TIS-100 | Programming puzzle game from SpaceChem dev & GAF's 77th Best GOTY 2015
- Thread starter More_Badass
- Start date
I've learnt 3 assembly languages over the course of my education: PIC assembly, x86 and PowerPC. I've also touched on 68k in my own free time. It's still important to learn these things now.
lordfuzzybutt
Member
I used JRO instead of JMP when I ran out of space in the nodes
I swear I'll rewrite these nodes soon, boss!
Write your labels and codes on the same line and you won't have to use JRO.
That's as far as I went. Don't think I'd try to raise it more because it took most of a weekend to wait for that result.
Looks like it was the same price during the autumn sale.
Hum ... IIRC I didn't do this because I used too long labels, but that might be a luxury not afforded in the cycle wars (plus the nodes are short enough that it shouldn't be an issue).
I'll add some people in the thread tonight and hack out some improvements this weekend. I'm expecting to be completely rolled, though, hah. My id is toppform if anyone wants to add me.
I completely misunderstood what I was doing with the pattern detector for several hours. Time well spent.
Go my signal comparator, go!
I'll admit, I made some evil laughter when I finished this...
not like this
d00d3n
Member
I'll admit, I made some evil laughter when I finished this...
Wow, that is great work. The corrupted notes in the middle are so annoying on that level.
249 Cycles in Signal Comparator. But no idea how to improve it anymore. And my current approach is really messy. lol
edit:
oh I'm at least first somewhere:
edit:
oh I'm at least first somewhere:
Yes, and you can hit F1 to see them in-game, albeit with less description.
Alrighty, cool.
I have no idea how you guys actually get that low... Any hint maybe? (Might look on my solution posted above)
Edit: I initially had more obvious solutions but this got me in the 300s
Edit2: thinking about it ... I have an idea.
The main thing that jumps out at me about your solution is:
your input node takes 6 cycles to handle an L or E input and 5 to handle a G while your output nodes take 6 to handle a G or E and 5 to handle an L. Once you get down to the ~240 range the solution's speed can be pretty sensitive to which kind(s) of input you can handle in 5 cycles and whether you're consistent in that handling throughout the process. Note that your "JLZ 3" is not doing much work.
Your solution is pretty similar to mine if that helps lol.
One thing I can say in general is that if you're comparing an input to be less than zero and equal to zero, then you don't need to check if it'll be greater than zero.
edit: There's probably a better way to do it though... Reddit high score for Signal Comparator is 195/6/71
Spent like 3 hours on Signal Comparator and cant get below the simplest, standard solution (278 cycles), but was far from it. I tried like 5 different approaches. I'm giving up on this one and i'll spoil myself with some posted solution here. I think i'm missing something obvious, we'll see in a second.
Ps. This game is killing me. I spent 8 hours on PC at work, i came home and after like 30minutes sit down to TIS and now its 11pm ;p
Ps. This game is killing me. I spent 8 hours on PC at work, i came home and after like 30minutes sit down to TIS and now its 11pm ;p
I noticed everyone else was faster than me on that but everything I tried just made it slower. ¯\_(ツ_/¯
Yeah, exactly the same.
I looked up the solution from ekim and I didnt know You could do for example JRO LEFT, without going through ACC. This changes everything.
Well, having a go.
Added most of the thread; steamid: http://steamcommunity.com/profiles/76561198032677017
Added most of the thread; steamid: http://steamcommunity.com/profiles/76561198032677017
Sometimes I could punch myself. Makes sense. This will probably just save me a few cycles tho. I have the feeling I could optimize the equal zero out more.
71 instructions for the high score on Reddit? That's sounds fishy.
I subconsciously realized but just consciously acknowledged that leaderboards for each challenge take best per program.
As in, cycle count, node count, and instruction count aren't necessarily all the same program.
I feel like this is something I KNEW, but every time I was looking at leaderboards it wasn't clicking until just now. I feel WAY better about some of my results now.
As in, cycle count, node count, and instruction count aren't necessarily all the same program.
I feel like this is something I KNEW, but every time I was looking at leaderboards it wasn't clicking until just now. I feel WAY better about some of my results now.
do all instructions take 1 cycle? Something isn't quite right.
Why does a MOV UP, ACC \ MOV ACC, DOWN take 50% longer, and not 100% longer, than MOV UP, DOWN?
e: okay... starting to get the hang of this. Can't wait to see the record slaughtered when more people accept the friend request @_@
Why does a MOV UP, ACC \ MOV ACC, DOWN take 50% longer, and not 100% longer, than MOV UP, DOWN?
e: okay... starting to get the hang of this. Can't wait to see the record slaughtered when more people accept the friend request @_@
RE: Signal Comparator
Take the standard solution and change the JEZ to JNZ for the node where you are testing for 0. This will cut a few cycles since most numbers are not zero and you can skip reading the JMP command on the != 0 cases.
I have no idea how people are getting 230 or lower though. Tried to revise my code for an hour, and all I did was cut 266 to 265. On to something else!
Take the standard solution and change the JEZ to JNZ for the node where you are testing for 0. This will cut a few cycles since most numbers are not zero and you can skip reading the JMP command on the != 0 cases.
I have no idea how people are getting 230 or lower though. Tried to revise my code for an hour, and all I did was cut 266 to 265. On to something else!
Most instructions take one cycle. Instructions that move something to L/R/U/D take at least two. The first cycle pushes the source to the destination port and then the instruction completes simultaneously with the receiving instruction in the other node.
Ah, so all instructions which push a value to I/O are blocking until the value is consumed?
That makes sense.
Hey, this is holding. I may actually be able to sleep.
I think i can get it down some more, though...
Wait, you can optimize the differential converter? How the hell did durante do that.
No, really.
160 means no node that goes over 4C, which means no read + 2x write, which means no 2x write. Ever. How do you split data? I parallelized the split, but that has just gotten it to 180, and even if i could somehow thread to x3 that's be 170, not 160... argh.
Ah, so all instructions which push a value to I/O are blocking until the value is consumed?
That makes sense.
Hey, this is holding. I may actually be able to sleep.
I think i can get it down some more, though...
Wait, you can optimize the differential converter? How the hell did durante do that.
No, really.
160 means no node that goes over 4C, which means no read + 2x write, which means no 2x write. Ever. How do you split data? I parallelized the split, but that has just gotten it to 180, and even if i could somehow thread to x3 that's be 170, not 160... argh.
There's a lot you can do with this one.
There's a lot you can do with this one.
Wait.
Wait.
...
What?
WHAT?
Okay, sleep is pushed back for a while...
Stumpokapow
listen to the mad man
you bought it for me, hahaha.
Yeah there were three people on my friends list with it on their wishlist. I picked you randomly. Didn't connect the names but that works out well as this seems to be a game that consumes mods and ex-mods
This game is great for teaching all kinds of concepts related to parallelization.
Often I made a much more involved, parallel version of something, only to have it end up slower due to synchronization overhead. Just like real life!
In unrelated news, I built an 18 instruction solution for "Histogram Viewer". And I see that Cyan got a lot closer to my instruction count in the second image test pattern.
Often I made a much more involved, parallel version of something, only to have it end up slower due to synchronization overhead. Just like real life!
In unrelated news, I built an 18 instruction solution for "Histogram Viewer". And I see that Cyan got a lot closer to my instruction count in the second image test pattern.
d00d3n
Member
Works without a hitch on my macbook air (2013 model).
Got my Differential Converter Cycle count down to 165 by using two threads. I'm happy now!
Here my solution if anyone is interested.
http://abload.de/img/unbenanntksszr.png
No idea how Gotchaye got to 127 though
edit: and down to 164. lol
edit: 163 - both outputs are now fed the same time. 127 seems impossible?!
Here my solution if anyone is interested.
http://abload.de/img/unbenanntksszr.png
No idea how Gotchaye got to 127 though
edit: and down to 164. lol
edit: 163 - both outputs are now fed the same time. 127 seems impossible?!
This is basically 90% of my efforts to "optimise" solutions in this game. Completely refactor the code the spread the load across as many nodes as possible only to end up a few hundred cycles slower.
I'm starting to find that what needs optimization is usually I/O buffers <.<
Thats my world,
- hey lets try rework it to use more nods and start it from completely different side
30 minutes later result is 150-200 cycles,
- lets optimize it more
1.5 hour later result - is 100 cycles slower from the simplest solution
WTF?!
---
Ok, now i'm in blank how exactly leaderboards are working. I should be there with 3516 cycles, but i'm not ;pI spent 2 hours making a fast histogram viewer to go with my tiny one. It's ... not tiny.
I have an idea of how to build a faster one yet, but the implementation effort is a deterrent.
--
Ps. I need to stop optimizing early levels and do ones i have not done yet ;p
Yeah, once you understand the timings of reads and writes it seems to become a matter of making sure your vital nodes are never waiting on I/O and making minimal writes. Poorly parallelized code (in my experience) generally gets stuck on nodes that make too many writes. One of my solutions is currently stuck on a node with nothing more than:
Code:
MOV UP, ACC
MOV ACC, LEFT
MOV ACC, DOWNNo I/O waits, but two writes means five cycles to execute, which turns an seemingly simple node into the program's main bottleneck.
Is that Signal Multiplexer? He. I had run into that bottleneck =\
So count me in as another who's been swept up by TIS-100. Although I was only currious after some chat about it in the Steam thread but about 3 hours after picking it up on Steam (and at a time my lower back was about to give way) I (un)fortunately noticed it got released on iOS yesterday. So that's the last 4 hours spent prodding the grey matter led flat out on my back in agony then. Still I still getting the smug sense of satisfaction when everything comes together even if I am 8 quid out of pocket buying it twice!
Its not wasted money though. You can still use Steam version to check leaderboards.
I personally think that this game loses a lot of 'playability and charm' without leaderboards.
I really hope that game will do reasonably well for Zachtronics and they will make sequel or at least spiritual successor. For example, better UI would be a starter. I'm really annoyed by a fact that You cant save and swap program versions on the fly and there are only 3 slots. Proper Steam Workshop integration would also do wonders to this game, as better leaderboards and for more aspects of code. Better debugging etc
I just built a solution to "Signal Pattern Detector" which actually uses every single available node, in an effort to break below the 3 cycles / output barrier.
After I finally got it to work It actually runs at 5 cycles / output :/
On the other hand, I'm pretty happy with my new & improved exposure mask viewer:
I really do like the image stuff best.
After I finally got it to work It actually runs at 5 cycles / output :/
On the other hand, I'm pretty happy with my new & improved exposure mask viewer:
I really do like the image stuff best.
More_Badass
Member
Wait, what? TIS released on mobile? Got a link?
How the hell did I miss that?
Its not wasted money though. You can still use Steam version to check leaderboards.
I personally think that this game loses a lot of 'playability and charm' without leaderboards.
I really hope that game will do reasonably well for Zachtronics and they will make sequel or at least spiritual successor. For example, better UI would be a starter. I'm really annoyed by a fact that You cant save and swap program versions on the fly and there are only 3 slots. Proper Steam Workshop integration would also do wonders to this game, as better leaderboards and for more aspects of code. Better debugging etc
Oh for sure and despite being well off the pace it does certainly give me some food for thought about where I'm being inefficient and where I need to improve on.
Well the release date was yesterday
https://itunes.apple.com/us/app/tis-100p/id1070879899?mt=8
Oh on iPad? Screw it, double dipping.
It will always look strange to me when the same game on mobile cost less than game on PC.
Btw there is no info about this release on official site or even their twitter. Hell their TIS-100 website still has this sentence ;p
"Now available on Steam Early Access for Windows, OS X, and Linux!"