From e20d5ae3622d5d656dc28d7a090aee76b08158b0 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Pablo=20Mart=C3=ADn?= Date: Sun, 7 Sep 2003 20:01:24 +0000 Subject: updating pdp to current version 0.12.2 svn path=/trunk/externals/pdp/; revision=936 --- TODO.EXP | 58 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 58 insertions(+) create mode 100644 TODO.EXP (limited to 'TODO.EXP') diff --git a/TODO.EXP b/TODO.EXP new file mode 100644 index 0000000..f3d1955 --- /dev/null +++ b/TODO.EXP @@ -0,0 +1,58 @@ + +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 + + -- cgit v1.2.1