lua-users home
lua-l archive

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


bertie wrote:
> On Thursday 28 September 2006 09:14, Nicolas Cannasse wrote:
> > Some of the Shootout benchmarks scripts are available in the
> > neko/src/benchs. In general, when meaningful computation is performed
> > (i.e. not "hello world") then Neko tends to be one of the fastest
> > dynamicly typed VM.
> Correct.


- You select a single, mostly irrelevant benchmark.
- You compare absurdly low runtimes in the millisecond range.
- You don't know how to write good Lua programs (ever heard of
- You fail to do proper interpreted/interpreted vs.
  compiled/compiled comparisons.
- You try (and fail) to copy the years of work that went in to
  the shootout benchmark set.
- The graphs are cute, but it's content that counts.

IMHO the Neko community should contribute a _complete_ set of
benchmarks to the shootout. Then everyone can do a proper

On the few semi-comparable benchmarks I tried, Lua easily beats
Neko -interp. And LuaJIT beats the hell out of Neko JIT (between
2x and 6x slower).

> Recursiveness test is also included in shootout.debian. It means a lot, 
> particulary to low-level programmers. 

Nope. The fibonacci micro-benchmark measures call frame setup and
teardown overhead and nothing else. This is only a small fraction
of the overhead of any realistic program.