Re: LuaJIT2 performance for number crunching

[Mike Pall <mikelu-1102@...>]

KHMan wrote:
>   "this is a major advance in computer science because"
> 
> Yes, you should try sending a paper somewhere and see if they let
> you keep such claims.

This is hyperbole, for sure.

> The HP Dynamo trace compiler was doing
> faster-than-compiled-original speeds years ago.

Oh no, not that again. Everybody is repeating that claim, but
nobody has actually bothered to read the paper! Here's my standard
rant on that:

  This is a common misinterpretation of the Dynamo paper: they
  compiled their C code at the _lowest_ optimization level and
  then ran the (suboptimal) machine code through Dynamo. So there
  was actually something left to optimize.

  Think about it this way: a 20% difference isn't unrealistic if
  you compare -O1 vs. -O3.

  But it's completely unrealistic to expect a 20% improvement if
  you'd try this with the machine code generated by a modern
  C compiler at the highest optimization level.

  Claiming that JIT compilers outperform static compilers, solely
  based on this paper, is utter nonsense.
  
  [In fact a very good JIT compiler _can_ outperform a very good
  static compilers under specific, but rare, circumstances. But
  this has more to do with extra specialization opportunities at
  runtime and is completely unrelated to this paper.]

--Mike