Details

source/blender/editors/sculpt_paint/paint_image_2d.c
388–391	Use `(for int y = ;` and similar, define variables where they are first initialized.
390–391	use `CLAMP`, avoid unnecessary `(int)` cast.
392–436	`float total_samples = 0;`
395–398	`bpos[0]` and `bpos[1]` can include `offset + aa_offset`, so more is computed in advance.
399	Computing `sin` and `cos` for every AA sample is slow. Compute them once in advance since rotation is fixed.
400–401	These two lines are equivalent to just doing `len_v2(xy_rot)`.
403	Division by zero because hardness=1.
411	The `65535.0f / (float)(aa_samples * aa_samples))` normalization factor could be computed once in advance then multiplied.
765–767	Explain that the padding is needed for anti-aliasing, it's not so much fixing artifacts as just providing the proper space for it.
805	I don't see a good reason to remove this, caching the curve mask seems like an optimization we should keep.

This revision now requires changes to proceed.Oct 7 2019, 4:44 PM

source/blender/editors/sculpt_paint/paint_image_2d.c
805	Any comment on this?

source/blender/editors/sculpt_paint/paint_image_2d.c
805	With antialiasing and radius < 1 the mask needs to be recalculated for every new step position, we can't rely on the previous mask diameter.

source/blender/editors/sculpt_paint/paint_image_2d.c
397–398	Rather than having a `final_offset`, you can do: bpos[0] = pos[0] - floorf(pos[0]) - offset - aa_offset; bpos[1] = pos[1] - floorf(pos[1]) - offset - aa_offset;
805	This is probably going to have a measurable performance impact for painting with big brushes. Do you think it would make sense to only do the subpixel positioning for small brushes, so we can cache the mask for big brushes?

source/blender/editors/sculpt_paint/paint_image_2d.c
805	If you are using a sharp falloff the antialiasing is quite noticeable. I personally prefer antialiasing to sightly better performance. Also, we we add the rest of the brush parameters in D5697, the mask will need to be recalculated for almost any feature that is enabled. We should probably try to optimize this by doing something else instead of caching the mask.

2D painting performance is probably not a major issue, we'll see if it shows up in profiles.

This revision is now accepted and ready to land.Oct 7 2019, 7:10 PM

This revision was automatically updated to reflect the committed changes.

Paint: 2D paint line improvements
ClosedPublic
Actions