/* test program for SSE by kaz @ kobe1995.net
	% c++ -g loop.C && ./a.out 999999
  update:
	2001/05/13 added check of kernel support for SSE
	2001/05/27 align 16(*8=128bit) for movaps
	2001/07/09 "movaps (%0),%%xmm0::g(A)" -> "movaps %0,%%xmm0::g(*A)"
*/
#include <stdio.h>
#include <sys/time.h>
#include <unistd.h>
#include <stdlib.h>
#include <limits.h>

int MMXcheck(void)
{
	int mmx_available;

	asm("push %eax");
	asm("push %edx");

	__asm__ __volatile__ (
		/* Get CPU version information */
		"movl $1, %%eax\n\t"
		"cpuid\n\t"
		"andl $0x800000, %%edx\n\t"
		"movl %%edx, %0"
		: "=q" (mmx_available)
		: /* no input */
	);

	asm("pop %edx");
	asm("pop %eax");

	return (mmx_available != 0) ? 1 : 0;
}


int SSEcheck(void)
{
	int sse_available;

	asm("push %eax");
	asm("push %edx");

	__asm__ __volatile__ (
		/* Get CPU version information */
		"movl $1, %%eax\n\t"
		"cpuid\n\t"
		"andl $0x2000000, %%edx\n\t"
		"movl %%edx, %0"
		: "=q" (sse_available)
		: /* no input */
	);

	asm("pop %edx");
	asm("pop %eax");

	return (sse_available != 0) ? 1 : 0;
}

/* 3D Now! check */
int TDNcheck(void)
{
	int TDN_available;

	asm("push %eax");
	asm("push %edx");

	__asm__ __volatile__ (
		/* Get CPU version information */
		"movl $0x80000001, %%eax\n\t"
		"cpuid\n\t"
		"andl $0x80000000, %%edx\n\t"
		"movl %%edx, %0"
		: "=q" (TDN_available)
		: /* no input */
	);

	asm("pop %edx");
	asm("pop %eax");

	return (TDN_available != 0) ? 1 : 0;
}

int main(int argc,char **argv){
static float __attribute__((aligned(16)))
	A[4]={1.,1.,1.,1.},
	B[4]={0.,1.,2.,3.},
	C[4];

	int i,sum=0,msum,n;

	double TClock;
	unsigned long TTime;
	unsigned int HTime0,HTime1,LTime0,LTime1;
	struct timeval Tim0,Tim1;

	printf("MMX %d	SSE %d	3D Now! %d\n",
		MMXcheck(),SSEcheck(),TDNcheck());
	if(argc!=2){
		fprintf(stderr,"usage: %s #\n",argv[0]);
		n=1000000;
//		exit(1);
	}else{
		n=atoi(argv[1]);
	}
	
	printf("A:%08x B:%08x C:%08x	n=%d\n",A,B,C,n);
	printf("float A=(");
	for(i=0;i<4;i++)	printf("%f,",A[i]);
	printf("\b)\n");

	printf("float B=(");
	for(i=0;i<4;i++)	printf("%f,",B[i]);
	printf("\b)\n");

	if(SSEcheck()){
		if(!system("exit `/sbin/sysctl -n hw.instruction_sse 2>/dev/null`")){
			fprintf(stderr,"lack of kernel support for SSE\n");
			exit(1);
		}
		asm("movaps %0,%%xmm0"					//SSE
		:
		:"g"(*A));
		asm("movaps %0,%%xmm1"					//SSE
		:
		:"g"(*B));
		gettimeofday(&Tim0,0);
		asm("push %eax");
		asm("push %edx");
		asm("rdtsc");
		asm("mov %%edx,%0"
		:"=g"(HTime0));
		asm("mov %%eax,%0"
		:"=g"(LTime0));

		for(i=n;i>0;i--){
//			asm("addps %xmm1,%xmm0");			//SSE
			asm("mulps %xmm1,%xmm0");			//SSE
		}
		asm("rdtsc");
		asm("mov %%edx,%0"
		:"=g"(HTime1));
		asm("mov %%eax,%0"
		:"=g"(LTime1));
		asm("pop %edx");
		asm("pop %eax");
		gettimeofday(&Tim1,0);
		asm("movaps %%xmm0,%0":"=g"(*C));			//SSE
		TClock=(double)(LTime1-LTime0)+
			(double)(UINT_MAX)*(double)(HTime1-HTime0);
		TTime=1000000*(Tim1.tv_sec-Tim0.tv_sec)+
			(Tim1.tv_usec-Tim0.tv_usec);
		printf("A:%08x B:%08x C:%08x\n",A,B,C);
		printf("float C=(");
		for(i=0;i<4;i++)	printf("%f,",C[i]);
		printf("\b)\n");
		printf("time: %ld(us)	clock: %.01f\n",TTime,TClock);
	}
	if (1) {
		asm("movaps %0,%%xmm0"					//SSE
		:
		:"g"(*A));
		asm("movaps %0,%%xmm1"					//SSE
		:
		:"g"(*B));
		gettimeofday(&Tim0,0);
		asm("push %eax");
		asm("push %edx");
		asm("rdtsc");
		asm("mov %%edx,%0"
		:"=g"(HTime0));
		asm("mov %%eax,%0"
		:"=g"(LTime0));

		for(i=n;i>0;i--){
//			A[0]+=B[0];A[1]+=B[1];A[2]+=B[2];A[3]+=B[3];	//x87
			A[0]*=B[0];A[1]*=B[1];A[2]*=B[2];A[3]*=B[3];	//x87
		}
		asm("rdtsc");
		asm("mov %%edx,%0"
		:"=g"(HTime1));
		asm("mov %%eax,%0"
		:"=g"(LTime1));
		asm("pop %edx");
		asm("pop %eax");
		gettimeofday(&Tim1,0);
		for(i=0;i<4;i++) C[i]=A[i];				//x87
		TClock=(double)(LTime1-LTime0)+
			(double)(UINT_MAX)*(double)(HTime1-HTime0);
		TTime=1000000*(Tim1.tv_sec-Tim0.tv_sec)+
			(Tim1.tv_usec-Tim0.tv_usec);
		printf("A:%08x B:%08x C:%08x\n",A,B,C);
		printf("float C=(");
		for(i=0;i<4;i++)	printf("%f,",C[i]);
		printf("\b)\n");
		printf("time: %ld(us)	clock: %.01f\n",TTime,TClock);
	}
}