GCC Bugzilla will be upgraded from version 4.4.9 to 5.0rc3 on Saturday, April 25, starting around 17:00 UTC. The upgrade process should only last a few minutes. Check bug 64968 for details.
Bug 19623 - ICE: compiling SPECfp2000 benchmark fails in 191.fma3d with -ftree-vectorizer
ICE: compiling SPECfp2000 benchmark fails in 191.fma3d with -ftree-vectorizer
Status: RESOLVED WORKSFORME
Product: gcc
Classification: Unclassified
Component: target
4.0.0
: P2 normal
: ---
Assigned To: Not yet assigned to anyone
: ice-on-valid-code, ssemmx
Depends on:
Blocks:
  Show dependency treegraph
 
Reported: 2005-01-25 14:32 UTC by Oliver Lau
Modified: 2005-07-23 22:49 UTC (History)
2 users (show)

See Also:
Host: i686-pc-linux-gnu
Target: i686-pc-linux-gnu
Build: i686-pc-linux-gnu
Known to work:
Known to fail:
Last reconfirmed:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Oliver Lau 2005-01-25 14:32:34 UTC
Compiler used: GCC 4.0 Snapshot as of 2005-01-16.


Optimization flags used:

OPTIMIZE   =  -march=opteron -O3 -funroll-loops -ftree-vectorize


Error doesn't occur with:

OPTIMIZE   =  -march=opteron -O3 -funroll-loops


-----------------------------------------------------------


Compiler output (taken from file 
/cpu2000/benchspec/CFP2000/191.fma3d/run/00000019/make.err)

-----------------------------------------------------------
 In file beam.f90:1379

        IF (AK .EQ. 0.0) GO TO 200
                                 1
 In file beam.f90:1447

 200    ENDDO
            2
Warning: Obsolete: GOTO at (1) jumps to END of construct at (2)
 In file beam.f90:1387

        IF ((Psi - AK*AK) .LE. 0.0) GO TO 200
                                            1
 In file beam.f90:1447

 200    ENDDO
            2
Warning: Obsolete: GOTO at (1) jumps to END of construct at (2)
 In file beam.f90:1397

            IF ((Psi - AK*AK) .LE. 0.0) GO TO 200
                                                1
 In file beam.f90:1447

 200    ENDDO
            2
Warning: Obsolete: GOTO at (1) jumps to END of construct at (2)
beam.f90: In function ‘beam_stress_integration’:
beam.f90:675: internal compiler error: in ix86_expand_fp_absneg_operator, at con
fig/i386/i386.c:8057
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.
specmake: *** [beam.o] Error 1
Comment 1 Richard Henderson 2005-02-12 20:07:45 UTC
I do not replicate this as of today.
Comment 2 Richard Henderson 2005-02-12 20:11:00 UTC
Well, we don't vectorize any loops as of today either...  Sigh.
Comment 3 Dorit Naishlos 2005-02-12 20:38:59 UTC
> Well, we don't vectorize any loops as of today either...  Sigh.

(just fyi:) I just checked on powerpc-darwin (but with NAG as fortran FE, not 
gfortran), and these were reported as vectorized:

beam.022195.c.t53.vect:beam.f90:946: note: LOOP VECTORIZED.
beam.022195.c.t53.vect:beam.f90:934: note: LOOP VECTORIZED.
beam.022195.c.t53.vect:beam.f90:1004: note: LOOP VECTORIZED.
beam.022195.c.t53.vect:beam.f90:992: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:701: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:702: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:703: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:704: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:705: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:706: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:715: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:716: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:717: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:718: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:719: note: LOOP VECTORIZED.
fma1.022301.c.t53.vect:fma1.f90:720: note: LOOP VECTORIZED.
fma2.022632.c.t53.vect:fma2.f90:5667: note: LOOP VECTORIZED.
partition.022706.c.t53.vect:partition.f90:975: note: LOOP VECTORIZED.
partition.022706.c.t53.vect:partition.f90:976: note: LOOP VECTORIZED.
partition.022706.c.t53.vect:partition.f90:977: note: LOOP VECTORIZED.
partition.022706.c.t53.vect:partition.f90:877: note: LOOP VECTORIZED.
strain.022728.c.t53.vect:strain.f90:99: note: LOOP VECTORIZED.
strain.022728.c.t53.vect:strain.f90:172: note: LOOP VECTORIZED.
strain.022728.c.t53.vect:strain.f90:163: note: LOOP VECTORIZED.
strain.022728.c.t53.vect:strain.f90:156: note: LOOP VECTORIZED.
strain.022728.c.t53.vect:strain.f90:237: note: LOOP VECTORIZED.
strain.022728.c.t53.vect:strain.f90:232: note: LOOP VECTORIZED.
strain.022728.c.t53.vect:strain.f90:227: note: LOOP VECTORIZED. 
Comment 4 Richard Henderson 2005-02-13 09:29:26 UTC
Actually, I misread the function I was supposed to be looking for loops in.
We do in fact vectorize 21 loops in beam_stress_integration.  So... WORKSFORME.