/*
 * Tested with:
 * gcc version 4.8.2 (Ubuntu 4.8.2-19ubuntu1)
 *
 * Without OMP, this works just fine with no special options:
 * $ gcc test.c -o test && ./test alpha bravo charlie
 * 0x7ffc32236700 ./test
 * 0x7ffc32236700 alpha
 * 0x7ffc32236700 bravo
 * 0x7ffc32236700 charlie
 *
 * Broken with OMP:
 * $ gcc -fopenmp test.c -o test && ./test alpha bravo charlie
 * 0x7f1c78b1ea40 charlie
 * 0x7f1c79b20a40 alpha
 * 0x7ffc450c4d30 ./test unaligned!
 * 0x7f1c7931fa40 bravo
 *
 * The rest assumes -fopenmp:
 *
 * Using -mavx2 and -DUSE_VTYPE (declare as vtype) seems to help.
 * Just -mavx2 (and declaring char with aligned attribute) does not help.
 * Just -mincoming-stack-boundary=4 doesn't help. I wasn't allowed to try lower.
 * Just -mpreferred-stack-boundary=5 obviously doesn't help.
 * Using both incoming & preferred doesn't help either (it got worse, now all
 *                                                      threads got misaligned).
 * -mstackrealign make no difference (even with incoming & preferred).
 *
 */

#include <stdio.h>
#include <string.h>
#ifdef _OPENMP
#include <omp.h>
#endif
#if USE_VTYPE && __AVX2__
#include <immintrin.h>
#endif

/*
 * I had to place this in a separate function to avoid having the unaligned
 * branch optimized away (even using -O0), since gcc is totally convinced it
 * simply can't be unaligned!
 */
void out(char *pt)
{
	if ((size_t)pt & 31)
		printf("%s unaligned!\n", pt);
	else
		puts(pt);
}

int main(int argc, char **argv)
{
	int i;

#ifdef _OPENMP
#pragma omp parallel for
#endif
	for (i = 0; i < argc; i++) {
#if USE_VTYPE && __AVX2__
		__m256i _pt[1024 / sizeof(__m256i)];
		char *pt = (void*)_pt;
#else
		__attribute__ ((aligned(32))) char pt[1024];
#endif

		sprintf(pt, "%p ", pt);
		strcat(pt, argv[i]);
		out(pt);
	}
	return 0;
}