arbitrary precision arithmetic library
C Assembly Makefile Other
Latest commit d4c9a98 Jan 12, 2017 @suiginsoft committed on GitHub Update README.md
Permalink
Failed to load latest commit information.
bench
check
src
.gitignore
.travis.yml
LICENSE
Makefile
README.md
TODO
config.def.h
config.def.x86_64.inc
config.mk
hebimath.h.in
internal.h
runtests.sh

README.md

hebimath - arbitrary precision arithmetic library

Build Status Coverity Scan Quality Gate License

Overview

Hebimath is a library for working with big numbers. It's in the same class of libraries as GNU MP, MPI, TomsFastMath, libzahl and others. As such, Hebimath shares many features with these libraries, but has some unique ones as well.

Hebimath has the following design features:

  • Written in C99 and assembly language for POSIX userspace platforms. Is forward compatible with C11.
  • High-level interface for large integers, rationals and floating-point numbers.
  • Low-level maths kernel functions that operate on natural numbers, optimized for common hardware platforms. Supports both compile-time and runtime function multi-versioning, depending on your needs and requirements.
  • Ability to override the memory allocator for each high-level number primitive.
  • Can override the default memory allocator for each thread.
  • Can use fixed-size buffers instead of an allocator when initializing a primitive, useful when using statically allocated memory or fast ad-hoc memory allocations on the stack.
  • Can setup a separate allocator for internal temporary/scratchpad allocations.
  • Smallest unit size for natural numbers and integers is 16 bytes. This larger unit size aids in shrinking the amount of code bloat for unrolled kernel functions.
  • Memory buffers are correctly aligned for use with SIMD instructions.

Hebimath adheres to these principles in its implementation:

  • Be compact, concise and readable. Favor small code but do not over do it. In the low-level kernel functions favor runtime speed without sacrificing reliability & security best practices.
  • Avoid unnecessary memory allocations and delay allocations until they are absolutely required for output variables.
  • Allocate all temporary memory upfront for algorithms using sharp upper bounds on the input argument lengths. No allocations/deallocations at each recursive or iterative level.
  • Avoid mutating output variables when errors occur and document what may be left in an unspecified state.
  • Load values from data structures into local variables if they are used across function call boundaries to avoid reloading them more than once. Useful on CPUs with lots of callee-saved registers.

What's with the name?

ヘビ 【Hebi】

  1. Snake. Serpent. Reptile. "Best watch out for that SICP snake!"

ヘビー 【Hebī】

  1. Heavy. Big. Huge. Of great weight; hard to lift or carry: "A heavy load."
  2. Extreme. Intense. Scary. Tough. "A heavy drinker."
  3. Applying power; a final burst of energy.

Hebimath