Vertical scrolling routine (MSX1)

thank you, posted here the results just to have infos about cycles. But I have to manage the cose 'cause every 8th frame the scroll has a bit of "indecision", i think because at the 8th frame it does the LDDR on RAM nametable and take a row from the map coded in RAM. So I'm thinking if it's worth to reserve in RAM a second nametable and LDDR it by 1/8 every frame and the swap with the actual nametable...

branch taken / not taken. 13 / 8.
when the jump is not done then you save 5 cycles.

Ty, never jr in a loop, then!

Ty. I already knew of Z80 Heaven. Other site seems interesting too. I've discovered an optimization by myself:
Instead of:
LD A,(HL) ;8
ADD A,128 ;8
LD (HL),A ;8

I use
SET 7,(HL); 17
To subtract 128 I use RES 7,(HL)

I use this for double buffering cause I have 31 tiles in every bank from 0 to 30 and the db tiles from 128 to 158

Xor stores the result in A so it’s not faster, you still need to ld (hl),a after.

Optimisations rarely do the same thing, but as long as what it does doesn’t matter for the end result that’s ok. In this case, if the range of input values is 0-127 and you alternate set 7 and res 7, you just use bit 7 as a buffer index.

The two sites mentioned above contain a nice overview of what kind of optimisations you can make and along which lines you need to think. However what I don’t like so much is that they say things like "never do", or this comment here:

  sla l
  rl h         ; I've actually seen this!
; >
  add hl,hl
; -> save 3 bytes and 5 T-states

As if the programmer is incredibly stupid for not making that optimisation...

I think these kind of recommendations should be taken with a big grain of salt and certainly don’t always need to be applied. Many of them obfuscate the code so if it’s not in a hot code path and you’re not pressed for memory space, there’s often something to be said to use the thing that most clearly expresses what you want to do. E.g. for ld a,0 vs. xor a, the former can be preferable. (Though the latter is such a commonly known optimisation that I guess it’s not a big deal either.)

Also as an example of "optimisations rarely do the same thing", the add hl,hl one does not set the zero flag, while xor a sets them all. Always be wary that Z80 instructions are not very consistent in terms of what flags they set.