During transforming an image, a matrix multiplication per pixel was done.
The matrix in itself is always linear so it could be replaced by two additions.
Before
After
Details
Details
- Reviewers
Sebastian Parborg (zeddb) - Group Reviewers
Video Sequencer - Commits
- rB2e19649bb91b: Sequencer: Performance image crop transform.
Diff Detail
Diff Detail
- Repository
- rB Blender
- Branch
- temp-sequencer-image-transform (branched from master)
- Build Status
Buildable 15031 Build 15031: arc lint + arc unit
Event Timeline
Comment Actions
Wow, I went from 20fps in the debug build to ~300fps.
The next bottle neck seems to be in nearest_interpolation_color.
I think you could remove the if (x1 < 0 || y1 < 0) { check on line 266 as that has already been done on line 255.
However from looking at the perf data the slow part in that function seems to be the array index lookup (unsigned char *)in->rect + ((size_t)in->x) * y1 * 4 + 4 * x1;
Don't know if this could also be optimized further.
I guess we could be clever and only calculate the ((size_t)in->x) * y1 * 4 once per line.
But that might make the function more complex
However the current code looks good and because of the huge performance uplift, I think it can be committed as is.
