kimcinoo
/
efl
forked from enlightenment/efl
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

evas png loader: optimize scale down logic. 

Previously, mannual scale down logic was too primitive,
it copied pixel data each channels. Obviously, it's ineffective.

We know the general case - 4 bytes channel which is the most usage,
If loader copies data per four bytes, instructions could be reduced.

When I load scale-downed image(original 8k), about 0.02 secs was reduced by this.
This commit is contained in:
Hermet Park 2019-04-29 19:55:38 +09:00
parent 49f22b60e5
commit 1f438bd1f0
1 changed files with 69 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

88
src/modules/evas/image_loaders/png/evas_image_load_png.c
View File

@ -223,7 +223,7 @@ static Eina_Bool
evas_image_load_file_data_png(void *loader_data,
Evas_Image_Property *prop,
void *pixels,
int *error)
int *error)
{
Evas_Loader_Internal *loader = loader_data;
Evas_Image_Load_Opts *opts;
@ -395,9 +395,8 @@ evas_image_load_file_data_png(void *loader_data,
}
else
{
unsigned char *src_ptr, *dst_ptr;
unsigned char *src_ptr;
int skip_row = 0, region_x = 0, region_y = 0;
dst_ptr = surface;
if (region_set)
{
@ -412,20 +411,49 @@ evas_image_load_file_data_png(void *loader_data,
for (skip_row = 0; skip_row < region_y; skip_row++)
png_read_row(png_ptr, tmp_line, NULL);
for (i = 0; i < h; i++)
//general case: 4 bytes pixel.
if (pack_offset == sizeof(DATA32))
{
png_read_row(png_ptr, tmp_line, NULL);
src_ptr = tmp_line + region_x * pack_offset;
for (j = 0; j < w; j++)
DATA32 *dst_ptr = (DATA32 *) surface;
DATA32 *src_ptr2;
for (i = 0; i < h; i++)
{
for (k = 0; k < (int)pack_offset; k++)
dst_ptr[k] = src_ptr[k];
dst_ptr += pack_offset;
src_ptr += scale_ratio * pack_offset;
png_read_row(png_ptr, tmp_line, NULL);
src_ptr2 = (DATA32 *) (tmp_line + region_x * pack_offset);
for (j = 0; j < w; j++)
{
*dst_ptr = *src_ptr2;
++dst_ptr;
src_ptr2 += scale_ratio;
}
for (j = 0; j < (scale_ratio - 1); j++)
png_read_row(png_ptr, tmp_line, NULL);
}
for (j = 0; j < (scale_ratio - 1); j++)
png_read_row(png_ptr, tmp_line, NULL);
}
else
{
unsigned char *dst_ptr = surface;
for (i = 0; i < h; i++)
{
png_read_row(png_ptr, tmp_line, NULL);
src_ptr = tmp_line + region_x * pack_offset;
for (j = 0; j < w; j++)
{
for (k = 0; k < (int) pack_offset; k++)
dst_ptr[k] = src_ptr[k];
dst_ptr += pack_offset;
src_ptr += scale_ratio * pack_offset;
}
for (j = 0; j < (scale_ratio - 1); j++)
png_read_row(png_ptr, tmp_line, NULL);
}
}
for (skip_row = region_y + h * scale_ratio; skip_row < image_h; skip_row++)
png_read_row(png_ptr, tmp_line, NULL);
}
@ -443,15 +471,37 @@ evas_image_load_file_data_png(void *loader_data,
src_ptr = pixels2 + (region_y * image_w * pack_offset) + region_x * pack_offset;
for (i = 0; i < h; i++)
//general case: 4 bytes pixel.
if (pack_offset == sizeof(DATA32))
{
for (j = 0; j < w; j++)
DATA32 *dst_ptr = (DATA32 *) surface;
DATA32 *src_ptr2 = (DATA32 *) src_ptr;
for (i = 0; i < h; i++)
{
for (k = 0; k < (int)pack_offset; k++)
dst_ptr[k] = src_ptr[k + scale_ratio * j * pack_offset];
dst_ptr += pack_offset;
for (j = 0; j < w; j++)
{
*dst_ptr = *src_ptr2;
++dst_ptr;
src_ptr2 += scale_ratio;
}
src_ptr2 += scale_ratio * image_w;
}
}
else
{
unsigned char *dst_ptr = surface;
for (i = 0; i < h; i++)
{
for (j = 0; j < w; j++)
{
for (k = 0; k < (int)pack_offset; k++)
dst_ptr[k] = src_ptr[k + scale_ratio * j * pack_offset];
dst_ptr += pack_offset;
}
src_ptr += scale_ratio * image_w * pack_offset;
}
src_ptr += scale_ratio * image_w * pack_offset;
}
free(pixels2);
}