pix: ; rdi, rsi, rdx
    push rdx
    push rdi
    push rsi ; stack = 16 * k + 8

    rdtsc ; answer in edx:eax

    mov edi, edx
    shl rdi, 32
    add rdi, rax

    call pixtime

    pop rsi
    pop rdi
    pop rdx

    ; preserve registers
    push rbx
    push rbp
    push r12
    push r13
    push r14
    push r15

    sub rsp, 8 ; stack = 16 * k + 8

loop_pix:
    mov rcx, 1
    xadd [rsi], rcx ; rax will equal current [rsi]
    cmp rcx, rdx ; rdx == max
    jae exit

    mov r10, rcx

    push rdi
    push rsi
    push rdx

    lea rdi, [rcx * 8]
    count_Pi rdi

    pop rdx
    pop rsi
    pop rdi

    shl r10, 2 ; multiply array index by 4
    shr rax, 32

    mov [rdi + r10], eax

    jmp loop_pix

exit:
    rdtsc ; answer in edx:eax

    mov edi, edx
    shl rdi, 32
    add rdi, rax

    call pixtime

    add rsp, 8

    pop r15
    pop r14
    pop r13
    pop r12
    pop rbp
    pop rbx

    ret