mini-gmp-plus

This is a fork of the mini version of the GNU Multi-Precision library with some patches/optimizations for usage in numerical geometry and easy-to-use CMakeList (tested under Windows, Linux, Mac).

My changes are listed in mini-gmp-plus-Changelog, and indicated by [Bruno Levy] tags in the sources, with the date and reason for the modification.

Design rationale

In numerical geometry, one manipulates numbers that are relatively small (a few 64-bit limbs), so Karatsuba multiplication is not needed, hence mini-gmp suffices. On the other hand, it is interesting to avoid dynamic allocation, hence numbers smaller than a certain threshold (that corresponds to 5 64-bits numbers by default) are allocated on the stack. Modern processors have some instructions that can significantly optimize some multi-precision operations, such as finding the number of leading and trailing zeroes in an integer, shifting 128 bit numbers, multiplying two 64-bit numbers and obtaining the 128-bits result, or adding 64-bit numbers and propagating the carry in multi-limb operations. While 64-bit add/subtract with carry propagation is recognized and optimized by modern compilers, it is not the case for the other operations, that are implemented by some compiler-dependent intrinsics. For this reason, I needed to create a new bitops64.h file to abstract them. These two modifications (local storage and 64 bits intrinsics) introduced a couple of very simple yet not completely trivial changes (I must confess I did not have them right the first time), so it is important - I would say vital - to be able to run the testsuite. For this reason I also created new CMakeLists.txt files as well as a github action that runs the (already existing, thank you !) mini-gmp testsuite under Linux,Mac and Windows.

Features and specificities

As compared to mini-gmp, mini-gmp-plus has the following differences:

• limitation: limb size is fixed as 64 bits
• numbers smaller than a certain size (5 limbs) are stored in the mpz_t structure for better multithreading (avoids most dynamic allocations)
• the file bitops64.h, not part of mini-gmp, contains some wrappers for efficient bit operations on 64-bit numbers, for GCC, Clang and VisualC++ using these compiler's intrinsics
• mini-gmp-plus is compiled as a dynamic library
• CMakeLists.txt optionally builds and runs non-regression tests using CTest, use cmake -DMINI_GMP_PLUS_WITH_TESTS=1 to compile and run the testsuite.
• The github action builds and runs the testsuite systematically. Under Windows, the github action installs gmp (through vcpkg) and declares it (through pkgconf) so that tests/CMakeLists.txt sees it. Note that gmp is only needed if you want to run the non-regression testsuite.

Bruno Lévy, November 2025