Someone broke complex arithmetic

Someone broke complex arithmetic

[Steve Kargl]

This is a heads up.  Someone has broken complex arithmetic
on mainline.  I've just found this problem and unfortunately
it will take me some time to cut the test program down to
something managable.  This could be a gfortran bug or it
may be a middle/back end bug.

Here's the output from a program that brought the problem
to my attention.  It uses downward recursion to compute
spherical Bessel functions.

NAG's F95 compiler
n     x          jn(x)         jn(cmplx(x,0))     
0 2.2900E+01 -3.4445E-02 -3.4445E-02  0.0000E+00
1 2.2900E+01  2.5337E-02  2.5337E-02  0.0000E+00
2 2.2900E+01  3.7765E-02  3.7765E-02  0.0000E+00
3 2.2900E+01 -1.7091E-02 -1.7091E-02  0.0000E+00
4 2.2900E+01 -4.2989E-02 -4.2989E-02  0.0000E+00

gfortran mainline
n     x          jn(x)         jn(cmplx(x,0))     
0 2.2900E+01 -3.4445E-02 -3.4445E-02  0.0000E+00
1 2.2900E+01  2.5337E-02  3.9638E-02  0.0000E+00
2 2.2900E+01  3.7765E-02 -5.0710E-02  0.0000E+00
3 2.2900E+01 -1.7091E-02  7.3035E-02  0.0000E+00
4 2.2900E+01 -4.2989E-02 -1.2032E-01  0.0000E+00

Notice the 3rd column is foobar.

Also note that I'm seeing this problem on i386-*-freebsd and
amd64-*-freebsd.

-- 
Steve

Re: Someone broke complex arithmetic

[Steve Kargl]

On Mon, Jul 18, 2005 at 07:31:27PM -0700, Steve Kargl wrote:
> 
> Here's the output from a program that brought the problem
> to my attention.  It uses downward recursion to compute
> spherical Bessel functions.
> 
> NAG's F95 compiler
> n     x          jn(x)         jn(cmplx(x,0))     
> 0 2.2900E+01 -3.4445E-02 -3.4445E-02  0.0000E+00
> 1 2.2900E+01  2.5337E-02  2.5337E-02  0.0000E+00
> 2 2.2900E+01  3.7765E-02  3.7765E-02  0.0000E+00
> 3 2.2900E+01 -1.7091E-02 -1.7091E-02  0.0000E+00
> 4 2.2900E+01 -4.2989E-02 -4.2989E-02  0.0000E+00
> 

gfortran in 4.0.1 gives the correct results shown above.


-- 
Steve

Re: Someone broke complex arithmetic

[Andrew Pinski]

On Jul 18, 2005, at 10:45 PM, Steve Kargl wrote:

> On Mon, Jul 18, 2005 at 07:31:27PM -0700, Steve Kargl wrote:
>>
>> Here's the output from a program that brought the problem
>> to my attention.  It uses downward recursion to compute
>> spherical Bessel functions.
>>
>> NAG's F95 compiler
>> n     x          jn(x)         jn(cmplx(x,0))
>> 0 2.2900E+01 -3.4445E-02 -3.4445E-02  0.0000E+00
>> 1 2.2900E+01  2.5337E-02  2.5337E-02  0.0000E+00
>> 2 2.2900E+01  3.7765E-02  3.7765E-02  0.0000E+00
>> 3 2.2900E+01 -1.7091E-02 -1.7091E-02  0.0000E+00
>> 4 2.2900E+01 -4.2989E-02 -4.2989E-02  0.0000E+00
>>
>
> gfortran in 4.0.1 gives the correct results shown above.

This is most likely the same problem as PR 22504.

Could you attach your program to that PR?

-- Pinski

Re: Someone broke complex arithmetic

[Steve Kargl]

On Mon, Jul 18, 2005 at 11:02:12PM -0400, Andrew Pinski wrote:
> 
> This is most likely the same problem as PR 22504.
> 
> Could you attach your program to that PR?
> 

I'm still cutting it down.  I'll attach it to
the PR when its much small than it is now.

-- 
Steve

Re: Someone broke complex arithmetic

[Gerald Pfeifer]

On Tue, 19 Jul 2005, FX Coudert wrote:
>> Don't folk run the gfortran testsuite???
> No. People don't regtest with gfortran enabled. That's a pity, since it only
> adds little time to the total build and testing time.

I believe on of the reasons people often do not build with gfortran 
enabled is that this requires additional libraries which are not part
of common system configurations and do not come with GCC either.

Gerald

Re: Someone broke complex arithmetic

[Steve Kargl]

On Tue, Jul 19, 2005 at 11:20:51AM -0700, Richard Henderson wrote:
> On Tue, Jul 19, 2005 at 10:31:13AM -0700, Steve Kargl wrote:
> > Someone broke optimization of complex arithmetic.  A 2005-06-01
> > mainline gives the expected answer.  A 2005-06-15 mainline is
> > broken.  I'll continue my binary search.  Fortunately, building
> > gcc on a dual opteron system with 12 GB of memory goes fairly
> > quick.
> 
> It's sure to be my complex reorg.  You can quit searching and just
> file a bug.
> 

Andrew Pinksia  suggested the bug is the same as PR 22504,
so I've attached my Fortran 90 code to that PR.

-- 
Steve

Re: Someone broke complex arithmetic

[Richard Henderson]

On Tue, Jul 19, 2005 at 10:31:13AM -0700, Steve Kargl wrote:
> Someone broke optimization of complex arithmetic.  A 2005-06-01
> mainline gives the expected answer.  A 2005-06-15 mainline is
> broken.  I'll continue my binary search.  Fortunately, building
> gcc on a dual opteron system with 12 GB of memory goes fairly
> quick.

It's sure to be my complex reorg.  You can quit searching and just
file a bug.


r~

Re: Someone broke complex arithmetic

[Steve Kargl]

On Tue, Jul 19, 2005 at 01:17:22PM -0400, Daniel Berlin wrote:
> On Tue, 2005-07-19 at 09:37 +0200, FX Coudert wrote:
> > > There are regressions involving complex aritmetic in the testsuite too:
> >  > FAIL: gfortran.dg/real_const_1.f (test for excess errors)
> >  > WARNING: gfortran.dg/real_const_1.f compilation failed to produce
> >  > executable
> > 
> > The regression appeared between 20050716 and 20050717 on i686-linux and 
> > i386-freebsd.
> 

This failure may be different than the problem I'm seeing.  

> > > Don't folk run the gfortran testsuite???
> > 
> > No. People don't regtest with gfortran enabled. That's a pity, since
> > it only adds little time to the total build and testing time.
> I do.
> 
> I noticed this on all my trees around that time, and since their has
> been churn in libgfortran, i assumed it was someone doing something they
> hadn't quite finished yet :)

The 3rd and 4th columns should be the same.

troutmask:sgk[207] gfc -o z zy.f90
troutmask:sgk[208] ./z
 0   2.2900E+01 -3.4445E-02 -3.4445E-02  0.0000E+00
 1   2.2900E+01  2.5337E-02  2.5337E-02  0.0000E+00
 2   2.2900E+01  3.7765E-02  3.7765E-02  0.0000E+00
 3   2.2900E+01 -1.7091E-02 -1.7091E-02  0.0000E+00
 4   2.2900E+01 -4.2989E-02 -4.2989E-02  0.0000E+00
troutmask:sgk[209] gfc -o z -O zy.f90
troutmask:sgk[211] ./z
 0   2.2900E+01 -3.4445E-02 -3.4445E-02  0.0000E+00
 1   2.2900E+01  2.5337E-02  3.9638E-02  0.0000E+00
 2   2.2900E+01  3.7765E-02 -5.0710E-02  0.0000E+00
 3   2.2900E+01 -1.7091E-02  7.3035E-02  0.0000E+00
 4   2.2900E+01 -4.2989E-02 -1.2032E-01  0.0000E+00
troutmask:sgk[213] gfc --version
GNU Fortran 95 (GCC 4.1.0 20050712 (experimental))

Someone broke optimization of complex arithmetic.  A 2005-06-01
mainline gives the expected answer.  A 2005-06-15 mainline is
broken.  I'll continue my binary search.  Fortunately, building
gcc on a dual opteron system with 12 GB of memory goes fairly
quick.

-- 
Steve

Re: Someone broke complex arithmetic

[Daniel Berlin]

On Tue, 2005-07-19 at 09:37 +0200, FX Coudert wrote:
> > There are regressions involving complex aritmetic in the testsuite too:
>  > FAIL: gfortran.dg/real_const_1.f (test for excess errors)
>  > WARNING: gfortran.dg/real_const_1.f compilation failed to produce
>  > executable
> 
> The regression appeared between 20050716 and 20050717 on i686-linux and 
> i386-freebsd.


> > Don't folk run the gfortran testsuite???
> 
> No. People don't regtest with gfortran enabled. That's a pity, si
> ce it 
> only adds little time to the total build and testing time.
I do.

I noticed this on all my trees around that time, and since their has
been churn in libgfortran, i assumed it was someone doing something they
hadn't quite finished yet :)

Re: Someone broke complex arithmetic

[FX Coudert]

> There are regressions involving complex aritmetic in the testsuite too:
 > FAIL: gfortran.dg/real_const_1.f (test for excess errors)
 > WARNING: gfortran.dg/real_const_1.f compilation failed to produce
 > executable

The regression appeared between 20050716 and 20050717 on i686-linux and 
i386-freebsd.

> Don't folk run the gfortran testsuite???

No. People don't regtest with gfortran enabled. That's a pity, since it 
only adds little time to the total build and testing time.

I asked geoffk on 2005-06-17 to add gfortran on the GCC Automated 
Testing System, and he's investigating. We can only hope this can be 
done soon.

FX