[tremor] A possible improvement for some of the processor where tables hit hard

[marc dukette]

Monty is right there appears to be an issue(which I do not quite understand)
with shifting 32 bits, it always ends up as -1.  If you use the constant 32
it works, but with a variable it always fails.  A friend mentioned something
about shift commands not being able to do a full 32 bit shift and that it
only works with the contant value of 32 becasue the ocmpile optimizes it to
0.

This would be a nice improvement for some processors if it could be made to
work.

----- Original Message -----
From: "Nicolas Pitre" <nico at cam.org>
To: "Monty" <xiphmont at xiph.org>
Cc: <tremor at xiph.org>
Sent: Saturday, October 19, 2002 10:02 PM
Subject: Re: [tremor] A possible improvement for some of the processor where
tables hit hard

<p>> On Wed, 16 Oct 2002, Monty wrote:
>
> > On Sun, Oct 13, 2002 at 04:01:56PM -0400, marc dukette wrote:
> > > I have replaced the mask table in bitwise.c with the following, and
while it
> > > has little to no effect on an SA processor it does appear to help the
> > > version we have running on MIPS, so it may be of value to other
processors
> > > as well:
> > >
> > > old:
> > >   unsigned long m= mask[bits];
> > >
> > > new:
> > >   unsigned long m= (((unsigned int)-1)>>(32-bits));
> > >
> > > and of course you can now remove the mask table.  I'm just throwing
this
> > > out, not sure if it's worth adding to the standard or not.
> >
> > Good basic idea, but not equivalent.  In the case where bits==0...
>
> What's wrong here?  0xffffffffUL >> (32 - 0) should be 0 like the lookup
> table.
>
> Or what about: unsigned long m = (1 << bits) - 1;
>
> The ARM assembly for this could be just this:
>
> mov r1, #1
> rsb r1, r1, r1, lsl r0
>
> where r0 = bits and r1 = m.  This would certainly be faster than a table
> lookup, but I doubt any compiler would generate that by itself though.
>
>
> Nicolas
>
> --- >8 ----
> List archives:  http://www.xiph.org/archives/
> Ogg project homepage: http://www.xiph.org/ogg/
> To unsubscribe from this list, send a message to 'tremor-request at xiph.org'
> containing only the word 'unsubscribe' in the body.  No subject is needed.
> Unsubscribe messages sent to the list will be ignored/filtered.

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'tremor-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.