Re: musl 0.9.5 release and new website

[Rich Felker <dalias@aerifal.cx>]

On Sat, Sep 15, 2012 at 03:53:41PM +0200, Szabolcs Nagy wrote:
> > Post-release priorities are adding the last of the missing Linux
> > syscalls, md5 crypt, and the long-elusive scheduling priority
> 
> i attached an md5 crypt implementation
> 
> i set the key limit to 30000 but that
> can be lowered
> 
> i'm not sure about the copyright notice in the middle
> as the code was rewritten from scratch so nothing
> was actually copied
> but the old bsd code is the specification itself
> and it is copyrighted:

The license is 2-clause BSD so it doesn't really matter.

> static uint32_t rol(uint32_t n, int k) { return (n << k) | (n >> (32-k)); }
> #define F(x,y,z) (z ^ (x & (y ^ z)))
> #define G(x,y,z) (y ^ (z & (y ^ x)))
> #define H(x,y,z) (x ^ y ^ z)
> #define I(x,y,z) (y ^ (x | ~z))
> #define FF(a,b,c,d,w,s,t) a += F(b,c,d) + w + t; a = rol(a,s) + b
> #define GG(a,b,c,d,w,s,t) a += G(b,c,d) + w + t; a = rol(a,s) + b
> #define HH(a,b,c,d,w,s,t) a += H(b,c,d) + w + t; a = rol(a,s) + b
> #define II(a,b,c,d,w,s,t) a += I(b,c,d) + w + t; a = rol(a,s) + b
> 
> [...]
> static void processblock(struct md5 *s, const uint8_t *buf)
> {
> 	uint32_t i, W[16], a, b, c, d;
> 
> 	for (i = 0; i < 16; i++) {
> 		W[i] = buf[4*i];
> 		W[i] |= (uint32_t)buf[4*i+1]<<8;
> 		W[i] |= (uint32_t)buf[4*i+2]<<16;
> 		W[i] |= (uint32_t)buf[4*i+3]<<24;
> 	}
> 
> 	a = s->h[0];
> 	b = s->h[1];
> 	c = s->h[2];
> 	d = s->h[3];
> 
> 	i = 0;
> 	while (i < 16) {
> 		FF(a,b,c,d, W[i],  7, tab[i]); i++;
> 		FF(d,a,b,c, W[i], 12, tab[i]); i++;
> 		FF(c,d,a,b, W[i], 17, tab[i]); i++;
> 		FF(b,c,d,a, W[i], 22, tab[i]); i++;
> 	}

This is more of the same old ugly manual unrolling. The file is small
as-is, but I think it could be a lot smaller with -Os (and same speed
as now with -O3) if the manual unrolling were removed.

I'm going to commit as-is though so I can include it in 0.9.6 (see the
other message I'm about to post).

Rich