malkia on June 18, 2015 | parent | favorite | on: How Naughty Dog Fit Crash Bandicoot into 2MB of RAM

I worked on the port of MGS to PC back in 1999-2000. Here is what I've learned:

- Models were not “skinned” as it was popular in the day. Some textures were covering only the front part of the body, others arms, etc. As such it was possible to use very little colors per texture (16) and use palettes (which is a very small “texture” in the graphics memory). If models were skinned they would've required all the colors used anywhere on the body, and would produce other unpleasant effects (different sampling frequency, especially on the shoulders, etc.) Konami's character modeling is top-notch.

- Music/Sound - this was enignma for us. We were never given their internal sound mixer, but the popular metal gear tune was “mod”-like with very short samples - all of this + game effects was fitting in a 512kb audio buffer (adpcm).

- Game used overlays for the executable part. About 600kb were a main/shared/common part, and if I'm not mistaken 100kb or a bit more were swapped (the overlay). The main part would declare entry-points to be reached, and the “swapped” overlay were like many .so/.dylib/.dll files that knew about the main part.

- TCL-like language was used to script the game, the radio, traps/objects in the game, etc. Each character would have a “main” like function that accepted (int argc, const char argv[]) and handled the arguments from there (these were directly from the TCL scripts). Ah, the whole thing used “C” only.

- So 600kb+100kb, leaves you about 1.0mb for objects, “scenerio” files to be loaded, etc. Since our port was more or less “wrap” the PSX libs as PC, we didn't have to change too much, just on the surface - a bit like patching here and there. - The game used a tricky pointer hack, basically on the PSX accessing a pointer with the highest-24-bit set means read it from the CPU cache, otherwise not (or maybe the other way around). This was used for example to indicate whether the C4 bomb was planted on the ground, or on the wall instead of keeping a booblean/bit flag for it. Possibly was used for some more things. Since it was 24-bit, that meant 16mb.

To work on Windows we had to ensure that we don't go above the 16mb (and the exe starts from 4MB), we also had all overlays for the game compiled-in rather than doing the swapping as the game did, but we had plenty of space even then to fit. It's possible that we might've messed up some of the AI tweaks, but no one complained, and we were young and did not care. Then I had something to find all places where these pointers were used and mask them out when they had to be read, but kept the 24-bit highest bit in there (okay, it's a bit like tagging I've learned much later when I did some Common Lisp).

- As we couldn't do shit about having the original mod-music working, we relied on couple of then popular MGS web-sites and “stole” from them the whole music piece, and other things which came as an audio “pre-rendered” form, and then played them directly from our game. Ah… So embarrased!

- On my part I'm really proud that I was able to do a global-hack where I kept the fixed-point coordinates sub-pixel precision, so our PC port did not “tremble” or “shake” like others to come. Basically on the PSX when you draw a triangle, the “chip” makes all numbers integer pixels, and each vertex “sticks” to a concrete pixel - this makes “shimmering” like feature, and I was able to get around it.

- The other team-mates were able to get software/hardware renderring (directx I think 3, or was it 5?…). Konami used quite a lot of rendering trick that were not available back then. For example the camo-suit basically used the framebuffer as texture, from the location where the character was rendered - so it looked a bit like shimmering!

- Two lessons learned from it - We've put much better high-res textures for the eyes (hired someone from Texas to do it for us), when we got the idea rejected by Hideo himself (by the phone), he told us (through the interpretter) that the game during normal game-play did not have any eye-movement, so higher-res textures would look like crap, while with a blurry texture your own eyes won't see it as a problem - it's really sometimes LESS is better.

- Another was from my boss back then. We had to have a very strict frame-rate - I thnink 30fps otherwise some things were not properly emulated. On some older machines we had the fps going below 15fps, due to the actual renderering, not game code - and since he had experience, he simply said - we'll just skip the drawing of frame then to gain some time. Now that seemed like thing that it should not work, but it did - and saved us from trying to do a non-constant frame-rate hacks.

- Another minor tidbid. The game reffered to it's files/chunks/etc. by using a 16-bit CRC, since there were quite of lot of objects - almost 32000 overall, there would be collisions, but the way Konami solved it was by simply renaming an object, rather than changing the hash-function, or something else. It puzzled us why some soldiers were called charaB or chara4 without other numbers, but when I got afraid of hash-collision, and saw that there are none (for all objects in the game) it kind of explained.

- Who knows how many other treasures we did not discover. In all working on it, made me love the “C” language more than “C++” then. The code was only with japanese comments, and early on I wrote a very simple translator - using the offline Star Dictionary (I've downloaded from somewhere), while it was not much usable, apart from really weird to understand (at first) code or algorithm, it also uncovered things like “CONTINEKU” “METARU GIRU SORIDU” (Continue, Metal Gear Solid), and at first I was like… are these folks writing english with japanese symbols?

- They had a dedicated “optimization” programmer - he basically went through the code, have found the hot-spots and turned them into assembly (mainly model t-junction extrapolations, splitting the model in pieces to fit in the small 1kb fast-cache, and few others). Hopefully he kept the original “C” code, and it was easier for us to choose the right part here and there.

qrmn on June 18, 2015 [-]

Fascinating insights. Lots of music on the PSX used a system like that, because that's a very natural fit for the PSX SPU. Tracker “modules” combine the sample data and tabulated sequencing data, but what you found more often on the PSX was separate sample wavetables and sequencing data closer to (i.e. literally converted from, and convertible back to) a MIDI format: it's smaller, timing-based, without all those pesky 00s wasting space. (It sounds better with the PSX reverb unit/buffer on top, of course.) It's actually very similar to what Minoru Akao did for the AKAO sound engine for the PSX Final Fantasy games, for example.

What did you think of the multi-tasking kernel/DMA bit in the “main” binary? (Or did you just remove that?) By the way, the VR missions mentioned above were released as a separate add-on disc in many regions (rather than the later release Integral which the PC port was). If you do happen to have an original and can't play it on a PS2/PS3 because it doesn't recognise that the 'lid' is open (because it's a tray/slot-loader)… try launching the other executable, it runs fine :)

malkia on June 18, 2015 [-]

I don't rememmber a lot of it. Konami were very late with delivering their (MTS?) system that was their audio/tasking thing, we were not given anything in advance. As such we've just found in the code where the samples and music had to be played, and as I've said above we “stole” (downloaded) the music data from the web-sites that had them (not sure how they've got all the effects, or it could be that we also found some waveforms in the source package).

I think at some point the “radio-codec” (okay, that's the actual Radio that Snake talks to the others) used this system - maybe a bit like fibers/threads, when a message comes, then switches. I'm not sure what exactly I did, and how much I understood it well (threads/fibers were not my thing back then… that much) but we've got it working.

qrmn on June 18, 2015 [-]

I do remember disassembling it and looking at it from the other way! (I was curious.) Pre-emptive multi-tasking kernels at that level weren't things you saw frequently on consoles then. Many people just hung things off the vertical blank instead. You can decode the audio bits - it's just Sony's special version of ADPCM - or use a cartridge (for those PlayStations old enough to have a cartridge port) and read out the SPU memory over X-Link. You didn't even need a debug model to do it (although you did need a handy parallel port and the ability to bit-bang, or run a DOS program).

The CODEC used CD-XA Mode 2 Form 2 (2532-byte sector, with less error-correction layers) ADPCM-compressed streaming audio, 1 of 8 channels, at a relatively low sample rate - which works fine for speech. Lots of PlayStation games used the same basic technique for music and voices (as well as FMVs, although the bulk of that data would have been MDEC-compressed video).

malkia on June 18, 2015 [-]

I've got more familiar later with PS2, as I move to another studio, and was an audio programmer there (Treyarch) and had to do a lot of “PS1” programming on the SPU. I was later reading an article on the TRON operating systems, and found that a lot of the primitives on the PS2 were based on it, even the scheduler to the point - https://en.wikipedia.org/wiki/TRON_project - but never got much of it. Oh, these were some exciting times! - the whole systems was there open for you to see (at least from the software level, and to some point HW).

wnevets on June 18, 2015 [-]

- Two lessons learned from it - We've put much better high-res textures for the eyes (hired someone from Texas to do it for us), when we got the idea rejected by Hideo himself (by the phone), he told us (through the interpretter) that the game during normal game-play did not have any eye-movement, so higher-res textures would look like crap, while with a blurry texture your own eyes won't see it as a problem - it's really sometimes LESS is better.

Could this be part of the reason why I didnt like the look of the gamecube port as much as the PS version?

Retra on June 18, 2015 [-]

That's one of the appeals of classic 8-bit-style artwork: characters features are so poorly defined that it's easy for the player to mentally substitute their own perspective for the character's. (The character is essentially imagined to feel how you feel.) This can give the character a kind of 'charm' that hi res doesn't really do.

malkia on June 18, 2015 [-]

The art in the games of Bitmap Brothers, and many other games - Star Control I/II, Heroes of Might and Magic I, II & III (all with different styles), the grotesque gothic view of Disciples ][ (beautiful!) and plenty of other 2D games. For anyone that enjoys it, here is some great art done with 8-bit palette cycling (cylcing dozen or more colors to achieve animation) - http://www.effectgames.com/demos/canvascycle/ - (select other bitmaps too)

hitchhiker999 on June 22, 2015 [-]

That's a great link, thanks! Something subtly 'exciting' about those images, probably the idea of adventure buried in there somewhere :)

malkia on June 18, 2015 [-]

We've been lucky somehow I guess to discover and talk about such issues in advance. But it could be that a lot of people might not be bothered by them. Even today's game do lack fine touches and are hovering the Uncanny Valley, but it seems people are used to it… The same way I still can't get used to watching Lord Of The Rings in 48fps :)

Jare on June 18, 2015 [-]

Hehe I always get a kick off of showing my wife Uncanny Valley examples. The other day she was really creeped by the E3 FIFA footage. I on the other hand must have developed a way of adapting, or teaching my brain what to enjoy, because I loved 48fps Hobbit despite noticing the artificialness that everyone else complains about.

Natsu on June 19, 2015 [-]

it also uncovered things like “CONTINEKU” “METARU GIRU SORIDU” (Continue, Metal Gear Solid), and at first I was like… are these folks writing english with japanese symbols?

It's pretty normal for Japanese people to write English words in katakana, especially in things like games. Many program menus are perfectly readable by English speakers if you can read katakana. It's something taught in every Japanese school, so being skilled in it makes you look intelligent. It was probably closer to KONEKUTTO and METARU GIRU SORIDO though.

brobinson on June 19, 2015 [-]

Close, it's “soriddo” (ソリッド). English loan words in Japanese are so fascinating to me. Here's an example: “limited slip differential” → リミテッド・スリップ・デフ (rimiteddo surippu defu) (The ・ is used to separate foreign words/names when a Japanese speaker would not be able to figure it out) This must be how Romance-language speakers feel when they see their words modified and incorporated into English. If you want to learn the Katakana syllabary, try this website I found recently: http://katakana.training There's also http://hiragana.training for the other syllabary.

Natsu on June 19, 2015 [-]

Yes, it's surprising how many things you can figure out just by being able to read hiragana and katakana. Though there are a lot of things that tend to be anomalous, like the insertion of small tsu characters in places an English speaker would not imagine a glottal stop, even assuming an English speaker who even knows what that is. Sometimes it really takes imagination. I have a family member who has an arcade game labeled “Hangly Man” (a Pac-Man clone). It took quite a while for it to dawn on me to reverse that back to kana (HANGURI) and figure out that it was meant to be “Hungry Man.”

brobinson on June 30, 2015 [-]

“Hangly Man”

That is quite amusing! I think the hardest word I've found for Koreans and Japanese to say is “parallel”.

mtinkerhess on June 18, 2015 [-]

That is some amazing info! I found this video where you can see the effects of the sub-pixel vertex precision issue: https://www.youtube.com/watch?v=HrFcYbwz_ws

malkia on June 18, 2015 [-]

Thank you for sharing this. This brings so many memories to me, as it was my first professional gig! I've put all my time into it, we were so fast back then, we've got the first level working on PC for two-three weeks. It took much more (7-8 months) to finish. At some point there was a deal whether we would allow load/save from any point in the game, but instead we proposed to include the VR Missions instead of it.

Early on due to my porting libraries I've introduced a severe bug, where the internal timer was 10x (or 100x?) faster, causing issues for loading/saving, this was resolved just weeks before shipping thanks to the other awesome Ukraine programmer. It was also a lesson for me to be less cocky, and take the blame sometimes.

coolnow on June 18, 2015 [-]

Thank you so much for this. MGS is my all time favourite series and i spent countless hours on the PS1 version and also the Integral version on PC.

nickpsecurity on June 18, 2015 [-]

Two things to take away for coding in general: clever CRC trick; dedicated optimization programmer. These might benefit other projects in the future.

malkia on June 20, 2015 [-]

The compilers back then were horrible (gcc2 something?). Our next project was the other way around - porting IHRA Drag Racing from PC to Playstation… now that was a bitch, and showed how inexperienced we were :) (it's much easier to port to something where you have better hardware and software all along).

The IHRA Drag Racing game (PC) version had a full simulation of the engine (valves, torque) - e.g. for any configuration it'll calculate right away the torque/gear ratio tables (okay, I was never a car buff, and my memory is very short here), but essentially there was an algorithm, which I later understood certain car-tuning services were using to adjust real cars! - I mean there were DOUBLES and lots of fortran-looking code written in “C”.

Where we failed was trying to reuse this code on the PSX. First there was no hardware floating point, and what took 10 seconds of calcuation on PC took 45 minutes on PS1 - unacceptable.

Again our boss made a crazy idea - why we don't precalculate some values and store them on the CD - way more limited than the PC, but still something. Not sure how the values were chosen, but overall the game was not a success - one magizine rated it, as one of the worst PSX games ever… I left the project early on, as I felt it wasn't doing good (and was feeling really bad since then about it, as I felt like deserting the person that took care of me, and brought me to US) - http://www.mobygames.com/game/playstation/ihra-motorsports-d…

nickpsecurity on June 21, 2015 [-]

I'm guessing the compilers got better with the modern game consoles? They should be great for latest since it's just multi-core x86. Regarding engine simulation, that's a trip that it was useful enough mechanics were using it. Kind of disturbing they were using it though… Yeah, precalculation is kind of the goto way to deal with this sort of things for many resource- or performance-constrained systems. Always worth remembering.

I looked up the game on eBay. I can't get a consistent price because everyone starts at $10 and works down from there. Still worth somewhere from $1-9 plus shipping. Your worst project is at least helping people with bill money. Not the worst outcome. ;)

deadlockgB on June 19, 2015 [-]

Hopefully he kept the original “C” code

You mean “Fortunately he kept the original “C” code? Great read anyway!

malkia on June 20, 2015 [-]

Thank you! And thanks for the correction, but it looks like I can't correct it anymore (no more editing allowed).