Video Chess disassembled and commented
nanochess.org
nanochess.org
IIRC, the game devs used a PDP-11 for much of their work, so that had a bit more RAM to work with to get started.
The problem seems to be more the 4KB of ROM, but only if one is trying to make it actually somewhat good, and also perhaps using 1979 tech for programming.
The photographs I've seen aren't all that interesting.
A VT-52 terminal and a hand-built cart wired into an RS-232 port, all on a long desk.
Hopefully there are more exciting photographs out there that I haven't seen.
Also check out the links provided at the end of the document.
Besides the PDP-11, I'm glad to see the beloved PDP-10 was also used by game developers back in the days.
IIRC, ours had 64KB of RAM in two banks.
Better call the puzzle training something else than Puzzle Rush though.
Thank you.
> Better call the puzzle training something else than Puzzle Rush though.
Why? I think it's the most recognised name for the sequence of puzzles becoming harder and harder, even if I removed the time limit (as that leads to not learning much :)
I'm no UX engineer at all, but the sounds just do not sound chess-y. Just a thought.
C64 (and other home computer) games had to load all code and data into RAM, which usually accounts for most RAM usage. Game consoles like the Atari 2600 stored the code and static data in a ROM cartridge.
For simple arcade-style games (like Pacman), 128 bytes of RAM is more then enough to manage the gameplay state.
(still of course: a chess engine in 4 KByte ROM and 128 bytes RAM is pretty damn impressive)
Bravo to OP for the disassembly and explanation, too!
Here’s a silly example: https://blog.jgc.org/2022/11/primality-testing-on-commodore-...
I think you might be "suffering" from the Dunning-Kruger Effect. Even if I accept that "all" the coders at that time were able to do this, that would just mean that the pool of people who were capable coders then is vastly more skilled than the typical coder these days.
When others say, "we had to", they are in part, communicating how thinking was done. The lower level details were often published down to the bits in registers and doing even simple things took one into tricks land more often than we may realize today.
Now that said, the VCS (2600) is a difficult machine. But, it was not above an average programmer to be successful as much as it did demand one go through the careful steps and thinking as told today. In fact, just doing what the machine was designed for; namely, "Combat" required that thinking.
I think you may be getting hung up on scope and scale. What I mean is the field was not as broad in some ways, and most everyone was a lot closer to the hardware than we mostly are today too.
Kids were jumping onto micros and literally figuring it out with the info they could glean from one another, manuals shipped with the machines (maybe), magazines and the like were the basic communications happening daily. I was one of those kids and in my peer group it was expected to understand the hardware well enough to follow code.
So yeah, overall it was just different in ways that may well prove confusing today, depending on what any of us may have chosen as our start point.
But holy cow, that chess game really pushed the limits!
(A side note I like to bring up occasionally: my first "computer" was "Atari BASIC" on my 2600. It had 63 bytes of RAM, I think. That didn't last long until my parents brought home a TS1000 with a luxurious 1KB of RAM.)
In 1985, me and another boy from school wrote a whole reliable network transport and stuffed it into unused memory on a Research Machines 480Z. We had 128 bytes for the receive buffer so in order to do anything we'd literally ship a packet across the network containing machine code and the stub code would CALL the receive buffer to perform tasks. Executable packets!
Having 4 kilobytes of ROM made it relatively easy, I think. If one ignores draws by repeated positions (which I think all small chess programs do), board state can easily be stored in 67 bytes or so (64 for the board proper plus four bits for whether various castlings are allowed, plus a byte indicating the square for en passant opportunity, plus a byte for counting for the 50-move rule), leaving 61 for the search algorithm.
I think https://en.wikipedia.org/wiki/1K_ZX_Chess, with 672 bytes of code and therefore 352 bytes of RAM for storing game state _and_ screen data might be more impressive.
I would have to see at what level both programs play to be sure, though.
Video Chess uses the lower nibble of 64 bytes for the board, and the upper bits of those for various other purposes.
You actually could condense further than that. Build the storage as an array of pieces rather than squares, so 32 storage locations. Each of those locations needs only 6 bits to indicate that piece's current square. (And even only 5 bits per bishop, although pawns need three more bits each to handle promotion and to which piece type.) Of course, writing an engine for that data structure is much more cumbersome, since it can't look at a square to find what piece is on it, instead it has to iterate over all the pieces to find out what's on a square.
Ignoring the move generation, this is also way harder to render the output with. The Atari 2600 has very few cpu cyles per line, and I don't know if you'd have enough time to iterate through all the pieces to see if and where they are on the current line, even given that several video lines are spent on each chess rank.
Of course it's possible to go even more than that - enumerate every reachable position (so remove situations like a pawn on the first or eighth rank, kings adjacent or in some check situation that couldn't have happened, pawns in positions that would require more captures than there are enemy pieces missing from the board, etc) and assign them all an index number, which requires log N storage size. And of course it would be practically impossible to do anything computationally with positions stored like that.
This link estimates an upper bound of around 10^46 reachable positions, so log2 of that would be only about 150 bits to enumerate them all.
https://math.stackexchange.com/questions/1406919/how-many-le...
As a neat side effect, this means the user can see all the moves that the engine is considering.
In the early 00's one enthusiast had the idea to make a simple BASIC compiler for the 2600. I was one of the early testers and the first thing I did was make a functional "BREAKOUT", bounce the ball off the bat to eliminate bricks in a wall game.
It took a couple hours and was, I think, the first (and buggy) playable program authored by someone other than the language author. And it was fun!
What the author did was kind of brilliant in that he packaged up several of the common tricks used to make games and presented them along with a simple BASIC.
Variables, for example, were just bytes and or individual bits, essentially addressing bits in a byte like an array.
J = %01101101
If J[2] = 0 then X = X +1
x=50
y=50
main2
COLUP0=28
COLUBK=02
player0:
%00011100
%00011000
%00011000
%00100000
%01011010
%01111100
%00100100
%00010000
%00011000
%00111100
%00011000
end
player0x=x
player0y=y
drawscreen
if joy0right then x=x+1
if joy0left then x=x-1
if joy0up then y=y-1
if joy0down then y=y+1
goto main2
The BASIC exposes the little bit of hardware to a programmer in a simple way that I found to be a lot of fun because the really hard parts of assembly language are not necessary, but can be written in-line just as easily as defining a sprite is:
asm
LDA J
ASL
ASL
.
.
.
end
Fred did a lot of great work along with some others to create both Batari Basic and the Harmony cartridge, both of which mean mere mortals can have a little fun programming for the 2600.
For instance your bitmaps for pieces/spaces are stored in ROM, the code called when a scanline is drawn goes and accessed the values in ROM (it's just a memory address) without needing it to ever living in RAM. The only part that needs to live in RAM is the board state that the drawing code uses to find the bitmaps to draw. There's no need to draw to an off-screen buffer in RAM that then needs to be drawn to the screen.
None of that is to say 128 bytes of RAM is somehow easy but the code and data in ROM can do a lot of heavy lifting without taking much RAM.
Only until you need more rom than address space available to the cartridge slot, and you need to add bank switching. Then you've got to keep track of what bank resources are in.
Keep in mind is these kind of things have a flat
memory space between RAM and ROM
While you know this, for the benefit of those reading: you could also use ROM for time/space tradeoff purposes -- for example instead of doing a bunch of trig calculations in real time for character movement purposes you could just have lookup tables in ROM.
the code called when a scanline is drawn goes and
accessed the values in ROM (it's just a memory address)
without needing it to ever living in RAM
For example the Sega Genesis/Megadrive has a scanline based video chip. No framebuffer, etc. But you've got to copy graphics data from ROM into the 64KB of video RAM before it can be drawn onscreen -- no rendering directly from ROM.
You can do this during the VBLANK interval (the time between frames) but there is not enough time to rewrite all 64KB so you've got to be crafty. Typical procedure was to stream animation data for the main character into VBLANK because you would like your main character to be animated as smoothly as possible. [1]
Contemporary systems like the Neo-Geo did not have that restriction and could render ROM data directly,[2] although it was several times more expensive than the Genesis and had many more address lines physically exposed on the ROM cartridge connector. May have used faster ROM chips too; dunno.
[1] https://rasterscroll.com/mdgraphics/graphical-effects/animat...
Contemporary systems like the Neo-Geo did not have that
restriction and could render ROM data directly
Games on the Genesis and SNES could therefore use video RAM as sort of a makeshift kludgy framebuffer for polygonal games or for fancier raster effects. The Neo-Geo could do no such thing.
The Neo-Geo might have been able to given some custom hardware on the cartridge itself, with a cartridge-based CPU rendering data to RAM on the cart that was seamlessly mapped into the cartridge's address space. I have seen some wacky emulator projects that work this way, like DOOM on NES: https://hackaday.com/2019/06/06/doom-on-the-nes/
On a Genesis, games spend most of vblank feeding the VDP chip with sprite and other graphics data, and then can run their game logic during the screen on time. With some effects requiring frobbing the VDP chip during the drawing phase.
However, the genesis and contemperaries still don't have a frame buffer, so their graphics chips are processing background tiles and sprites pretty much just in time. I see some notes that the Genesis VDP does some of this work earlier than most others, so something like changing sprite position data in vram won't be effective until the next frame, whereas contemporaries can reuse a sprite on the top and bottom of a frame by repositioning it in the middle. Otoh, the Atari 2600 doesn't really have sprites, you have to program the TIA with patterns each line.
For the Genesis you've got two independent background layers. They are made of tiles. You also have 80 sprites which are also tiles.
For each sprite, you give the system an X,Y position and a pointer to VRAM for the graphics data.
For each background layer, you give it a scroll offset and a list of pointers to VRAM for the graphics data. Very similar to sprites, but you move the whole layer at once so each tile doesn't get its own X,Y.
You don't have to do this on every frame. It will just keep drawing the same thing over and over until you change something. You don't have to do steps 1 through 3 on every single frame - you just make the changes you need. If the game is paused or the player hasn't moved, for example, you might not do anything at all.
You have a limited amount of time during VBLANK (the time between frames) to do graphics work. This is where you update sprite/background positions, transfer data, change palettes, whatever.
You have an even more limited amount of time during HBLANK (the time between individual lines in a frame) to do things as well. A common use for this is updating the background scroll position in the middle of the frame. This is how you get "line scrolling" effects like the cool pseudo-3D floor effect in Street Fighter II. You can also change palette colors. This is how you get the transparent water effect in Sonic the Hedgehog even through the Genesis doesn't have true alpha transparency - you change the palette part of the way down the screen during VBLANK.
MicroChess says hold my 924 byte beer: http://www.benlo.com/microchess/index.html
MicroChess is also believed to be the first home computer software that was available for sale to the public.
I have this on my KIM-1, and I find watching it play itself kind of soothing.
1K ZX Chess's code takes up only 672 bytes in memory,[2] but implements chess rules except for castling, promotion, and en passant, including a computer opponent.[3] It was the smallest implementation of chess on any computer at the time.
672 bytes RAM, No ROM. No promotion is a killer though
I love the 2600 breakdowns and the hacks to get things working.
True story: One time I was at a party and my parents introduced me to some friends of theirs and they said their son worked on graphics drivers. I said I couldn't handle low-level work like that. My parents glared at me and told me I was being rude. Once I figured out the disconnect I explained that low-level means closer to the CPU and it's much more complicated - not that it's less desirable work.
I'm trying to get it to the speed-runnable state before Mysterium next week, we'll see how that goes
Edit: Really? Downvotes? Certainly it must have one. You could play someone on chess.com and find out?
I was wondering the same thing. It is interesting to see the exact feature in past engines with limited resources that overcomes the threshold and moves an engine from the bottom to the higher ranks, and how historically what ideas were in place and which were culled that moved the engines in that direction.
The rules are the same with three major differences.
I) Black has the option of transposing any two pieces on their back rank before play begins. If they do they must transpose the parallel pieces on their other side to make their setup symmetrical (unless they transpose the king and queen). White must adjust their setup to mirror Black's, and then play can commence.
This creates 11 possible setups:
1)b>RNBQKBNR w>RNBQKBNR
2)b>RNBKQBNR w>RNBKQBNR
3)b>RBNQKNBR w>RBNQKNBR
4)b>BNRQKRNB w>BNRQKRNB
5)b>NRBQKBRN w>NRBQKBRN
6)b>RNQBBKNR w>RNQBBKNR
7)b>RNKBBQNR w>RNKBBQNR
8)b>RQBNNBKR w>RQBNNBKR
9)b>RKBNNBQR w>RKBNNBQR
10)b>QNBRRBNK w>QNBRRBNK
11)b>KNBRRBNQ w>KNBRRBNQ
Castleling is only possible in the first five set-ups, and will always be with the piece that occupies the Rook's original square.
II) Pieces can be demoted by moving them to an original pawn square (on one's own side of the board) and declaring a demotion. However, there must be at least one of their pawns captured that they can "exchange" it for. Note that the new pawn could deliver checkmate/stalemate.
III) Pawns can only be promoted to pieces currently captured. If no pieces are currently captured, the pawn may not move to the promotion rank (as if a piece was blocking it). Note that it may still deliver checkmate/stalemate in this position, as normal.
The thinking behind the first rule is to make book opening memorization less of an issue (without making things too complicated or unusual) as well as giving Black compensation for moving second. The thinking behind demotions is to make games that would normally be draws due to lack of material (or skill) winning. The thinking behind the last rule is to make over-the-board play easier (e.g. no confusion about whether a piece is a pawn or queen), though it also potentially makes "two for one" exchanges more attractive for the person capturing two pieces (I recently saw somewhere that there's some evidence that when the queen is exchanged for two rooks or the rook for two minor pieces, the person losing two pieces typically does better, though nominally they have less in material value).