Re: [knotwell@f5.com] Re: [PATCH] Re: startup speed (or lack thereof)

[Dirk Herrmann <dirk at ida dot ing dot tu-bs dot de>]

On 20 Mar 2000, Jim Blandy wrote:

> From: knotwell@f5.com
[...]
> I spent some time in symbols.c (specifically,scm_intern_obarray_soft).  I 
> made some small changes that make the code a bit easier to understand
> while providing a slight optimization. 
>   
>     scm_hash = scm_strhash (tmp, i, SCM_LENGTH(obarray));
>   
>    retry_new_obarray:
>     for (lsym = SCM_VELTS (obarray)[scm_hash]; SCM_NIMP (lsym); lsym = SCM_CDR (lsym))
>       {
> !       SCM a = SCM_CAAR(lsym);
> !       if (SCM_LENGTH(a) != len) 
>   	goto trynext;
> ! 
> !       tmp = SCM_UCHARS(a);
> !       for(i=len;i--;)
>   	if (((unsigned char *) name)[i] != tmp[i])
> ! 	  goto trynext; 
> ! 
> !       a = SCM_CAR(lsym);
> !       SCM_REALLOW_INTS;
> !       return a;
> ! 
> !       trynext:;
>       }
>   
>     if (obarray == scm_symhash)

What about the following solution, which even does avoid using gotos and
relies on memcmp to perform the character field comparison?  (The use of
memcmp might cause a performance degradation due to function calling
overhead.  For longer symbols, however, memcmp may gain due to a more
efficient implementation.  Since it is a standard library call, it might
even get inlined by the compiler automatically.  Thus, there are just too
many possibilities for me to tell whether it is a good idea to use it
here.  OTOH it will only be called when it is already known that two
symbols of the same length are to be compared.)

  for (lsym = SCM_VELTS (obarray)[scm_hash]; SCM_NNULLP (lsym); lsym = SCM_CDR (lsym))
    {
      SCM symbol_value_pair = SCM_CAR (lsym);
      SCM symbol = SCM_CAR (symbol_value_pair);
      if (SCM_LENGTH (symbol) == len) {
	unsigned char *symbol_chars = SCM_UCHARS (symbol);
	if (memcmp (name, symbol_chars, len) == 0) {
	  SCM_REALLOW_INTS;
	  return symbol_value_pair;
	}
      }
    }

If nobody objects, I will apply this patch.

Best regards
Dirk Herrmann