Solaris studio OpenMP runtime library name

Question

Hello:
Imagine I'm using some functions from the OpenMP API as omp_get_wtime(), for example. If I use gcc compiler I need to pass to the linker the order -lgomp as libgomp.a|so is the GNU OpenMP implementation (I know that I can pass -fopenmp instead the -lgomp, but I want to pass -lgomp for example in order to use omp_get_wtime() but not to activate other OpenMP directives in the program). If I use the Intel compiler, the order should be -liomp5 -lpthread and for PGI compilers -lpgmp
But what is the name of the solaris sturio OpenMP runtime library? Which flag should I pass to the linker in order to use the API functions of the OpenMP implementation? I know -xopenmp flag for activate OpenMP, but I want to link only the runtime library
Thanks

Answer 1

Thank you again for your answer:
You are right, I should say my platform. I'm using an Intel Core i5 processor running Debian GNU/Linux (64 bits). I know that the compiler supports only Oracle and RHEL linuxes, but I have downloaded from Oracle and installed the file SolarisStudio12.3-linux-x86-bin.tar.bz2 in my Debian box without problems. The compiler seems to run very well, and the obtained results from the generated binaries are correct, so I suppose that the compiler runs too in Debian.
Inspecting in detail the -xdryrun output I have found the actual path of libmtsk.so in my insyallation: /opt/sun/solarisstudio12.3/prod/lib/compilers/rtlibs/usr/lib/amd64
Why this path is not loaded by default by the compiler as when I use -xopenmp, I don't know. Now, my program runs.
The reason to use omp_get_wtime() function when I not use OpenMP pragmas is because not exists (at least I don't know) any method to obtain millisecond precision using ISO C99. I know functions as clock_gettime() but is only POSIX complaint. The OpenMP is not C99 of course, but OpenMP is present in almost all platforms and comes with the majority of moder compilers. If anyone knows an ISO C99 method to obtain millisecond (or better) precision to compute time, I would like to know about :)
About your suggestion of compile the program with -xopenmp flag and run with OMP_NUM_THREADS=1, the execution of a program compiled without OpenMP and with OpenMP+OMP_NUM_THREADS=1 is not the same. The latter one is slower due to OpenMP in complicated loops with barriers or other OpenMP orders as flush() for example, so I need to compile pure non-OpenMP code. Other case is, for example, when you want to use the break clause inside a for loop that is possible to stop before the last iteration. In OpenMP the break sentence cannot be used and you should to use dirty tricks to simulate it. This makes a bit slower the OpenMP execution but with the correct number of processors is faster than serial code, but definitively running it with OMP_NUM_THREADS=1 is not a good idea.
Thanks again for your time