topic Strange linker warnings on Mac OS X 10.7 in Intel® oneAPI Math Kernel Library

Strange linker warnings on Mac OS X 10.7

jd_weeks — Mon, 30 Jan 2012 22:29:57 GMT

We just upgraded to MKL 10.3 update 8. One of our engineers is using Mac OS X 10.7 with Xcode 4.1 and gcc 4.2.1.

He is seeing lots of linker warnings like these (just a sample for about 2600 warnings):

ld: warning: could not create compact unwind for _dqform_: cfa had negative offset (dwarf might contain epilog)

ld: warning: could not create compact unwind for _dfdjc1_: register saved more than once (might be shrink wrap)

ld: warning: could not create compact unwind for _mkl_lapack_spotrf: r15 is saved too far from return address

ld: warning: could not create compact unwind for _mkl_vml_serv_openmp_get_num_procs: does not use RBP or RSP based frame

ld: warning: could not create compact unwind for _mkl_lapack_cgbtrf: stack subq instruction is too different from dwarf stack size

Most of the warnings are the "too far from return address" warnings.

What can I do about this?

Thanks,

John Weeks

Strange linker warnings on Mac OS X 10.7

barragan_villanueva_ — Tue, 31 Jan 2012 04:13:21 GMT

Hi,

It's because of SnowLeopard (darwin10) unwind feature changes.. but its still ld: warnings.

It is recommended:

add -no_compact_unwind to the link line. This tells the linker to not
produce compact unwind information from the dwarf unwind info in .o
files. Then at runtime the darwin unwinder will fallback and use the
slow dwarf unwind info.

So please try to add option "-Wl,-no_compact_unwind" for linking.

We are using the IPP library

Johannes_F_ — Thu, 14 Apr 2016 10:11:41 GMT

We are using the IPP library in out application.

Unfortunately, the linker option -no_compact_unwind disables try/catch in my application. Somewhere else I found -keep_dwarf_unwind. If additionally giving this option, exceptions work again, also, everything else seems to work.

I'm not at all understanding what these options mean. Does any more knowlegeable person advise against doing so?