x86汇编牛顿平方根算法寄存器为1＃IND

我试图在下面编写C程序的汇编版本。它的意思是取9的平方根。直到第3次通过我的循环，所有东西似乎都在起作用，然后当我的上，下值被压入堆栈时，它们的上，下值就变成1＃IND。我不知道为什么，因为它正确地打印在标签whileLoop下方，但是我在第二次在标签notAccurateEnough之后直接打印时得到了错误的值，即使这两个标签之间的值没有变化并且也能正常工作在之前的迭代中。
下面的屏幕截图显示了如何成功打印一次，然后在同一迭代中第二次打印失败。

汇编代码：

.586
.MODEL FLAT
.STACK 4096                 
n=550
.data
    PUBLIC _sqrtX

upper REAL4 9.0
accuracy REAL4 0.00001
lower REAL4 1.0
guess REAL4 1.0
multValue REAL4 1.0
num REAL4 9.0
two DWORD 2

.code

_sqrtX  PROC

        finit   ;initialize fpu

        whileLoop:
        fld upper    ;push upper onto stack
        fld lower    ;push lower onto stack
        fsub         ;subtract upper-lower

        fld accuracy          ;push accuracy onto stack
        fcom                  ;compare accuracy with upper-lower

        fstsw ax              ;copy condition code bits to AX
        sahf                  ;shift condition code bits to flags
        jna notAccurateEnough ;jump if accuracy is greater
        jmp endnewt           ;end algorith, is less than accuracy

        notAccurateEnough:
        ;**************************
        ;************  Logic to implement:   guess = (lower + upper) / 2;
        ;**************************
        fld upper      ;push upper onto stack
        fld lower      ;push lower onto stack
        fadd           ;(lower + upper)
        fidiv two      ;divide by 2
        fstp guess     ;set guess to first stack element. guess = (lower + upper) / 2
        ;**************************

        ;**************************
        ;************  Logic to implement:  if((guess * guess) > number)
        ;**************************
        fld guess ;push guess onto stack as first element
        fld guess ;push guess onto stack again, it is now first and second element
        fmul      ;multiply first and second element
        fcom num  ;compare 9 to product of guess*guess:
        fstsw ax  ;copy condition code bits to AX
        sahf      ;shift condition code bits to flags
        jna SetLower ;if guess is less than num, call set lower and set lower = guess;
        ;*** else, guess > num
        fld guess ;push guess onto the stack
        fstp upper ;set upper equal to top of stack: upper = guess
        jmp whileLoop
        ;**************************

        setLower:
        fld guess   ;push guess onto stack
        fstp lower  ;store guess in lower
        jmp whileLoop

        endnewt:
        mov eax, 0
        ret

_sqrtX ENDP
END

我尝试在Assembly中编写的C算法：

double sqrtX (double x)
{
    const double ACCURACY=0.00001;
    double lower, upper, guess;
    double number;

    number = x;
    lower = 1;
    upper = number;

    while ((upper-lower) > ACCURACY)
    {
        guess = (lower + upper) / 2;
        if((guess * guess) > number)
        {
            upper = guess;
        }
        else
        {
            lower = guess;
        }
    }
    return (lower + upper)/2;
}