How to use floating point numbers in x86-64 assembly?

Question

I'm struggling with some x86-64 assembly, and floating point numbers are giving me a headache. For instance, when I run this code :

section .data
     omega:  dq 2.0
     omega2: dq 3.0

section .text
global func

func: push rbp
      mov rgp, rsp

      FINIT

      FLD qword [omega]
      FLD qword [omega2]
      FADD st0, st0

      mov rsp, rbp
      pop rbp
      ret

This function is called from a C code like that : printf("%Lf \n", func() ); Unfortunately the result is some bizarre number... I tried adding two integers using FIADD, and it worked fine. I dug through a ton of material already, but maybe someone here can point me to decent FPU tutorial, or share her/his experience and wisdom :)

Wrapping up the essentials:

language: x86-64 assembler
assembler: nasm v. 2.09.04 installed from repositories
compiler (for C): gcc v. 4.5.2 (installed with Ubuntu)
OS: Ubuntu 11.04 64bit on Oracle VM
Host OS: Windows 7 SP1 64bit
Processor: Intel i5 - 2430M 64bit (checked twice :D )
Problem: FPU can't add two numbers :(

Just in case : in the end i hope to use FSINCOS and other fancy FPU instructions, but seeing as even simple addition fails...

Thanks all in advance!

One way to figure out the intricacies is to compile some simple C code, and look at the resulting machine code with a disassembler. — Oliver Charlesworth, Jun 03 '12 at 23:29
Well, it seems that it used SSE instruction set. Maybe I will try this approach tomorrow. — Losiowaty, Jun 03 '12 at 23:43
You don't use the legacy FPU in x86_64. At least you shouldn't. Better use the scalar SSE instructions. — Gunther Piez, Jun 04 '12 at 00:19
why don't use SSE/SSE2, it's much faster and you don't get into trouble like the stack problem — phuclv, Jan 23 '14 at 07:15

score 4 · Accepted Answer · answered Jun 05 '12 at 02:09

Ok, so in the end it turned out that my problems were related to the fact, that FPU registers are organized as a stack, and i didn't pay enough attention to it which resulted in garbage and unwanted leftovers. Switching from "standard" to "pop" versions of instructions helped!

Anyway - thanks to all who bothered to read, it is much appreciated! :)

If anyone is interested - I was doing a class assignment where I had to compute motion of one planet from the view of another (geocentric model) in assembly. The final result translated to processing can be viewed here.

score 0 · Answer 2 · answered Jun 03 '12 at 23:36

0

The L qualifier to printf %f makes that argument a long double (80-bit float), which is not the assembly data type. Remove the L and it will default to a double (64-bit float) which is what you are computing.

Also, it should be

  FADD st1, st0

to add the two numbers together. Otherwise it will double the second value.

answered Jun 03 '12 at 23:36

wallyk

56,922
16
83
148

Well, the problem still stands. – Losiowaty Jun 03 '12 at 23:44
@Losiowaty: Does the `C` code have a declaration for `func()`? – wallyk Jun 03 '12 at 23:46
It does, otherwise I would get a linker error during compilation. And you are right with the addition formula, but still, it doesn't matter what i try to add/sub/div/mul. – Losiowaty Jun 04 '12 at 07:33
@Losiowaty: What is the declaration? Especially the return type. – wallyk Jun 04 '12 at 07:46
I tried all of them : float, double, long double, but it goes like this: `extern double func();` – Losiowaty Jun 04 '12 at 07:48
@Losiowaty: `double func(void);` is the most correct. What C compiler is it? You may have to look for a compiler option to return floating point values in the FPU. – wallyk Jun 04 '12 at 09:00

How to use floating point numbers in x86-64 assembly?

2 Answers2