It doesn't take long before you get sick of doing data by hand, and often the time spent in creating tools is repaid many-times-over in the increase in productivity and capability those tools deliver. Sometimes a tool is a 'hack' in that it's not professionally produced, it has bugs, and it isn't user-friendly. But until you've tried creating bitmap graphics by hand a bit-at-a-time (and I'm sure that some of you have already done this by now), you won't really appreciate something - anything! - that can make the process easier. Having prepared you for the fairly shocking quality of this, I now point you towards http://www.atari2600.org/DASM/fsb.exe FSB stands for "Full-Screen-Bitmap", and it's the tool I use for generating the data for those spiffy Interleaved ChronoColour Full-Screen-Bitmaps. But it's able to be used for monochrome playfields, too.
The tool (Windows-only, sorry - if you're on a non-Windows platform then you may need to write your own) is run from a DOS command-line. It takes three graphics files as input (representing the RED, GREEN, and BLUE components of a colour image) and spits-out data which can be used to display the original data on an Atari 2600. For now we're not really at the level of drawing colour bitmaps - but we'll get there shortly. First, let's examine how to use FSB to generate data for simple bitmap displays.
As noted, FSB takes three graphics files as input. Let's simplify things, and pass the utility only one file. This equates to having exactly the same data for red, green, and blue components of each pixel - and hence the image will be black and white (specifically, it will be two-colour). That's the capability of the '2600 playfield display, remember! It's only through trickery that there ever appear to be more than two colours on the screen at any time. That trickery being either time-based or position-based changing of the background and playfield colours to give the impression of more colours.
Actually, I cheated a bit - if we pass only one file, the utility will process it, then have a fit when it can't find the others. As I said, it's a bit of a hack. But sometimes, hacking is OK. Sometime, I'll get a round tuit and fix it up.
Right, let's get right into it. Create yourself a graphic file with a 40 x 192 pixel image, just 2-colours. It doesn't really HAVE to be two colours for the utility to work, but the utility will only process the pixels as on or off. It's difficult to create good-looking images in such low-resolution and odd aspect ratio. Remember, with a graphics package you're probably drawing with square(ish) pixels, so your 40 x 192 image probably looks narrow and tall. On the '2600 it will be pretty much full-screen. That is, the pixels are 'stretched' to roughly 8x their width. So, if you like, use your paint program's capabilities to draw in that aspect ratio. Doesn't matter how you do it, as long as your final image is just 40 pixels across, 192 deep.
Once you have the image, save it as either a .BMP, a .JPG or a .PNG file. I don't support .GIF as the idea of software patents is abhorrent to me. Having said that, I actually am the inventor of one particular piece of patented software (It's true! Look it up - that's exercise 1 for today) so you just never know when I'm serious or not, do you? Once we have that image file, we can feed it into the utility...
Navigate to where you've placed the utility .exe file, and type (without the quotes) "FSB".
You'll see something like this...
D:Atari 2600ToolsFullScreenBitmapDebug>fsb FSB -- Atari 2600 Colour Image Generatorv0.02 Copyright (c)2001 TwoHeaded Software Contains paintlib code. paintlib is copyright (c) 1996-2000 Ulrich von Zadow Usage: FSB [-switch] RED_FILE GREEN_FILE BLUE_FILE Switches... RED_FILE File with red component of image (2-colour) GREEN_FILE File with green component of image (2-colour) BLUE_FILE File with blue component of image (2-colour) v Toggle verbose output ON/OFF nNAME set output filename prefix. Defaults to IMAGE Input files may be .BMP, .JPG, or .PNG format. Reading File: IMAGE Unrecognised and/or unsupported file type.
If you see that, then the utility is working fine. Ignore the various error messages - as I said, it's a hack and incomplete. But it does work well-enough for our purposes. If there's much demand/usage and I'm embarassed enough I'll clean it up.
This time, let's pass an image to it... let's assume we saved our file as test.png in the same directory.
Type (without... you know the drill)... "FSB test.png"
D:Atari 2600FullScreenBitmapGraphics>fsb test.png FSB -- Atari 2600 Colour Image Generatorv0.02 Copyright (c)2001 TwoHeaded Software Contains paintlib code. paintlib is copyright (c) 1996-2000 Ulrich von Zadow Reading File: test.png Bitmap size: 40 x 170 pixels ........ ........ ........ ........ ........ ........ ........ ........ ........ ........ ........ ........ ........ ........ >>>lots of stuff cut from here!! <<< .*..*.*. .*.*.*.* ..*.*.*. **.*..*. ...*.*.* *.*.*.*. .*..*..* .*.*.*.* *.*.*.*. ..*..*.. Reading File: IMAGE Unrecognised and/or unsupported file type.
You'll see a WHOLE LOT MORE of those lines with dots and asterisks. This is my debugging visual output of the graphics as the utility is converting the data. Strictly speaking this is not necessary. But as I said, it's a hack. Increasingly I'm feeling I need to fix this *sigh*. That's one of the problems with releasing your tools for others to use. Please IGNORE that last line saying Unrecognised and/or unsupported file type. It's uh... a feature. Anyway, plunging on...
Remember in the previous sessions how we determined that an asymmetrical playfield was created by writing to playfield registers PF0, PF1, and PF2, and then with exquisite timing writing again to those registers before the scanning of the electron-beam across the scanline got to display them again? In essence, there are 6 bytes of data for each scanline (two of each of the three playfield registers). Although 4 bits in playfield 0 aren't used, and there's a potential saving there of 8 bits total (ie: one byte per line) we're not going to delve into that sort of saving here. Let's just accept that the utility will convert the 40-bit wide image into 'segments' such that we really have data for PF0, PF1, PF2 for the left side of each scanline, and more data for those registers for the right side of each scanline.
Some of the examples presented by our astute readers have already shown formidable asymmetrical playfield solutions - so good, in fact, that I'm not going to trouble with an 'official' asymmetrical playfield solution for these tutorials. Take one of the already-presented solutions and use that.
What I would like to discuss, though, is just how the data for a full-screen-bitmap should be presented. We can organise our data into 192 scanlines, each having 6 bytes of data - or we could organise it into 6 columns, each having 192 bytes of data. The first method is more intuitive (to me, anyway) but it is a much more inefficient way to store our data from the 6502's perspective. In fact, to use the first method correctly we would need to use an addressing-mode of the 6502 that I haven't introduced yet - so let's just look at how the utility spits out the data and hopefully as time goes by you will come to trust my wisdom and perhaps even understand WHY we did it this way
A hint: When using an index register, you can address 256 bytes from any given base-address. That is, the index register can range from 0 to 255, and that register is added to the base address when doing absolute indexed addressing to give you a final address to read from or write-to. Now consider if we had our data organised as 192 lines, each being 6 bytes long... we could do the following...
ldx #0 ; index to the PF data ldy #0 ; line number ALine lda PFData,x ; PF0 data sta PF0 lda PFData+1,x ; the next byte of data (assembler calculates the +1 when assembling) sta PF1 lda PFData+2,x ; the next sta PF2 ; delays here, as appropriate lda PFData+3,x ; PF0 data, right side sta PF0 lda PFData+4,x ; the next sta PF1 lda PFData+5,x ; the next sta PF2 txa clc adc #6 tax ; increment pointer by one line (6 bytes of data) sta WSYNC ; wait till next line iny cpy #192 bne ALine
The above code essentially assumes that the data for the screen is in a single table consisting of 6 bytes per scanline, and that the scanlines are stored consecutively. Can you see the problem with this?
It's a bit obscure, but the problem is when we get to scanline #43. At or about that point, the index register used to access the data will be 42 x 6 (=252) and we come to add 6 to it. So we get 258, right? Wrong! Remember, our registers are 8-bits only, and so we only get the low 8-bits of our result - and so 252 + 6 = 2 (think of it in binary: %11111100 +
%00000110 = %100000010 (9 bits) and the low 8 bits are %00000010 = 2 decimal). So at line 43, instead of accessing data for line 43 we end up accessing data for line 0 again - but worse yet, not from the start of the line, but actually two bytes 'in'. Urk! This is a fundamental limitation of absolute indexed addressing - you are limited to accessing data in a 256-byte area from your base address. There are addressing-modes which allow you to get around this, but they're slower - and besides, it's better to reorganise your data rather than using slow code.
OK, so now let's consider if each of the bytes of the playfield (all 6 of them) were stored in their own tables. Think of the screen being organised into 6 columns each of 192 bytes (the depth of the screen). Since each table is now <256 bytes in size, we can easily access each one of them using absolute indexed addressing. As an added bonus, they can all be accessed using just the one index register which can ALSO double as our line-counter. Like this...
ldx #0 ; line # ALine lda PF0Data,x ; PF0 left sta PF0 lda PF1Data,x ; PF1 left sta PF1 lda PF2Data,x ; PF2 left sta PF2 ; delay as appropriate lda PF3Data,x ; PF0 right sta PF0 lda PF4Data,x ; PF1 right sta PF1 lda PF5Data,x ; PF2 right sta PF2 sta WSYNC inx cpx #192 bne ALine
The above code assumes that there are 6 tables (PF0Data - PF5Data) containing 'strips' or 'columns' of data making up our screen. We COULD have had just a single table with the first 192 bytes being column 0, the next being column 1, etc., and letting the assembler calculate the actual address from the base address like this (snippet...)
ldx #0 ; line # ALine lda PFData,x ; column 0 - PF0 left sta PF0 lda PFData+192,x; column 1 - PF1 left sta PF1 lda PFData+384,x; column 2 - PF2 left ; delay, etc. lda PFData+384+192,x; column 3 - PF0 right ; etc.
What it's important to understand here is that the "+192" etc., is *NOT* done by the 6502. Remember how our assembler converts labels to their actual values (using the symbol table)? Likewise it converts expressions to their actual values - and in this case it will take the value of 'PFData' and add to it 192, and put the resulting 16-bit value as the 2-byte address following the lda op-code. Remember, the 6502 absolute addressing mode is simply given a base address to which it adds the index register to get a final address from which data is retrieved (lda) or to which it is stored (sta).
The above example with the manual-offset from the base address (that is, where +n was added) is functionally identical to the example where there were 6 separately named tables. In both cases, the data is assumed to be strips of 192 bytes, each strip being one of the columns representing the values to put into each of the 6 playfield registers (given that there are 6 writes to three registers per-line, I think of the three registers as 6 separate registers).
So that's exactly what FSB does. It creates 6 tables, each representing a 'strip' of 192 lines of data for a single register. Those tables are saved to a .asm file with the same prefix as the input file, and contents like this (abridged)...
screen screen_STRIP_0 .byte 240 .byte 240 .byte 240 .byte 240 ;188 more bytes here screen_STRIP_1 ;192 bytes here screen_STRIP_2 ;192 bytes here screen_STRIP_3 ;192 bytes here screen_STRIP_4 ;192 bytes here screen_STRIP_5 ;192 bytes here ;end
For space purposes that has been heavily abridged. The file was produced from a source-file called 'screen.jpg' - as you can see, the filename prefix has been used to create labels to identify the whole table ('screen') and also to identify each of the strips ('screen_STRIP_0', etc). So you can use either of the access methods described above, if you wish. Remember, if this file were assembled, the values of the symbols 'screen' and 'screen_STRIP_0' would be identical as they will be at the same address in the binary.
So, we have a DASM-compatible file which contains a text-form version of the graphics file. How do we include this data into our source, so that we may display the data as an image? It's pretty easy - and in fact we've already encountered the method when we included the 'vcs.h' and 'macro.h' files.
We just use the include dasm pseudo-op.
include "screen.asm" ; or whatever your generated file is
When you use the include pseudo-op, DASM actually inserts the contents of the file you specify right then and there into that very spot into the sorce-code it is assembling. So be careful about where you enter that include pseudo-op. Don't put it in the middle of your kernel-loop, for example! Put it somewhere at the beginning or end of your code segment, where it won't be executed as 6502 code. For example, after the jump at the end of your kernel, which goes back to the start of the frame.
1. Create a circle as a 40 x 192 image and save it as a .JPG, .PNG or .BMP. Convert it to source-code through FSB to create source-code data. Can you think of good ways to draw circles in such an odd screen-size? Hint - make the size of your image the LAST step in the draw process!
2. Take one of the asymmetric playfield demos from the last session and convert it to display the data generated in step 1.
3. Set the playfield colour to a RED for one frame, then the next frame set it to a GREEN, and for the third frame set it to a BLUE. What effect do you see? What colour does the circle appear to be? Why? If you haven't cottoned-on yet, this is leading towards colour-bitmap technology - we may cover that in a future session. By using different colours over time, we can trick the eye to seeing a different colour than those we actually use.
4. How can this temporal colour change be used to display a range of colours? This is tricky, so don't worry if you can't understand it. Hint: don't just change the colour each frame! What else can you change?
5. All our discussions about bitmap graphics have revolved around the use of asymmetrical (mirrored) playfields. Yet some (not many!) games use non-mirrored playfields. What timing problems can you see when using non-mirrored playfields for bitmap graphics - and why on earth would you want to do this?
Don't forget to remind me to answer these questions next session!
Until then, enjoy!
Changes in bold made 12/July/2003 - AD