[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 7 Apr 2013 11:23:27 +0200
From: Szabolcs Nagy <nsz@...t70.net>
To: musl@...ts.openwall.com
Subject: Re: [PATCH] String: expand to word size && refactor ||
refactor
* Nathan McSween <nwmcsween@...il.com> [2013-04-06 20:38:16 +0000]:
> diff --git a/src/string/memccpy.c b/src/string/memccpy.c
> index b85009c..f032030 100644
> --- a/src/string/memccpy.c
> +++ b/src/string/memccpy.c
> @@ -1,32 +1,36 @@
> #include <string.h>
> -#include <stdlib.h>
> #include <stdint.h>
> -#include <limits.h>
>
> -#define ALIGN (sizeof(size_t)-1)
> -#define ONES ((size_t)-1/UCHAR_MAX)
> -#define HIGHS (ONES * (UCHAR_MAX/2+1))
> -#define HASZERO(x) ((x)-ONES & ~(x) & HIGHS)
> +#define HIGHS(x) ((x) - ((x)*0-1) / 255 * 128)
> +#define LOWS(x) (((x)*0-1) / 255)
> +#define has_char(x, c) ((x) - LOWS(x) & ~(x) & HIGHS(x) ^ LOWS(x) * (c))
the meaning of the new HIGHS is unclear
the original macros are easier to read, HIGHS is probably not
needed and i think size_t can be hardcoded, so
#define ONES ((size_t)-1/255)
#define HASZERO(x) ((x)-ONES & ~(x) & ONES*128)
a has_char can be useful for clarity (but the original code
made sure ones*c is only calculated once) and then i'd use
#define HASCHAR(x,c) HASZERO((x) ^ ONES*(unsigned char)(c))
> -void *memccpy(void *restrict dest, const void *restrict src, int c, size_t n)
> +void *memccpy(void *restrict d, const void *restrict s, int c, size_t n)
> {
> - unsigned char *d = dest;
> - const unsigned char *s = src;
> - size_t *wd, k;
> + unsigned char *cd = (unsigned char *)d;
> + const unsigned char *cs = (const unsigned char *)s;
> + size_t *wd;
> const size_t *ws;
>
> c = (unsigned char)c;
> - if (((uintptr_t)s & ALIGN) == ((uintptr_t)d & ALIGN)) {
> - for (; ((uintptr_t)s & ALIGN) && n && (*d=*s)!=c; n--, s++, d++);
> - if ((uintptr_t)s & ALIGN) goto tail;
> - k = ONES * c;
> - wd=(void *)d; ws=(const void *)s;
> - for (; n>=sizeof(size_t) && !HASZERO(*ws^k);
> - n-=sizeof(size_t), ws++, wd++) *wd = *ws;
> - d=(void *)wd; s=(const void *)ws;
> - }
> - for (; n && (*d=*s)!=c; n--, s++, d++);
> -tail:
> - if (*s==c) return d+1;
> - return 0;
> +
> + if ((uintptr_t)s % sizeof(size_t) != (uintptr_t)d % sizeof(size_t)
> + || n < sizeof(size_t))
> + goto bytewise;
> +
> + for (; ; *cd = *cs, cd++, cs++)
> + if (*cs == c) return cd + 1;
if ((*cd++ = *cs++) == c) return cd; is more idiomatic
with your code
the c at the end is not copied, this is a bug
the n limit is not accounted for, this is a bug
the code below is never reached, this is another bug
> + for (wd = (size_t *)d, ws = (const size_t *)s
> + ; !has_char(*ws, c) && n >= sizeof(size_t)
> + ; ws++, wd++, *wd = *ws);
> +
> + cd = (unsigned char *)wd;
> + cs = (const unsigned char *)ws;
> +
> +bytewise:
> + for (; *cs != c; *cd = *cs, cs++, cd++, n--)
> + if (!n) return NULL;
> +
> + return cd + 1;
> }
i think you should test and benchmark before submitting
such code
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
