[racket-dev] can we write these four lines of C in performant racket?

On Mon, Jul 25, 2011 at 7:51 AM, Matthew Flatt <mflatt at cs.utah.edu> wrote:
>
> Here are some timings for 1000 iterations on 2^20-element inputs
> (32-bit mode, Mac Book Pro 2.53 GHz):
>
>  C as above, gcc -02          :  1409
>  C with indirections, gcc -O2 :  4041
>  C as above, gcc -O0          :  6425
>  C with indirections, gcc -O0 :  8480
>  Racket fxvector (more direct):  8883
>  Racket                       : 11248
>
> I can tweak the JIT in small ways to make a small improvement:
>
>  Racket with JIT tweaks       : 10670

What do the JIT tweaks do to the performance of the fxvector version?
-- 
sam th
samth at ccs.neu.edu