Re^3: Simple arithmetic? (And the winner is ... )

And no sooner have I said that, and I think of a way to use _BitScanForward64() that improves upon the original by 70+%:

C:\test\C>gcm
anonyM: gcm for  s=2147483648 & r=1 to 1073741824 took:8.535316994670
anonyM: gcm2 for s=2147483648 & r=1 to 1073741824 took:2.271266720696
[download]

Pre-masking the scanned var, avoids the later subtraction being conditional:

U64 gcm2( U64 max, U64 lcm ) {
    I32 b;

    _BitScanForward64( &b, lcm & 0xfff );
    lcm <<= ( 12 - b );
    return ( max / lcm ) * lcm;
}
[download]

With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority". I'm with torvalds on this

In the absence of evidence, opinion is indistinguishable from prejudice. Agile (and TDD) debunked

Comment on Re^3: Simple arithmetic? (And the winner is ... ) Select or Download Code

Replies are listed 'Best First'.
Re^4: Simple arithmetic? (And the winner is ... ) by oiskuu (Hermit) on Mar 09, 2015 at 21:52 UTC
Fixed version with a quick demonstration of the problem: #include <stdio.h> #include <stdint.h> #include <stdlib.h> uint64_t gcm2( uint64_t max, uint64_t lcm ) { int b = __builtin_ctzl( lcm & 0xfff ); lcm <<= ( 12 - b ); return ( max / lcm ) * lcm; } uint64_t gcm(uint64_t max, uint64_t lcm) { lcm <<= 12 - __builtin_ctzl(lcm \| 0x1000); return max - (max % lcm); } int main(int argc, char *argv[]) { unsigned long max = 12345 << 10; while (argc-- > 1) { unsigned long v = strtoul(argv[argc], NULL, 10); printf("gcm2(%lu,%lu) = %lu\n", max, v, gcm2(max, v)); printf("gcm(%lu,%lu) = %lu\n", max, v, gcm(max, v)); } return 0; } [download] $ ./a.out 12 77 4096 gcm2(12641280,4096) = 0 gcm(12641280,4096) = 12640256 gcm2(12641280,77) = 12615680 gcm(12641280,77) = 12615680 gcm2(12641280,12) = 12632064 gcm(12641280,12) = 12632064	[reply] [d/l]

Replies are listed 'Best First'.

Re^4: Simple arithmetic? (And the winner is ... )
by oiskuu (Hermit) on Mar 09, 2015 at 21:52 UTC

Fixed version with a quick demonstration of the problem:

#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>

uint64_t gcm2( uint64_t max, uint64_t lcm ) {

    int b = __builtin_ctzl( lcm & 0xfff );
    lcm <<= ( 12 - b );
    return ( max / lcm ) * lcm;
}

uint64_t gcm(uint64_t max, uint64_t lcm)
{
    lcm <<= 12 - __builtin_ctzl(lcm | 0x1000);
    return max - (max % lcm);
}


int main(int argc, char *argv[])
{
    unsigned long max = 12345 << 10;
    while (argc-- > 1) {
        unsigned long v = strtoul(argv[argc], NULL, 10);

        printf("gcm2(%lu,%lu) = %lu\n", max, v, gcm2(max, v));
        printf("gcm(%lu,%lu) = %lu\n", max, v, gcm(max, v));
    }
    return 0;
}
[download]

$ ./a.out 12 77 4096
gcm2(12641280,4096) = 0
gcm(12641280,4096) = 12640256
gcm2(12641280,77) = 12615680
gcm(12641280,77) = 12615680
gcm2(12641280,12) = 12632064
gcm(12641280,12) = 12632064

[reply]
[d/l]