Experimental Physics and
Industrial Control System

[email protected] (Steve Lewis) · Thu, 21 Nov 96 16:01:43 PST

1) The Tornado version of VxWorks has much cleaner makefiles for the 5.3
kernel--a lot like the EPICS ones, as a matter of fact.  Unfortunately,
they have re-arranged things, so the old "gnu" is now under "hosts" and
the config is under target.  Note that I am not using any of Tornado.
Just setting the "5.2" option true in configAll.h gives all the old
kernel properties back (shell, loader, etc.).

They have also broken the $HOST_ARCH.$TARG_ARCH scheme; you now have
to use "sun4-solaris2", "sun4-sunos4", and "parisc-hpux9" in the paths.
It is not clear to me how you mix two target types without an entire
copy of the (250 MB) installation.  250 MB!!  See below for snippet
from startup/Site.cshrc... add "x86-win32" if you care.

2) Looking in $(VX_DIR)/target/h/make at make.MC680X0gnu, they seem to
use a slightly different choice of cc68k options than we do in
CONFIG_ARCH.mvxxx.  In particular, they use -fno-builtin 
everywhere; -fstrength-reduce for NORMAL; "-O" for NORMAL and TARGET
up to 68040 and "-O2" for 68060.  "DRIVER" never has "-O".  Since we
do use "-O" that corresponds to their NORMAL ("application").

Without -fno-builtin, cc68k complains a lot that memcpy, fabs, and
other inlines are incompatible with "strings.h".  The gnu manual (v2.6, p.18)
says you lose a little performance.  Does WRS insist you make a
library call for all of them?  Any opinions?

EPICS does boot and run with our old options.

	--Steve

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System