Shorter encoding when a register has known value

Index > Main > Shorter encoding when a register has known value

Author

Thread

l4m2

Joined: 15 Jan 2015
Posts: 674

l4m2 21 Jul 2015, 00:50

when eax equals to 0, people still use

Code:

lea ecx, [edx*4]

but

Code:

[eax+4edx]

shorter won't ?

21 Jul 2015, 00:50

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20298
Location: In your JS exploiting you and your system

revolution 21 Jul 2015, 09:36

It uses 4 fewer bytes. But if you ever change the preceding code and the value in EAX is not zero then you also need to update the LEA. It is a potential bug trap and relies upon remembering such finicky details, and/or having great comments to explain what is happening.

Last edited by revolution on 21 Jul 2015, 13:07; edited 1 time in total

21 Jul 2015, 09:36

Feryno

Joined: 23 Mar 2005
Posts: 509
Location: Czech republic, Slovak republic

Feryno 21 Jul 2015, 13:00

I like to use LEA instead of ADD/SUB when I need not to destroy flags and also in case of some multiplication with some numbers like 3, 4, 5, 8, 9
in 64 bit mode the shortest encoding is like

Code:

lea ecx,[rax+rdx*4]

21 Jul 2015, 13:00

PeExecutable

Joined: 26 Jun 2015
Posts: 181

PeExecutable 21 Jul 2015, 13:16

Sometimes it is beneficial to use a lea, and some other below, sometimes an add or a inc instruction helps. You have to reorder and replace them and time the code until you find the combo that is faster. Do not rely in your understanding only, rely in timing. It's amazing how the processor tricks you, it has become a smarter unit over the years. Similarly, be careful to rely in timing if you have hyper threading enabled.

21 Jul 2015, 13:16

l4m2

Joined: 15 Jan 2015
Posts: 674

l4m2 21 Jul 2015, 13:26

not only lea. I try to have static variables but direct address costs 3 more bytes than

21 Jul 2015, 13:26

l4m2

Joined: 15 Jan 2015
Posts: 674

l4m2 21 Jul 2015, 13:32

I just found that the title has been changed from assume to Shorter encoding when a register has known value. assume because in some compiler assume can be used to optimize

21 Jul 2015, 13:32

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20298
Location: In your JS exploiting you and your system

revolution 21 Jul 2015, 13:45

l4m2 wrote:

I just found that the title has been changed from assume to Shorter encoding when a register has known value. assume because in some compiler assume can be used to optimize

I changed the title because "assume" didn't convey any information about the topic. But you are free to change it again at any time if you feel it should something else. Although, I would urge you to say more about what you are suggesting, because by itself just having "assume" in the title is ambiguous.

21 Jul 2015, 13:45

l4m2

Joined: 15 Jan 2015
Posts: 674

l4m2 21 Jul 2015, 13:57

Code:

assume eax 0
  lea ecx, [edx*4]  ;8D0C90
endassume

21 Jul 2015, 13:57

l4m2

Joined: 15 Jan 2015
Posts: 674

l4m2 21 Jul 2015, 14:02

Code:

foo:push esi
    mov esi, var
    assume esi var
      add [var1], eax
      xor [var3], ecx
    endassume
    pop esi
    ret
var:
var1  dd ?
var2  dd ?
var3  dd ?

21 Jul 2015, 14:02

shutdownall

Joined: 02 Apr 2010
Posts: 517
Location: Munich

shutdownall 22 Jul 2015, 10:46

l4m2 wrote:

Code:

assume eax 0
  lea ecx, [edx*4]  ;8D0C90
endassume

Sounds nice but how will you assure, that eax is 0 ?
You can "assume" whatever you want but one very important register ist restricted to use - doesn't make much sense for me. The only way to assure is to paste xor eax,eax first but then you have only one byte less (and eax destroyed !). Wink

22 Jul 2015, 10:46

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20298
Location: In your JS exploiting you and your system

revolution 22 Jul 2015, 14:59

I suppose one could have this:

Code:

;...
  cmp eax,0 ;or test eax,eax, or whatever
  jnz .somewhere
  assume eax 0
    lea ecx, [edx*4]  ;8D0C90
  end assume
;...

22 Jul 2015, 14:59

l4m2

Joined: 15 Jan 2015
Posts: 674

l4m2 23 Jul 2015, 01:36

revolution wrote:

I suppose one could have this:

Code:

;...
  cmp eax,0 ;or test eax,eax, or whatever
  jnz .somewhere
  assume eax 0
    lea ecx, [edx*4]  ;8D0C90
  end assume
;...

besides

Code:

ecx=edx*8
eax=0

can

23 Jul 2015, 01:36

l4m2

Joined: 15 Jan 2015
Posts: 674

l4m2 24 Jul 2015, 15:29

Quote:

Sounds nice but how will you assure, that eax is 0 ?

You can "assume" whatever you want but one very important register ist restricted to use - doesn't make much sense for me. The only way to assure is to paste xor eax,eax first but then you have only one byte less (and eax destroyed !).

Just let ecx=edx*4 these code have same length:

Code:

xor ecx,ecx
lea ecx,[ecx+4edx]

lea ecx,[4edx+eax]
sub ecx,eax

mov ecx,edx
shl ecx,2

lea ecx,[2edx]
shl ecx,1 ;or maybe add ecx,ecx instead

and this code is shorter but maybe slower(?)

Code:

imul ecx,edx,4

Last edited by l4m2 on 25 Jul 2015, 03:31; edited 1 time in total

24 Jul 2015, 15:29

shutdownall

Joined: 02 Apr 2010
Posts: 517
Location: Munich

shutdownall 24 Jul 2015, 16:16

You are talking here only about partial optimizations.
In fact you can do much more with optimized datastructures than just keeping an eye on single instructions. Cool

24 Jul 2015, 16:16

l4m2

Joined: 15 Jan 2015
Posts: 674

l4m2 25 Jul 2015, 03:25

shutdownall wrote:

You are talking here only about partial optimizations.
In fact you can do much more with optimized datastructures than just keeping an eye on single instructions.

Always lots of instruction when needed (especially for global variables)

25 Jul 2015, 03:25

< 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