User account creation filtered due to spam.

Bug 48841

Summary: [regression] lot more libgomp testsuite failures compared to 4.4.5
Product: gcc Reporter: Hin-Tak Leung <htl10>
Component: libgompAssignee: Not yet assigned to anyone <unassigned>
Status: RESOLVED FIXED    
Severity: normal CC: ro
Priority: P3    
Version: 4.4.6   
Target Milestone: ---   
Host: Target: alphaev68-dec-osf5.1a
Build: Known to work: 4.4.5, 4.6.1
Known to fail: 4.4.6 Last reconfirmed: 2011-05-02 10:34:32
Attachments: The full soon-to-post-to-gcc-testresults test summary

Description Hin-Tak Leung 2011-05-01 22:21:24 UTC
Created attachment 24160 [details]
The full soon-to-post-to-gcc-testresults test summary

My 4.4.5 result, on the same machine:
http://gcc.gnu.org/ml/gcc-testresults/2010-12/msg01338.html

4.4.5:
                === libgomp Summary ===

# of expected passes            2463
# of unexpected failures        9
# of unsupported tests          2

         === libgomp Summary ===

# of expected passes        1610
# of unexpected failures    466
# of unsupported tests        2

so something has gone very wrong with 4.4.6.
Comment 1 Hin-Tak Leung 2011-05-01 22:46:34 UTC
attachment posted as:
http://gcc.gnu.org/ml/gcc-testresults/2011-05/msg00074.html
after prepending with some notes.

Mentioned the issue but forgot to mention the actual bug number, but I am sure somebody will find this report if they need to.
Comment 2 Richard Biener 2011-05-02 10:34:32 UTC
Please look into the testsuite log file and investigate why we are not able
to create the executable.
Comment 3 Rainer Orth 2011-07-19 14:06:58 UTC
Apart from that, why are you wasting your time with GCC 4.4 which I don't test
any longer?  GCC 4.5 and 4.6 should be fine and have seen lots of bug fixes.

  Rainer
Comment 4 Hin-Tak Leung 2011-08-02 11:03:23 UTC
> Apart from that, why are you wasting your time with GCC 4.4 which I don't test
any longer?  GCC 4.5 and 4.6 should be fine and have seen lots of bug fixes.

4.5 does not build correctly.
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44959
Comment 5 ro@CeBiTec.Uni-Bielefeld.DE 2011-08-02 11:09:56 UTC
> --- Comment #4 from Hin-Tak Leung <htl10 at users dot sourceforge.net> 2011-08-02 11:03:23 UTC ---
>> Apart from that, why are you wasting your time with GCC 4.4 which I don't test
> any longer?  GCC 4.5 and 4.6 should be fine and have seen lots of bug fixes.
>
> 4.5 does not build correctly.
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44959

Please follow the directions I gave in that PR.  Start with a standard
configure; make setup, no bootstrap-lean4, no relative paths to the
source dir.  This works just fine for me, so instead of insisting on
trying untested paths, first determine if the regular one works.  Even
gcc 4.5 is old by now...

	Rainer
Comment 6 Hin-Tak Leung 2011-08-03 22:26:13 UTC
(In reply to comment #5)

> Please follow the directions I gave in that PR.  Start with a standard
> configure; make setup, no bootstrap-lean4, no relative paths to the
> source dir.  This works just fine for me, so instead of insisting on
> trying untested paths, first determine if the regular one works.  Even
> gcc 4.5 is old by now...

Okay... did the plain configure and make and no relative path, and watching my 4.6.1 "make -k check" - I'll be summiting the result later to the testsuite mailing list; but it is currently spilling a lot of 

--------------------
Unaligned access pid=353493 <a.28.1.exe> va=0x14000ad5e pc=0x3ff81006d00 ra=0x3ff81006c44 inst=0xb3e90000
-------------------

messages, with a different <?.exe>. and I looked, and it is currently doing make check inside libgomp . So it looks like there was some kind of alignment regression after 4.4.5 .
Comment 7 Hin-Tak Leung 2011-08-03 23:08:47 UTC
Please ignore last comment 6. With 4.6.1:

		=== libgomp Summary ===

# of expected passes		2586
# of unsupported tests		3

So the problem is not seen in 4.6.1.
Comment 8 ro@CeBiTec.Uni-Bielefeld.DE 2011-08-04 10:55:04 UTC
> Okay... did the plain configure and make and no relative path, and watching my
> 4.6.1 "make -k check" - I'll be summiting the result later to the testsuite
> mailing list; but it is currently spilling a lot of 
>
> --------------------
> Unaligned access pid=353493 <a.28.1.exe> va=0x14000ad5e pc=0x3ff81006d00
> ra=0x3ff81006c44 inst=0xb3e90000
> -------------------
>
> messages, with a different <?.exe>. and I looked, and it is currently doing
> make check inside libgomp . So it looks like there was some kind of alignment
> regression after 4.4.5 .

This is not a regression, but a bug inside librt.  Already reported (PR
libgomp/45351) and worked around for gcc 4.7.0.

	Rainer
Comment 9 ro@CeBiTec.Uni-Bielefeld.DE 2011-08-04 10:56:24 UTC
> Please ignore last comment 6. With 4.6.1:
>
>         === libgomp Summary ===
>
> # of expected passes        2586
> # of unsupported tests        3
>
> So the problem is not seen in 4.6.1.

It is, but the unaligned accesses are fixed up by the kernel, so this
just gives (admittedly ugly) warnings, no errors/failures.

	Rainer
Comment 10 Hin-Tak Leung 2011-08-05 23:44:55 UTC
Most strange -re-run 4.4.6 and got a different answer from previous
(http://gcc.gnu.org/ml/gcc-testresults/2011-05/msg00074.html) - every part of the resulting summary file is identical, except the libgomp section.
Here is the old 4.4.5, new 4.4.6, and the 4.6.1 from a few days ago:

4.4.5:
                === libgomp Summary ===

# of expected passes            2463
# of unexpected failures        9
# of unsupported tests          2

4.4.6
		=== libgomp Summary ===

# of expected passes		2511
# of unexpected failures	9
# of unsupported tests		2

4.6.1
        === libgomp Summary ===

# of expected passes        2586
# of unsupported tests        3

These numbers make a lot more sense. The machine has not been upgraded nor used all these times, so the only two changes are (1) use of absolute path in configure, (2) during this run (in 4.6.1) I noticed /tmp was a bit small (it failed quite early on with libnumber or libiberty) so set TMIDIR to a different disk.

I don't think it is the path, but it is possible I just didn't notice a not-enough temp space error; the other possibillity is some transient/intermittent error (the alignment?). Anyway, since 4.6.1 seems to work and fixes most of the regressions or what not, I am happy to let this one go.
Comment 11 ro@CeBiTec.Uni-Bielefeld.DE 2011-08-09 12:16:59 UTC
> I don't think it is the path, but it is possible I just didn't notice a
> not-enough temp space error; the other possibillity is some
> transient/intermittent error (the alignment?). Anyway, since 4.6.1 seems to

I don't think the alignment issue ever caused failures for me.

> work and fixes most of the regressions or what not, I am happy to let this one
> go.

Fine, thanks for checking.

	Rainer