[Bug target/31394] cos() returns wrong value unless -O0 is used

[Bug target/31394] cos() returns wrong value unless -O0 is used

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31394

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

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

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-01-19 05:01:17 UTC ---
This is a bug in Solaris's libc where sincos does not produce the same results
as sin() and cos() separately would.

[Bug c/31394] New: cos() returns wrong value unless -O0 is used

[sdirkse at gams dot com]

I have some code that involves computing a cosine via cos(), specifically b =
cos(v).  Depending on the context of the cos call, I get correct and incorrect
answers.  In the example code's main program, I get a correct result regardless
of the compilation flags.  In the MATHNEW_funceval routine, correctness depends
on the flags.

Just a note on "correctness" since these are floating-point operations:  IMHO
the incorrect result is far enough away from the true result to remove all
doubt, but I can say more if necessary.  In the output below, "b =" is correct,
but "b2=" is incorrect.

Here's the output I get:
sunamd1:/export/home/gams/lang/cosbug$gmake  && ./cosbug
gcc -m64 -Wall -w -m64 -fwrapv -O1 -save-temps -v -o cosbug cosbug.c -lm
Using built-in specs.
Target: i386-pc-solaris2.10
Configured with: ../configure --build=i386-pc-solaris2.10 --with-gnu-as
--with-as=/usr/sfw/bin/gas --without-gnu-ld -with-ld=/usr/ccs/bin/ld
--with-gmp=/usr/local --with-mpfr=/usr/local --enable-languages=c,c++,fortran
--enable-shared
Thread model: posix
gcc version 4.3.0 20070301 (experimental)
 /usr/local/libexec/gcc/i386-pc-solaris2.10/4.3.0/cc1 -E -quiet -v -imultilib
amd64 cosbug.c -m64 -mtune=generic -Wall -w -fwrapv -O1 -fpch-preprocess -o
cosbug.i
ignoring nonexistent directory
"/usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0/../../../../i386-pc-solaris2.10/include"
#include "..." search starts here:
#include <...> search starts here:
 /usr/local/include
 /usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0/include
 /usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0/include-fixed
 /usr/include
End of search list.
 /usr/local/libexec/gcc/i386-pc-solaris2.10/4.3.0/cc1 -fpreprocessed cosbug.i
-quiet -dumpbase cosbug.c -m64 -mtune=generic -auxbase cosbug -O1 -Wall -w
-version -fwrapv -o cosbug.s
GNU C version 4.3.0 20070301 (experimental) (i386-pc-solaris2.10)
        compiled by GNU C version 4.3.0 20070301 (experimental).
GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096
Compiler executable checksum: 15fd974b8d70fc2c218783c6076c49f0
 /usr/sfw/bin/gas --traditional-format -V -Qy --64 -s -o cosbug.o cosbug.s
GNU assembler version 2.15 (i386-pc-solaris2.10) using BFD version 2.15
 /usr/local/libexec/gcc/i386-pc-solaris2.10/4.3.0/collect2 -V -Y
P,/lib/64:/usr/lib/64 -Qy -o cosbug /usr/lib/amd64/crt1.o /usr/lib/amd64/crti.o
/usr/lib/amd64/values-Xa.o
/usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0/amd64/crtbegin.o
-L/usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0/amd64
-L/usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0/../../../amd64 -L/lib/amd64
-L/usr/lib/amd64 -L/usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0
-L/usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0/../../.. cosbug.o -lm -lgcc
-lgcc_eh -lc -lgcc -lgcc_eh
/usr/local/lib/gcc/i386-pc-solaris2.10/4.3.0/amd64/crtend.o
/usr/lib/amd64/crtn.o
ld: Software Generation Utilities - Solaris Link Editors: 5.10-1.482
0 = 0
x0=     -635.4813998334932
y = -7.036102677544457e-09
z =      635.4813998264574
 *** |t| < mu: return smoother
v =     -3.141592653589793
b = -1.224646799147353e-16
*** |t| < mu: return smoother
v2=     -3.141592653589793
b2=   -0.04318444754676655
sunamd1:/export/home/gams/lang/cosbug$

Now using the .i file I get:
sunamd1:/export/home/gams/lang/cosbug$gcc -m64 -o xxx cosbug.i -lm
sunamd1:/export/home/gams/lang/cosbug$./xxx
0 = 0
x0=     -635.4813998334932
y = -7.036102677544457e-09
z =      635.4813998264574
 *** |t| < mu: return smoother
v =     -3.141592653589793
b = -1.224646799147353e-16
*** |t| < mu: return smoother
v2=     -3.141592653589793
b2= -1.224646799147353e-16
sunamd1:/export/home/gams/lang/cosbug$gcc -m64 -O1 -o xxx cosbug.i -lm
sunamd1:/export/home/gams/lang/cosbug$./xxx
0 = 0
x0=     -635.4813998334932
y = -7.036102677544457e-09
z =      635.4813998264574
 *** |t| < mu: return smoother
v =     -3.141592653589793
b = -1.224646799147353e-16
*** |t| < mu: return smoother
v2=     -3.141592653589793
b2=   -0.04318444754676655


-- 
           Summary: cos() returns wrong value unless -O0 is used
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: sdirkse at gams dot com
 GCC build triplet: i386-pc-solaris2.10
  GCC host triplet: i386-pc-solaris2.10
GCC target triplet: i386-pc-solaris2.10


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31394

[Bug target/31394] cos() returns wrong value unless -O0 is used

[dominiq at lps dot ens dot fr]

------- Comment #2 from dominiq at lps dot ens dot fr  2007-03-29 18:50 -------
This bug reminds me PR30980 and PR31161, though they were reported only for g++
and gfortran (they were fixed on 2007-03-16). Could you look at them to see if
the bug you have reported is not a duplicate? If yes, you should update your
build.

Could you be also kind enough to look at the related optimization problem
PR31249 and see it affects also your platform?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31394

[Bug target/31394] cos() returns wrong value unless -O0 is used

[sdirkse at gams dot com]

------- Comment #3 from sdirkse at gams dot com  2007-03-30 21:36 -------
I installed gcc 20070329 and the problem I was having is solved.  I suppose
that makes it a duplicate of PR30980.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31394

[Bug target/31394] cos() returns wrong value unless -O0 is used

[dominiq at lps dot ens dot fr]

------- Comment #4 from dominiq at lps dot ens dot fr  2007-03-30 21:56 -------
> I suppose that makes it a duplicate of PR30980.

It would have been better to check it directly before updating.  PR30980 was
related to g++ and gfortran and not gcc, so it seems that your platform
(i386-pc-solaris2.10?) differs from Darwin at least.

Now, could you please time the C test of PR31249, with and without -O and
report what you get:

#include <math.h>
#include <stdio.h>

int main()
{

  long    n = 1000000;
  long    i;
  double mo = -1.0;
  double pi = acos(mo);
  double sc = 0.0;
  double ss = 0.0;
  double  t = 0.0;
  double dt = pi/n;

  printf("%.17g \n", pi);
  printf("%.17g \n", dt);
  for (i=0; i< 40*n; i++) {
    sc += cos(t);
    ss += sin(t);
    t += dt;
  }
  printf("%.17g %.17g \n", sc, ss);
}

It will take less than a couple of minutes and see if -O is a real optimization
or not on your platform.
TIA


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31394