Re: qcam on NetBSD

To: linux-connectix@crynwr.com
Subject: Re: qcam on NetBSD
From: nelson@crynwr.com (Russell Nelson)
Date: Fri, 12 Jan 96 13:36 EST
From crynwr.com!nelson Fri Jan 12 13: 46:28 1996
In-Reply-To: <199601121620.KAA09249@lanshark.mnscorp.com>
References: <m0takbN-000G3CC@ns.crynwr.com> <199601121620.KAA09249@lanshark.mnscorp.com>

Thomas Davis writes:
 > Russ Nelson writes:
 > 
 > Where the code gets slow is the decoding of the 6bpp/uni-directional mode
 > and the 4bpp/bi-directional mode.  The nybble packing just kills things..
 > (but makes for a fast transfer..)

Problem is that the stupid 386 doesn't have a barrel shifter (shifts
any number of bits in one cycle).  Problem is the 6bpp mode has to
convert three 8-bit bytes into four 6-bit bytes, right?  So burn some
memory.  Build one 256-byte table and two 64Kbyte tables.  Index into
them using the three bytes in turn.  The 4bpp mode is even easier.

Hand-coded, optimized assembler will get you twice the speed of a C
compiler's code.

-russ <nelson@crynwr.com>    http://www.crynwr.com/~nelson
Crynwr Software   | Crynwr Software sells packet driver support | PGP ok
11 Grant St.      | +1 315 268 1925 voice | Good government is a contradiction
Potsdam, NY 13676 | +1 315 268 9201 FAX   | in terms.  Ask me for details.

Follow-Ups:

Re: qcam on NetBSD

From: Jani Mattsson <jmattsso@niksula.hut.fi>

References:

Re: qcam on NetBSD

From: nelson@crynwr.com (Russell Nelson)

Re: qcam on NetBSD

From: Thomas Davis <Thomas.Davis@mnscorp.com>

Prev: Re: qcam on DOS
Next: Re: assembly language and unrolling loops
Index: Mail Index
Thread: Mail Thread Index