benchmark of SIMD

inner_ip_benchmark.c

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <arpa/inet.h>

typedef int v4si __attribute__ ((vector_size (16)));

//                 /8          /10         /12         /16
const v4si mask = {0x000000FF, 0x0000C0FF, 0x0000F0FF, 0x0000FFFF};

//                10.0.0.0     100.64.0.0  172.16.0.0  192.168.0.0
const v4si netw = {0x0000000A, 0x00004064, 0x000010AC, 0x0000A8C0};

#ifdef NO_SSE
__attribute__ ((__target__("no-sse")))
#endif
static inline int priv(int ip) {
	v4si r = ((ip & mask) == netw);
	return !!(r[0] | r[1] | r[2] | r[3]);
}

static const unsigned long N = 1000000;
static void usage(char *cmd) {
	printf("usage: %s <ip> [loop_count=%lu]\n", cmd, N);
}

int main(int argc, char *argv[]) {
	if (argc < 2) {
		usage(argv[0]);
		return -1;
	}

	size_t n = argc > 2 ? atol(argv[2]) : N;
	int ip = inet_addr(argv[1]);

	clock_t start = clock();
	for (unsigned long i = 0; i < n; ++i) {
		priv(ip);
		__asm__("");
	}
	clock_t end = clock();

	unsigned long cost = end - start;
	printf("inner ip: %u, cost %.5f seconds, %.5f nano seconds per each\n", priv(ip),
			((double) cost) / CLOCKS_PER_SEC,
			((double) cost) * 1000000000 / CLOCKS_PER_SEC / n);
	return 0;
}

run.sh

#!/bin/bash

ARCH=(
native
nocona
core2
nehalem
westmere
bonnell
silvermont
knl
sandybridge
ivybridge
haswell
broadwell
)

gcc -O0 -DNO_SSE inner_ip_benchmark.c -o inner_ip_benchmark

echo "./inner_ip_benchmark ${1:-0.0.0.0} ${2:-100000000}"
./inner_ip_benchmark ${1:-0.0.0.0} ${2:-100000000}
echo

for arch in ${ARCH[@]}; do
	gcc -O3 -march=$arch inner_ip_benchmark.c -o inner_ip_benchmark.$arch
done

for arch in ${ARCH[@]}; do
	echo "./inner_ip_benchmark.$arch ${1:-0.0.0.0} ${2:-100000000}":
	./inner_ip_benchmark.$arch ${1:-0.0.0.0} ${2:-100000000}
	echo
done