Uhh... ;-) Well, looking at the C code now, it's probably pretty easy to convert the smaller routine ("smart" rendering) to some BASIC-lookalike:

10 for y=1 to 208:for x=1 to 88
20 poke d,peek(s):d=d+1:s=s+1
30 c1=peek(s):s=s+1
40 c2=peek(s):s=s+1
50 c3=peek(s)
60 if c1=c3 and c1<>c2 then poke d,c2 else poke d,c1
70 d=d+1:next x:s=s+8:next y

Now d represents the memory address of the display frame buffer and s points to the MSX offscreen bitmap as described before. Does it make more sense now? ;-)

If you still prefer assembler, here's the same bit in assembly with a little bit of unrolling already done (now r1 points to MSX frame and r2 to the display frame buffer)... oh, just noticed that this code and the BASIC one I wrote earlier expect that the MSX frame pointer has the "excess border" adjustment done already when entering the routine.

    mov r3, #208
y_loop:
    mov r4, #176
    ldrh r7, [ r1 ], #2
x_loop:
    strh r7, [ r2 ], #2
    ldrh r5, [ r1 ], #2
    ldrh r6, [ r1 ], #2
    ldrh r7, [ r1 ], #2
    cmp r5, r7
    movne r6, r5
    cmp r5, r6
    strneh r6, [ r2 ], #2
    streqh r5, [ r2 ], #2
    strh r7, [ r2 ], #2
    ldrh r5, [ r1 ], #2
    ldrh r6, [ r1 ], #2
    ldrh r7, [ r1 ], #2
    cmp r5, r7
    movne r6, r5
    cmp r5, r6
    strneh r6, [ r2 ], #2
    streqh r5, [ r2 ], #2
    subs r4, r4, #4
    bne x_loop
    add r1, r1, #14
    subs r3, r3, #1
    bne y_loop

Yes, makes very much more sense

I have to make some tests, if this new method looks ultimate cool, I tell you that you have to implement it.

Shit... can't stop writing this BASIC code... I'm infected! ;-) The "resample" routine in BASIC could be something like this (same s and d variables as before):

10 defint a-z:for y=1 to 208:for x=1 to 88
11 c1=peek(s):s=s+1:c2=peek(s):s=s+1:c3=peek(s):s=s+1
12 r1=c1/2048:g1=(c1/32) and 63:b1=c1 and 31
13 r2=c2/2048:g2=(c2/32) and 63:b2=c2 and 31
14 r3=c3/2048:g3=(c3/32) and 63:b3=c3 and 31
15 r2=r2+r3:g2=g2+g3:b2=b2+b3
16 r1=(r1+r2)/3:g1=(g1+g2)/3:b1=(b1+b2)/3
17 poke d,(r1*2048) or (g1*32) or b1:d=d+1
18 c3=peek(s):s=s+1
19 r3=c3/2048:g3=(c3/32) and 63:b3=c3 and 31
20 r2=(r2+r3)/3:g2=(g2+g3)/3:b2=(b2+b3)/3
21 poke d,(r2*2048) or (g2*32) or b2:d=d+1
22 next x:s=s+8:next y

One important thing to note though: in this code, as well as the BASIC code I wrote here earlier, poke and peek are supposed to return 16bit values, not 8bit values. Also the s and d addresses when increased by 1 move to the next 16bit value in memory... so in 8bit addressing you would need to change all s and d increases from 1 to 2 plus make the peek and poke read 2 bytes and construct a 16bit value. Sorry :-)