Very interesting chunk of code and very interesting performance results.

weird.cc

#include <cstdio>

const int Size = 256;
static int opti_killer = 2;

int test1() {
	int *buf = new int[Size*Size*2];
	int out = 0;
#define bufget(buf, x, y, z) buf[y*Size+x+((z)%2)]
	for (int z = 0; z < Size; z++) {
	for (int y = 0; y < Size; y++) {
	for (int x = 0; x < Size; x++) {
		if (z == 0) {
			bufget(buf, x, y, z) = y*Size+x;
		} else {
			bufget(buf, x, y, z) = bufget(buf, x, y, z-1)+opti_killer;
		}
		if (z == Size-1 && y == Size-1 && x == Size-1) {
			out = bufget(buf, x, y, z);
		}
	}}}
#undef bufget
	delete [] buf;
	return out;
}

int test2() {
	int *buf = new int[Size*Size*Size];
	int out = 0;
#define bufget(buf, x, y, z) buf[((z)*Size+y)*Size+x]
	for (int z = 0; z < Size; z++) {
	for (int y = 0; y < Size; y++) {
	for (int x = 0; x < Size; x++) {
		if (z == 0) {
			bufget(buf, x, y, z) = y*Size+x;
		} else {
			bufget(buf, x, y, z) = bufget(buf, x, y, z-1)+opti_killer;
		}
		if (z == Size-1 && y == Size-1 && x == Size-1) {
			out = bufget(buf, x, y, z);
		}
	}}}
#undef bufget
	delete [] buf;
	return out;
}

int test1_n(int n) {
	int out = 0;
	for (int i = 0; i < n; i++) {
		out += test1();
	}
	return out;
}

int test2_n(int n) {
	int out = 0;
	for (int i = 0; i < n; i++) {
		out += test2();
	}
	return out;
}

const int N = 100;

int main(int argc, char **argv) {
	if (argc == 1)
		return 0;
	if (argc < 999)
		opti_killer = 1;
	switch (argv[1][0]) {
	case '1':
		printf("starting 1\n");
		printf("%d\n", test1_n(N));
		break;
	case '2':
		printf("starting 2\n");
		printf("%d\n", test2_n(N));
		break;
	default:
		printf("unknown option: %c\n", argv[1][0]);
	}
	return 0;
}

/*
[nsf @ ~]$ g++ -O3 -o main main.cc
[nsf @ ~]$ time ./main 1
starting 1
6579000

real    0m2.636s
user    0m2.633s
sys     0m0.003s
[nsf @ ~]$ time ./main 2
starting 2
6579000

real    0m0.908s
user    0m0.490s
sys     0m0.417s
[nsf @ ~]$ clang++ -O3 -o main main.cc
[nsf @ ~]$ time ./main 1
starting 1
6579000

real    0m2.176s
user    0m2.177s
sys     0m0.000s
[nsf @ ~]$ time ./main 2
starting 2
6579000

real    0m1.938s
user    0m1.503s
sys     0m0.433s

[nsf @ ~]$ g++ --version
g++ (GCC) 4.8.2
[nsf @ ~]$ clang++ --version
clang version 3.3 (tags/RELEASE_33/final)

intel i5-3470 "Ivy Bridge"
*/