/* Written by Clifford Wolf <clifford@clifford.at>, http://www.clifford.at/
 *
 * Short example program to calculate the APPROXIMATED sum of the first
 * 1 000 000 multiples of pi (including pi itself) brute-force, as example
 * for how to waste cpu cycles in SSE instructions.  ;-)
 *
 * Note that a significant higher setting for LIMIT would just return bogus
 * results because we are running into the limitations of single-precision
 * floating point then...
 *
 * Compile: gcc -O2 -msse demo.c -o demo
 * SSE Performance: time ./demo +1000
 * FPU Performance: time ./demo -1000
 */

#define LIMIT 1000000
#define PI 3.1416

typedef float v4sf __attribute__ ((mode(V4SF)));

#define GCC_VERSION (__GNUC__ * 10000 + \
		__GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__)

float waste_time_sse()
{
        float buf[4] = { PI*1, PI*2, PI*3, PI*4 };
        v4sf counter, sum, step;
	unsigned int i;

        sum = counter = __builtin_ia32_loadups(buf);

	buf[0] = buf[1] = buf[2] = buf[3];
        step = __builtin_ia32_loadups(buf);

	for (i=1; i < LIMIT/4; i++)
#if GCC_VERSION >= 30300
		sum += (counter += step);
#else
	/* gcc versions prior to 3.3.0 did not overload math operators */
	{
		counter = __builtin_ia32_addps(counter, step);
		sum = __builtin_ia32_addps(counter, sum);
	}
#endif

        __builtin_ia32_storeups(buf, sum);

        return buf[0]+buf[1]+buf[2]+buf[3];
}

float waste_time_fpu()
{
	float sum = PI, counter = PI;
	unsigned int i;

	for (i=1; i < LIMIT; i++) 
		sum += (counter += PI);

	return sum;
}

int main(int argc, char ** argv)
{
	int i = argc == 2 ? atoi(argv[1]) : 0;
	if ( !i ) {
		printf("SSE: %f\n", waste_time_sse());
		printf("FPU: %f\n", waste_time_fpu());
	}
	while (i>0) { waste_time_sse(); i--; }
	while (i<0) { waste_time_fpu(); i++; }
	return 0;
}