Re: Efficient output for integer types

On Sat, Sep 21, 2019 at 03:36:21AM +0100, Andrew Gierth wrote:
> >>>>> "David" == David Fetter <david@fetter.org> writes:
> 
>  David> +    /* Compute the result string. */
>  David> +    if (value >= 100000000)
>  David> +    {
>  David> +        const    uint32 value2 = value % 100000000;
>  David> +
>  David> +        const uint32 c = value2 % 10000;
>  David> +        const uint32 d = value2 / 10000;
>  David> +        const uint32 c0 = (c % 100) << 1;
>  David> +        const uint32 c1 = (c / 100) << 1;
>  David> +        const uint32 d0 = (d % 100) << 1;
>  David> +        const uint32 d1 = (d / 100) << 1;
>  David> +
>  David> +        char *pos = a + olength - i;
>  David> +
>  David> +        value /= 100000000;
>  David> +
>  David> +        memcpy(pos - 2, DIGIT_TABLE + c0, 2);
>  David> +        memcpy(pos - 4, DIGIT_TABLE + c1, 2);
>  David> +        memcpy(pos - 6, DIGIT_TABLE + d0, 2);
>  David> +        memcpy(pos - 8, DIGIT_TABLE + d1, 2);
>  David> +        i += 8;
>  David> +    }
> 
> For the 32-bit case, there's no point in doing an 8-digit divide
> specially, it doesn't save any time. It's sufficient to just change
> 
>  David> +    if (value >= 10000)
> 
> to              while(value >= 10000)

Done.

> in order to process 4 digits at a time.
> 
>  David> +        for(int i = 0; i < minwidth - len; i++)
>  David> +        {
>  David> +            memcpy(str + i, DIGIT_TABLE, 1);
>  David> +        }
> 
> Should be:
>                         memset(str, '0', minwidth-len);

Done.

Best,
David.
-- 
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate