towards a faster isdigit()

[Paul Eggert]

steve at taumet.com (Stephen Clamage) writes:

>The macro you suggest requires a test and jump, and on some modern RISC
>machines the penalty is very high.  For example, I tried a loop which
>only tested for isdigit and incremented a global based on the test.  On
>a Sun-4 (SPARC), the program using your macro took 44% longer than the one
>using the usual ctype macro.  On a DECStation (MIPS), your macro took
>46% longer.

That's odd.  I tried several systems and got speedups for the proposed
isdigit() every time, if anything even more for RISC than CISC.  E.g.:

    prop    trad    speedup
     2.5     4.5    1.8     Sparcstation 1 SunOS 4.1.1 cc -O
     3.6     5.4    1.5     DECstation 3100 Ultrix V4.0 (Rev. 179) cc -O
     9.7    13.3    1.4     Sun-3/260 SunOS 4.1 cc -O

prop = time with proposed ``#define isdigit(c) ((unsigned)((c)-'0') < 10))''
trad = time with traditional isdigit() in <ctype.h>
speedup = trad/prop = how much faster the proposed isdigit() is

All times are the sum of the user+system CPU time in seconds,
and are measured with the command `time ./a.out 10000000' under csh.
Here's the program I used.

	#include <stdio.h>

	#if traditional
	#   include <ctype.h>
	#else
	#   define isdigit(c) ((unsigned)((c) - '0')  <  10)
	#endif

	main(argc, argv)
		int argc;
		char **argv;
	{
		char *p = argv[1];
		int i = atoi(p), c = *p;
		while (0 <= --i)
			if (isdigit(c))
				c = ++*p;
		printf("%s\n", p);
		return 0;
	}