Question on the level Differential Converter

[Pixel Knight]

So the first two levels were ridiculously easy, but after that, the third level, really upped the difficulty I felt.

The problem is, everyone else seemed to solve the level in about 250 cycles, whereas mine took 472, which is WAAAY outside the realm of being an outlier. So I want to ask about my logic, on that level.

It seems like the method I used was way too slow compared to everyone else's.

Basically, I have the first nodes read in the inputs, then save it to the BAC, then move the inputs down to the next node. I then first swap over the A input to the right, then the B input to the left, and then perform the calculations, then continue with each value.

What is the easier way to do this that takes far less cycles? I feel like if I am not seeing something so simple, I am in for big trouble later in the game.

[Cheese]

My first time solution was similar to yours. Especially early in the game you tend to make the problems more complex than they are. If you play further in levels, redoing the early levels will give much better cycle times and use less nodes.

My tip is that instead of looking at what you must do, look at the outputs. The way I'd do it now is with just one substraction for 201 cycles

[Pixel Knight]

One subtraction with two separate outputs? That is possible?

[Cheese]

I used SUB instruction only one time for IN.A-IN.B. For the other output I used NEG since the other output is always -1 times the first output to get the result IN.B-IN.A.

[Pixel Knight]

Ahh, yeah. That is a smart method. I noticed that, and thought about using the that operation, but then decided against it, ultimately. It sounds like that is actually a much better method.

[天冷手抖，天热眼花]

Originally posted by Doomcheese:

My first time solution was similar to yours. Especially early in the game you tend to make the problems more complex than they are. If you play further in levels, redoing the early levels will give much better cycle times and use less nodes.

My tip is that instead of looking at what you must do, look at the outputs. The way I'd do it now is with just one substraction for 201 cycles

Hey guys, do you manage to get the cycles, nods and sentences all best optimized at one time?

I kinda forgot my algorithm, but I remember I tried 2 ways, once got best node & sentence but around 240-250 cycles, the other got best node & cycles but more sentence. Just can't get all 3 done...

[Cheese]

Originally posted by Pixel Knight:

Ahh, yeah. That is a smart method. I noticed that, and thought about using the that operation, but then decided against it, ultimately. It sounds like that is actually a much better method.

Both get the job done! Looking at the scores too early can take away a bit of the fun. If you want to get good scores, I suggest you come back to earlier puzzles when you're further in the game.

[Cheese]

Originally posted by FunQie:

Hey guys, do you manage to get the cycles, nods and sentences all best optimized at one time?

Often short efficient takes least amount of nodes too, I only care about cycles if I try to optimize.

[noob_Lurker]

Originally posted by Ghostbird:

It's not always possible, but I think for this level it is.

Are you sure about that?

There's usually a trade-off between speed and resources, and for this level i have two programs saved: a "201 cycles / 5 nodes / 11 instructions" (which i named Fastest), and a "240 cycles / 5 nodes / 10 instructions" (which i named Optimal Mem)...

[Calisthra]

I do like seeing what others' stats are *after I complete a puzzle. This one had me at 202/6/13 before seeing this post. When I went back to the project the obvious error of moving the data early was staring at me. Saved me 1 node and 2 instructions at a glance without losing speed.

[kschang77]

World record is about 133/6/36. It's TRIPLE pipelined.

My personal best prior to peeking was 181/6/23, which was only parallelized. (dual pipelined)