mips-sgi-irix6.2 RTL check testsuite failures

[Kaveh R. Ghazi]

 > From: Clinton Popetz <cpopetz@cygnus.com>
 > 
 > On Wed, Jul 12, 2000 at 06:35:38PM -0500, Clinton Popetz wrote:
 > > On Wed, Jul 12, 2000 at 10:32:20AM -0400, Kaveh R. Ghazi wrote:
 > > > The following are testsuite failures exposed by RTL checking on
 > > > mips-sgi-irix6.2:
 > > > gcc.c-torture/execute/struct-ret-1.c:40: RTL check: expected code `reg', have `parallel'
 > > > gcc.c-torture/unsorted/s.c:24: RTL check: expected code `reg', have `parallel'
 > > > gcc.c-torture/unsorted/structret.c:26: RTL check: expected code `reg', have `parallel'
 > > > g++.old-deja/g++.pt/alignof.C:14: RTL check: expected code `reg', have `parallel'
 > > > g++.old-deja/g++.pt/t21.C:4: RTL check: expected code `reg', have `parallel'
 > > > g++.old-deja/g++.pt/t21.C:4: Internal compiler error in `mips_expand_prologue', at config/mips/mips.c:7007
 > > > g++.old-deja/g++.pt/t28.C:4: RTL check: expected code `reg', have `parallel'
 > > > g++.old-deja/g++.pt/t28.C:4: Internal compiler error in `mips_expand_prologue', at config/mips/mips.c:7007
 > > 
 > > These were all from the same bug...the Irix abi has a bizarre
 > > requirement to break up an argument struct into chunks when it
 > > contains a double, and mips_expand_prologue couldn't handle this.
 > 
 > I decided against this patch, because we don't need the information
 > the loop is calculating when compiling for that ABI anyway.  This
 > patch make us only calculate the information when we need it (but it
 > also does a sanity check when we are calculating it.)


Well, I tested your first patch over last night and it worked great. :-)
But you're withdrawing it so....

I'll try your replacement as well as the fix for 950612-1.c and let
you know tomorrow how it does.

		--Kaveh
--
Kaveh R. Ghazi			Engagement Manager / Project Services
ghazi@caip.rutgers.edu		Qwest Internet Solutions