if lab2-lab1>8
  macro foo
    do something
  end macro
end if
lab1:
some code
lab2:    

In fact, I am so satisfied with these macros and how flexible they are that I no longer feel any pressure to create fasm 2 with internal x86 encoder, I just want to write more and more macros.

But what is the asymptotic behavior of fasmg on big sources? This can be much more important than the speed of compilation on particular (newer or older) processors.

N = 4000000

db 'test0 = 0',13,10
rept N, i:0,j:1
        db 'test',`j,' = (test',`i,'*33797+1) and 0FFFFFFFFh',13,10
        if %=%%
                db 'dd test',`j
        end if
end rept    

N               fasm 1.71.58    fasm g.hll54

100000          0.3 s           0.2 s
200000          0.6 s           0.3 s
300000          1.1 s           0.6 s
400000          1.5 s           0.8 s
600000          2.3 s           1.3 s
800000          3.1 s           1.8 s
1000000         4.1 s           2.1 s
2000000         8.7 s           4.1 s
4000000         out of memory   8.1 s    

N = 4000000
P = 2

rept 1, v:P
        db 'P = ',`v,13,10
end rept
db 'if P > 2',13,10
db '    PASS = PASS + (1-PASS/(P-2))',13,10
db 'end if',13,10
rept N, i:0,j:1
        db 'test',`j,' = (test',`i,'*33797+1) and 0FFFFFFFFh',13,10
end rept
db 'test0 = 0',13,10
rept 1, v:N
        db 'dd test',`v,13,10
end rept    

      fasm 1.71.57               fasm g.hnq73
2 passes, 10.9 seconds     2 passes, 11.3 seconds
3 passes, 11.2 seconds     3 passes, 16.7 seconds
4 passes, 11.6 seconds     4 passes, 21.7 seconds
5 passes, 11.8 seconds     5 passes, 27.1 seconds
6 passes, 12.0 seconds     6 passes, 32.0 seconds
7 passes, 12.3 seconds     7 passes, 37.6 seconds
8 passes, 12.4 seconds     8 passes, 42.3 seconds    

      fasm 1.71.57               fasm g.hnq73
1 passes, 10.7 seconds     1 pass, 5.9 seconds    

It's important to note that the comparison is made for the initial pass only. To make the comparison fair it's important to compare subsequent passes as well (...)

What about the kinds of loop(1rept,1repeat,1hardloop,grept,ghardloop)

As I mentioned repeatedly...

But since there is no simple correspondence between the length of source text and the number of passes, I saw it as an independent problem from the asymptotic data that JohnFound asked for.

adding another pass has little effect on the overall time

nice to see you show up again to give a sober comment. And I'm glad that you still find some time to play with fasm (1 or g)

The asymptotic timing behaviour can be expressed by a multivariate function.

Unfortunately I don't. I hope I will — maybe, in a few months — as I still have community-supportive tasks on my todo-list. But thank you for the motivation.

I think that fasmg approach is generally more in line with the expectations of users.

Thank you for your feedback Tomasz. Again I don't want to hijack this thread so maybe we can move this discussion to another one, but I do wonder how you handle internally macros like "include" or inclusion of binary data from external files. I'm asking since I think it is easier to implement in old fasm approach.