So I've run into a couple conflicts while merging this, I'm working on
fixing them now.

One of them involves 2 of my patches, one in which I cleared LDFLAGS in
testsuite/test-code-or1k/configure.ac, and another where I clear CCASFLAGS
(also in that file).

Jeremy, you asked in another message why I cleared these flags.  My
reasoning is this:  First, we're already clearing CFLAGS in that file.
 Second, there can actually be problems if LDFLAGS *aren't* cleared,
because the LDFLAGS get carried down from the top-level configure.  *Those*
LDFLAGS are likely intended to be used *only* when compiling the simulator,
and not when cross-compiling the testsuite code.  One potential problem is
the inclusion of -L flags while compiling the testsuite, which may end up
causing the linker to attempt to link against a library native to the host,
which causes the compile to fail.  The reason I cleared CCASFLAGS was
because (up until recently) I compiled or1ksim with -m32 in my CFLAGS.
 Since CCASFLAGS is initialized from CFLAGS, this caused -m32 to be passed
to cross-compile commands, when it shouldn't be.

In summary, the problem at the moment is there's no way to specify which
CFLAGS/LDFLAGS/CCASFLAGS are to be used when compiling the simulator, and
which are to be used when compiling the testsuite.  Some values for these
flags can break the compile.  I thought it was best to force some safe and
consistent flags when compiling the testsuite.

Incidentally, I fixed the bug that was leading me to add -m32 to my
compiles (the "<invalid>" printout for some instructions when tracing).  I
tracked that down to the insn_extract function, which sometimes will
sign-extend an intermediate value when it shouldn't be.  That fix is in
another patch in my repo on the or1k-master branch.

-Pete




On Fri, Apr 5, 2013 at 1:27 PM, Jeremy Bennett
<[email protected]>wrote:

> Hi Peter,
>
> Thanks for all this. This looks good. Please commit them upstream (or
> send me a pull request).
>
> The other thing I would ask is if you can merge them into the OR32
> branch. I know it's not sexy, but most ordinary users are still using
> the or32 version, and divergence will just store up nightmares for the
> future. I think these commits are all clear of the or32/or1k divide, so
> should just apply cleanly.
>
> Many thanks,
>
>
> Jeremy
>
> On Thu, 2013-04-04 at 15:34 -0400, Peter Gavin wrote:
> > Hi guys,
> >
> >
> > I just pushed some patches to my repo that implement the AECR/AESR
> > bits from the new architecture manual for the arithmetic instructions,
> > as well as a few other small things.  I updated the testsuite to
> > match, everything seems to be working correctly.
> >
> >
> > Please review :)
> >
> >
> > https://github.com/pgavin/or1ksim/commits/or1k-master
> >
> >
> >
> > -Pete
> > _______________________________________________
> > Openrisc mailing list
> > [email protected]
> > http://lists.opencores.org/listinfo/openrisc
>
> --
> Tel:      +44 (1590) 610184
> Cell:     +44 (7970) 676050
> SkypeID: jeremybennett
> Email:   [email protected]
> Web:     www.embecosm.com
>
> _______________________________________________
> OpenRISC mailing list
> [email protected]
> http://lists.openrisc.net/listinfo/openrisc
>
_______________________________________________
OpenRISC mailing list
[email protected]
http://lists.openrisc.net/listinfo/openrisc

Reply via email to