// means roughly we can monitor the time duration with up to 200 cycles * SAMPLE_SIZE
#define SAMPLE_SIZE    0x40000
unsigned short* a;


#define HTT	0x10000000

int main() {
	int i, j, k;
	unsigned int sample_size_in_byte;
	
	unsigned int cpu_eax, cpu_ebx, cpu_ecx, cpu_edx;
	
	sample_size_in_byte = SAMPLE_SIZE * sizeof(short);
	a = (unsigned short *)malloc(sample_size_in_byte);
	for (i=0; i< SAMPLE_SIZE; i++)
		a[i] = 0;
	
asm volatile (
      	"cpuid"
	: "=a"(cpu_eax), "=b"(cpu_ebx), "=c"(cpu_ecx), "=d"(cpu_edx)
	: "a"(0x80000000)
);

	printf("\nGot CPU info: EAX %x,  EBX %x, ECX %x, EDX %x\n", cpu_eax, cpu_ebx, cpu_ecx, cpu_edx);

asm volatile (
	"cpuid"
	: "=a"(cpu_eax), "=b"(cpu_ebx), "=c"(cpu_ecx), "=d"(cpu_edx)
	: "a"(2)
);
	printf("\nGot cache info: EAX %x,  EBX %x, ECX %x, EDX %x\n", cpu_eax, cpu_ebx, cpu_ecx, cpu_edx);
	return 0;
	
	
	/*	

	if (fork()==0) {
		execl("/usr/bin/openssl", "rsautl", "-in", "zzz_test", "-out","zzz_out_z", 
			"-inkey", "priv_10k.key", "-sign", (char *)0 );
		exit(0);	
	}
	
	*/
	
	
asm volatile  (
       "
	movl 	%0, %%ecx;
        movl 	%1, %%ebx;
	subl 	$0x4000, %%ebx;   //L1 cache size   one way size  should be 0x800
	rdtsc;
	mov	%%eax, %%esi;
	xor 	%%edi, %%edi;

loop:
	prefetcht2	0x4800(%%ecx, %%edi, 1);
	
	add	0x0000(%%ecx, %%edi, 1), %%cx;
	//imul	$1, %%ecx;
	add	0x0800(%%ecx, %%edi, 1), %%cx;
	//imul	$1, %%ecx;
	add	0x1000(%%ecx, %%edi, 1), %%cx;
	//imul	$1, %%ecx;
	add	0x1800(%%ecx, %%edi, 1), %%cx;
	//imul	$1, %%ecx;
	add	0x2000(%%ecx, %%edi, 1), %%cx;
	imul	$1, %%ecx;
	add	0x2800(%%ecx, %%edi, 1), %%cx;
	imul	$1, %%ecx;
	add	0x3000(%%ecx, %%edi, 1), %%cx;
	imul	$1, %%ecx;
	add	0x3800(%%ecx, %%edi, 1), %%cx;
	imul	$1, %%ecx;
	
	rdtsc;
	sub 	%%esi, %%eax;
	mov	%%ax, (%%ecx, %%edi, 1);
	add 	%%eax, %%esi;
	imul	$1, %%ecx;
	
	add	$0x40, %%edi;	//L1 cache line size (64 bytes)
	test	$0x7c0, %%edi;	//L1 cache set size  (2^ (bits in 0x7C0))  the lower 0s are determinted by L1 cache line size 
	jnz	loop;  		//32 sets
	
	sub	$0x7FE, %%edi;	//start from another timing measurement(2 bytes) in the cache line 
	test	$0x3E, %%edi;	//one cache line can hold up to (L1 cache line size)/(2 bytes each timing measurement)
	jnz	loop;  		//32 * 32 sets
	
	add	$0x7C0, %%edi;	//one-way data cache size (time to move to another timing measurement way)
	sub 	$0x800, %%ebx;  //each way of cache can contain 2k bytes
	jge	loop;"
	:  
	:  "m"(&(a[0])), "m"(sample_size_in_byte)   // /* the starting address of sample array and the sample size */
	: "eax", "ebx", "ecx", "edx", "esi", "edi", "memory" 
);
	
	for (i=0; i<(sample_size_in_byte/2048); i++) 
		for (j=0; j<32; j++) {
			printf("\n");
			for (k=0; k<32; k++)
				printf("%u\t", a[i*1024 + j + 32*k]);
		} 
}