On Thu, 17 Mar 2011 19:03:04 +0100, Sylvere Teissier <[email protected]>
wrote:
Just a random idea to even more improve simulation speed in GHDL:
Using Link Time Optimisation to inline parts of the GRT library.
-GCC version >4.5 support Link Time Optimisation (LTO)
-Alternatively LLVM code can be generated from a GCC frontend using
the DragonEgg project (http://dragonegg.llvm.org/ [2]), LLVM is also
a
compiler middle-end/backend that support LTO.
SVN r116 and r118 are commits about LLVM, but I don't know if it
still
works on recent versions of GCC/DragonEgg.
hmmmm interesting, though i'm not sure of the speedup,
maybe 20% ? look at the ratio of development effort vs
actual speedup : if you can cut 20% of simulation time
with just a few compiler switches, great. Otherwise,
i don't know... compiling with -Os or -O2 is already
a good habit.
the bulk of the code runtime is spent doing things you wouldn't
think are so critical (if you are used to programming in C).
I have seen that range checking (in array indexing)
and managing local dynamic arrays takes a lot of CPU cycles.
Using integers cuts those corners but it's just the beginning
of the story.
Keep us informed of your advances and ideas !
yg
_______________________________________________
Ghdl-discuss mailing list
[email protected]
https://mail.gna.org/listinfo/ghdl-discuss
