Newton - Raphson inversion algorithm in assembly

问题

I am trying to implement Newton - Raphson inversion algotihm in assembly according to this equation:

Xn+1 = Xn(2-b*Xn)

My function is:

.data
const:
    .int 2
.text
.global inversion
inversion:
    pushl   %ebp
    movl    %esp, %ebp
    fldl    8(%ebp)         # load b
    fldl    12(%ebp)        # load X0
inv:    
    fst     %st(2)          # save Xn
    fmul    %st(1), %st(0)      # b*Xn
    fsubr   const           # 2-b*Xn
    fmul    %st(2), %st(0)      # Xn(2-b*Xn)
    fcomi   %st(2), %st(0)      # check if it the same result as before
jne inv                 # jump
                    # it should return the st(0)
leave
ret

And my C program:

#include <stdio.h>

extern float inversion (float a, float b);

int main(void)
{
    float x = inversion (5,0.1);
    printf ("%f\n", x);
    return 0;
}

My problem is that this function doesn't return anything, and when I am debuggin it with gdb it shows that my function never stops, because values are never the same. In my opinion it's load bad numbers, it's what I am reading after loading data on stack (st(0), st(1), st(2)):

st0            2.6342588374767957140994886365053441e-314    (raw 0x3bed9ee6666680000000)
st1            5.2386907379892329317261356891322066e-11 (raw 0x3fdce6666a0500000000)
st2            2.6342588374767957140994886365053441e-314    (raw 0x3bed9ee6666680000000)

Thanks for all your responses.

回答1:

Yes, gdb pointed you in the right direction: you are loading the arguments wrong. You used fldl where the l means double when dealing with floating point instructions. I guess you thought the l meant 4 bytes. To load 4 byte floats you need the flds instruction.

Furthermore, fsubr expects a floating point number but you give it an integer. Solution: change .int 2 to .float 2. Also change to fsubrs for safety (although that is the default).

Finally, you are leaving 3 values on the fpu stack, which is not according to the expected calling convention and can cause problems later. You should only leave the return value there.

来源：https://stackoverflow.com/questions/23501773/newton-raphson-inversion-algorithm-in-assembly

标签

assembly

x86

att