LuaJIT and C OpenMP trials

Makefile

# make with different optimisation levels:
# make O_LEVEL=2 will make with -O2

CC=gcc
O_LEVEL=0
OMP=-fopenmp
CFLAGS=-std=c99 -Wall -pedantic $(OMP) -O$(O_LEVEL)
NL=10
SZ=64

test-$(O_LEVEL): test.c
	$(CC) $(CFLAGS)        -shared -fPIC test.c -o libtest.so
	$(CC) $(CFLAGS) -DMAIN               test.c -o test
	time ./test          $(NL) $(SZ)
	time luajit test.lua $(NL) $(SZ)

test.c

#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#include <math.h>
#include <time.h>

// for running from lua:
// gcc -std=c99 -O3 -Wall -pedantic -fopenmp -shared -fPIC test.c -o libtest.so

// for running from C:
// gcc -std=c99 -O3 -Wall -pedantic -fopenmp -DMAIN test.c -o test

#ifdef _OPENMP
#include <omp.h>
#else
static int
omp_get_num_procs(void)
{
  return 1;
}
#endif

static double
dot_prod(int n, double a[n], double b[n])
{
  double sum = 0;

#ifdef _OPENMP
#pragma omp parallel for reduction(+:sum)
#endif
  for (int i = 0; i < n; ++i)
    sum += a[i] * b[i];

  return sum;
}

void
do_parallel(int NL, int SZ)
{
  double *a = malloc(SZ * sizeof(double));
  double *b = malloc(SZ * sizeof(double));
  double sum = 0;
  assert(a && b);

  for (int i = 0; i < SZ; ++i)
    a[i] = b[i] = sqrt(i+1);

  double t0 = clock();

  for (int l = 0; l < NL; ++l)
    sum += dot_prod(SZ, a, b);

  double dt = (clock()-t0)/CLOCKS_PER_SEC;

  free(a);
  free(b);

  printf("sum = %g, time = %.3f, threads = %d\n", sum, dt/omp_get_num_procs(), omp_get_num_procs());
}

#ifdef MAIN

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

int main(int argc, char **argv)
{
  int NL = 10;
  int SZ = 64;

  if (argc >= 2) NL = atoi(argv[1]);
  if (argc >= 3) SZ = atoi(argv[2]);

  for (int l = 0; l < 10; ++l)
    do_parallel(NL, SZ);

  exit(EXIT_SUCCESS);
}

#endif

test.lua

local ffi = require "ffi"
local testLib = ffi.load("./libtest.so")

ffi.cdef[[
  void do_parallel(int,int);
]]

local NL = tonumber(arg[1]) or 10
local SZ = tonumber(arg[2]) or 64

for l=1,10 do
    testLib.do_parallel(NL, SZ)
end

os.exit(0)