blender
/
blender
136
406
Fork 590
Code Issues 6.2k Pull Requests 764 Projects 19 Wiki Activity

Fix T101211: Image jitters when scaling by large values 

Issue was caused by imprecise math due to using float numbers.
Use double instead.

No negative performance impact was observed.

Reviewed By: jbakker

Differential Revision: https://developer.blender.org/D16517
This commit is contained in:
Richard Antalik 2022-11-27 21:33:05 +01:00
parent 5a504a0d9d
commit 3a65d2f591
Notes: blender-bot 2024-01-16 13:34:02 +01:00 
Referenced by issue #101211, VSE Image Strip transform jitters under extreme scaling, unlike the smooth transform strip and compositor transform node.
Referenced by pull request #117160, ImBuf: no need to use double precision inside IMB_transform
Referenced by commit f16dbd0360, ImBuf: no need to use double precision inside IMB_transform
3 changed files with 67 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
source/blender/blenlib/BLI_math_vector.h
View File

@ -133,6 +133,7 @@ MINLINE void add_v4_v4v4(float r[4], const float a[4], const float b[4]);
MINLINE void add_v3fl_v3fl_v3i(float r[3], const float a[3], const int b[3]);
MINLINE void sub_v2_v2(float r[2], const float a[2]);
MINLINE void sub_v2_v2_db(double r[2], const double a[2]);
MINLINE void sub_v2_v2v2(float r[2], const float a[2], const float b[2]);
MINLINE void sub_v2_v2v2_db(double r[2], const double a[2], const double b[2]);
MINLINE void sub_v2_v2v2_int(int r[2], const int a[2], const int b[2]);
@ -193,6 +194,7 @@ MINLINE void madd_v2_v2fl(float r[2], const float a[2], float f);
MINLINE void madd_v3_v3fl(float r[3], const float a[3], float f);
MINLINE void madd_v3_v3v3(float r[3], const float a[3], const float b[3]);
MINLINE void madd_v2_v2v2fl(float r[2], const float a[2], const float b[2], float f);
MINLINE void madd_v2_v2db_db(double r[2], const double a[2], const double b[2], double f);
MINLINE void madd_v3_v3v3fl(float r[3], const float a[3], const float b[3], float f);
MINLINE void madd_v3_v3v3db_db(double r[3], const double a[3], const double b[3], double f);
MINLINE void madd_v3_v3v3v3(float r[3], const float a[3], const float b[3], const float c[3]);

12
source/blender/blenlib/intern/math_vector_inline.c
View File

@ -463,6 +463,12 @@ MINLINE void sub_v2_v2(float r[2], const float a[2])
r[1] -= a[1];
}
MINLINE void sub_v2_v2_db(double r[2], const double a[2])
{
r[0] -= a[0];
r[1] -= a[1];
}
MINLINE void sub_v2_v2v2(float r[2], const float a[2], const float b[2])
{
r[0] = a[0] - b[0];
@ -697,6 +703,12 @@ MINLINE void madd_v2_v2v2fl(float r[2], const float a[2], const float b[2], floa
r[1] = a[1] + b[1] * f;
}
MINLINE void madd_v2_v2db_db(double r[2], const double a[2], const double b[2], double f)
{
r[0] = a[0] + b[0] * f;
r[1] = a[1] + b[1] * f;
}
MINLINE void madd_v3_v3v3fl(float r[3], const float a[3], const float b[3], float f)
{
r[0] = a[0] + b[0] * f;

100
source/blender/imbuf/intern/transform.cc
View File

@ -22,19 +22,19 @@ struct TransformUserData {
/** \brief Destination image buffer to write to. */
ImBuf *dst;
/** \brief UV coordinates at the origin (0,0) in source image space. */
float start_uv[2];
double start_uv[2];
/**
* \brief delta UV coordinates along the source image buffer, when moving a single pixel in the X
* axis of the dst image buffer.
*/
float add_x[2];
double add_x[2];
/**
* \brief delta UV coordinate along the source image buffer, when moving a single pixel in the Y
* axes of the dst image buffer.
*/
float add_y[2];
double add_y[2];
/**
* \brief Cropping region in source image pixel space.
@ -54,39 +54,45 @@ struct TransformUserData {
private:
void init_start_uv(const float transform_matrix[4][4])
{
float start_uv_v3[3];
float orig[3];
zero_v3(orig);
mul_v3_m4v3(start_uv_v3, transform_matrix, orig);
copy_v2_v2(start_uv, start_uv_v3);
double start_uv_v3[3];
double orig[3];
double transform_matrix_double[4][4];
copy_m4d_m4(transform_matrix_double, transform_matrix);
zero_v3_db(orig);
mul_v3_m4v3_db(start_uv_v3, transform_matrix_double, orig);
copy_v2_v2_db(start_uv, start_uv_v3);
}
void init_add_x(const float transform_matrix[4][4])
{
double transform_matrix_double[4][4];
copy_m4d_m4(transform_matrix_double, transform_matrix);
const int width = src->x;
float add_x_v3[3];
float uv_max_x[3];
zero_v3(uv_max_x);
double add_x_v3[3];
double uv_max_x[3];
zero_v3_db(uv_max_x);
uv_max_x[0] = width;
uv_max_x[1] = 0.0f;
mul_v3_m4v3(add_x_v3, transform_matrix, uv_max_x);
sub_v2_v2(add_x_v3, start_uv);
mul_v2_fl(add_x_v3, 1.0f / width);
copy_v2_v2(add_x, add_x_v3);
mul_v3_m4v3_db(add_x_v3, transform_matrix_double, uv_max_x);
sub_v2_v2_db(add_x_v3, start_uv);
mul_v3db_db(add_x_v3, 1.0f / width);
copy_v2_v2_db(add_x, add_x_v3);
}
void init_add_y(const float transform_matrix[4][4])
{
double transform_matrix_double[4][4];
copy_m4d_m4(transform_matrix_double, transform_matrix);
const int height = src->y;
float add_y_v3[3];
float uv_max_y[3];
zero_v3(uv_max_y);
double add_y_v3[3];
double uv_max_y[3];
zero_v3_db(uv_max_y);
uv_max_y[0] = 0.0f;
uv_max_y[1] = height;
mul_v3_m4v3(add_y_v3, transform_matrix, uv_max_y);
sub_v2_v2(add_y_v3, start_uv);
mul_v2_fl(add_y_v3, 1.0f / height);
copy_v2_v2(add_y, add_y_v3);
mul_v3_m4v3_db(add_y_v3, transform_matrix_double, uv_max_y);
sub_v2_v2_db(add_y_v3, start_uv);
mul_v3db_db(add_y_v3, 1.0f / height);
copy_v2_v2_db(add_y, add_y_v3);
}
};
@ -104,7 +110,7 @@ class BaseDiscard {
/**
* \brief Should the source pixel at the given uv coordinate be discarded.
*/
virtual bool should_discard(const TransformUserData &user_data, const float uv[2]) = 0;
virtual bool should_discard(const TransformUserData &user_data, const double uv[2]) = 0;
};
/**
@ -117,7 +123,7 @@ class CropSource : public BaseDiscard {
*
* Uses user_data.src_crop to determine if the uv coordinate should be skipped.
*/
bool should_discard(const TransformUserData &user_data, const float uv[2]) override
bool should_discard(const TransformUserData &user_data, const double uv[2]) override
{
return uv[0] < user_data.src_crop.xmin || uv[0] >= user_data.src_crop.xmax ||
uv[1] < user_data.src_crop.ymin || uv[1] >= user_data.src_crop.ymax;
@ -134,7 +140,7 @@ class NoDiscard : public BaseDiscard {
*
* Will never discard any pixels.
*/
bool should_discard(const TransformUserData & /*user_data*/, const float /*uv*/[2]) override
bool should_discard(const TransformUserData & /*user_data*/, const double /*uv*/[2]) override
{
return false;
}
@ -202,12 +208,12 @@ class BaseUVWrapping {
/**
* \brief modify the given u coordinate.
*/
virtual float modify_u(const ImBuf *source_buffer, float u) = 0;
virtual double modify_u(const ImBuf *source_buffer, double u) = 0;
/**
* \brief modify the given v coordinate.
*/
virtual float modify_v(const ImBuf *source_buffer, float v) = 0;
virtual double modify_v(const ImBuf *source_buffer, double v) = 0;
};
/**
@ -215,12 +221,12 @@ class BaseUVWrapping {
*/
class PassThroughUV : public BaseUVWrapping {
public:
float modify_u(const ImBuf * /*source_buffer*/, float u) override
double modify_u(const ImBuf * /*source_buffer*/, double u) override
{
return u;
}
float modify_v(const ImBuf * /*source_buffer*/, float v) override
double modify_v(const ImBuf * /*source_buffer*/, double v) override
{
return v;
}
@ -231,7 +237,7 @@ class PassThroughUV : public BaseUVWrapping {
*/
class WrapRepeatUV : public BaseUVWrapping {
public:
float modify_u(const ImBuf *source_buffer, float u) override
double modify_u(const ImBuf *source_buffer, double u) override
{
int x = int(floor(u));
@ -242,7 +248,7 @@ class WrapRepeatUV : public BaseUVWrapping {
return x;
}
float modify_v(const ImBuf *source_buffer, float v) override
double modify_v(const ImBuf *source_buffer, double v) override
{
int y = int(floor(v));
y = y % source_buffer->y;
@ -284,24 +290,24 @@ class Sampler {
static const int ChannelLen = NumChannels;
using SampleType = std::array<StorageType, NumChannels>;
void sample(const ImBuf *source, const float u, const float v, SampleType &r_sample)
void sample(const ImBuf *source, const double u, const double v, SampleType &r_sample)
{
if constexpr (Filter == IMB_FILTER_BILINEAR && std::is_same_v<StorageType, float> &&
NumChannels == 4) {
const float wrapped_u = uv_wrapper.modify_u(source, u);
const float wrapped_v = uv_wrapper.modify_v(source, v);
const double wrapped_u = uv_wrapper.modify_u(source, u);
const double wrapped_v = uv_wrapper.modify_v(source, v);
bilinear_interpolation_color_fl(source, nullptr, r_sample.data(), wrapped_u, wrapped_v);
}
else if constexpr (Filter == IMB_FILTER_NEAREST && std::is_same_v<StorageType, uchar> &&
NumChannels == 4) {
const float wrapped_u = uv_wrapper.modify_u(source, u);
const float wrapped_v = uv_wrapper.modify_v(source, v);
const double wrapped_u = uv_wrapper.modify_u(source, u);
const double wrapped_v = uv_wrapper.modify_v(source, v);
nearest_interpolation_color_char(source, r_sample.data(), nullptr, wrapped_u, wrapped_v);
}
else if constexpr (Filter == IMB_FILTER_BILINEAR && std::is_same_v<StorageType, uchar> &&
NumChannels == 4) {
const float wrapped_u = uv_wrapper.modify_u(source, u);
const float wrapped_v = uv_wrapper.modify_v(source, v);
const double wrapped_u = uv_wrapper.modify_u(source, u);
const double wrapped_v = uv_wrapper.modify_v(source, v);
bilinear_interpolation_color_char(source, r_sample.data(), nullptr, wrapped_u, wrapped_v);
}
else if constexpr (Filter == IMB_FILTER_BILINEAR && std::is_same_v<StorageType, float>) {
@ -317,8 +323,8 @@ class Sampler {
true);
}
else {
const float wrapped_u = uv_wrapper.modify_u(source, u);
const float wrapped_v = uv_wrapper.modify_v(source, v);
const double wrapped_u = uv_wrapper.modify_u(source, u);
const double wrapped_v = uv_wrapper.modify_v(source, v);
BLI_bilinear_interpolation_fl(source->rect_float,
r_sample.data(),
source->x,
@ -329,8 +335,8 @@ class Sampler {
}
}
else if constexpr (Filter == IMB_FILTER_NEAREST && std::is_same_v<StorageType, float>) {
const float wrapped_u = uv_wrapper.modify_u(source, u);
const float wrapped_v = uv_wrapper.modify_v(source, v);
const double wrapped_u = uv_wrapper.modify_u(source, u);
const double wrapped_v = uv_wrapper.modify_v(source, v);
sample_nearest_float(source, wrapped_u, wrapped_v, r_sample);
}
else {
@ -341,8 +347,8 @@ class Sampler {
private:
void sample_nearest_float(const ImBuf *source,
const float u,
const float v,
const double u,
const double v,
SampleType &r_sample)
{
BLI_STATIC_ASSERT(std::is_same_v<StorageType, float>);
@ -461,8 +467,8 @@ class ScanlineProcessor {
{
const int width = user_data->dst->x;
float uv[2];
madd_v2_v2v2fl(uv, user_data->start_uv, user_data->add_y, scanline);
double uv[2];
madd_v2_v2db_db(uv, user_data->start_uv, user_data->add_y, scanline);
output.init_pixel_pointer(user_data->dst, 0, scanline);
for (int xi = 0; xi < width; xi++) {
@ -472,7 +478,7 @@ class ScanlineProcessor {
channel_converter.convert_and_store(sample, output);
}
add_v2_v2(uv, user_data->add_x);
add_v2_v2_db(uv, user_data->add_x);
output.increase_pixel_pointer();
}
}