I forgot to mention;
"" But magically turning single-threaded CPU hogs into multithreaded processes that run on
your graphics card isn't going to happen anytime soon, I'm afraid. ""
Of course not. Wild guess, but i believe that 50% of what is in a normal desktop installation,
in a foreseeable future, wont see any vectorization and only mild parallelization if any at
all.
Start with the obvious ones
So benchmarks, of which i urled one in my first post show possibilities of more than 2 orders
of magnitude improvement, for hand tuned and for that code base that can benefice higly.
All in all, if only 50% of the code base installed gets improvements of only a little more
than 1 order of magnitude average, in overall the improvement can still be in the various of
thens of % (Amdahl law).
And here is the beauty of source code!. It shouldn't just be a recompile. FOSS permits always
some sort of hand tunning... start with the obvious, start conservative... i don't think that
everything will be possible... but in the end it can prove more than worthed.
Posted Apr 21, 2008 6:59 UTC (Mon) by nix (subscriber, #2304)
[Link]
If you use OpenMP (which has a really nasty #pragma-based syntax but works
quite well, and which recent GCCs have native support for) then there is
no need to recompile at all: you just compile with OpenMP, and the program
itself probes for the number of cores at startup. I see no reason why an
enhanced OpenMP runtime couldn't probe for a GPGPU and run bits on there
if it wanted to, all without any need for this elaborate IR-shipping you
seem to be discussing.
Red Hat: no desktop products coming
Posted Apr 21, 2008 17:01 UTC (Mon) by mmarq (guest, #2332)
[Link]
Others also obvious targets are Intel open sourced IBB, and AMD open sourced "framewave"...
and you can bet that similar approaches would be getting here like the rain.