#ifdef HAVE_CONFIG_H
#  include "config.h"
#endif

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include <platon/str/strmmx.h>

/* Function for align string length to multiple of 8 
 */
static int align(const int n)
{
	return ((n >> 3) + (n % 8 != 0)) << 3;
}

char *PLATON_FUNC(strcpymmx)(void *dest, const void *src)
{
	int len = strlen(src);
	char *ret = PLATON_FUNC(memcpymmx)(dest, src, len);
	ret[len] = '\0';
	return ret;
}

char *PLATON_FUNC(strncpymmx)(void *dest, const void *src, size_t n)
{
	char *ret = PLATON_FUNC(memcpymmx)(dest, src, n);
	ret[n] = '\0';
	return ret;
}

#if defined(__sparc__)

/* XXX: this should be done probably through macros, if possible */

void *PLATON_FUNC(memcpymmx)(void *dest, const void *src, size_t n)
{
	return memcpy(dest, src, n);
}

void *PLATON_FUNC(memsetmmx)(void *s, int c, size_t n)
{
	return memset(s, c, n);
}

int PLATON_FUNC(memcmpmmx)(const void *s1, const void *s2, size_t n)
{
	return memcmp(s1, s2, n);
}

#else

/* Function copy memory regions using MMX */
void *PLATON_FUNC(memcpymmx)(void *dest, const void *src, size_t n)
{
	if ((n % 8) != 0)
		n = align(n);
	n = n >> 3;
	__asm__ __volatile__("1:\n\t"
			"movq (%%eax),%%mm0\n\t"
			"movq %%mm0,(%%ebx)\n\t"
			"addl $8,%%eax\n\t"
			"addl $8,%%ebx\n\t"
			"loop 1b\n\t"::"a"(src), "b"(dest), "c"(n)
			:"memory");
	return dest;
}

/* Function set every item of buffer to char 'c' using MMX */
void *PLATON_FUNC(memsetmmx)(void *s, int c, size_t n)
{
	char set[8];
	set[0] = set[1] = set[2] = set[3] = set[4] = set[5] = set[6] = set[7] = c;
	if ((n % 8) != 0)
		n = align(n);
	n = n >> 3;
	__asm__ __volatile__("movq (%%ebx),%%mm0\n\t"
			"1:\n\t"
			"movq %%mm0,(%%eax)\n\t"
			"addl $8,%%eax\n\t"
			"loop 1b\n\t"::"a"(s), "b"(&set), "c"(n)
			:"memory");
	return s;
}

/* Function compare two memory regions using MMX, returns 0 or 1 */
int PLATON_FUNC(memcmpmmx)(const void *s1, const void *s2, size_t n)
{
	if ((n % 8) != 0)
		n = align(n);
	n = n >> 3;
	__asm__ __volatile__("1:\n\t"
			"movq (%%eax),%%mm0\n\t"
			"movq (%%ebx),%%mm1\n\t"
			"addl $8,%%eax\n\t"
			"addl $8,%%ebx\n\t"
			"pcmpeqw %%mm0,%%mm1\n\t"
			"movd %%mm1,%%edx\n\t"
			"cmpl $0xFFFFFFFF,%%edx\n\t"
			"jne 2f\n\t"
			"loop 1b\n\t"
			"jmp 3f\n\t"
			"2:\n\t"
			"movl $1,%%eax\n\t"
			"leave\n\t"
			"ret\n\t" "3:\n\t"::"a"(s1), "b"(s1), "c"(n));
	return 0;
}

#endif /* #if defined(__sparc__) */

#if defined(SELF) || defined(SELFTEST) || defined(SELF_MMX)

int main(int argc, char *argv[])
{
	char *pole1, *pole2;
	int n, bsize;
	register int i;

	if (argc < 3) {
		fprintf(stderr, "Use: %s <buffersize> <memory>\n", argv[0]);
		return 1;
	}

	bsize = atoi(argv[1]);
	pole1 = (char *) malloc(bsize);
	pole2 = (char *) malloc(bsize);
	n = atoi(argv[2]) / bsize;
	PLATON_FUNC(memsetmmx)(pole1, 'A', bsize);
	for (i = 0; i <= n; i++)
		PLATON_FUNC(memcpymmx)(pole2, pole1, bsize);

	return 0;
}

#endif