Re: 19980418 on IRIX-6.5

Re: 19980418 on IRIX-6.5

[Martin Knoblauch]

Hi Jim,

 yes, I suspected something like that. The
assembly code sequence looks as if it is doing
"the right thing", but the register used are
totally bogus. 

 The funny thing is that the 4 other complex
tests seem to run OK, but that may be just by
chance.

Martin
===
------------------------------------------------------
Martin Knoblauch
email: knobi@rocketmail.com or knobi@sirius-cafe.de
www:   http://home.sirius-cafe.de/~knobi







---Jim Wilson <wilson@cygnus.com> wrote:
>
> 	 The testsuites failed until I inserted "-lpthread"
> 	into the "*lib" section of the specs file. Is
> 	there a better place for it?
> 
> It would go in LIB_SPEC in the mips/config/iris6.h
file.  It perhaps
> needs to be conditional on the --enable-threads
option somehow, since
> we don't necessary want to link it in by default.
> 
> 	 After that I got execution failures from
complex-5.c
> 	(explicit call to abort)
> 
> This fails on all 64 bit machines.  It is a
fundamental problem with the
> complex type support.
> 
> Jim
> 
> 

_________________________________________________________
DO YOU YAHOO!?
Get your free @yahoo.com address at http://mail.yahoo.com

Re: 19980418 on IRIX-6.5

[Jim Wilson]

	 The funny thing is that the 4 other complex
	tests seem to run OK, but that may be just by
	chance.

The specific problem here, if you care, is that complex foo fails if foo
is less than or equal to half the word size.  This is because the code that
accesses the individual parts of a complex value fails if both parts end up
in the same register.  Hence complex float fails on 64 bit machines.
The same problem can be seen on 32 bit machines by writing a testcase that
uses complex short.

Jim

Re: 19980418 on IRIX-6.5

[Jim Wilson]

	 The testsuites failed until I inserted "-lpthread"
	into the "*lib" section of the specs file. Is
	there a better place for it?

It would go in LIB_SPEC in the mips/config/iris6.h file.  It perhaps
needs to be conditional on the --enable-threads option somehow, since
we don't necessary want to link it in by default.

	 After that I got execution failures from complex-5.c
	(explicit call to abort)

This fails on all 64 bit machines.  It is a fundamental problem with the
complex type support.

Jim

Re: 19980418 on IRIX-6.5

[Jeffrey A Law]

  In message < 19980421075151.20772.rocketmail@web4.rocketmail.com >you write:
  >  Lots of warnings from the stage2 and stage3
  > cycles. Is someone actually working on getting
  > rid of them ? :-)
Yes.  If you think there's lots now, you should have been around
when we first turned extra warnings on :-) :-)

  >  After that I got execution failures from complex-5.c
  > (explicit call to abort) and compilation failures
  > (internal compiler error) from some c++ files. Iwill
  > post a bug report on those later.
I believe complex-5 and a few other complex related tests are
expected to fail on mips & alpha platforms.

jeff

19980418 on IRIX-6.5

[Martin Knoblauch]

Hi,

 19980418 built successfully on the current
IRIX-6.5 beta. Configured with "--with-stabs",
"--enable-haifa" and "--enable-threads=posix".

 Lots of warnings from the stage2 and stage3
cycles. Is someone actually working on getting
rid of them ? :-)

 The testsuites failed until I inserted "-lpthread"
into the "*lib" section of the specs file. Is
there a better place for it?

 After that I got execution failures from complex-5.c
(explicit call to abort) and compilation failures
(internal compiler error) from some c++ files. Iwill
post a bug report on those later.

Martin
===
------------------------------------------------------
Martin Knoblauch
email: knobi@rocketmail.com or knobi@sirius-cafe.de
www:   http://home.sirius-cafe.de/~knobi







_________________________________________________________
DO YOU YAHOO!?
Get your free @yahoo.com address at http://mail.yahoo.com