This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Regression in LAPACK with -O3

On Mac OSX 10.3.9, I applied the patch on the latest 4.2 snapshot. Indeed
it did fix the problem and I got:

with -O3 -unroll-loops
cgd.out: CGV drivers:     62 out of   1092 tests failed to pass the threshold
ded.out: DES:    2 out of  3264 tests failed to pass the threshold
ded.out: DSX:    2 out of  3494 tests failed to pass the threshold
dgd.out: DXV drivers:    200 out of   5000 tests failed to pass the threshold
sgd.out: SXV drivers:     37 out of   5000 tests failed to pass the threshold
ssep.out: SST:    2 out of  4662 tests failed to pass the threshold
ssep.out: SST drivers:      1 out of  14256 tests failed to pass the threshold
zgd.out: ZGV drivers:     61 out of   1092 tests failed to pass the threshold
zgd.out: ZXV drivers:     24 out of   5000 tests failed to pass the threshold

with -O1
cgd.out: CGV drivers:     62 out of   1092 tests failed to pass the threshold
ded.out: DES:    2 out of  3264 tests failed to pass the threshold
ded.out: DSX:    2 out of  3494 tests failed to pass the threshold
dgd.out: DXV drivers:    200 out of   5000 tests failed to pass the threshold
sgd.out: SXV drivers:     37 out of   5000 tests failed to pass the threshold
ssep.out: SST:    2 out of  4662 tests failed to pass the threshold
ssep.out: SST drivers:      1 out of  14256 tests failed to pass the threshold
zgd.out: ZGV drivers:     61 out of   1092 tests failed to pass the threshold
zgd.out: ZXV drivers:     24 out of   5000 tests failed to pass the threshold

with -O0
csep.out: CST:    1 out of  4662 tests failed to pass the threshold
csep.out: CST drivers:      1 out of  11664 tests failed to pass the threshold
csep.out: CST:    1 out of  4662 tests failed to pass the threshold
csep.out: CST drivers:      1 out of  11664 tests failed to pass the threshold
csep.out: CST drivers:      1 out of  11664 tests failed to pass the threshold
dgd.out: DXV drivers:    200 out of   5000 tests failed to pass the threshold
sgd.out: SXV drivers:     37 out of   5000 tests failed to pass the threshold
ssep.out: SST:    1 out of  4662 tests failed to pass the threshold
ssep.out: SST drivers:      1 out of  14256 tests failed to pass the threshold
zgd.out: ZXV drivers:     24 out of   5000 tests failed to pass the threshold

For comparison, Jerry DeLisle found
cgd.out: CGV drivers:     67 out of   1092 tests failed to pass the threshold
csep.out: CST:    2 out of  4662 tests failed to pass the threshold
csep.out: CST drivers:      1 out of  11664 tests failed to pass the threshold
dgd.out: DXV drivers:    200 out of   5000 tests failed to pass the threshold
sgd.out: SXV drivers:     37 out of   5000 tests failed to pass the threshold
ssep.out: SST:    1 out of  4662 tests failed to pass the threshold
ssep.out: SST drivers:      1 out of  14256 tests failed to pass the threshold
ssvd.out: SBD:      1 out of   5510 tests failed to pass the threshold
zgd.out: ZGV drivers:     66 out of   1092 tests failed to pass the threshold
zgd.out: ZXV drivers:     24 out of   5000 tests failed to pass the threshold

and with -O1 before patch I got
cgd.out: CGV drivers:     62 out of   1092 tests failed to pass the threshold
ded.out: DES:    2 out of  3264 tests failed to pass the threshold
ded.out: DSX:    2 out of  3494 tests failed to pass the threshold
dgd.out: DXV drivers:    200 out of   5000 tests failed to pass the threshold
sgd.out: SXV drivers:     37 out of   5000 tests failed to pass the threshold
ssep.out: SST:    2 out of  4662 tests failed to pass the threshold
ssep.out: SST drivers:      1 out of  14256 tests failed to pass the threshold
zgd.out: ZGV drivers:     61 out of   1092 tests failed to pass the threshold
zgd.out: ZXV drivers:     24 out of   5000 tests failed to pass the threshold

Now, I still have a problem with 'xeigtstc < sep.in' which triggers an 
infinite loop but for -O0 (so no results for csep.out, except with -O0). 
I have traced it to slarrb.f at line
   60       CONTINUE
when RIGHT is a NaN. This is not the first time I see this problem (the
first time was in *lartg.f with g95 and -ffast-math) and I think the 
problem is latent in several places in LAPACK which are not protected
against NaN.

Thanks for the quick fix.

Dominique
Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]