fabs/rotate-me-fast
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

Fill a temp buffer for each line.

Browse source 
We don’t write into the memory right away, we first write to a temp
buffer that we blit into the rotated image.
This commit is contained in:
Fabien Freling 2014-07-04 08:23:06 +02:00
parent e3e0d3c20a
commit 049228b377
2 changed files with 72 additions and 42 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
Makefile
View file

@ -1,5 +1,5 @@
CXX = clang++ CXX = clang++
CXXFLAGS = -std=c++11 -W -Wall -O3 -ffast-math #-Werror CXXFLAGS = -std=c++11 -W -Wall -O3 -ffast-math -Werror -g
BUILD_DIR=/tmp BUILD_DIR=/tmp
all: rotation.cpp all: rotation.cpp

112
rotation.cpp
View file

@ -534,12 +534,13 @@ DPoint get_mapped_point(Image const& src, Point const& p, double const rotation)
} }
inline inline
void rotate_pixel(Image const& src, Image& rotated, void rotate_pixel(Image const& src, /*uint8_t* rotated,*/
DPoint const& src_rotated_point, Point const& rot_point, DPoint const& src_rotated_point, /*Point const& rot_point,*/
unsigned int const src_limit, unsigned int const rot_limit) unsigned int const src_limit, unsigned int const rot_limit,
uint8_t* rotate_buffer, unsigned int rot_index)
{ {
unsigned int src_index = ((int) src_rotated_point.y * src.width + (int) src_rotated_point.x) * 3; unsigned int src_index = ((int) src_rotated_point.y * src.width + (int) src_rotated_point.x) * 3;
unsigned int rot_index = (rot_point.y * rotated.width + rot_point.x) * 3; // unsigned int rot_index = (rot_point.y * rotated.width + rot_point.x) * 3;
// Out-of-bounds check // Out-of-bounds check
if (src_index >= src_limit if (src_index >= src_limit
@ -563,7 +564,10 @@ void rotate_pixel(Image const& src, Image& rotated,
// special case if we can directly map the src to the dest // special case if we can directly map the src to the dest
if (x_delta == 0 && y_delta == 0) if (x_delta == 0 && y_delta == 0)
{ {
memcpy(&rotated.buffer[rot_index], &src.buffer[src_index], 3 * sizeof (uint8_t)); // memcpy(&rotated.buffer[rot_index], &src.buffer[src_index], 3 * sizeof (uint8_t));
rotate_buffer[rot_index] = src.buffer[src_index];
rotate_buffer[rot_index + 1] = src.buffer[src_index + 1];
rotate_buffer[rot_index + 2] = src.buffer[src_index + 2];
return; return;
} }
@ -589,9 +593,9 @@ void rotate_pixel(Image const& src, Image& rotated,
top_left = _mm_add_ps(top_left, bottom_left); top_left = _mm_add_ps(top_left, bottom_left);
// convert float values to uint8_t // convert float values to uint8_t
rotated.buffer[rot_index] = top_left[3]; rotate_buffer[rot_index] = top_left[3];
rotated.buffer[rot_index + 1] = top_left[2]; rotate_buffer[rot_index + 1] = top_left[2];
rotated.buffer[rot_index + 2] = top_left[1]; rotate_buffer[rot_index + 2] = top_left[1];
} }
Image* rotate(Image const& src, double angle) Image* rotate(Image const& src, double angle)
@ -643,46 +647,72 @@ Image* rotate(Image const& src, double angle)
unsigned int const src_limit = src.width * src.height * 3; unsigned int const src_limit = src.width * src.height * 3;
unsigned int const rot_limit = rotated->width * rotated->height * 3; unsigned int const rot_limit = rotated->width * rotated->height * 3;
for (int y_i = 0; y_i <= (int) origin_nb_steps; ++y_i) // for (int y_i = 0; y_i <= (int) origin_nb_steps; ++y_i)
// {
// // first column origin
// Point const rot_origin(tl.x + y_i * rotated_step.x, tl.y + y_i * rotated_step.y);
// Point rot_point(rot_origin.x, rot_origin.y);
// DPoint rot_delta(0.0, 0.0);
//
// Point previous = rot_origin;
//
// for (int x_i = 0; x_i <= (int) line_nb_steps; ++x_i)
// {
//
// Point const delta(rot_point.x - tl.x, rot_point.y - tl.y);
// DPoint src_rotated_point(src_tl.x + delta.x * src_delta_x.x + delta.y * src_delta_y.x,
// src_tl.y + delta.x * src_delta_x.y + delta.y * src_delta_y.y);
//
// rotate_pixel(src, *rotated, src_rotated_point, rot_point, src_limit, rot_limit);
//
// if (previous.x != rot_point.x && previous.y != rot_point.y)
// {
// int y_slope = rot_point.y > previous.y ? 1 : -1;
// int tmp_y = rot_point.y;
// rot_point.y = previous.y;
//
// src_rotated_point.x -= y_slope * src_delta_y.x;
// src_rotated_point.y -= y_slope * src_delta_y.y;
//
// rotate_pixel(src, *rotated, src_rotated_point, rot_point, src_limit, rot_limit);
//
// rot_point.y = tmp_y;
// }
//
// previous = rot_point;
//
// rot_delta.x += bresenham.x;
// rot_point.x = rot_origin.x + (int) rot_delta.x;
//
// rot_delta.y += bresenham.y;
// rot_point.y = rot_origin.y + (int) rot_delta.y;
// }
// }
DPoint const rot_origin_in_src_grid = get_mapped_point(*rotated, Point(0, 0), -rotation);
DPoint const rot_origin_in_src = convert_img_coord_precision(src, rot_origin_in_src_grid);
unsigned int const buffer_size = rotated->width * 3;
uint8_t buffer[buffer_size];
for (unsigned int y = 0; y < rotated->height; ++y)
{ {
// first column origin memset(buffer, 0, buffer_size);
Point const rot_origin(tl.x + y_i * rotated_step.x, tl.y + y_i * rotated_step.y); DPoint src_rotated_point(rot_origin_in_src.x + y * src_delta_y.x,
Point rot_point(rot_origin.x, rot_origin.y); rot_origin_in_src.y + y * src_delta_y.y);
DPoint rot_delta(0.0, 0.0);
Point previous = rot_origin; for (unsigned int x = 0; x < rotated->width; ++x)
for (int x_i = 0; x_i <= (int) line_nb_steps; ++x_i)
{ {
if (src_rotated_point.x >= 0 && src_rotated_point.x < src.width
Point const delta(rot_point.x - tl.x, rot_point.y - tl.y); && src_rotated_point.y >= 0 && src_rotated_point.y < src.height)
DPoint src_rotated_point(src_tl.x + delta.x * src_delta_x.x + delta.y * src_delta_y.x,
src_tl.y + delta.x * src_delta_x.y + delta.y * src_delta_y.y);
rotate_pixel(src, *rotated, src_rotated_point, rot_point, src_limit, rot_limit);
if (previous.x != rot_point.x && previous.y != rot_point.y)
{ {
int y_slope = rot_point.y > previous.y ? 1 : -1; rotate_pixel(src, src_rotated_point, src_limit, rot_limit, buffer, x * 3);
int tmp_y = rot_point.y;
rot_point.y = previous.y;
src_rotated_point.x -= y_slope * src_delta_y.x;
src_rotated_point.y -= y_slope * src_delta_y.y;
rotate_pixel(src, *rotated, src_rotated_point, rot_point, src_limit, rot_limit);
rot_point.y = tmp_y;
} }
previous = rot_point; src_rotated_point.x += src_delta_x.x;
src_rotated_point.y += src_delta_x.y;
rot_delta.x += bresenham.x;
rot_point.x = rot_origin.x + (int) rot_delta.x;
rot_delta.y += bresenham.y;
rot_point.y = rot_origin.y + (int) rot_delta.y;
} }
memcpy(rotated->buffer + buffer_size * y, buffer, buffer_size);
} }
return rotated; return rotated;