FPU Conditional Testing; What's wrong with this code?

Index > Main > FPU Conditional Testing; What's wrong with this code?

Author

Thread

fasmnub

Joined: 26 Jan 2010
Posts: 14

fasmnub 18 Feb 2010, 08:05

..apart from it being very unoptimized (it's clunky so I can understand what's happening).

This is a series of If float_variable=<>??? ... Endif type statements I wrote to help me learn this stuff, but it doesn't behave as expected. Can anyone spot what's going wrong? I commented it so you can tell what I'm trying to do here..

Code:

;--------------------------------------------If x#=1.0
mov  eax, dword [v_x]
push eax
mov  dword eax, 1.0f
pop  ebp
MOV  dword [FPV], ebp
MOV  dword [FPVn], eax
FLD    dword [FPV]
FCOMP  dword [FPVn]
FNSTSW ax
mov eax,0
TEST   ah,40h
JE   cl_2
cmp  eax, 0
jz   cl_1
;--------------------------------------------Condition is true; y#=0
mov  dword eax, 0f
mov  [v_y], eax
;--------------------------------------------EndIf
cl_1:
;--------------------------------------------If x#>1.0
mov  eax, dword [v_x]
push eax
mov  dword eax, 1.0f
pop  ebp
MOV  dword [FPV], ebp
MOV  dword [FPVn], eax
FLD    dword [FPV]
FCOMP  dword [FPVn]
FNSTSW ax
mov eax,0
TEST   ah,41h
JNE   cl_4
cmp  eax, 0
jz   cl_3
;--------------------------------------------Condition is true; x#=1.0
mov  dword eax, 1.0f
mov  [v_x], eax
;--------------------------------------------EndIf
cl_3:
;--------------------------------------------If x#>=1.0
mov  eax, dword [v_x]
push eax
mov  dword eax, 1.0f
pop  ebp
MOV  dword [FPV], ebp
MOV  dword [FPVn], eax
FLD    dword [FPV]
FCOMP  dword [FPVn]
FNSTSW ax
mov eax,0
TEST   ah,1h
JNE   cl_6
cmp  eax, 0
jz   cl_5
;--------------------------------------------Condition is true; y#=0
mov  dword eax, 0f
mov  [v_y], eax
;--------------------------------------------EndIf
cl_5:
;--------------------------------------------If x#<0
mov  eax, dword [v_x]
push eax
mov  dword eax, 0f
pop  ebp
MOV  dword [FPV], ebp
MOV  dword [FPVn], eax
FLD    dword [FPV]
FCOMP  dword [FPVn]
FNSTSW ax
mov eax,0
TEST   ah,1h
JE   cl_8
cmp  eax, 0
jz   cl_7
;--------------------------------------------Condition is true; x#=0
mov  dword eax, 0f
mov  [v_x], eax
;--------------------------------------------EndIf
cl_7:
;--------------------------------------------If x#<=0
mov  eax, dword [v_x]
push eax
mov  dword eax, 0f
pop  ebp
MOV  dword [FPV], ebp
MOV  dword [FPVn], eax
FLD    dword [FPV]
FCOMP  dword [FPVn]
FNSTSW ax
mov eax,0
TEST   ah,41h
JE   cl_10
cmp  eax, 0
jz   cl_9
;--------------------------------------------Condition is true; y#=0
mov  dword eax, 0f
mov  [v_y], eax
;--------------------------------------------EndIf
cl_9:
;--------------------------------------------If x#=0.5
mov  eax, dword [v_x]
push eax
mov  dword eax, 0.5f
pop  ebp
MOV  dword [FPV], ebp
MOV  dword [FPVn], eax
FLD    dword [FPV]
FCOMP  dword [FPVn]
FNSTSW ax
mov eax,0
TEST   ah,40h
JE   cl_12
cmp  eax, 0
jz   cl_11
;--------------------------------------------Condition is true; y#=0.5
mov  dword eax, 0.5f
mov  [v_y], eax
;--------------------------------------------EndIf
cl_11:
;--------------------------------------------If x#<>0
mov  eax, dword [v_x]
push eax
mov  dword eax, 0f
pop  ebp
MOV  dword [FPV], ebp
MOV  dword [FPVn], eax
FLD    dword [FPV]
FCOMP  dword [FPVn]
FNSTSW ax
mov eax,0
TEST   ah,40h
JNE   cl_14
cmp  eax, 0
jz   cl_13
;--------------------------------------------Condition is true; y#=1
mov  dword eax, 1f
mov  [v_y], eax
;--------------------------------------------EndIf
cl_13:

I'm mainly concerned about the FCOMP, FNSTSW, TEST, JE stuff.. any help, code samples or related links would be hugely appreciated.

Thanks!

Edit to add: This isn't the whole code, just the bit that's not working, FPV and FPVn are declared elsewhere and used for conditional testing. The specific problem is that the condition tested is either always true, or never true.. so I've misunderstood something fundamental about FPU stuff. :(

18 Feb 2010, 08:05

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20299
Location: In your JS exploiting you and your system

revolution 18 Feb 2010, 09:05

Hint: Look at the sahf opcode. It probably does everything you want without needing all those complicated 'test' instructions.

18 Feb 2010, 09:05

LocoDelAssembly
Your code has a bug

Joined: 06 May 2005
Posts: 4624
Location: Argentina

LocoDelAssembly 18 Feb 2010, 16:55

Or he could just take a look to FCOMI/FCOMIP, I think that very few still-working old computers won't work with it.

18 Feb 2010, 16:55

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20299
Location: In your JS exploiting you and your system

revolution 18 Feb 2010, 17:00

BTW: Why do you have the 'mov eax,0' after every fnstsw?

Code:

FNSTSW ax
mov eax,0  ;<---- What is this for?

18 Feb 2010, 17:00

bitshifter

Joined: 04 Dec 2007
Posts: 796
Location: Massachusetts, USA

bitshifter 18 Feb 2010, 21:15

Here a snippet...

Code:

        fld     [value]
        fld     [max] ; 60.0 for example (>0)
        fcom
        fnstsw  ax
        sahf
        jc      .clamp
        fstp    st0
        fld     [min] ; -60.0 for example (<0)
        fcom
        fnstsw  ax
        sahf
        jc      .noclamp
    .clamp:
        fxch
    .noclamp:
        fstp    st0
        fstp    [value]

Also you can gain speed in some cases by just checking sign bit...

18 Feb 2010, 21:15

fasmnub

Joined: 26 Jan 2010
Posts: 14

fasmnub 18 Feb 2010, 21:55

@bitshifter, @revolution

perfect! just what i needed, thanks both!

18 Feb 2010, 21:55

LocoDelAssembly
Your code has a bug

Joined: 06 May 2005
Posts: 4624
Location: Argentina

LocoDelAssembly 18 Feb 2010, 22:27

Sorry to insist but is really so problematic FCOMI?

Code:

        fld     [value]
        fld     [max] ; 60.0 for example (>0)
        fcomi
        jc      .clamp

        fstp    st0
        fld     [min] ; -60.0 for example (<0)
        fcomi
        jc      .noclamp

    .clamp:
        fxch

    .noclamp:
        fstp    st0
        fstp    [value]

18 Feb 2010, 22:27

bitshifter

Joined: 04 Dec 2007
Posts: 796
Location: Massachusetts, USA

bitshifter 19 Feb 2010, 02:28

Thanks Loco!
No need to apologize when posting nicer code.
I wish you would insist a little more often Smile

Also, if you have SSE2 and can find some way to have
the need to min/max 4 floats at the same time (even in less clocks)

Code:

; data is assumed to be aligned on 16 byte boundary!
movaps xmm0,[_4values]
maxps  xmm0,[_4mins]
minps  xmm0,[_4maxes]
movaps [_4results],xmm0

So to make it work, the data need to be like this...

Code:

align 16
_4values  dd 0.0,0.0,0.0,0.0
_4mins    dd 0.0,0.0,0.0,0.0
_4maxes   dd 0.0,0.0,0.0,0.0
_4results dd 0.0,0.0,0.0,0.0
; TODO: Give values something valid...

Have fun

19 Feb 2010, 02:28

bitshifter

Joined: 04 Dec 2007
Posts: 796
Location: Massachusetts, USA

bitshifter 19 Feb 2010, 14:44

Ooh, i just cant stop playing with this...
So now we use a conditional move (no jumping)

Code:

      fld     [value] 
    fld     [max]
        fcomi   st0,st1
        fcmovnb st0,st1
        fxch
        fstp    st0
        fld     [min]
        fcomi   st0,st1
        fcmovb  st0,st1
        fstp    [result]
        fstp    st0

Just for fun Smile

19 Feb 2010, 14:44

LocoDelAssembly
Your code has a bug

Joined: 06 May 2005
Posts: 4624
Location: Argentina

LocoDelAssembly 19 Feb 2010, 15:27

And seems that fcmov was introduced in PPro too so you are not adding extra hardware requirements Very Happy

19 Feb 2010, 15:27

< Last Thread | Next Thread >

Forum Rules:

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You can download files in this forum