libpvn

A small portable C library with several utility functions.

(... work in progress ...)

This software is a supplementary material for:

• the paper doi:10.1016/j.cam.2024.116003,
• the preprint arXiv:2407.13116.

Some functions have been adapted from those in the repositories JACSD and VecJac.

Building

The library has been successfully built using:

compiler	uname	platform
clang(1)	Darwin	x86_64
clang(2)	FreeBSD	amd64
clang(2)	FreeBSD	arm64
gcc(3)	Darwin	x86_64
gcc(2)	FreeBSD	amd64
gcc(2)	FreeBSD	arm64
gcc(4)	Linux	aarch64
gcc(4)	Linux	ppc64le
gcc(4)	Linux	x86_64
icx(5)	Linux	x86_64
nvc(6)	Linux	x86_64

Recent versions of the compilers have been provided by or used on:

1. Apple (clang 13.0.0),
2. FreeBSD (clang 18.1.5 and GCC 13.2.0),
3. Homebrew (GCC 14.2.0_1),
4. openSUSE Tumbleweed (GCC 14.2.1),
5. Intel oneAPI (2025.0.0),
6. NVIDIA HPC SDK (24.11).

Examples of building the library:

cd src
# query the building options (GNU make is necessary)
make help
# the output should be something like:
# make [COMPILER=clang|gcc|icx|nvc|icc] [COMPILER_PREFIX=...] [COMPILER_SUFFIX=...] [MARCH=...] [NDEBUG=0|1|2|3|...] [PRINTOUT=ERR|OUT] [VECLEN=...] [CR_MATH=...] [OPENMP=...] [PROFILE=...] [SAFE=...] [DYNAMIC=dylib|so] [QUADMATH=-lquadmath] [all|clean|help]
# where gcc is the default compiler to be used on Linux, and clang is otherwise
#
# a release build with icx on x86_64 Linux with a high optimization level
make COMPILER=icx NDEBUG=3 clean all
#
# a release build with the Homebrew's gcc on x86_64 macOS with a low optimization level
make COMPILER=gcc COMPILER_SUFFIX=-14 NDEBUG=g clean all
#
# a debug build with clang on FreeBSD (note the usage of gmake instead of make)
gmake clean all

Using

Include the pvn.h header file in your C/C++ sources where needed and link with the libpvn.a static library, as shown in GNUmakefile. There are no mutable global/static variables present, unless profiling has been enabled, but some static constant lookup tables are used. The functions can safely be called from multiple threads on unrelated data without locking. You might wish to tune the compiler flags to match yours, especially if you are building a dynamic library, a static executable, and/or want to use OpenMP.

Fortran (column-major) array order is assumed for the functions that operate on matrices.

A function with the name ending with an underscore (_) should be callable from Fortran without an explicit interface, just by declaring it EXTERNAL (without the underscore).

Several correctly-rounded mathematical functions provided by the CORE-MATH project are used. Their slightly modified implementations included here will be taken if CR_MATH is not set. Otherwise, set the CR_MATH variable in a [g]make invocation to the cloned core-math source code directory path. If the object files have not been prepared beforehand, the source files will be compiled in either case. The object files will be integrated into libpvn.a for easier re-use by other software linked with it.

The MARCH option can be used to indicate another CPU architecture than native (for clang, gcc, and nvc) or Host (for icx and icc).

The OPENMP option enables OpenMP and its content is appended to the compiler's flags. Set it to true if no additional compiler flags are desired. Bear in mind, however, that this will introduce a dependency on the OpenMP runtime library!

If long double is not the 128-bit floating-point datatype, the PVN_QUADMATH macro should be set automatically to the name of a library implementing quadruple precision arithmetic, unless COMPILER=clang is used.

The SAFE variable lists, as one or more comma-separated short names, the components for which a safe implementation is requested:

• cma makes the complex multiplication and FMA safe from unwarranted overflow;
• ran requests that all random numbers (not only those explicitly created as such) lie in the range [xxx_MIN*2^p,xxx_MAX/4], where p=0 by default;
• sv2 indicates that hypotf and hypot are correctly rounded and can be used for the singular value decomposition.

So, SAFE=cma,sv2, e.g., requests the cma and sv2 safe implementations, but not the ran one.

The other options are for testing, debugging, and profiling, and should not be set unless their effects are fully understood.

Caveat: certain parts of the library will not work on big-endian systems! Also, several functions will not work on Windows, and probably on other untested systems as well.

Examples

Several examples and tests are built as *.exe executables in the src and the etc subdirectories and are meant to be run from there.

Documentation

...is a work in progress.

Run doxygen in the doc subdirectory to generate the HTML documentation.

This work has been supported in part by Croatian Science Foundation under the project IP-2014-09-3670 (MFBDA).