[Bug middle-end/42220] New: [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[Bug middle-end/42220] New: [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[dominiq at lps dot ens dot fr]

On at revision 154712 I see the following failures:

FAIL: gfortran.dg/complex_intrinsic_5.f90  -O3 -fomit-frame-pointer
-funroll-loops  execution test
FAIL: gfortran.dg/complex_intrinsic_5.f90  -O3 -fomit-frame-pointer
-funroll-all-loops -finline-functions  execution test

[karma] f90/bug% gfc -m64 -O3 -funroll-loops
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90
[karma] f90/bug% a.out
check4:
   z=.0000000 + I .83298129
zref=.38187021 + I 1.0719848

Diff: -.38187021 + I*-.23900348  eps=.23841858E-06
Abort
[karma] f90/bug% gfc -m64 -O3 -funroll-all-loops
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90
[karma] f90/bug% a.out
check4:
   z=.0000000 + I .83298129
zref=.38187021 + I 1.0719848

Diff: -.38187021 + I*-.23900348  eps=.23841858E-06
Abort

The last successful revision is 154405 (configured with my build of mpc, while
the failing one use the fink package).


-- 
           Summary: [4.5 Regression] FAIL:
                    gfortran.dg/complex_intrinsic_5.f90  -O3 -funroll*-loops
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: dominiq at lps dot ens dot fr
 GCC build triplet: powerpc-apple-darwin9
  GCC host triplet: powerpc-apple-darwin9
GCC target triplet: powerpc-apple-darwin9


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2009-11-29 16:33 -------
So this is a mpc / fink bug, not a gcc one.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[dominiq at lps dot ens dot fr]

------- Comment #2 from dominiq at lps dot ens dot fr  2009-11-29 16:56 -------
> So this is a mpc / fink bug, not a gcc one.

I have forgotten to say that the failure occurs only with -funroll*-loops,
-O[1-3], and -m64 options.
Without -funroll*-loops the test pass. BTW I do not see any loop in the code.


-- 

dominiq at lps dot ens dot fr changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |UNCONFIRMED
         Resolution|INVALID                     |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[ebotcazou at gcc dot gnu dot org]

------- Comment #3 from ebotcazou at gcc dot gnu dot org  2009-11-29 17:15 -------
> I have forgotten to say that the failure occurs only with -funroll*-loops,
> -O[1-3], and -m64 options.
> Without -funroll*-loops the test pass. BTW I do not see any loop in the code.

Very likely revision 154688:
  http://gcc.gnu.org/ml/gcc-cvs/2009-11/msg00911.html

Can you attach the files generated by -fdump-rtl-ce3 -fdump-rtl-rnreg-details?


-- 

ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bernds at gcc dot gnu dot
                   |                            |org, ebotcazou at gcc dot
                   |                            |gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[dominiq at lps dot ens dot fr]

------- Comment #4 from dominiq at lps dot ens dot fr  2009-11-29 17:27 -------
Created an attachment (id=19178)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19178&action=view)
ce3 file


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[dominiq at lps dot ens dot fr]

------- Comment #5 from dominiq at lps dot ens dot fr  2009-11-29 17:29 -------
Created an attachment (id=19179)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19179&action=view)
rnreg file


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[dominiq at lps dot ens dot fr]

------- Comment #6 from dominiq at lps dot ens dot fr  2009-11-29 17:30 -------
> Can you attach the files generated by -fdump-rtl-ce3 -fdump-rtl-rnreg-details?

I have reduced the test to

module test
  implicit none
  real(4), parameter :: eps4 = epsilon(0.0_4)*2.0_4
  real(8), parameter :: eps8 = epsilon(0.0_8)*2.0_8
  interface check
    procedure check4
  end interface check
contains
  SUBROUTINE check4(z, zref)
    complex(4), intent(in) :: z, zref
    if (    abs (real(z)-real(zref)) > eps4 &
        .or.abs (aimag(z)-aimag(zref)) > eps4) then
      print '(a,/,2((2g0," + I ",g0),/))', "check4:","   z=",z,'zref=',zref
      print '(a,g0," + I*",g0,"  eps=",g0)', 'Diff: ', &
                                 real(z)-real(zref), &
                                 aimag(z)-aimag(zref), eps4
!      call abort()
    end if
  END SUBROUTINE check4
end module test

PROGRAM ArcTrigHyp
  use test
  IMPLICIT NONE
  complex(4), volatile :: z4

!!!!! ZERO !!!!!!

  ! z = 0
  z4 = cmplx(0.0_4, 0.0_4, kind=4)

  ! Exact: 0
  call check(asin(z4), cmplx(0.0_4, 0.0_4, kind=4))
  ! Exact: Pi/2 = 1.5707963267948966192313216916397514
  call check(acos(z4), cmplx(1.57079632679489661920_4, 0.0_4, kind=4))
  ! Exact: 0
  call check(atan(z4), cmplx(0.0_4, 0.0_4, kind=4))
  ! Exact: 0
  call check(asinh(z4), cmplx(0.0_4, 0.0_4, kind=4))
  ! Exact: I*Pi/2 = I*1.5707963267948966192313216916397514
  call check(acosh(z4), cmplx(0.0_4, 1.57079632679489661920_4, kind=4))
  ! Exact: 0
  call check(atanh(z4), cmplx(0.0_4, 0.0_4, kind=4))


!!!!! POSITIVE NUMBERS !!!!!!

  ! z = tanh(1.0)
  z4 = cmplx(0.76159415595576488811945828260479359_4, 0.0_4, kind=4)

  ! Numerically: 0.70502684355523799494171984544790700*I
  call check(acosh(z4), cmplx(0.0_4, 0.70502684355523799494171984544790700_4,
kind=4))
  ! Exact: 1
  call check(atanh(z4), cmplx(1.0_4, 0.0_4, kind=4))


  ! z = I*tanh(1.0)
  z4 = cmplx(0.0_4, 0.76159415595576488811945828260479359_4, kind=4)

  ! Numerically: I*0.70239670712987482778422106260749699
  call check(asin(z4), cmplx(0.0_4, 0.70239670712987482778422106260749699_4,
kind=4))

END PROGRAM ArcTrigHyp

Note that the errors fluctuate, depending on the check lines commented. I'll
attach the asked files for this reduced test.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[hjl dot tools at gmail dot com]

------- Comment #7 from hjl dot tools at gmail dot com  2009-11-29 23:25 -------
It may be related to PR 42202.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[dominiq at lps dot ens dot fr]

------- Comment #8 from dominiq at lps dot ens dot fr  2009-12-04 14:21 -------
> It may be related to PR 42202.

If I am not mistaken this has been fixed by revision 154944 (see
http://gcc.gnu.org/ml/gcc-testresults/2009-12/msg00343.html ).

However this pr is not fixed at revision 154970.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
   Target Milestone|---                         |4.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[dominiq at lps dot ens dot fr]

------- Comment #9 from dominiq at lps dot ens dot fr  2009-12-08 10:06 -------
I have bootstrapped gcc revision 155054 with the mpc version for which the
tests passed (revision 154405) and the tests are still failing.

The minimal set of options to make the test in comment #6 abort is "-O1
-funroll-loops -m64 -fomit-frame-pointer". The test pass if I omit
-fomit-frame-pointer.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[rguenth at gcc dot gnu dot org]

------- Comment #10 from rguenth at gcc dot gnu dot org  2009-12-11 23:09 -------
Completely unclear what this bug is about.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[dominiq at lps dot ens dot fr]

------- Comment #11 from dominiq at lps dot ens dot fr  2009-12-11 23:23 -------
> Completely unclear what this bug is about.

What is unclear?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -O3 -funroll*-loops

[rguenth at gcc dot gnu dot org]

------- Comment #12 from rguenth at gcc dot gnu dot org  2009-12-14 13:50 -------
We don't fold anything here (because of the volatileness of z4).  Thus this
can't be a mpc issue but is probably libm issue.

I don't know how PPC implements _Complex float and if the HW has excess
precision and thus we get extra rounding/truncating when spilling values.
-funroll-loops triggers -frename-registers which again would hint at
Bernds change.

Assuming that the testcase from comment #6 is what this is all about
(i?86-linux is fine w/ and w/o -frename-registers and w/ and w/o volatile).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers -fomit-frame-pointers

[rguenth at gcc dot gnu dot org]

------- Comment #13 from rguenth at gcc dot gnu dot org  2009-12-14 14:01 -------
If no calls remain in the assembly as dominiq suggests then the
*call_value_nonlocal_darwin64 pattern must be bogus.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dje at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #14 from dominiq at lps dot ens dot fr  2009-12-14 20:42 -------
When compiled with '-m64 -O1 -frename-registers' the code in comment #6 fails,
but passes with ''-m64 -O1'.

> If no calls remain in the assembly as dominiq suggests then the
> *call_value_nonlocal_darwin64 pattern must be bogus.

I do not see "call", but in both cases

...
        bl L_casinf$stub
...

which I understand as an other form of call. So the darwin math lib seems to
work correctly when -frename-registers is not used. I'll attach the assembly
generated with/without the option.


-- 

dominiq at lps dot ens dot fr changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |UNCONFIRMED
            Summary|[4.5 Regression] FAIL:      |[4.5 Regression] FAIL:
                   |gfortran.dg/complex_intrinsi|gfortran.dg/complex_intrinsi
                   |c_5.f90  -m64 -O -frename-  |c_5.f90  -m64 -O -frename-
                   |registers -fomit-frame-     |registers
                   |pointers                    |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #15 from dominiq at lps dot ens dot fr  2009-12-14 20:48 -------
Created an attachment (id=19298)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19298&action=view)
assembly generated with -frename-registers


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #16 from dominiq at lps dot ens dot fr  2009-12-14 20:54 -------
Created an attachment (id=19299)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19299&action=view)
assembly generated without -frename-registers


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[ebotcazou at gcc dot gnu dot org]

------- Comment #17 from ebotcazou at gcc dot gnu dot org  2010-01-11 11:31 -------
Still present?


-- 

ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #18 from dominiq at lps dot ens dot fr  2010-01-11 12:20 -------
> Still present?

Yes!-(If it has not been fixed meanwhile, it still fails at revision 155621).


-- 

dominiq at lps dot ens dot fr changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |UNCONFIRMED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #19 from dominiq at lps dot ens dot fr  2010-01-30 22:23 -------
The test fails also on powerpc-unknown-linux-gnu (see
http://gcc.gnu.org/ml/gcc-testresults/2010-01/msg02790.html ).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #20 from bernds_cb1 at t-online dot de  2010-02-16 17:40 -------
Sorry I've seen this so late; the mails I got have been hidden in my unread
fortran folder so far.  Need to change the filters.

Comment #9 suggests you can reproduce this without -frename-registers.  Is this
correct?

What I'd be looking at is if mpc has functions written in assembly that maybe
clobber registers they shouldn't.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #21 from dominiq at lps dot ens dot fr  2010-02-16 18:13 -------
> Comment #9 suggests you can reproduce this without -frename-registers.  Is this
> correct?

>From comment #12:

> -funroll-loops triggers -frename-registers which again would hint at
> Bernds change.

I think the answer is no: the test passes with '-m64 -O1' but fails with '-m64
-O1 -frename-registers'.

For the record I am using GMP version 4.3.1, MPFR version 2.4.1, MPC version
0.8. If needed I can try to update to mpc 0.8.1 (so far I failed to update GMP,
both 4.3.2 and 5.0.1).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[mmitchel at gcc dot gnu dot org]

------- Comment #22 from mmitchel at gcc dot gnu dot org  2010-02-17 16:52 -------
I don't think we really know enough yet to understand whether this is a bug, or
if it is a bug, where the bug might lie.  So, we certainly can't make it P1,
ignoring even the fact that this test is in Fortran.  But, it would be good if
we could understand the situation better.


-- 

mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #23 from bernds_cb1 at t-online dot de  2010-02-17 22:13 -------
Created an attachment (id=19900)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19900&action=view)
Possible fix.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #24 from bernds_cb1 at t-online dot de  2010-02-17 22:14 -------
Would you mind testing the attached patch?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #25 from dominiq at lps dot ens dot fr  2010-02-18 08:00 -------
> Would you mind testing the attached patch?

Apparently the patch in comment #23 does not fix the problem (incremental
update of gcc and partial test):

make -k check-gfortran RUNTESTFLAGS="dg.exp=complex*.f90
--target_board=unix'{-m32,-m64}'"

Running target unix/-m32
Using /sw/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /sw/share/dejagnu/config/unix.exp as generic interface file for target.
Using /opt/gcc/gcc-4.5-work/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /opt/gcc/gcc-4.5-work/gcc/testsuite/gfortran.dg/dg.exp ...

                === gfortran Summary for unix/-m32 ===

# of expected passes            104
Running target unix/-m64
Using /sw/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /sw/share/dejagnu/config/unix.exp as generic interface file for target.
Using /opt/gcc/gcc-4.5-work/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /opt/gcc/gcc-4.5-work/gcc/testsuite/gfortran.dg/dg.exp ...
FAIL: gfortran.dg/complex_intrinsic_5.f90  -O3 -fomit-frame-pointer
-funroll-loops  execution test
FAIL: gfortran.dg/complex_intrinsic_5.f90  -O3 -fomit-frame-pointer
-funroll-all-loops -finline-functions  execution test

                === gfortran Summary for unix/-m64 ===

# of expected passes            102
# of unexpected failures        2

                === gfortran Summary ===

# of expected passes            206
# of unexpected failures        2
/opt/gcc/darwin_buildw/gcc/testsuite/gfortran/../../gfortran  version 4.5.0
20100215 (experimental) [trunk revision 156774p2] (GCC) 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #26 from bernds_cb1 at t-online dot de  2010-02-18 11:51 -------
Created an attachment (id=19905)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19905&action=view)
A patch to help debug the problem

I'll need some help since on my system a compiler targetting
powerpc-apple-darwin9 produces rather different output in the dumps than what
is attached to this bug.  The powerpc-linux failures look unrelated as they
appear even at -O0.

If the problem really is in regrename, the attached patch should help debug it.
 If you wish to help, please follow these instructions.

You'll need to use a kind of binary search using the -fdbg-cnt=rnreg:N option,
where N is an integer.  Start with something small, maybe even 1, then double
it until the failure appears.  Then do a binary search between the last number
that produced a working binary, and the one that did not.

You should end up with a value of N such that -fdbg-cnt=rnreg:N produces a
working executable, and -fdbg-cnt=rnreg:N+1 does not.  Please attach the .rnreg
dumps and assembly files for both.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #27 from dominiq at lps dot ens dot fr  2010-02-18 12:17 -------
> You'll need to use a kind of binary search using the -fdbg-cnt=rnreg:N option,
> where N is an integer.  Start with something small, maybe even 1, then double
> it until the failure appears.  Then do a binary search between the last number
> that produced a working binary, and the one that did not.

Should I use '-fdbg-cnt=rnreg:N' to build gcc or only when testing
complex_intrinsic_5.f90?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #28 from bernds_cb1 at t-online dot de  2010-02-18 12:21 -------
Only when building the testcase.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #29 from dominiq at lps dot ens dot fr  2010-02-18 13:42 -------
In order to compile gcc/regrename.c, I had to add

+  enum debug_counter rnreg;

Is this right?

Then compiling the test with

gfc -fdbg-cnt=rnreg:1 -m64 -O1 -frename-registers
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90

I get the following error

f951: error: Can not find a valid counter:value pair:
f951: error: -fdbg-cnt=rnreg:1
f951: error:           ^

What is missing?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[steven at gcc dot gnu dot org]

------- Comment #30 from steven at gcc dot gnu dot org  2010-02-18 13:53 -------
It looks like there should be a patch to dbgcnt.def.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #31 from dominiq at lps dot ens dot fr  2010-02-18 14:06 -------
> It looks like there should be a patch to dbgcnt.def.

Does this mean that I should remove the line

+  enum debug_counter rnreg;

I have added, and add a line

DEBUG_COUNTER (rnreg)

in dbgcnt.def?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #32 from bernds_cb1 at t-online dot de  2010-02-18 14:17 -------
Created an attachment (id=19908)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19908&action=view)
Additional patch on top of the previous one

Sorry about that.  Yes, you'll need to add that in dbgcnt.def, or just apply
this additional patch.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #33 from dominiq at lps dot ens dot fr  2010-02-18 14:22 -------
> Sorry about that.  Yes, you'll need to add that in dbgcnt.def, or just apply
> this additional patch.

This recompiles most of gcc!-(it will take a couple hours on my poor G5!-).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #34 from dominiq at lps dot ens dot fr  2010-02-18 15:08 -------
And the winner is N=137!

[karma] f90/bug% gfc -fdbg-cnt=rnreg:137 -m64 -O1 -frename-registers
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90
dbg_cnt 'rnreg' set to 137
[karma] f90/bug% a.out
[karma] f90/bug% gfc -fdbg-cnt=rnreg:138 -m64 -O1 -frename-registers
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90
dbg_cnt 'rnreg' set to 138
[karma] f90/bug% a.out
check4:
   z=.0000000 + I .0000000
zref=1.0000000 + I .0000000

Diff: -1.0000000 + I*.0000000  eps=.23841858E-06
Abort


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #35 from bernds_cb1 at t-online dot de  2010-02-18 15:32 -------
Okay, great.  Could you attach the .rnreg dumps and assembly output for both
values?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #36 from dominiq at lps dot ens dot fr  2010-02-18 15:38 -------
> Could you attach the .rnreg dumps

How do I get them?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #37 from dominiq at lps dot ens dot fr  2010-02-18 15:40 -------
Created an attachment (id=19911)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19911&action=view)
Assembly for -fdbg-cnt=rnreg:137

Command used

gfc -S -fdbg-cnt=rnreg:137 -m64 -O1 -frename-registers
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #38 from dominiq at lps dot ens dot fr  2010-02-18 15:41 -------
Created an attachment (id=19912)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19912&action=view)
Assembly for -fdbg-cnt=rnreg:138

Command line

gfc -S -fdbg-cnt=rnreg:138 -m64 -O1 -frename-registers
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #39 from bernds_cb1 at t-online dot de  2010-02-18 15:52 -------
(In reply to comment #36)
> > Could you attach the .rnreg dumps
> 
> How do I get them?
> 

-fdump-rtl-rnreg-details


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #40 from dominiq at lps dot ens dot fr  2010-02-18 15:58 -------
Created an attachment (id=19914)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19914&action=view)
.rnreg for -fdbg-cnt=rnreg:137

Command used

fc -fdump-rtl-rnreg-details -fdbg-cnt=rnreg:137 -m64 -O1 -frename-registers
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #41 from dominiq at lps dot ens dot fr  2010-02-18 15:59 -------
Created an attachment (id=19915)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19915&action=view)
.rnreg for -fdbg-cnt=rnreg:138

Command line

gfc -fdump-rtl-rnreg-details -fdbg-cnt=rnreg:138 -m64 -O1 -frename-registers
/opt/gcc/_gcc_clean/gcc/testsuite/gfortran.dg/complex_intrinsic_5.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #42 from bernds_cb1 at t-online dot de  2010-02-18 18:13 -------
Created an attachment (id=19917)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19917&action=view)
Another test patch that attempts to fix the problem.

Could you test whether this fixes it?


-- 

bernds_cb1 at t-online dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #19900|0                           |1
        is obsolete|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #43 from dominiq at lps dot ens dot fr  2010-02-18 18:44 -------
The compilation of gcc/regrename.c fails with

...
cc1: warnings being treated as errors
../../gcc-4.5-work/gcc/regrename.c: In function 'build_def_use':
../../gcc-4.5-work/gcc/regrename.c:1113:6: error: array subscript has type
'char'

The line is

            has_dup[recog_data.dup_num[i]] = true;

trying

has_dup[(int) recog_data.dup_num[i]] = true;


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #44 from dominiq at lps dot ens dot fr  2010-02-18 19:10 -------
The patch in comment #43 with the fix in comment #44 works for the limited
tests I am able to do right now.  I can do a "full" test with a fresh bootstrap
of gcc and fortran, but it will take a full day, so I'ld prefer to do it during
the week-end.

Thanks for the patch.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #45 from dominiq at lps dot ens dot fr  2010-02-22 11:21 -------
Bootstrapped and regtested on powerpc-apple-darwin9 (see
http://gcc.gnu.org/ml/gcc-testresults/2010-02/msg02098.htm ) and
x86_64-apple-darwin10 (see
http://gcc.gnu.org/ml/gcc-testresults/2010-02/msg02102.html) with the patch in
comment #43 with the fix in comment #44.

Everything I can look at seems to work as expected.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[dominiq at lps dot ens dot fr]

------- Comment #46 from dominiq at lps dot ens dot fr  2010-03-01 13:34 -------
Anything else I can do for this pr?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds at gcc dot gnu dot org]

------- Comment #47 from bernds at gcc dot gnu dot org  2010-03-07 15:20 -------
Subject: Bug 42220

Author: bernds
Date: Sun Mar  7 15:20:12 2010
New Revision: 157263

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=157263
Log:
        PR rtl-optimization/42220
        * regrename.c (scan_rtx) <case STRICT_LOW_PART, ZERO_EXTRACT>:
        Use verify_reg_tracked to determine if we should use OP_OUT rather
        than OP_INOUT.
        (build_def_use): If we see an in-out operand for a register that we
        know nothing about, treat is an output if possible, fail the block if
        not.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/regrename.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[rguenth at gcc dot gnu dot org]

------- Comment #48 from rguenth at gcc dot gnu dot org  2010-03-07 15:35 -------
Fixed.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220

[Bug middle-end/42220] [4.5 Regression] FAIL: gfortran.dg/complex_intrinsic_5.f90 -m64 -O -frename-registers

[bernds_cb1 at t-online dot de]

------- Comment #49 from bernds_cb1 at t-online dot de  2010-03-08 23:06 -------
This fix caused a SPEC regression (see bug 42216).  Could you test the patch I
attached to #42216, on top of current mainline, to see whether it does not
cause your problem to reappear?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42220