NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/sys/amiga/programmer/2552

Internet Message Format | 1996-08-05 | 3.4 KB

Path: grafix.xs4all.nl!john.hendrikx
Date: Wed, 31 Jan 96 21:00:02 GMT+1
Newsgroups: comp.sys.amiga.programmer
Distribution: world
Subject: Re: Amiga doesn`t need Planar!
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
From: [e-mail] (John Hendrikx)
Message-ID: <[e-mail]>
Organization: Grafix Attack BBS Holland

In a message of 30 Jan 96 Michael Van Elst wrote to All:

 >> So, what do you think is likely nowadays?  Memory is slow these days.

 MVE> As I recall your average SVGA system has faster memory than your CPU.

Yes, but a planar system needs more memory accesses on average because of more
severe alignment restrictions than Chunky, but also because the memory it needs
to access isn't located closely together like with Chunky.  In other words
Planar can't take advantage of wider memory-busses as good as Chunky can.

 >> That would have to be a pretty wide line,

 MVE> Why ?

 >> and you'd better hope it doesn't
 >> overlap a WORD or LONG boundary (or whatever width is used to access the
 >> gfx-memory).

 MVE> Why ? As you know there is already a funnel shifter.

Okay here's why, I'll replace the line with thin rectangles to make it easier
to explain:

You draw a rectangle at (0,0)-(15,99) on a 16-bit orientated planar display of
8 bitplanes deep (the best case).  This requires 8x100 WORD accesses (only
writes as I assume there is masking hardware).

You draw a rectangle at (1,0)-(16,99) on a 16-bit orientated planar display of
8 bitplanes deep (a bit worse case).  This requires 2x8x100 WORD accesses (as
it crosses a memory word boundary, no funnel shifter is gonna fix that
problem).

You draw a rectangle at (15,0)-(18,99) (only 4 pixels wide) on a 16-bit
orientated planar display of 8 bitplanes deep (a very bad case).  This requires
2x8x100 WORD accesses.

Now chunky:

First the best case:  You draw a rectangle at (0,0)-(15,99) on a 8-bit chunky
display with 16-bit memory bus.  This requires 8x100 WORD accesses

And the worst case:  You draw a rectangle at (1,0)-(16,99) on a 8-bit chunky
display with 16-bit memory bus.  This requires 9x100 WORD accesses.

And finally the case with the 4 pixels wide rectangle, at its worst case in
chunky:  You draw a rectangle at (15,0)-(18,99) on a 8-bit chunky display with
16-bit memory bus.  This requires 3x100 WORD accesses.

As you can see the best planar gets is equal speed, but only under very, VERY
specific circumstances.  Chunky wins hands down in the last case however.  So
as I said, the line would have to be very thick to get good speeds on planar.
Drawing a horizontal line would of course be better on planar (although still
equally fast in the best case), but as in that case the speed is about the same
it is not really interesting.

 >> The
 >> only chance Planar has to be faster is when using display-depths Chunky
 >> does not directly support (ie, 5-bit deep)

 MVE> You are still thinking about one CPU that does all.

The point is that if it takes more memory accesses to do a specific action than
no matter what hardware you throw at it the number of memory accesses will
remain higher.

Grtz John

-----------------------------------------------------------------------
 [e-mail]   TextDemo/FastView/Etc... development
-----------------------------------------------------------------------
-- Via Xenolink 1.981, XenolinkUUCP 1.1