10 Dec
2002
10 Dec
'02
2:23 p.m.
Steve Harris wrote:
yep, i was thinking about direct code generation (relocation
if you will) assuming it'd give the best execution speed. rt
modification of such constructs can easily set your thoughts
spinning for a few hours.
nope, i meant
dynamic updates on a realtime (lock-free)
code path; it's an interesting problem with, afaict, no
obviously elegant solutions.
Argh! I was thinking of dumping the code and rebuilding (hopefully keeping
the state). Doing it that way would be interesting, but much harder. Youd
have to either use a lot of function calls or do some hard code relocation
stuff I think. No, I imagine it will be noticably slower, however I
think CPU's are
getting to the kind of power where its feasable to use it for real.
i have yet to notice a power increase in the local cpu. ;)
The dynamic compiliation will win you some speed back.
I dont think the implementation is really that hard, the UI would be the
most complex part, as always. Its kindof a pipedream anyway, as none of us
has enough free time to tackle it at the moment.
sigh, yes. i've been thinking about a simplistic dsp-only
scripting language for similar purposes lately, but concluded
that for the time being gcc producing dynamically reloaded
.so's would be the way to go (unfortunately saol produces
executables, hint ;).
tim