[Date Prev][Date Next][Thread Prev][Thread Next]
- Subject: Re: Benchmark shootout shows LuaJIT 2.0 (was Re: [ANN] LuaJIT-2.0.0-beta1)
- From: Mike Pall <mikelu-0911@...>
- Date: Sun, 1 Nov 2009 19:24:24 +0100
Wim Couwenberg wrote:
> The pidigits benchmark is nagging me a bit, in particular i'm not sure
> if my "matrix optimizations" used in #5 are in fact optimizations when
> using lgmp. (I did those for a Lua only implementation.) For
> example, as shown here
> version #7 is still faster and that one doesn't use the matrix trick.
> This is on x64, no JIT, but still i'm wondering why the difference...
Back when I looked at this I also wondered about some of the
timings. Since the pidigits benchmark deteriorated into a GMP
optimization festival, I'm not too worried about it anymore.
After all it only spends something like 1% of its time in Lua.
But to get the Lua/LuaJIT scores up, it would be nice to
investigate what algorithmic changes the better performing GMP
programs made and simply copy that.