flat assembler
Message board for the users of flat assembler.
 Home   FAQ   Search   Register 
 Profile   Log in to check your private messages   Log in 
flat assembler > Main > [fasmg] Optimal relocation

Author
Thread Post new topic Reply to topic
marste



Joined: 05 May 2015
Posts: 42
[fasmg] Optimal relocation
Here I’m again to see if fasmg can again exceed my expectations… Cool

In my little program I’ve some code that I can relocate. More: I need to relocate it in order to respect some constrain that I have (e.g. some label is within some address range), and depending on the parameters that I select for compilations this pieces should be relocated differently in order to respect the constraints.

I’ve seen examples in which fasmg resolve some second degree equation finding automatically values fitting a constraint, so I was thinking something as including the pieces (many little ones) of code in many places with a condition fitting just in one place, e.g.:


Code:

Relocatablecode.inc:
  if PIECE_1_LOC=THISLOC
    Relocatable code…
  end if

mainfile.asm:
  …code…
  THISLOC=1
  Include Relocatablecode.inc
  …more code…
  THISLOC=2
  Include Relocatablecode.inc
  ..etc..




But not sure how to let fasmg calculate PIECE_1_LOC… And maybe there are different way also to have the same result!…

Additionally if possible (and I really this is difficult) there are other conditions (e.g. some label address mod 256 equal 0) that are not strictly required but if possible a nice to have (I can even add some “weights” to this conditions in order to have a value to minimize or maximize for example).

Hope I was clear enough but anyway available for any more clarification! Smile

Thank you!
Post 19 Sep 2017, 09:09
View user's profile Send private message Reply with quote
Tomasz Grysztar
Assembly Artist


Joined: 16 Jun 2003
Posts: 6633
Location: Kraków, Poland
If you use multiple conditions to place constraints on what can be assembled where, fasm[g] is going to look for a solution until it finds one. Unfortunately, even if a solution exists, there is no guarantee that fasm is going to find it (and the existence of a solution is one of the undecidable problems). You may get the "could not generate code within the allowed number of passes" error.

One of the possible obstacles is the "oscillator problem", please read about it.

What I would suggest is to write the IF blocks to place various constraints, and keep adding them until you get the "could not generate code" error. If that happens, we may look at the specific problem to see how we can guide fasmg to find the right solution with some additional flags and conditions (like in the case of the "oscillator problem").
Post 19 Sep 2017, 09:33
View user's profile Send private message Visit poster's website Reply with quote
marste



Joined: 05 May 2015
Posts: 42
Do you suggest to start with something like this?


Code:

Relocatablecode.inc:
  if PIECE_1_LOC = THISLOC
    Relocatable code…
    PIECE_1_LOC = THISLOC
  end if
  if PIECE_2_LOC = THISLOC
    Other relocatable code…
    PIECE_2_LOC = THISLOC
  end if

mainfile.asm:
  …code…
  THISLOC = 1
  Include Relocatablecode.inc
  …more code…
  THISLOC = 2
  Include Relocatablecode.inc
  THISLOC = 3
  Include Relocatablecode.inc
  …etc…
  assert PIECE_1_LOC
  assert PIECE_2_LOC
  …end


Post 19 Sep 2017, 13:45
View user's profile Send private message Reply with quote
Tomasz Grysztar
Assembly Artist


Joined: 16 Jun 2003
Posts: 6633
Location: Kraków, Poland
I have prepared a very simplified example that demonstrates some of the ideas. I have used numbered macros to define pieces of code that get moved around:

Code:
macro piece_1
     first:
     db 'first',13,10
end macro

macro piece_2
     second:
     db 'second',13,10
end macro

macro piece_3
     third:
     db 'third',13,10
end macro

macro piece_4
     fourth:
     db 'fourth',13,10
end macro


macro place_piece number
        repeat 1n:number
                piece_#n
                macro piece_#n          ; do not assemble same piece again
                end macro
        end repeat
end macro


repeat 4
        ; if address is aligned, place piece that prefers alignment:
        if $ and 3 = 0
                place_piece 2
        end if
        if $ and 3 = 0
                place_piece 4
        end if
        ; otherwise keep placing pieces consecutively:
        place_piece %
end repeat

Or:

Code:
NUMBER_OF_PIECES = 100

repeat NUMBER_OF_PIECES
        macro piece_#%
                p#%:
                db `%,13,10
        end macro
end repeat

macro place_piece number
        repeat 1n:number
                piece_#n
                macro piece_#n          ; do not assemble same piece again
                end macro
        end repeat
end macro


repeat NUMBER_OF_PIECES
        ; if we are sufficiently near piece 30, place ones that need to be near it
        if p30 - 100 < $ & $ < p30 + 100
                place_piece 14
        end if
        if p30 - 100 < $ & $ < p30 + 100
                place_piece 28
        end if
        if p30 - 100 < $ & $ < p30 + 100
                place_piece 49
        end if
        if p30 - 100 < $ & $ < p30 + 100
                place_piece 99
        end if
        place_piece %
end repeat

; we can use assertion to force assembly to fail when some end up too far from each other to be acceptable
assert p14 - p30 < 100 | p30 - p14 < 100


The main problem here is that need to write rules that are easy to apply, like above. For any constraints that you have, you need to have an idea what kind of reordering may help. In the worst case you may actually need to test every possible order.
Post 19 Sep 2017, 15:40
View user's profile Send private message Visit poster's website Reply with quote
marste



Joined: 05 May 2015
Posts: 42
The proposed approach might be good if you have some condition that can determine “without search” the optimal solution.

I’ve a different scenario. To simplify, imagine that I should fit the more piece I can below a certain address. The if then can read something like “if $<limit then place piece”, but this is not guaranteeing an optimal solution.

As for example to simplify: imagine the address is 20 and the pieces are (in order) 8 7 6 6 long, with the proposed method 8 and 7 will fit while the 6s will not find anymore available space (5 is remaining); while an optimal solution should fit the 8 and the two 6 blocks (leaving just the 7 outside).
Post 20 Sep 2017, 08:37
View user's profile Send private message Reply with quote
Tomasz Grysztar
Assembly Artist


Joined: 16 Jun 2003
Posts: 6633
Location: Kraków, Poland
As I noted above, if you do not have at least a good heuristic to begin with, you are left with no other option than trying all permutations. This is viable when there are just few pieces, for larger numbers you would have to keep fingers crossed that it finds an acceptable solution within early tries (you could randomize generated permutations instead of using the lexicographical order).
Post 20 Sep 2017, 08:45
View user's profile Send private message Visit poster's website Reply with quote
Tomasz Grysztar
Assembly Artist


Joined: 16 Jun 2003
Posts: 6633
Location: Kraków, Poland
I have prepared an example that looks through all the permutations to find the solution in the simplified case you mentioned:

Code:
macro piece_1
        db 8 dup '1'
end macro

macro piece_2
        db 7 dup '2'
end macro

macro piece_3
        db 6 dup '3'
end macro

macro piece_4
        db 6 dup '4'
end macro

if PIECES_FITTING >= 3                  ; if we found the solution,
        PERMUTATION = PERMUTATION       ;  keep the permutation
else if ~ PERMUTATION
        PERMUTATION = 1234h             ; this is the permutation we start with
else
; the following code mutates the permutation to the next one lexicographically
        PERM = PERMUTATION
        BUFFER = 0
        while PERM & PERM and 0Fh > BUFFER and 0Fh
                BUFFER = (BUFFER shl 4or (PERM and 0Fh)
                PERM = PERM shr 4
        end while
        INDEX = 0
        while PERM and 0Fh < (BUFFER shr (INDEX+4)) and 0Fh
                INDEX = INDEX + 4
        end while
        TMP = PERM and 0Fh
        PERM = (PERM and not 0Fhor ((BUFFER shr INDEXand 0Fh)
        BUFFER = (BUFFER and not (0Fh shl INDEX)) or (TMP shl INDEX)
        PERM = (PERM shl ((bsr BUFFER shr 2 + 1shl 2)) or BUFFER
        PERMUTATION = PERM
end if

COUNT = 0
repeat 4
        repeat 1i:(PERMUTATION shr ((%%-%)*4)) and 0Fh        ; extract digit from permutation
                piece_#i
        end repeat
        if $ <= 20
                COUNT = COUNT + 1               ; count pieces that fit in the first 20 bytes
        end if
end repeat
PIECES_FITTING := COUNT

If the solution cannot be found, assembly is never going to finish and will abort upon reaching the limit of passes.

As I mentioned in the previous post, instead of looking through the permutations consecutively you could also try generating them in some pseudo-random order. When the space of acceptable solutions is a large set, such method could give results faster.
Post 20 Sep 2017, 09:50
View user's profile Send private message Visit poster's website Reply with quote
marste



Joined: 05 May 2015
Posts: 42
For me permutations will be good (I have a condition flexible enough), just that with 6 "mobile pieces" 720 will be the overall possiblities (I can anyway pilot more result and reduce search).

Anyway with condition that have to be explicit (and not something as max among possiblities) the solution cannot find optimal ones in general as e.g. in the given example having lists as: 8,7,6,8 (8+8 optimal) or 4,4,4,4,4,4 (4+4+4+4+4 optimal).

But I think we are already much much better than a "standard" assembler!!! Smile
Post 20 Sep 2017, 12:23
View user's profile Send private message Reply with quote
Display posts from previous:
Post new topic Reply to topic

Jump to:  


< 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


Powered by phpBB © 2001-2005 phpBB Group.

Main index   Download   Documentation   Examples   Message board
Copyright © 2004-2016, Tomasz Grysztar.