Re: [PATCH 5/7] inflate pt1: cleanup Huffman table code

[Matt Mackall <mpm@selenic.com>]

On Fri, Feb 24, 2006 at 01:52:05PM -0800, John Reiser wrote:
> > Index: 2.6.16-rc4-inflate/lib/inflate.c
> > ===================================================================
> > --- 2.6.16-rc4-inflate.orig/lib/inflate.c	2006-02-22 17:16:07.000000000 -0600
> > +++ 2.6.16-rc4-inflate/lib/inflate.c	2006-02-22 17:16:08.000000000 -0600
> > @@ -117,12 +117,12 @@
> >     an unused code.  If a code with e == 99 is looked up, this implies an
> >     error in the data. */
> >  struct huft {
> > -	u8 e;			/* number of extra bits or operation */
> > -	u8 b;			/* number of bits in this code or subcode */
> >  	union {
> > -		u16 n;		/* literal, length base, or distance base */
> > -		struct huft *t;	/* pointer to next level of table */
> > -	} v;
> > +		u16 val; /* literal, length base, or distance base */
> > +		struct huft *next; /* pointer to next level of table */
> > +	};
> > +	u8 extra; /* number of extra bits or operation */
> > +	u8 bits; /* number of bits in this code or subcode */
> >  };
> 
> How aggressive do you want to be?  About 3.7 years ago in
> http://freshmeat.net/projects/gzip_x86/  I published:
> -----
> typedef unsigned short huft_ndx;  /* index>>1 */
> struct huft {
>   uch e;                /* number of extra bits or operation */
>   uch b;                /* number of bits in this code or subcode */
>   union {
>     ush n;              /* literal, length base, or distance base */
>     huft_ndx t;         /* index>>1 of next level of table */
>   } v;
> };
> -----
> which makes 4==sizeof(struct huft) and enables manipulation by 32-bit
> fetch, exposing 'e' [extra] and 'b' [bits] in x86 byte registers
> for fewer shift-and-mask operations.  The struct huft can be managed
> as a stack of maximum length 1014.

Interesting. The current stuff is mostly obvious cleanups, but I'd
consider something like that after I get the current queue merged.

-- 
Mathematics is the supreme nostalgia of our time.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/