ravi@mcnc.UUCP (03/22/87)
[]
I was sent some neat info on Fortran compilers for the ST, which
Moshe suggested I post; I think it'll be useful to those (like me) who
are interested in Fortran compilers for the ST but had no real info.
Existing fortran users probably know this, but I think it's worth
posting for the sake of Those Not in the Know..
What follows is a condensed (!) form of messages sent to me by Wes
Cobb at Brandeis. The comments on Absoft and Prospero Fortrans are by
Wes. The benchmarks were run by him and Ulrich Lang of the Univ. of
Stuttgart. Thanks guys!
-ravi
ps: My comments in []
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
To whet your apetite (heh heh), an intro:
" we are all currently using absoft fortrn v2.2 since it seems to be
almost 100% source compatible with vax fortran and it also kicks the
living daylights out of everything else we've seen for the st.
absofts fortran whetstone: 65,000
prospero fortran whetsone: 56,000
megamax whetstone : 14,000
lattice whetstone : 7,000
atari basic whetstone : 800 "
(Hit n now to skip the rest)
ABSOFT:
------
" compiler offers complete support for all of the atari capabilities:
ports, mouse, joystick, graphics, printer, etc.
* it is VERY small - the largest `pass` of the compiler is only 46 K.
and the entire language PLUS libraries will EASILY fit on one single
sided floppy ( it takes 221 K ).
* it allows one to compile fortran programs to as68 source code for
easy interfacing with c programs.
* it has a full-screen debugger.
* it allows BOTH static and dynamic linking. the dynamic linked code
is *tiny* and ( when run from ramdisk ) is almost as fast as the static
linked code.
* it works with all ramdisk programs that i know of.
* the Absoft compiler supports F77 and F66 and also has many extensions
which are part of the upcoming F8x standard.
most vax-fortran code compiles AS IS under the F77 dialect. the
F8x extensions include: dowhile, case-switch, argv-argc, facilities
for pointer manipulations, and various string handling
extensions.
there are a couple of things i dont like about the compiler -
1. the librarian provided is not terribly smart. it doesnt look
backwards for references so you sometimes have to link to a
library 2 or 3 times if you have been silly about loading
the modules into it.
2. it is not compatible with [Breckmeyer's Multitasking] shell
v1.0 ( i dont know about v1.1 )
[compiler seems to work but crashes on exiting]
the people .. at absoft say they already have the driver code
for use with the 68881 - they are just waiting for someone to
produce an ST floating point board.
[Well, out there? You guys been bad or something, or did Santa
get held up last year? How come no-one's made one of these yet?
I guess there isn't any place to put it though.. Is there any
hope for one? Anyone know of someone making one?]
PROSPERO:
----------
Prospero is a very literal implementation of the ANSI F77 standard
with NONE of the usual extensions one finds on VAXae and mainframes.
it is so minimal that it doesnt even have do-enddo - tho with your
ratfor preprocessor this may not be a relevant worry... it does NOT
allow dynamic linking. it does NOT have LOC or BIT manipulation
functions. it only allows 6 chars in names. it only allows UPPER case.
[doesn't sound that hot.. not worth bothering about?]
***********************************************************************
benchmark data [sources are available]:
vax 8600 atari st atari st
vax fortran absoft fortran prospero F77
optimized < dynamic > < static >
benchmark seconds seconds seconds
--------- --------- --------- ---------
time1 2.09 210.0 320
time2a 3.78 405.0 552
time2b 4.02 409.4 560
time3 2.16 86.4 164
time4 0.36 16.5 12
time5 5.43 112.3 88
time6 5.19 73.3 46
whetstone "4160000" "65000" ++ "56250" ( for I = 99 )
trig_test 4.69 303.3 290
math_test 2.18 206.6 224
loop_test 3.56 284.9 472
flops 1060000 10380 - ( no answer yet )
++ On the Whetstone benchmark Absoft is somewhat faster in F66 ( 65000 )
than it is in F77 ( 57500 ).
[ Whew. ]