Problems with floating point instruction

[flo8464]

Hi,
was trying to reverse a lua function and had problems because of some floating-point instruction which aren't listed in my reference and I have problems to find a explaination of them. Never worked with fpi before.

Here is the complete function:

.text:004F3460 ; =============== S U B R O U T I N E =======================================
.text:004F3460
.text:004F3460 ; Attributes: bp-based frame
.text:004F3460
.text:004F3460 lua_HasNewMail  proc near               ; DATA XREF: .data:01040224o
.text:004F3460
.text:004F3460 var_8           = qword ptr -8
.text:004F3460 arg_0           = dword ptr  8
.text:004F3460
.text:004F3460                 push    ebp
.text:004F3461                 mov     ebp, esp
.text:004F3463                 fld     flt_1040124
.text:004F3469                 fabs
.text:004F346B                 fcomp   ds:flt_9A0C58
.text:004F3471                 fnstsw  ax
.text:004F3473                 test    ah, 5
.text:004F3476                 jp      short loc_4F3493
.text:004F3478                 fld1
.text:004F347A                 mov     eax, [ebp+arg_0]
.text:004F347D                 sub     esp, 8
.text:004F3480                 fstp    [esp+8+var_8]
.text:004F3483                 push    eax             ; int
.text:004F3484                 call    lua_pushNumber
.text:004F3489                 add     esp, 0Ch
.text:004F348C                 mov     eax, 1
.text:004F3491                 pop     ebp
.text:004F3492                 retn
.text:004F3493 ; ---------------------------------------------------------------------------
.text:004F3493
.text:004F3493 loc_4F3493:                             ; CODE XREF: lua_HasNewMail+16j
.text:004F3493                 mov     ecx, [ebp+arg_0]
.text:004F3496                 push    ecx
.text:004F3497                 call    lua_pushnil
.text:004F349C                 add     esp, 4
.text:004F349F                 mov     eax, 1
.text:004F34A4                 pop     ebp
.text:004F34A5                 retn
.text:004F34A5 lua_HasNewMail  endp
.text:004F34A5
.text:004F34A5 ; ---------------------------------------------------------------------------

This is where I have my problems, thats how I understand it:

fld flt_1040124
Ok, loading a value.

fabs
Casting it to an unsigned number

fcomp ds:flt_9A0C58
Comparing it to another value

fnstsw ax
Not listed in my reference

test ah, 5
Higher bits of ax are compared to 5

jp short loc_4F3493
Jump if parity flag set

Can please someone explain that piece of code a bit? Too bad I don't have the new HexRays with FPU-Support, makes everything clearer I think.

[Oowafas]

fnstsw ax stores the FPU status word in the ax register. When the upper bits are tested against 5 it's checking the C2 and C0 bits of the FPU status word. The parity flag is set if there are an odd number of bits after the bitwise and operation done by test, so if either the '4' bit or the '1' bit is set it will take the jp branch. C0 is set if st (abs(flt_1040124)) < source (flt_9A0C5. Neither bit is set if st >= source, and both are set if the operation is invalid. So the branch will only happen if C0 is set (st<source).

Edit in response to posts below : AoA is a good asm resource also, download that in HTML.

[Cypher]

Get a copy of the Intel Developer Reference Manuals (or w.e they're called), they have an instruction set reference.

Anyway, here:

signed int __cdecl lua_HasNewMail(int a1)
{
  signed int result; // eax@2

  if ( fabs(flt_1040124) >= 0.0000002384185791015625 )
  {
    sub_801A30(a1);
    result = 1;
  }
  else
  {
    sub_801A50(a1, 1.0);
    result = 1;
  }
  return result;
}

[amadmonk]

Get a copy of the Intel Developer Reference Manuals (or w.e they're called), they have an instruction set reference.

I second that.

You can get it from here for free.

[Cypher]

I second that.

You can get it from here for free.

I ordered hardcopies ages ago (which I still have), and I checked again recently because a friend wanted to order them and couldn't find out how. I had a look and it seems that they no longer mail out hard copies of the manuals.

Is that still true? It seems like it based on their website....

If anyone wants to try though, email their 'Literature department' or something (look it up, can't remember the name) and ask if they'll send you them. They did for me, but like I said, it seems that since then they've removed the SKU ordering numbers from their website.