比较Intel x86程序集中的两个数字（nasm）

Question

I have the following assembly code:

%include 'rw32.inc'

[segment .data use32]  

a dd 2.0
b dd 1.0  
[segment .code use32]

prologue                ; macro -- inicialization
  fld dword [a]
  fld dword [b]
  fcom
  jc greater
  jmp less
greater:
  fxch
less:
  call WriteDouble
  call WriteNewLine
  epilogue                ; macro -- termination

rw32.inc is utility library, which includes program initialization and termination and other functions. WriteDouble - prints out st0 on the screen WriteNewLine - just inserts newline.

and I want this to compare numbers 'a' and 'b' and print the greater one. My logic is this: get both numbers on stack. By fcom set flags. If carry flag is 1, number 'a' is greater, so it needs to be switched with 'b', so it is on top of the stack. Otherwise 'b' is greater and just print out result.

But it seems that program never jumps to label 'greater'. How can this be fixed, please? Thank you.

Answer 1

This why the nice people at Intel gave us fcomi(p) :

Performs an unordered comparison of the contents of registers ST(0) and ST(i) and sets the status flags ZF, PF, and CF in the EFLAGS register according to the results

The fstsw ax \\ sahf method is ancient, and doesn't even work on all "non-ancient" CPUs (some older x64 processors miss sahf ). The use of wait is a sign of truly ancient code, back from the time when dinosaurs still roamed the earth and the FPU was a coprocessor.

Answer 2

FPU math should never afect to CPU flags like carry and zero!

So copy flags from FPU to CPU flag register after fcomp instruction and than check the carry and zero flags like:

fld qword ptr [a]
fcomp qword ptr [b]
wait                 ;wait FPU
fstsw ax             ;copy FPU flags to ax
sahf                 ;copy ax to CPU flags
jbe LessOrEqu        ;do less or equal
...