benchmark of a simple 2D pixel pattern

bench.c

/* bench.c : XOR pattern without any display/output to get a baseline benchmark */
#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <sys/time.h>

#define BGRX8888(r, g, b) \
	((uint32_t)((unsigned char)(b)) << 24 | \
	(uint32_t)((unsigned char)(g)) << 16 | \
	(uint32_t)((unsigned char)(r)) << 8)
#define XRGB8888(r, g, b) \
	((uint32_t)((unsigned char)(r)) << 16 | \
	(uint32_t)((unsigned char)(g)) << 8 | (unsigned char)(b))

int main()
{
	unsigned i;
	uint32_t *buf;
	unsigned width = 640, height = 480;
	unsigned x, y;
	unsigned count = 10000;
	size_t rowbytes = width * sizeof(*buf);
	size_t buf_len = height * rowbytes;
	struct timeval start_tv, end_tv;
	double dur;

	buf = malloc(buf_len);
	gettimeofday(&start_tv, NULL);
	for (i = 0; i < count; i++) {
#if 0 /* memset - should be about 25x faster */
		memset(buf, 0, buf_len);
#else
		uint32_t *p = buf;
		for (y = 0; y < height; y++) {
			for (x = 0; x < width; x++)
				p[x] = BGRX8888(((x ^ y) + i) / 7,
				        (x ^ y) + i, ((x ^ y) + i) / 3);
			p = (void*)((char*)p + rowbytes);
		}
#endif
	}
	gettimeofday(&end_tv, NULL);

	if (end_tv.tv_usec < start_tv.tv_usec) {
		end_tv.tv_sec--;
		end_tv.tv_usec += 1000000;
	}
	end_tv.tv_usec -= start_tv.tv_usec;
	end_tv.tv_sec -= start_tv.tv_sec;
	dur = end_tv.tv_usec / 1000000. + end_tv.tv_sec;
	printf("total:%g\n", dur);
	printf("fps:%g\n", count / dur);
	return 0;
}