|
|
+ // this should work all the time, but the above saves some trig operations
|
|
|
|
|
+ FLOATTYPE roll = catan2(-c1, c2);
|
|
|
|
|
+ csincos(roll, &sr, &cr);
|
|
|
|
|
+ hpr[2] = rad_2_deg(roll);
|
|
|
|
|
+ ch = (cr * c3) + (sr * (xz + wy));
|
|
|
|
|
+ sh = (cr * c4) + (sr * (yz - wx));
|
|
|
|
|
+ cp = (cr * c2) - (sr * c1);
|
|
|
}
|
|
}
|
|
|
|
|
+ sp = yz + wx;
|
|
|
|
|
+ hpr[0] = rad_2_deg(catan2(sh, ch));
|
|
|
|
|
+ hpr[1] = rad_2_deg(catan2(sp, cp));
|
|
|
|
|
+
|
|
|
} else {
|
|
} else {
|
|
|
- // this should work all the time, but the above saves some trig operations
|
|
|
|
|
- FLOATTYPE roll = catan2(-c1, c2);
|
|
|
|
|
- csincos(roll, &sr, &cr);
|
|
|
|
|
- hpr[2] = rad_2_deg(roll);
|
|
|
|
|
- ch = (cr * c3) + (sr * (xz + wy));
|
|
|
|
|
- sh = (cr * c4) + (sr * (yz - wx));
|
|
|
|
|
- cp = (cr * c2) - (sr * c1);
|
|
|
|
|
|
|
+ // The code above implements quat-to-hpr for CS_zup_right only.
|
|
|
|
|
+ // For other coordinate systems, someone is welcome to extend the
|
|
|
|
|
+ // implementation; I'm going to choose the lazy path till then.
|
|
|
|
|
+ FLOATNAME(LMatrix3) mat;
|
|
|
|
|
+ extract_to_matrix(mat);
|
|
|
|
|
+ FLOATNAME(LVecBase3) scale;
|
|
|
|
|
+ decompose_matrix(mat, scale, hpr, cs);
|
|
|
|
|
+ return hpr;
|
|
|
}
|
|
}
|
|
|
- sp = yz + wx;
|
|
|
|
|
- hpr[0] = rad_2_deg(catan2(sh, ch));
|
|
|
|
|
- hpr[1] = rad_2_deg(catan2(sp, cp));
|
|
|
|
|
|
|
+
|
|
|
|
|
|
|
|
#ifndef NDEBUG
|
|
#ifndef NDEBUG
|
|
|
if (paranoid_hpr_quat) {
|
|
if (paranoid_hpr_quat) {
|