fastest way to code an one pixel sinus scroller

[ross]

Originally Posted by a/b

Is this incremental, like you lower/raise the whole 16-pix column (or 2 halves, as you can use d=a+b) and then adjust individual columns or groups of columns up/down by 1 row from the previous frame, as needed?
I did a 3D bitmap rotation around y axis in a similar way. Basically, I precalculated delta blits needed to morph a frame into the next one, and then merged individual colums that needed shifting for the same amount horizontally, as long as they had the same height (3D perspective stretch/shrink), into groups that were then processed as a single blit. And then blitted in new colums that had a different height.
So a bunch of blit masks...

I write a little more during the day when I find a moment

Originally Posted by Jobbo

a/b

Where can I see that effect, it sounds awesome!!!

I always loved those rare rotation effects like in Brian the Lion that used an iterative vertical skew.

Actually, yours sounds like the rotating skull in Sanity Interference.

Yeahh, Brian the Lion rotation is great,
I always wanted to write something like this...

Originally Posted by a/b

Not to derail this thread... Coming (not so ) soon (TM) when I finally finish my trackmo. It looks ok, it's just a transitional effect. Yeah, you could compare it to the skull in Interference. But I'm using it as a "64x200 sliding window" on 320x200 picture while preparing something else in the background. And I'm also doing vertical scaling of parts of the picture in real-time because there wouldn't be enough room to prescale the whole picture. So while it's pretty optimized (50 fps and it only does ~8% of blits that a copy 1 by 1 approach would do) there's still enough room do to other stuff, it's not a main effect.

[ross]

I will probably say things that are too simplified for someone and verbose, so be patient, I try to make myself understood in the best possible way, but I am not very good in English..
The terms I use may not be the most appropriate so correct me in case.

We are all used to positional notation for writing numbers, since we only have a finite amount of 'symbols' available.
We have a 'digit', an 'index' and a 'radix', and any number can be written as (nx .. n2 n1 n0)b = (nx*b^x)+(n2*b^2)+(n1*b^1)+(n0*b^0).
In our mind immediately the position defines us the index, because we process the number as a whole.

But suppose we are in a system where the digit is 'shown' only in a deferred way, in practice you will know its value when you have already received the information for a previuos state (which define also an implicit index).
In practice a system in which the position is not indicated in a 'parallel' way, as when the number is displayed in full, but only sequentially.
Is the same principle as for serial communications: if you want to know a number between 0 and (b^x)-1 you have to go through x states,
in which each value has no meaning except through the knowledge (feedback) of the previous states.

Now let's go back to what is our purpose by making a simple example with a matrix of 16*16 points (bit), in which we self-impose a maximum displacement, with respect to an initial point, of 15 positions (therefore with 4 state we are able to indicate the displacement, including also the 'non-movement').
This give us a max slope of 45 degree in the bitmatrix: it's an arbitrary choice, I could use a lower or higher slope, just by changing the number of states.
If we take every single column and use it as a reference for the previous state then we can move it by b^x positions in each subsequent state (this is a key point!).
Yes, in every single state we do not know the final position but at the end of the deferred summation we have the final position because we have retroactively inherited the partial result.
We just need a system that allows us to decide when 'reuse' a value from a previous state (basically shifting it to the new state), so a selection mask between the 'do not change state' or 'change state to previous at b^(x-1)'.

This is defined with a logical operation: D=B*C+A*C' (minterm $D8).
Where C is the mask that decide if I grab b from the previous state at (B) or I maintain the current state at (A).
Then D become the new state (the future B..), (A and D need to point to same location and B to the old A/D location). oops! The blitter can do it
[A and D are updated changing the pointers to the new ^x position, so not linearly but exponentially, logb(y_offset)=x].
So for each state (so for each blitter operation on the column) I need a mask that decides whether or not I should use the feedback of the previous operation (hence my use of the term).
You can notice that if the mask is full 0 I can completely skip the operation! (because state is already defined and is not needed an update, is like inserting a 0 in a positional system; you can skip the operations where there 0*b^x).

Said in a simpler way and for a binary system: with 4 state I can arbitrarily move (shift) each column between 0 and 15 (2^4-1) positions.
Basically 4 times faster (minimum, because I can also skip some blitter operation if mask is 0) than doing it for every single column, not bad
How to calculate the mask(s)?
Well, as mentioned in a previous message, it can be seen as a C2P operation because in practice it's a bit rotation in a square matrix (you need to spread the state bit (that define the shift) 'vertically' to the mask bits, therefore existing algorithms can also be used.

Pheew, I don't even want to reread.., so I hope I wrote stuff sensibly and not too much bullsh*t.
For sure I forgot something, but in any case the key points are there.

Cheers!

[a/b]

So in more practical terms... You first shift columns by 1 pixel, then by 2, then by 4, and finally by 8. If a column has to be shifted by 15 pixels, it's affected by all 4 shifts/blits and is contained in all 4 masks. If none of the columns needs to be shifted by more than say 3 pixels, you are done in 2 shifts (1+2).

[ross]

Originally Posted by a/b

So in more practical terms... You first shift columns by 1 pixel, then by 2, then by 4, and finally by 8. If a column has to be shifted by 15 pixels, it's affected by all 4 shifts/blits and is contained in all 4 masks. If none of the columns needs to be shifted by more than say 3 pixels, you are done in 2 shifts (1+2).

[a/b]

Didn't think of that immediatelly, it's a really neat idea and quite logical. I was sure though it's some kind of incremental approach.
Now, I've been thinking a bit more about what I guessed earlier. Wouldn't that be even faster? If you are guaranteed a slope of 45 degress or less, then each column can only move 1 row up or down between consecutive frames. Meaning you can process the whole 16-pixel column in 1 or 2 blits, or... (!!) the entire screen in 2 blits if you use a dynamic mask as C (have two single row screen wide mask for each sine table entry, one for +1 and another for -1).

[DanScott]

Originally Posted by a/b

Didn't think of that immediatelly, it's a really neat idea and quite logical. I was sure though it's some kind of incremental approach.
Now, I've been thinking a bit more about what I guessed earlier. Wouldn't that be even faster? If you are guaranteed a slope of 45 degress or less, then each column can only move 1 row up or down between consecutive frames. Meaning you can process the whole 16-pixel column in 1 or 2 blits, or... (!!) the entire screen in 2 blits if you use a dynamic mask as C (have two single row screen wide mask for each sine table entry, one for +1 and another for -1).

Depends how "fast" you are moving the sine wave through the image... if you move it slowly, then yes, you can guarantee only 1 pixel movement each way with a 45 degree max slope...but... it's going to look quite slow and unpleasing

Also if you are going to actually "scroll" the image right to left (rather than just sine a static image) then this also will increase max y deltas... and of course complicate how you actuall do the scroll (and bring in new data at the correct place on the right)

[ross]

Originally Posted by a/b

Didn't think of that immediatelly, it's a really neat idea and quite logical. I was sure though it's some kind of incremental approach.
Now, I've been thinking a bit more about what I guessed earlier. Wouldn't that be even faster? If you are guaranteed a slope of 45 degress or less, then each column can only move 1 row up or down between consecutive frames. Meaning you can process the whole 16-pixel column in 1 or 2 blits, or... (!!) the entire screen in 2 blits if you use a dynamic mask as C (have two single row screen wide mask for each sine table entry, one for +1 and another for -1).

Well, this can work for some specific effect, but as Dan said for *very* limited sine speeds

Originally Posted by DanScott

Also if you are going to actually "scroll" the image right to left (rather than just sine a static image) then this also will increase max y deltas... and of course complicate how you actuall do the scroll (and bring in new data at the correct place on the right)

Of course this method works and is born for scrolling images, I have a version somewhere that does it even using a playfield made entirely of sprites
I went a bit crazy to calculate the positions so that the curve could be displayed without having more than 8 present on the screen on the same line but that covered all the x.

[Jobbo]

Ross, thanks for sharing. Do you have any released demos from the past that I can watch? You always have some good insight on lots of topics, but I don't know of your work.

[ross]

Originally Posted by Jobbo

Ross, thanks for sharing. Do you have any released demos from the past that I can watch? You always have some good insight on lots of topics, but I don't know of your work.

Nah, nothing good from me
I like keep a low profile on the scene for personal reasons.
So I only do it as a mental exercise and to have some fun.

[morbid]

Originally Posted by a/b

So in more practical terms... You first shift columns by 1 pixel, then by 2, then by 4, and finally by 8. If a column has to be shifted by 15 pixels, it's affected by all 4 shifts/blits and is contained in all 4 masks. If none of the columns needs to be shifted by more than say 3 pixels, you are done in 2 shifts (1+2).

This is very similar to my sinescroll in the demo 'alt cure'.
I first copy one word from the source bitmap to the highest point on the curve within that word.
I then slide pixel columns down with a mask based on their delta y for 1, 2, 4, 8 and 16 pixels.
I of course only do the sliding if it is needed, and all masks and deltas are precalculated just before the effect starts.

[Int42]

Very interesting tech talk here. I don't understand all of that, some things are to advanced for me, but I'll try to learn it.
Difficult to learn for me are things like calculating or rotating stuff with the blitter, or do thinks like zoom with it. I only used the blitter to copy blocks, draw lines and fill areas, but I'm miles away from understanding those delta calculation things or this 1, 2, 4, 8 slide description above, can anybody draw a little sketch, with paint or so, to visual that please?

I'm very impressed about this forum, sorry for my really bad english, I'm an old german native speaker and in my english lessons at school I dreamed to much about the Amiga^^

[Antiriad_UK]

@Morbid *penny drops* ahhh I get it now

[DanScott]

Originally Posted by Antiriad_UK

@Morbid *penny drops* ahhh I get it now

yep.. took me until post #42 before I finally got it 100%...

I think there's other things that can "benefit" from this technique too

[ross]

Ok, then it's time to cancel that disgraceful gifanim at message #28

[Antiriad_UK]

Originally Posted by DanScott

yep.. took me until post #42 before I finally got it 100%...

I think there's other things that can "benefit" from this technique too

#42 was still gibberish to me!

[Antiriad_UK]

Originally Posted by ross

Ok, then it's time to cancel that disgraceful gifanim at message #28

No, leave it It's helpful.

[ross]

Originally Posted by Antiriad_UK

No, leave it It's helpful.

I can re-post this:

[REMOVED]

Dan was sick for 2 days when he saw it


Even that removed, looking at it gives me a headache

[DanScott]

I'm just always sick :P

[ross]

Originally Posted by DanScott

yep.. took me until post #42 before I finally got it 100%...

Originally Posted by Antiriad_UK

#42 was still gibberish to me!

ROTFL
https://en.wikipedia.org/wiki/Phrase...erything_is_42

And the thread is by Int42.
it cannot be accidental

[morbid]

Originally Posted by Antiriad_UK

@Morbid *penny drops* ahhh I get it now

I took me years to figure it out...

Originally Posted by DanScott

I think there's other things that can "benefit" from this technique too

Now, this sounds interesting! Do you have any suggestions?