| [AD] [ alleg-Bugs-2633140 ] optimise memory blenders |
[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
Bugs item #2633140, was opened at 2009-02-24 21:14
Message generated for change (Comment added) made by tjaden
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=105665&aid=2633140&group_id=5665
Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Core Library
Group: 4.9
Status: Open
Resolution: None
Priority: 3
Private: No
Submitted By: Peter Wang (tjaden)
Assigned to: Nobody/Anonymous (nobody)
Summary: optimise memory blenders
Initial Comment:
The memory blenders are not optimised at all. This needs to be done some time, possibly after 5.0, as long as it's clear in the release notes that's the case.
----------------------------------------------------------------------
>Comment By: Peter Wang (tjaden)
Date: 2009-03-11 10:43
Message:
Two ideas:
1. Dynamically generate blender routines at runtime, either
native code or threaded code. An interesting challenge, but
needs to be portable and maintainable.
2. One giant function containing all possible code paths.
Based on the known blending modes and alpha values, etc. set
some flags to control which code paths are taken, e.g. do we
read in the src colour or not? Do we read the dest colour
or not? Do we multiply this with that?
The hope is that the processor's branch prediction will
quickly be able to predict correctly which path to take at
each switch. Should do some benchmarks to see if this is
the case.
----------------------------------------------------------------------
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=105665&aid=2633140&group_id=5665