|
|
Blitz Devel : |
From: Julian Cummings (cummings_at_[hidden])
Date: 2002-10-03 10:59:15
Hi Oliver,
Thanks very much for this report. We are aware of the problem with a couple
of the testsuite codes under SGI CC. These are due to compiler bugs that were
reported long ago but never fixed. There is a workaround for the problem in
minsumpow, which is to assign the intermediate result from the sum() function
to an Array, and then apply the maxIndex() function to that temporary Array.
A similar workaround can be done in peter-nordlund-3.cpp. We are going to add
these workarounds with a #if test for use of SGI CC, just so the tests will pass.
I wasn't aware of the problems in the benchmark and examples areas, but I will
look into them. I did recently add the tinyvec-et.h header with the idea that one
could include this header if expression template machinery was truly needed,
since it takes more time to parse and compile this extra code. I tried to fix all
the codes that used to just include tinyvec.h for this, but clearly missed one.
I also noticed and recently fixed some issues regarding proper configuration for
64-bit vs. n32 format builds on SGI. This might address some of the linking
problems you saw, but I will look into these and double-check.
Thanks again for all the information.
Regards, Julian C.
Dr. Julian C. Cummings Office: PB-111
Caltech/CACR, MC 158-79 Phone: 626-395-2543
1200 E. California Blvd. Fax: 626-584-5917
Pasadena, CA 91125
-----Original Message-----
From: blitz-dev-admin_at_[hidden] [mailto:blitz-dev-admin_at_[hidden]]On Behalf Of Oliver Schoenborn
Sent: Friday, September 27, 2002 2:41 PM
To: blitz-dev_at_[hidden]
Subject: [Blitz-dev] cvs head on SGI: report
Hello, I don't know if the following may be useful but here is the result of doing a cvs checkout of blitz:
Built on SGI IRIX64 6.5.13m using CC 7.3.1.2m (let me know if you need the config.log):
1.. make: no problem;
2.. make check-examples: all pass, except that cfd seems to never end. At t=10s it was still evolving, last output before I killed it was:
Average magnitude of r is 5.76355e-09
CG: Iter 0 rho = 9.18325e-12
Velocity field: norm = [0 0 1.02321e-26 ]
Maximum velocity is 2.92633e-13 m/s
Set time step to 0.01 s
Iteration 971 Time = 9.701 s
3.. make check-benchmarks:
a.. tinydaxpy.cpp: I had to add '#include "tinyvec-et.h"' to get it to link
b.. acou3df.f and acou3df2.f both had about four lines of code that ended with a ";", causing the f77 compiler to fail; eliminating the ";" worked
c.. neither acoustic nor acou3d link; CC complains about missing symbols, which seem to be fortran symbols, but I couldn't find from which library
d.. qcd compiled but core dumped at length = 56234:
Core from signal SIGBUS: Bus error
qcdf(M = '3fe01767'X, res = '10039630'X, src = '10039698'X, V = 1, iters = 500000) ["qcdf.f":11, '1001c858'X]
4.. make check-testsuite: everything works (all tests pass), except that minsumpow.cpp and peter-nordlund-3.cpp fail to compile due to template instantiation problem. The problem is with maxIndex and minIndex: a line of code that has minIndex(A) is fine, but the line of code that has minIndex(fn(A)), where fn is pow or sqr, fails to compile due to "unexpected function as template parameter".
HTH,
Oliver
_______________________________________________
Blitz-dev mailing list
Blitz-dev_at_[hidden]
http://www.oonumerics.org/mailman/listinfo.cgi/blitz-dev