lua-users home
lua-l archive

[Date Prev][Date Next][Thread Prev][Thread Next] [Date Index] [Thread Index]

On Jul 9, 2013, at 10:52 AM, Luiz Henrique de Figueiredo wrote:

> However, I think that Roberto tested that program on a machine for which
> the conversion int <-> float is costly. I'm not sure modern machines have
> this problem.
> Bottom line, the work version is meant to be tested and feedback sent. Thanks.

I've got a big pile of embedded Linux boxes of various architectures. What I don't have is a reproducible test load. (I currently believe my core problems on the smaller machines is fun memory hierarchy effects[1] and they're largely not Lua's problem.) If you have a suggestion  for a self-contained set of work1/5.2/5.1 C/Lua code I'll obsessively test it in a bunch of toolchains, ABIs, ISAs, etc etc. I'd be happy to put a bunch of them in a network "rubber room" with ssh access to a configured x86_64 OpenWrt build host if it would be helpful. Tools like the malta MIPS simulators can be a lot more useful in analysis but a first pass on "real hardware" is always satisfying.


[1]: I repeat myself, but: 1) The goal of an operating system is to use as much memory as possible. 2) Q: How much memory is this Lua MIPS process using? A: Two pages. The location of the instruction pointer, and the destination of any load or store.