[PATCH/RFA] MIPS breakage with MDMX changes

[PATCH/RFA] MIPS breakage with MDMX changes

[Jason R Thorpe]

[-- Attachment #1: Type: text/plain, Size: 1200 bytes --]

Hi Chris...

I'm seeing some new testsuite failures for gcc when used with a freshly
CVS update'd binutils ... The failure mode I'm seeing is:

/usr/local/gnu/mipseb-unknown-netbsd/bin/ld: /amd/yeah-baby/u1/hack/gnu-toolchai
n/gcc/build/mipseb-mipseb/gcc/libgcc.a(unwind-dw2.o): uses different e_flags (0x
8000002) fields than previous modules (0x2)^M

That flag appears to be:

/* Use MDMX multimedia extensions */
#define EF_MIPS_ARCH_ASE_MDMX   0x08000000  

...which my R4400 certainly does not have :-)

Gas sets that bit like so:

  if (file_ase_mdmx)
    elf_elfheader (stdoutput)->e_flags |= EF_MIPS_ARCH_ASE_MDMX;

file_ase_mdmx is set via mips_opts.ase_mdmx, which is initialized to -1,
which appears to mean that file_ase_mdmx will be set to -1 unless an
explicit -mno-mdmx is used, thus causing the bit to be erroneously set.

It looks like the mips3d stuff has the same problem (although no bits get
set in that case).  I think the default handling of ase_mips3d and ase_mdmx
should be like the attached patch.

	* config/tc-mips.c (mips_after_parse_args): Fix default handling
	of mips_opts.ase_mips3d and mips_opts.ase_mdmx.

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>

[-- Attachment #2: mips-patch --]
[-- Type: text/plain, Size: 837 bytes --]

Index: tc-mips.c
===================================================================
RCS file: /cvs/src/src/gas/config/tc-mips.c,v
retrieving revision 1.137
diff -u -r1.137 tc-mips.c
--- tc-mips.c	4 Jun 2002 16:33:55 -0000	1.137
+++ tc-mips.c	6 Jun 2002 21:44:00 -0000
@@ -10614,10 +10614,10 @@
 
   /* If the selected architecture includes support for ASEs, enable
      generation of code for them.  */
-  if (mips_opts.ase_mips3d == -1 && CPU_HAS_MIPS3D (mips_arch))
-    mips_opts.ase_mips3d = 1;
-  if (mips_opts.ase_mdmx == -1 && CPU_HAS_MDMX (mips_arch))
-    mips_opts.ase_mdmx = 1;
+  if (mips_opts.ase_mips3d == -1)
+    mips_opts.ase_mips3d = CPU_HAS_MIPS3D (mips_arch) ? 1 : 0;
+  if (mips_opts.ase_mdmx == -1)
+    mips_opts.ase_mdmx = CPU_HAS_MDMX (mips_arch) ? 1 : 0;
 
   if (file_mips_gp32 < 0)
     file_mips_gp32 = 0;

Re: [PATCH/RFA] MIPS breakage with MDMX changes

[Daniel Jacobowitz]

On Thu, Jun 06, 2002 at 02:44:31PM -0700, Jason R Thorpe wrote:
> Hi Chris...
> 
> I'm seeing some new testsuite failures for gcc when used with a freshly
> CVS update'd binutils ... The failure mode I'm seeing is:

I checked in a fix for this earlier today - and apparently Thiemo
posted one last week without being noisy about it.  We've got you
covered :)

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [PATCH/RFA] MIPS breakage with MDMX changes

[Jason R Thorpe]

On Thu, Jun 06, 2002 at 05:44:25PM -0400, Daniel Jacobowitz wrote:

 > I checked in a fix for this earlier today - and apparently Thiemo
 > posted one last week without being noisy about it.  We've got you
 > covered :)

Oh, geez, you checked it in right after I update'd.  D'oh!  Thanks :-)

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>

Re: [PATCH/RFA] MIPS breakage with MDMX changes

[cgd]

At Thu, 6 Jun 2002 14:44:31 -0700, Jason R Thorpe wrote:
> I'm seeing some new testsuite failures for gcc when used with a freshly
> CVS update'd binutils ... The failure mode I'm seeing is:
> 
> /usr/local/gnu/mipseb-unknown-netbsd/bin/ld: /amd/yeah-baby/u1/hack/gnu-toolchai
> n/gcc/build/mipseb-mipseb/gcc/libgcc.a(unwind-dw2.o): uses different e_flags (0x
> 8000002) fields than previous modules (0x2)^M

As noted, this has been fixed already.  Sorry!


I'm actually kinda puzzled by this...

i've been doing N builds (full GCC check runs, over the last few days)
using current gcc and current binutils (before Daniel put his patch
in) for mips-elf, mips64-elf, mipsisa32-elf ... and I never saw this
problem.

now i'm wondering why not!



chris

Re: [PATCH/RFA] MIPS breakage with MDMX changes

[Jason R Thorpe]

On Thu, Jun 06, 2002 at 04:55:11PM -0700, cgd@broadcom.com wrote:

 > i've been doing N builds (full GCC check runs, over the last few days)
 > using current gcc and current binutils (before Daniel put his patch
 > in) for mips-elf, mips64-elf, mipsisa32-elf ... and I never saw this
 > problem.
 > 
 > now i'm wondering why not!

Well, if your full build includes newlib, etc. then all your .o's are
going to have the same flags... but in my case, not all the .o's did
(e.g. /usr/lib/crt0.o, etc.).

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>

Re: [PATCH/RFA] MIPS breakage with MDMX changes

[cgd]

At Thu, 6 Jun 2002 17:03:36 -0700, Jason R Thorpe wrote:
> Well, if your full build includes newlib, etc. then all your .o's are
> going to have the same flags... but in my case, not all the .o's did
> (e.g. /usr/lib/crt0.o, etc.).

OK, yeah, that makes sense.


thanks,

chris

Re: [PATCH/RFA] MIPS breakage with MDMX changes

[Daniel Jacobowitz]

On Thu, Jun 06, 2002 at 04:55:11PM -0700, cgd@broadcom.com wrote:
> At Thu, 6 Jun 2002 14:44:31 -0700, Jason R Thorpe wrote:
> > I'm seeing some new testsuite failures for gcc when used with a freshly
> > CVS update'd binutils ... The failure mode I'm seeing is:
> > 
> > /usr/local/gnu/mipseb-unknown-netbsd/bin/ld: /amd/yeah-baby/u1/hack/gnu-toolchai
> > n/gcc/build/mipseb-mipseb/gcc/libgcc.a(unwind-dw2.o): uses different e_flags (0x
> > 8000002) fields than previous modules (0x2)^M
> 
> As noted, this has been fixed already.  Sorry!
> 
> 
> I'm actually kinda puzzled by this...
> 
> i've been doing N builds (full GCC check runs, over the last few days)
> using current gcc and current binutils (before Daniel put his patch
> in) for mips-elf, mips64-elf, mipsisa32-elf ... and I never saw this
> problem.
> 
> now i'm wondering why not!

Cross testsuites will not catch it, with a full build tree.  Jason saw
it because he was running either a native testsuite or a testsuite
against installed libraries - which didn't have the bogus marking.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [PATCH/RFA] MIPS breakage with MDMX changes

[Jason R Thorpe]

On Thu, Jun 06, 2002 at 08:02:04PM -0400, Daniel Jacobowitz wrote:

 > Cross testsuites will not catch it, with a full build tree.  Jason saw
 > it because he was running either a native testsuite or a testsuite
 > against installed libraries - which didn't have the bogus marking.

Yup, native.

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>