Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!psuvax1!wuarchive!sdd.hp.com!news.cs.indiana.edu!arizona.edu!uazhe0.physics.arizona.edu!zazula
From: zazula@uazhe0.physics.arizona.edu (RALPH ZAZULA)
Newsgroups: comp.sys.next
Subject: Why does my DSP seem so SLOW????
Message-ID: <1991Mar28.173226.488@arizona.edu>
Date: 28 Mar 91 22:29:56 GMT
Reply-To: zazula@uazhe0.physics.arizona.edu
Distribution: usa,local
Organization: U of Arizona Experimental Elementary Particle Physics - Tucson,
Lines: 30
Nntp-Posting-Host: uazhe0.physics.arizona.edu
News-Software: VAX/VMS VNEWS 1.3-4   Lines: 30

So I'm working on this neural-net problem and figure that the matrix
multiplicatons I'm doing would be ideally suited for the DSP.  I was
curious just how much of a speedup I would get so I did a naive timing
study using a modified
/NextDeveloper/Examples/DSP/ArrayProcessing/matrix/matrix.c

What I did was to loop over the DSPAPmtm call and the 'manual'
calculation of the matrix-multiply and used time(3) to get the
times.  With a loop of 1000 iterations, the DSP takes 21 sec
and the 030 takes 1 sec.  What??? I said...  So, maybe the loop
got optimized out???  Nope.  If I omit the -O qualifier *and*
add the loop index to the matrix elements, thus insuring that the
entire loop gets executed, I get the same results.

So why does it appear that the DSP is running 20 (twenty) times slower
than the 030/math-chip combo?

NOTE that I did not include the DSPAPWrite/ReadFloatArray calls in 
the timing loop.  These are documented by NeXT as being inefficient.

Ralph

   |----------------------------------------------------------------------|
   | Ralph Zazula                               "Computer Addict!"        |
   | University of Arizona                 ---  Department of Physics     |
   |   UAZHEP::ZAZULA                            (DecNet/HEPNet)          |
   |   zazula@uazhe0.physics.arizona.edu         (Internet)               |
   |----------------------------------------------------------------------|
   |   "You can twist perceptions, reality won't budge."  - Neil Peart    |
   |----------------------------------------------------------------------|