sw_engine raster: image rastering optimization #239
Conversation
The alpha value calculation pulled out outside the inner loop to reduce the number of unnecessary operations. Added local variables to reduce the number of costly multiplications performed in a loop.
To improve the readability of the file.
mgrudzinska
force-pushed
the
mgrudzinska/Image_raster_opt_transl
branch
from
5ce5276
to
3c55b58
Compare
Functions for rastering the translated images added (no rotation or scaling).
mgrudzinska
force-pushed
the
mgrudzinska/Image_raster_opt_transl
branch
from
3c55b58
to
64293a7
Compare
| @@ -71,17 +71,21 @@ static bool _inverse(const Matrix* transform, Matrix* invM) | |||
| } | |||
|
|
|||
|
|
|||
| static bool _identify(const Matrix* transform) | |||
| static int _identify(const Matrix* m) | |||
There was a problem hiding this comment.
This does more than identify() check.... bad.
| else return _rasterImage(surface, image->data, image->w, image->h, bbox); | ||
| } | ||
| //Translation | ||
| if (translucent) return _rasterTranslucentImage(surface, image->data, image->w, image->h, opacity, bbox, -transform->e13, -transform->e23); |
There was a problem hiding this comment.
Why don't reuse the existing method? We can just offset the region xy...
_rasterTranslucentImage(SwSurface* surface, uint32_t *img, uint32_t w, uint32_t h, uint32_t opacity, const SwBBox& region);
There was a problem hiding this comment.
my problem is that I can't see how to use a region because both pictures and shapes/svgs may use the same rastering functions. For ex an svg with clipping and opacity will use _translucentImage, and a translucent picture (pixels) uses it as well.
For a picture I can easily use region to call the same function for the identity transform and the translation, but it's not what I want to do for an svg...
Am I missing something?
Adding functions for rastering the translated images (no rotation and scaling).
This pr contains 3 commits for readabilty.
I introduced changes reducing the number of multiplications performed in loops.
I changed the order of the functions, sticking to the rule (for each block): first identity, then for translated images and finally for general transformations.
In the last commit I introduced all functions for rastering the translated images.
Issue Tickets (if any) : Optimization Image Raster #206
Tests or Samples : draft version of OptRasterTest - calls all raster functions for images
https://github.com/mgrudzinska/thorvg/tree/mgrudzinska/Test_image_raster_opr_trans