xclanbsrch_pt segfaults on a OS X 32-bit build (i386 = ia32 arch).
./xclanbsrch_pt -oc res/ATL_ctGetNB_geqrf
maxN = 2508
FINDING NB FOR M=25, N=25, LDA=25:
nb=12: 381.22 979.44 759.28 --> 979.44 MFLOP.
NB refinement search:
*** i=1, nbB=12, NB=76, mul1=4, mul2=0 ***
nb=1: 610.17 620.48 623.13 --> 623.13 MFLOP.
nb=2: 780.12 786.49 790.72 --> 790.72 MFLOP.
nb=3: 871.72 890.60 897.03 --> 897.03 MFLOP.
nb=4: 953.62 925.60 967.95 --> 967.95 MFLOP.
nb=5: 924.64 953.51 958.18 --> 958.18 MFLOP.
nb=6: 618.90 973.43 969.86 --> 973.43 MFLOP.
nb=7: 989.50 1003.72 1008.30 --> 1008.30 MFLOP.
nb=8: 614.56 936.46 913.39 --> 936.46 MFLOP.
nb=9: 988.61 984.69 1004.85 --> 1004.85 MFLOP.
nb=10: 662.51 998.31 1000.06 --> 1000.06 MFLOP.
nb=11: 965.59 979.44 994.40 --> 994.40 MFLOP.
nb=12: 977.50 983.84 987.19 --> 987.19 MFLOP.
FOR M=25, N=25, LDA=25, BEST NB=7
FINDING NB FOR M=2508, N=2508, LDA=2508:
nb=76: 17750.43 --> 17750.43 MFLOP.
nb=152: 18294.10 --> 18294.10 MFLOP.
nb=228: 18384.34 --> 18384.34 MFLOP.
nb=304: 17962.58 --> 17962.58 MFLOP.
nb=380: 18022.88 --> 18022.88 MFLOP.
nb=456: 17463.22 --> 17463.22 MFLOP.
nb=532: 17348.23 --> 17348.23 MFLOP.
NB refinement search:
*** i=84, nbB=152, NB=76, mul1=8, mul2=0 ***
nb=88: 17408.39 --> 17408.39 MFLOP.
Bus error
and gdb says this:
Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00000000
[Switching to process 64569 thread 0x49f]
0x00000000 in ?? ()
(gdb) bt
#0 0x00000000 in ?? ()
#1 0x00096bce in ATL_cpKBmm ()
Hmm. It will be a bit before I can be definitive, but I know there are errors in the complex QR routines that ATLAS has recently started supporting. The author is debugging right now. I suspect, but do not know, that when they are fixed, it will fix this problem.
As soon as I have a fixed QR, I'll post to this bug report and ask you to try again (since the install works for us).
Many thanks,
Clint
Clint,
thanks for looking into this. I more or less unofficially took over the responsibility of porting Atlas to the Macports project. That's why I'm reporting you all the bugs I find on each new version.
I'll be happy to test a patch as soon as you've it available.
Thanks again and have a nice Sunday
Vincent
I believe this error is fixed in 3.9.42. Can you confirm?
Thanks,
Clint
No response, assuming fixed. Reopen if problem still occurs.
You're probably right. Sorry for overlooking this.
no response, assuming fixed.