aboutsummaryrefslogtreecommitdiff
path: root/TODO.EXP
blob: f3d19554e5b06ec367bbe839d568de1a14dd6cf9 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58

experimental stuff

todo:

* modify 2D rotation to be used as arbitrary rotation (givens)
* add 3D rotation
* frame rate limited delay line (burst delay?)
* optimize resampling code (mipmapped packed 16bit format?)


* cache slicer

pdp_slice_cut & pdp_slice_glue seem to have a significant impact on performance
for long chains with simple (memory rate limited) operators. especially when a lot
of passive packets are involved. todo: make a vertical slicer + mmx transpose object
to use this for horizontal and vertical processors too (vertical processors: more efficient
because of alignment). find out why this doesn't work in threads + think about a possibility
to do first layout a buffer structure, and then do incremental processing (outside of pd/pdp:
for constructing optimized signal processors using some kind of intermediate language
and automatic slice dimension selection)
NOTE: solve this as a forth script for specific abstractions.

* new image packet types

*) PDP_IMAGE_MCHP

multi channel stuff. this packet type is to solve some problems with the subsampled
chroma planes in PDP_IMAGE_YV12. although it is fast, it is best to have an equally 
sampled, arbitrary number of channels image type available. 


*) PDP_IMAGE_PCK4

the main reason for this package type is in resampling code to have single pixel 
access mmx routines. but is it worth it? probably not.. mipmap image packet for resampling?


* new vectors, matrices & gsl packets

general matrix math is a nice thing to have. wrapping gsl is a possibility. it can use
atlas for fast matrix math. finish better type handling first.


* opengl -> see also opengl/TODO

move opengl stuff to dpd.
todo:
- think about pbufs / textures : isolate system specific stuff


* better type handling

core is in place. todo:
- better (tree) search algorithm (breadth first + cost)
- add more converters