Issue 90 Issue 91 Contents Issue 92

HOW the HELL ... Does a Sprite Routine Work?

ASCII code nightmares? Disillusioned with disassemblers? Baffled by bytes? If you’re having problems with programming, whether they alliterate or not, Andrew Hewson is your man. Drop him a line and he’ll be investigating the problem before you can say “Ram Dos Buffer Interface Edge Connector”

From the people who write in with questions for this column, a lot frequently ask ‘How do you go about designing a sprite routine’? The problem with sprite routines is that there are many different types of routine that are really specific to the game in which it is operating.

Take something like ‘URIDIUM’, for example. It was based around a scrolling background with sprites moving over the surface of the spaceship. Due to the highly complex nature of the scrolling system, the sprite routine was optimised to work in that game only. A simple platform game, on the other hand, is a lot simpler than a full-blown 3D-type game and correspondingly the sprite routine is different in each case.

We shall look at the principles involved this month and next month I will have a super smooth sprite routine for you to try out.


Ok! What is flicker or shear? First, I will explain flicker. Your television monitor works on the principle of a raster scan. When you are watching a broadcast programme on TV, the picture you are watching is updated 25 times every second through the action of a pin prick of light (an electron beam) which traces out the picture starting at the top left corner and scanning line by line as shown in figure 1.

When the beam reaches the bottom right corner, the beam is switched off for a fraction of a second and it is directed back to the top left corner ready for the next scan whereupon it is switched back on again. The beam scans at a rate of 64 microseconds per line for 625 lines deep. The Spectrum interrupt incidentally is synchronised to occur at the beam ‘flyback’.

Before I go into any more detail, please note that the actual TV scan is a little bit more complicated in that the scan is done at a rate of 50 times a second with two scans of 312.5 lines per scan. Also, although there are 625 lines in total, only around 575 are visible. As far as our Spectrum is concerned, we can observe the display system like this:

  1. At the time of a Spectrum interrupt, the TV beam is at the top left corner of the screen. The first part of the beam outputs is the border at the top of the screen.
  2. Approximately 3.5 milliseconds later, the beam is scanning the screen memory and is updating our screen. The screen is 192 lines with a piece of border at each end.
  3. The screen memory scan takes approximately 12.5 milliseconds.
  4. The beam is now at the bottom border section of the display and is going to take another 3.5 milliseconds or so.
  5. Once the beam has reached the bottom right corner (after 312.5 lines), it is blanked and flies back to the top left hand comer ready for another frame update.

The above five steps are carried out 50 times a second using only 312.5 lines unlike the broadcast TV which uses the other ‘in between’ set of 312.5 lines. If you get a magnifying glass and look at the monitor screen of your Spectrum while it is displaying something you will be able to see a small black gap in between each pixel row. This rather complicated sounding system is based around the findings of early cinema pioneers who discovered that 25 frames per second was enough to convey the appearance of a moving picture without your eye detecting the effect of separate pictures — hence TV evolved along similar lines (no pun intended!)

Right, back to the plot. If we copy the relevant sprite data to the screen memory very quickly (i.e. we are outputting a sprite) then, providing the TV beam is scanning a part of the screen memory which is after our sprite or well before it, the sprite will be updated perfectly when the beam next reaches it. In other words, the perfect sprite system would be able to write the sprite data to the screen memory in an infinitely short period at time — during frame flyback say. Once the beam starts updating the screen then all the sprites would be output as well.

However, the ideal sprite system is just a pipe dream on our Spectrum. The cause of flicker is that our eye can see the background poking through all or part of the sprite in between the updating of the next sprite frame. To prevent the background ‘showing through’ we have to design the sprite routine around the principle of ‘never erase the sprite’.

‘But how can we make a sprite move around the screen if we don’t erase the old frames?’ The answer is that we draw the picture to be output in a separate workspace area of the memory. After the picture is prepared, we output this workspace screen to the main ‘visible’ screen — overwriting what was there before at a rate of anything quicker than about 60 milliseconds. As most of the screen will consist of the some data, the only parts of the screen to change will be the sprites in their new positions. The new positions will typically be no more than 1 character square away from their previous positions and the net change to the screen will be very small. More importantly is the fact that sprites are never erased.

See figure 2 for a graphic illustration of the method. This workspace screen method was used in many games including Manic Miner and Jet Set Willy where the press at the time thought the animation and sprite output was exceptionally good.

We are getting quite technical this month with talk of rasters and electron beams and have not quite finished yet as we have not looked at shear. If you dig out your grubby copy of Manic Miner and can remember how to play it, look at the sprites very carefully and on some screens you will see the main character ‘Willy’ lean slightly with the appearance that there is a split through his torso with the upper and lower halves slightly skew wiff (figure 3). This effect is shear. The TV scanning system is responsible again but admittedly the effect is less of a problem than flicker.

What happens here is that the workspace screen is transferred to the main screen as described but the TV beam is racing through outputting the screen faster than the copy-workspace-screen routine. If this happens then any part of the screen which changes after the raster has passed through, will only be output to the TV monitor 20 milliseconds later on the next scan. On the Commodore 64 for example it has a register commonly called the ‘raster compare value’. There is a counter in the video circuitry which is compared with this ‘raster compare’ value. If the two numbers match then an interrupt is generated.

On the C64 this Is used to split the screen into two or more different modes vertically at a predetermined point. No such luxury on the Spectrum though! So we are really stuck with shear but we can minimise it by doing the copy routine when the beam is just starting its scan i.e. do a HALT immediately before the copy.

So now we know how to reduce the little ‘bugs’ which spoil sprite routines we can break the actual routine down to several stages:

  1. Initialise Workspace A with the background screen.

Now we do the following operations in a loop:

  1. Copy Workspace A to workspace B.
  2. Draw in our sprites to Workspace B.
  3. Copy Workspace B to the main screen.
  4. Move sprite co-ordinates etc. and loop back to No 2.

The only tricky bit is that our sprite draw routine and the copying at various workspaces must be done at lightning speed. All the shuffling of data involved in this system is wasted time in that we want it to happen in zero time. Lots of data are being moved around with this system and most of it doesn’t change - an obvious time waste. However, look on the bright side, all the erasing is done automatically and in converting a program designed this way for other computers, only the final copy routine needs changing. Get your assemblers ready for next month and try out the sprite routine.


Figure 1.

Figure 2.

Figure 3.

Previous article in series (issue 90)
Next article in series (issue 92)

Issue 90 Issue 91 Contents Issue 92

Sinclair User
October 1989
Transcribed by Richard Milne