I am re-implementing Knuth's program P from Fascicle 1: generate the first 500 primes. The program generates the first 25 primes without problems. That code is below:
$ cat progp.S
/* print the first 500 primes */
#define n %bx
#define j %r12
#define k %r13
#define pk %r14d
.data
fmt: .asciz "%d\n"
x: .space 1000
.text
.globl main
.type main, @function
main:
pushq %rbp
movq %rsp, %rbp
xorq %rbx, %rbx
movw $2, x
movw $3, n
movq $1, j
Mtwo:
movw n, x(,j,2)
incq j
Mthree:
cmpq $500, j
je end
Mfour:
addw $2, n
Mfive:
movq $1, k
Msix:
movzwl x(,k,2), pk
movzwl n, %eax
xorq %rdx, %rdx
divl pk
cmpl $0, %edx
je Mfour
Mseven:
cmpl pk, %eax
jle Mtwo
Meight:
incq k
jmp Msix
end:
xorq j, j
loop:
leaq fmt, %rdi
movzwl x(,j,2), %esi
call printf
incq j
cmpq $25, j
je bye
jmp loop
bye:
movl $0, %edi
callq exit
leave
ret
.size main,.-main
.end
If you decrease the comparison in Mthree to 25, then the program is fine. Anything higher and the program faults or hangs in printf.
I am assembling it with:
cc -static progp.S
I can also add that without the call to printf, it succesfully generates the first 500 primes as can be seen from putting a breakpoint at "end" in gdb.
$ gdb ./a.out
(gdb) b end
Breakpoint 1 at 0x400531
(gdb) run
Starting program: /home/ben/src/hg/asm/knuth/a.out
Breakpoint 1, 0x0000000000400531 in end ()
(gdb) p $rbx
$1 = 3571
As soon as I try to call printf, it faults, so I assume I am doing something stupid with the stack.
printf expects in x86-64 a value in %rax that tells the function the amount of the floating point arguments. In your case there are no such arguments, so clear %rax (clearing %eax also clears %rax). Your program seems to run fine with the change:
...
loop:
leaq fmt, %rdi
movzwl x(,j,2), %esi
xor %eax, %eax
call printf
...