jk@utastro.UUCP (John Krist) (02/08/86)
I thought that an update review of the DSI-32 board from
Definicon Systems, Inc. might be of interest to those who are
still looking into this product. I have found new things since
my original review from December. But first..
** DON'T FLAME ON ME !! I DON'T WANT TO HEAR IT !! HAVE A
QUESTION? ASK !! FEEL LIKE INSULTING AND QUOTING SALES
HANDOUTS? KEEP IT TO YOURSELF !! This is going to be a
review, and I don't want to find my mailbox filled with
blah, blah, blah about how this or that system or chip
is better and that I am stupid, as has happened before.
Whew! Ok, on to the fun stuff...
Let's talk about compilers. There has been talk about
accuracy of the compilers in this newsgroup before, so I won't
say anything on that.
The documentation doesn't include any references to
compiler switches which are there. Some of these affect the
results considerably. One (-x88) suppresses generation of slow
and wasteful code to get around old FPU bugs. This speeds up
the Fortran Whetstone by about 20%. Another (-x71) makes all
Fortran calls for single precision variables be made in single
precision (the default is to convert to double precision and
back again. Is this teh UNIX default also?). All this gives a
Whetstone time of 3.56 seconds (about 280k whetstones/sec).
The Pascal compiler seems to have trouble with the field
width specifier in the write statements. The following code,
WRITELN ( 'h':6 )
produces "hhhhhh" instead of " h". I have had very little
experience with the Pascal, so there could very well be more
bugs.
I have had not done much with the C compiler, and I have
not found any bugs.
The compilers and libraries have been updated about four
times in the past three months. Updates are $15.
It seems that much of the blame for whatever problems are
in the compilers lies with Green Hills Software. They do not
want to provide much information on the compilers (the Definicon
people found the extra switches by searching through the source
code) and do not even provide full reference manuals for the
languages (I use the manual from my Digital Research F77,
which, by the way, is the worst compilers imaginable).
There are errors in both the manual and read.me files
concerning the SVC interrupt calls to the MS-DOS shell. The
read.me file is incorrect regarding the SVC 20 call (call IBM
interrupts). The entry in the manual is correct. Interrupt
calls ARE available through the _BDOS call from any of the
languages. Also, there are calls available from any language to
move memory to and from the IBM ram and DSI ram, input and
output to a port, and make C library calls.
I have the graphics package, which supports the IBM color
graphics card and the Hercules graphics card (720x348). It is
easy to set up the memory so that many frames can be generated
and downloaded for animation purposes. The animation results
are acceptable, and I get about 8-14 frames per second (very
rough guess). The problem is that the downloading of the frame
can be seen (those familiar with the way the Apple loads in a
picture by BLOADing it know exactly how it looks, only it
happens fast but perceptibly). I suppose, though, that things
would be better on an AT, since I have the regular PC, and the
memory fetching would be faster.
The really big problem is the time it takes to generate the
pixel in memory. With the source code (in C) included, one can
see that for every pixel, the following is used :
temp = 0x2000 * (row % 4) + 90 * (row / 4) + (col / 8)
Ugh! Ack! This takes FOREVER! I'm trying to find a better way,
but I'm not too hot on pixel mapping and such. If someone out
there knows the secret, HELP! Of course, with the way the card
works, you can't have cursors.
Let's talk about support. First word of advice; don't try
calling them except between 2 pm and 3 pm California time (the
fact that they're in California readily explains this). They
don't seem to hang around long, but when you get them, they know
what they are talking about.
The bad thing is that they keep really quiet about updates,
bugs, etc. If you're lucky, they will send you updates from out
of the blue and ask for $15. Don't pay the money and they won`t
send you any more updates. Don't expect to get bug reports, and
if you can get on the semi-support BBS (I tried an hour each day
for a week to no avail) then the bug reports you will find are
the ones the customers have entered on the BBS. This is the
weakest point of the entire DSI-32 board. If you want
information, you have to call, wait 10 minutes for someone to
find the right person, and then find out that you were not sent
the already-a-month-old update. Don't write (it took 6 weeks to
get a reply by writing).
In all, I still think this is a great board, and the
compilers are superior to the ones for the PC. The only big
problem is the lack of bug reports, and this is something that
Definicon should take seriously.
John Krist
U. Texas Astronomy Dept.
jk@utastro.UUCP
{allegra,ihnp4}!{noao,ut-sally}!utastro!jk