Date: Fri, 3 Feb 95 14:35 MST
From: mat AT ardi DOT com (Mat Hostetter)
To: THE MASKED PROGRAMMER <badcoe AT bsa DOT bris DOT ac DOT uk>
Cc: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Re: 256x256 look up optimisation
References: <0098B6F8 DOT 8C10FFC5 DOT 12 AT bsa DOT bristol DOT ac DOT uk>

>>>>> "badcoe" == THE MASKED PROGRAMMER <badcoe AT bsa DOT bris DOT ac DOT uk> writes:

    badcoe> Sorry about the delay to this but I've been away having a
    badcoe> baby (or rather my wife has):

Congratulations!  I hope things went well.

    >> This is a much different statement than your original question,
    >> which unnecessarily brought up two dimensional arrays.

    badcoe> Surely a 256x256 array and a 64K array are the same thing
    badcoe> (unless you think of 256x256 as table[256][256] but I
    badcoe> thought I said that in the first posting (which may not be
    badcoe> the one you were reading)).

The difference is that in one case your program may have two separate
indices (one for x and one for y), which may be computed in any number
of different ways; in the other case you have an array of 16 bit
indices lying around.  You can of course transform one to the other,
but at a performance cost you can't afford.

As a reductio ad absurdum, describing the problem as "I have a sixteen
dimensional array with sixteen indices that are either 0 or 1" is
likely to elicit suggestions which are not helpful, even though this
also describes a "64K array".  If you're mapping an array of 16 bit
values to an array of 8 bit values through a lookup table, just say
so.

    >> I have a question for you: does %es refer to conventional
    >> memory (so you can access the VGA frame buffer), or is all of
    >> this taking place in PM memory?

    badcoe> Do you mean %esi ?  It's in PM memory, I think (in
    badcoe> retrospect) that all my references to RM and PM were
    badcoe> spurious.

Actually I meant %es, as in the selector.  By changing %es to refer to
conventional memory you could write directly to the VGA board and
maybe avoid an extra copy.  I was wondering if you were using stosb to
get a %es override "for free", or just to get a postincrement of %edi.

    >> I noticed a few efficiency problems with your code,

    badcoe> What were they ?

Pentium and i486 things, like AGI stalls and use of non-pairable
instructions.  I wrote up a specific list of inefficiencies, but I
don't remember if I posted it or not.  I don't know how to optimize
your code for the i386 in particular.

-Mat