Hi Saba, I am hearing from other LAPACKers and essentially they are
confirming the same difficulty as you are relating. They have had no
problem on MAC OS to build shared libraries with C++, but as soon they use
gfortran, things seem to be eerie indeed. Julien.
On 8/15/13 10:58 AM, "Langou, Julien" <Julien.Langou@Domain.Removed> wrote:
Hi Saba, no idea on my end. When I see a static build works, but a dynamic
build that does not, I try to see if what I am linking against with the
dynamic library is really what I think I am linking with. Besides this
basic check, I do not know, Julien.
On 8/12/13 2:11 PM, "Saba Sehrish" <ssehrish@Domain.Removed> wrote:
I am trying to build a shared lapack library on my mac book pro with OS X
As a first step I build libblas.dylib, I am aware of the fact that Mac
does have a default build for that but for some of the system setups and
experiments we need to have our own working build on Mac. So using system
default is not being considered as one of the options.
I am using the source tar ball from net lib website and version
lapack-3.4.2. After making a few modifications to the Makefile esp.
changing the library extension to dylib and then using the -dynamic,
-fPIC and -O2 flags with gfortran. I get no errors when I build
Then the next step I try is
And when I look at the level 2 and level 3 tests, the output files have
similar errors such as
ROUTINES PASS COMPUTATIONAL TESTS IF TEST RATIO IS LESS THAN 16.00
RELATIVE MACHINE PRECISION IS TAKEN TO BE 1.2E-07
** On entry to SGEMV parameter number 1 had an illegal value
I want to know if this is some known error, or what is causing tests 2
and 3 to fail.
Let me know if you need any more information.
Using a static build (libblas.a) on Mac I didn't get any error in tests.
Lapack mailing list