[Bug ada/53974] New: Nearlyt all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[Bug ada/53974] New: Nearlyt all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

             Bug #: 53974
           Summary: Nearlyt all tests fail with ADA.CALENDAR.TIME_ERROR :
                    a-calend.adb:603
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: ada
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: danglin@gcc.gnu.org
              Host: hppa2.0w-hp-hpux11.11
            Target: hppa2.0w-hp-hpux11.11
             Build: hppa2.0w-hp-hpux11.11


=== acats tests ===
Running chapter a ...
splitting /test/gnu/gcc/objdir/gcc/testsuite/ada/acats/tests/a/a22006b.ada
into:   a22006b.adbBUILD a22006b.adb
gnatmake --GCC="/test/gnu/gcc/objdir/gcc/xgcc -B/test/gnu/gcc/objdir/gcc/"
-gnatws -O2 -I/test/gnu/gcc/objdir/gcc/testsuite/ada/acats/support a22006b.adb
-largs
 --GCC="/test/gnu/gcc/objdir/gcc/xgcc -B/test/gnu/gcc/objdir/gcc/"
/test/gnu/gcc/objdir/gcc/xgcc -c -B/test/gnu/gcc/objdir/gcc/ -gnatws -O2
-I/test/gnu/gcc/objdir/gcc/testsuite/ada/acats/support a22006b.adb
gnatbind -I/test/gnu/gcc/objdir/gcc/testsuite/ada/acats/support -x a22006b.ali
gnatlink a22006b.ali -O2 --GCC=/test/gnu/gcc/objdir/gcc/xgcc
-B/test/gnu/gcc/obj
dir/gcc/
RUN a22006b


raised ADA.CALENDAR.TIME_ERROR : a-calend.adb:603
FAIL:   a22006b

# ./xgcc -B./ -v
Reading specs from ./specs
COLLECT_GCC=./xgcc
COLLECT_LTO_WRAPPER=./lto-wrapper
Target: hppa2.0w-hp-hpux11.11
Configured with: ../gcc/configure --with-gnu-as --with-as=/opt/gnu/bin/as
--enable-shared --with-local-prefix=/opt/gnu --prefix=/opt/gnu/gcc/gcc-4.8
--with-gmp=/opt/gnu/gcc/gmp --enable-threads=posix --enable-debug=no
--disable-nls --disable-lto --without-cloog --without-ppl
--enable-languages=c,c++,objc,fortran,java,ada,obj-c++
Thread model: posix
gcc version 4.8.0 20120713 (experimental) [trunk revision 189459] (GCC)

[Bug ada/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Nearly all tests fail with  |[4.8 Regression] Nearly all
                   |ADA.CALENDAR.TIME_ERROR :   |tests fail with
                   |a-calend.adb:603            |ADA.CALENDAR.TIME_ERROR :
                   |                            |a-calend.adb:603

--- Comment #1 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-15 15:20:30 UTC ---
Revision 189245 was ok.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-07-18
                 CC|                            |ebotcazou at gcc dot
                   |                            |gnu.org
          Component|ada                         |target
     Ever Confirmed|0                           |1

--- Comment #2 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-07-18 10:15:23 UTC ---
I have seen the same error on SPARC.  The fix is

        * config/sparc/sparc.md (adddi3_insn_sp32): Add earlyclobber.

so you might want to try something similar on line 4923 of pa.md.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #3 from dave.anglin at bell dot net 2012-07-19 00:33:48 UTC ---
On 18-Jul-12, at 6:15 AM, ebotcazou at gcc dot gnu.org wrote:

> I have seen the same error on SPARC.  The fix is
>
>        * config/sparc/sparc.md (adddi3_insn_sp32): Add earlyclobber.
>
> so you might want to try something similar on line 4923 of pa.md.


Eric, thanks very much!  Even if this doesn't fix the bug, this is  
obviously
a bug.

Dave
--
John David Anglin    dave.anglin@bell.net

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #4 from dave.anglin at bell dot net 2012-07-20 12:16:37 UTC ---
On 18-Jul-12, at 6:15 AM, ebotcazou at gcc dot gnu.org wrote:

> I have seen the same error on SPARC.  The fix is
>
>        * config/sparc/sparc.md (adddi3_insn_sp32): Add earlyclobber.
>
> so you might want to try something similar on line 4923 of pa.md.


I don't think there is an earlyclobber problem in the insn at 4923.

I reviewed pa.md for missing earlyclobbers and found a few.   Some
are likely not problems because of alternative ordering.  In any case,
fixing them didn't change the situation.

I presume the problem was introduced with this change:

2012-05-15  Hristian Kirtchev  <kirtchev@adacore.com>

         * a-calend.adb (Day_Of_Week): The routine once again treats

The Month value extracted by __gnat_split is wrong and it fails this
comparison:

    0x000105bc <+68>:    ldo -1(r21),r22
    0x000105c0 <+72>:    cmpib,<< b,r22,0x10618 <ada__calendar__split+160>

(gdb) p $r22
$15 = 12

Have to run,
Dave
--
John David Anglin    dave.anglin@bell.net

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #5 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-07-20 12:26:33 UTC ---
> I don't think there is an earlyclobber problem in the insn at 4923.

SPARC has essentially the same pattern (it's a splitter though) and there was a
missing early clobber because there is no guarantee that DImode registers all
start on an even register in 32-bit mode: it's false for DImode arguments.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #6 from dave.anglin at bell dot net 2012-07-20 13:28:47 UTC ---
On 7/20/2012 8:26 AM, ebotcazou at gcc dot gnu.org wrote:
> SPARC has essentially the same pattern (it's a splitter though) and there was a
> missing early clobber because there is no guarantee that DImode registers all
> start on an even register in 32-bit mode: it's false for DImode arguments.
I believe that the definition of HARD_REGNO_MODE_OK on PA prevents 
partially overlapping
DImode registers.  See comment comment for HARD_REGNO_MODE_OK .  So, 
clobbering %R0
can't affect %1.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #7 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-07-20 15:01:49 UTC ---
> I believe that the definition of HARD_REGNO_MODE_OK on PA prevents 
> partially overlapping
> DImode registers.  See comment comment for HARD_REGNO_MODE_OK .  So, 
> clobbering %R0 can't affect %1.

Sure, just like on SPARC, but DImode arguments can nevertheless be unaligned.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #8 from dave.anglin at bell dot net 2012-07-20 17:20:29 UTC ---
On 7/20/2012 11:01 AM, ebotcazou at gcc dot gnu.org wrote:
> Sure, just like on SPARC, but DImode arguments can nevertheless be unaligned.
I've never seen this.  In addition to HARD_REGNO_MODE_OK, the definition of
CANNOT_CHANGE_MODE_CLASS prevents mode changes to larger modes
when it is greater than UNITS_PER_WORD.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #9 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-07-20 17:42:19 UTC ---
> I've never seen this.  In addition to HARD_REGNO_MODE_OK, the definition of
> CANNOT_CHANGE_MODE_CLASS prevents mode changes to larger modes
> when it is greater than UNITS_PER_WORD.

On SPARC, this is specified by the 32-bit ABI: if you have

  void foo(int32_t a, int64_t b)

b will be loaded into DImode %o1, which is SImode %o1 & SImode %o2.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #10 from dave.anglin at bell dot net 2012-07-20 18:48:25 UTC ---
On 7/20/2012 1:42 PM, ebotcazou at gcc dot gnu.org wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974
>
> --- Comment #9 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-07-20 17:42:19 UTC ---
>> I've never seen this.  In addition to HARD_REGNO_MODE_OK, the definition of
>> CANNOT_CHANGE_MODE_CLASS prevents mode changes to larger modes
>> when it is greater than UNITS_PER_WORD.
> On SPARC, this is specified by the 32-bit ABI: if you have
>
>    void foo(int32_t a, int64_t b)
>
> b will be loaded into DImode %o1, which is SImode %o1 & SImode %o2.
>
On PA, b will be loaded  into the second DImode call argument register which
is the r23/r24 register pair.  HARD_REGNO_MODE_OK is defined in a manner
that is consistent with the calling conventions.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

--- Comment #11 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-27 18:16:46 UTC ---
Adding early clobbers to the various 32-bit DImode patterns in pa.md
didn't make any difference.

The bug appears to have been introduced in revision 188786.  The
following four fails were introduced in this revision:

RUN c96005a
,.,. C96005A ACATS 2.5 12-07-26 09:57:57---- C96005A CHECK THAT THE ADDITION
AND SUBTRACTION FUNCTIONS FOR
                VALUES OF TYPE TIME WORK CORRECTLY.   * C96005A DIFFERENCE OF
TIMES IS INCORRECT - (A)1.
   * C96005A DIFFERENCE OF TIMES IS INCORRECT - (A)2.
raised ADA.CALENDAR.TIME_ERROR : a-calend.adb:301
FAIL:   c96005a

RUN c96005f
,.,. C96005F ACATS 2.5 12-07-26 09:58:05---- C96005F CHECKING PACKAGE CALENDAR
+ AND - FUNCTIONS.
   * C96005F FAILURE IN ADDING 1 DAY TO 'CURR_DAY1'.   * C96005F FAILURE IN
ADDING 1 DAY TO 'CURR_DAY2'.
   * C96005F FAILURE IN ADDING 1 DAY TO 'CURR_DAY3'.
   * C96005F FAILURE IN SUBTRACTING 1 DAY FROM 'CURR_DAY1'.
   * C96005F FAILURE IN SUBTRACTING 1 DAY FROM 'CURR_DAY2'.
   * C96005F FAILURE IN SUBTRACTING 1 DAY FROM 'CURR_DAY3'.
**** C96005F FAILED ****************************.
FAIL:   c96005f

RUN c96008a

,.,. C96008A ACATS 2.5 12-07-26 09:58:13
---- C96008A CHECK MISCELLANEOUS FUNCTIONS IN THE PACKAGE CALENDAR.   - C96008A
TIME_OF AND SPLIT ARE NOT INVERSES WHEN SECONDS IS A                NON-MODEL
NUMBER - (A).   * C96008A SPLIT(TIME_OF) CHANGED VALUE OF YR - (A).
   * C96008A SPLIT(TIME_OF) CHANGED VALUE OF MO - (A).   * C96008A
SPLIT(TIME_OF) CHANGED VALUE OF DY - (A).
   * C96008A SPLIT(TIME_OF) CHANGED VALUE OF SEC - (A).   * C96008A TIME_OF()
DID NOT ZERO SECONDS - (C).
   * C96008A YEAR() DID NOT RETURN CORRECT VALUE - (D).   * C96008A MONTH() DID
NOT RETURN CORRECT VALUE - (D).
   * C96008A DAY() DID NOT RETURN CORRECT VALUE - (D).
   * C96008A SECONDS() DID NOT RETURN CORRECT VALUE - (D).
**** C96008A FAILED ****************************.
FAIL:   c96008a

RUN c96008b
,.,. C96008B ACATS 2.5 12-07-26 09:58:16
---- C96008B CHECK THAT TIME_OF() ADVANCES DAY.   * C96008B TIME_OF DID NOT
CONVERT 86_400 SECONDS TO A DAY.
   * C96008B DAY OF NOW2 INCORRECT.   * C96008B TIME_OF DID NOT ADVANCE DAY.
**** C96008B FAILED ****************************.
FAIL:   c96008b

After revision 189441, almost all acats tests fail with the
ADA.CALENDAR.TIME_ERROR.

[Bug target/53974] [4.8 Regression] Nearly all tests fail with ADA.CALENDAR.TIME_ERROR : a-calend.adb:603

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53974

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |DUPLICATE

--- Comment #12 from John David Anglin <danglin at gcc dot gnu.org> 2012-07-27 18:19:26 UTC ---
Duplicate of 53823.

*** This bug has been marked as a duplicate of bug 53823 ***