FXdiv

Header-only library for division via fixed-point multiplication by inverse

On modern CPUs and GPUs integer division is several times slower than multiplication. FXdiv implements an algorithm to replace an integer division with a multiplication and two shifts. This algorithm improves performance when an application performs repeated divisions by the same divisor.

Features

• Integer division for uint32_t, uint64_t, and size_t
• Header-only library, no installation or build required
• Compatible with C99, C++, OpenCL, and CUDA
• Uses platform-specific compiler intrinsics for optimal performance
• Covered with unit tests and microbenchmarks

Example

#include <fxdiv.h>

/* Division of array by a constant: reference implementation */
void divide_array_c(size_t length, uint32_t array[], uint32_t divisor) {
  for (size_t i = 0; i < length; i++) {
    array[i] /= divisor;
  }
}

/* Division of array by a constant: implementation with FXdiv */
void divide_array_fxdiv(size_t length, uint32_t array[], uint32_t divisor) {
  const struct fxdiv_divisor_uint32_t precomputed_divisor =
    fxdiv_init_uint32_t(divisor);
  for (size_t i = 0; i < length; i++) {
    array[i] = fxdiv_quotient_uint32_t(array[i], precomputed_divisor);
  }
}

Status

Project is in alpha stage. API is unstable. Currently working features:

Platform	uint32_t	uint64_t	size_t
x86-64 gcc	Works	Works	Works
x86-64 MSVC	Works	Works	Works
x86 gcc	Works	Works	Works
x86 MSVC	Works	Works	Works
ARMv7 gcc	Works	Works	Works
PPC64 gcc	Works	Works	Works
PNaCl clang	Works	Works	Works
Asm.js clang	Works	Works	Works
CUDA	Untested	Untested	Untested
OpenCL	Untested	Untested	Untested

References

• Granlund, Torbjörn, and Peter L. Montgomery. "Division by invariant integers using multiplication." In ACM SIGPLAN Notices, vol. 29, no. 6, pp. 61-72. ACM, 1994. Available: gmplib.org/~tege/divcnst-pldi94.pdf