Re: [Libclc-developers] new function: clc_strrev

[<bo...@me...>]

Hallvard B Furuseth wrote:
 > Bjørn Augestad writes:
 >
 >>Some sales pitches for clc_strrev. :-)
 >
 >
 > OK.
 >
 > In that case, here is a smaller version.  I added clc_strnrev() too
 > since I imagine that can be useful sometimes, or at least faster when
 > we already know strlen.

How about renaming clc_strnrev() to clc_memrev()? That's what it does 
and suddenly we have a general purpose memory reverser. It can even be 
used to swap bytes in integers.


 >
 > BTW, I did as you said and replaced #include "clc_*" with <clc_*>:-)

That was in the user documentation. We must use "" to avoid that we 
include headers from a previous version already installed. Nice try, 
though ;-)

 >
 > Note: The 'if(len)' is a very slight optimization, it can be omitted
 > (and the end-- moved) at the price of an extra loop for uneven-length
 > strings.
 >
 > /* $Id$ */
 > /*
 >  * Copyright(c) 2003, Hallvard B Furuseth <h.b...@us...>

I compared the two versions and optimized mine a little. Turned out that 
mine was slightly faster on a 600MHz Pentium running Linux. Then I tried 
it on an Athlon 1800+ running Windows XP. Then your version was faster. 
Finally Windows *crashed*. :-( Oh well, your version looks better so we 
keep that one. Please update libclc/src/string/clc_strrev.c

Speaking of optimizations, any opinions on code like this?

int clc_ultostr(char *ptr, size_t size, unsigned long num, int base)
{
     const char *sym = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
     char* sp;

#ifdef CLC_FAST
     if(base == 2) {
         sp = ptr;
         size--;
         do {
             *sp++ = num & 0x01 ? '1' : '0';
             num >>= 1;
         } while(num > 0 && size-- > 0) ;

         if(num == 0) {
             *sp = '\0';
             clc_strrev(ptr);
         }

         return num == 0;
     }
#endif
    .... /* regular version goes here */
}

The CLC_FAST version is 4 times faster than the regular one, but the 
library will be bigger. Nice for those who likes to print bitpatterns?

Strangely enough division by bitshifting is still faster than / 2. (gcc 
3.2.1 -O3 -NDEBUG)

-- 
boa

Please join the libclc-developers list
at http://lists.sourceforge.net/lists/listinfo/libclc-developers