tmaq
/
tornavis
1
0
Fork 0
Code Issues 2 Pull Requests 29 Packages Projects 24 Releases Wiki Activity

Fix #116880: Rotation discrepancy when converting a matrix to euler 

Converting a near-identity matrix to a euler could be off by over
4.5 degrees when the result of the hypotenuse calculated from the
matrix was small (around 2e-6).

Resolve by increasing the epsilon ~20x to 3.75e-05 which results
in ~1/20th the error.

This was tested against many generated near-identity matrices
(which tend to cause arithmetic error) to ensure this change doesn't
cause worse results in other cases (see report for details).
This commit is contained in:
Campbell Barton 2024-01-08 23:32:31 +11:00
parent 214f942258
commit 5b00141144
1 changed files with 10 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
source/blender/blenlib/intern/math_rotation.c
View File

@ -21,6 +21,14 @@
# define QUAT_EPSILON 0.0001
#endif
/**
* The threshold for using a zeroed 3rd (typically Z) value when calculating the euler.
* NOTE(@ideasman42): A reasonable range for this value is (0.0002 .. 0.00002).
* This was previously `16 * FLT_EPSILON` however it caused imprecision at times,
* see examples from: #116880.
*/
#define EULER_HYPOT_EPSILON 0.0000375
void unit_axis_angle(float axis[3], float *angle)
{
axis[0] = 0.0f;
@ -1392,8 +1400,7 @@ static void mat3_normalized_to_eul2(const float mat[3][3], float eul1[3], float
BLI_ASSERT_UNIT_M3(mat);
if (cy > 16.0f * FLT_EPSILON) {
if (cy > (float)EULER_HYPOT_EPSILON) {
eul1[0] = atan2f(mat[1][2], mat[2][2]);
eul1[1] = atan2f(-mat[0][2], cy);
eul1[2] = atan2f(mat[0][1], mat[0][0]);
@ -1727,7 +1734,7 @@ static void mat3_normalized_to_eulo2(const float mat[3][3],
cy = hypotf(mat[i][i], mat[i][j]);
if (cy > 16.0f * FLT_EPSILON) {
if (cy > (float)EULER_HYPOT_EPSILON) {
eul1[i] = atan2f(mat[j][k], mat[k][k]);
eul1[j] = atan2f(-mat[i][k], cy);
eul1[k] = atan2f(mat[i][j], mat[i][i]);