Experimental Physics and
Industrial Control System

Paul Banta <[email protected]> · Sun, 9 Jun 1996 14:08:41 -0400

>>>>> " " == Pete Jemian <[email protected]> writes:

    >> the mv162 has the 68LC040 CPU with no hardware floating point
    >> support.
    >> 
    >> To run EPICS (which uses floating point calculations), you must
    >> have a VxWorks kernel with the software floating point
    >> emmulation installed.  Epics needs to be built knowing that
    >> software FP is to be used, this is done using the -msoft-float
    >> compiler flag.
    >> 
    >> Building VxWorks for software FP support requires the
    >> fpSoft040.c file for 5.1.1 and a new library file
    >> libMC68040gnuvx.
    >> 
    >> The flag INCLUDE_SW_FP also needs to be set in configAll.h.
    >> 
    >> hope this helps,
    >> 
    >> Dave Barker.
    >> 
     > Hmmm...

     > My VxWorks v5.2 does not have a file fpSoft040.c Is this a
     > difference between 5.1.1 and 5.2 or could it be a BSP option?

No it is not the difference.  The fpSoft040.c file is a hack fix from
Wind River for the 5.1 BSP.  Also you have to run a script that extracts
the software emulation floating point routines from the (WRS supplied) 
68020 libs and puts them into the 68040 libs.

However, all that drivel is beside the point if you are using 5.2.
WindRiver now has two BSPs for the 162: mv162 and mv162lc.  The names
point out the obvious.  Additionally, there is a new set of libs to use
just for the 68LC040: libMC68LC040gnudrv.a and libMC68LC040gnuvx.a.

When we got our 5.2 upgrade from WindRiver we did not get these new BSPs.
Only when I asked for them did I get them.

     > Also, my mv162 is a MVME 162-042.  Don't they have the floating
     > point support built-in?  (It was left out on the LC to speed
     > the 68040 chip to market?)

Motorola has more versions of the 162 than flavors of coffee in one of
those trendy little coffee dives in Berkeley.  I don't remember what the
-042 signifies.  Take a look at the cpu.  It will probably have printed on it 
"68040" or "68LC040".

No matter what you have, though, it sounds like your kernel does not have
the floating point routines it needs. (EPICS uses the -msoft-float flag in
the 162 compilation by default, I think).  One way to check for the routines
is to use the nm68k utility to take a look at what is actually in the kernel.
Here is the output I get from our 5.2 release:

Kernel for the 167:
codahp{banta}56: nm68k vxWorks | grep eqdf2
codahp{banta}57: cd ../mv162lc

Kernel for the 162lc:
codahp{banta}59: !nm
nm68k vxWorks | grep eqdf2 
000717b0 T ___eqdf2

I suggest making sure you have the most up to date BSPs for the 162 from 
WindRiver.  Then if you have a 68LC040 you can use the mv162lc BSP to get
a kernel with the software emulation floating point routines.  If you have
a 68040, then you can rebuild EPICS without the -msoft-float flag for the
162.  Since you're at ANL you should probably talk to John Winans.  I
bet he has this sorted out for ANL.

(Then have a nice coffee in honor of the 162.  Any flavor would do but
Hazelnut is a pleasant choice.)

Good Luck,

-- 
Paul Banta                              voice: 804-249-5829
CEBAF, MS 12H                             fax: 804-249-5800
12000 Jefferson Avenue                  email: [email protected]
Newport News, VA  23606

(the lab (I mean, "facility") formerly known as CEBAF)

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System