LAPACK Archives

[Lapack] CBLAS test suite fails on Mac OS X

Dear all;

I've had problems getting the test suite for CBLAS to behave on Mac OS
X.  It would seem that the test suite relies on tricking the runtime
linker to resolve certain symbols (xerbla) in the legacy BLAS library
differently compared to what was done when the library was originally
linked.  This doesn't work on Darwin--at least not without playing with
linker flags, and even after playing I couldn't get it to work reliably.

I therefore propose to modify the test suite such that for each
invocation of a subroutine, it spawn a child process and the exit status
of the child is used to determine whether the routine succeeded or not.
 This seems to work fine on the systems I have access to (Darwin/ppc,
Darwin/ppc64, Darwin/x86_64, Linux/x86_64 and FreeBSD/x86).  The
drawback is that since this makes use of fork(3), it will have most
likely have issues on Windows.

The patch is big because all subroutine calls in the test suite are
wrapped.  The actual code to make this work is a 20-odd line macro in
cblas_test.h.  I'd be happy to discuss this further with anybody interested.

I don't know where to go with this; hence I'm posting it to both lapack@
and blast-comm at .  Please let me know if there's some place better to
turn to.

// Best wishes; Johan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cblas-reference-20030223-testfork.patch.gz
Type: application/gzip
Size: 30582 bytes
Desc: not available
Url : 
http://lists.eecs.utk.edu/mailman/private/lapack/attachments/20110124/c1e622a4/cblas-reference-20030223-testfork.patch-0001.bin
 

<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or