format ELF64

extrn	printf
extrn	puts
extrn	scanf
public	main

section '.text' executable

main:
	push	rbp
	mov	rbp,rsp

        finit

        ; to determine how many clocks it takes to count clocks
        cpuid                   ; force all previous instructions to complete
        rdtsc                   ; read time stamp counter
        shl     rdx,32          ; shift rdx left 32 places
        or      rax,rdx         ; timestamp in rax
        mov     r11,rax         ; r11 stores the first timestamp
        
        ; get the second timestamp
        cpuid                   ; force all previous instructions to complete
        rdtsc                   ; read time stamp counter
        shl     rdx,32          ; shift rdx left 32 places
        or      rax,rdx         ; timestamp in rax

        sub     rax,r11         ; get difference in timestamps
        mov     r11,rax         ; r11 stores clocks required to count clocks

        ; begin benchmarking
        cpuid
        rdtsc
        shl     rdx,32
        or      rax,rdx
        mov     r10,rax         ; r10 stores first timestamp

        ; code to benchmark goes here
        nop
        nop

        ; end code to benchmark

        ; end benchmarking, take timestamp
        cpuid
        rdtsc
        shl     rdx,32
        or      rax,rdx         ; rax stores second timestamp

        sub     rax,r10         ; get difference in timestamps
        sub     rax,r11         ; subtract out time to count clocks

        ; print results
        mov     rsi,rax         ; rsi is first argument (the number of clocks)
        mov     edi,msg         ; rdi is second argument (location of msg)
        xor     eax,eax         ; eax = 0 (number of floating point arguments)
        call    printf          ; call printf and print results

	leave
	ret

section '.data' writable align 16

msg     db      "%lu clock cycles",0xA,0