some small imporvements

[TheMasterofBlubb]

Hi, i found your Video on the interwebs and decided to give you some tips.

1. Dont use std::vector on ESP32(embed in general), when adding elements to vectors and exceeding its initial size it will resize the vector which includes a copy process of all elements. For embed std::deque is more fitting as it works with chunks that can be added or removed. Imagine linked list but with memory blocks that store arrays. In general though only use vectors/deque if your final size is unknown, otherwise use arrays.

2. Dont use std::vectorstd::vector, this usually end is getting an element from the inner vector to be a dereferencing of the entry in the outer vector. A way better and more performant solution is to use a multidimensional array (either a c array int[][] or an std::array std::array<std::array<int, 4>, 4> ), this claims a fixed size block of memory, where any access is a simple memory read.

3. Apply Nr.1+2 to your placeRandom method

4. std::string on embed. This is a very nasty thing where you might get weird issues on longer runtimes and even slowdowns. Strings have the same reallocation mechanism as vectors, but when constructing one you kinda have to live with that, but there is a fallpit.
   string a = "A" + "B" will be optimized to not make issues.
   string a = "A"; string b= "B"; a = a+b;
   Now you might think that this is fine. but it isnt. What it does is create 2 strings, including memory allocation. then create a third string, including memory allocation, but thats still acceptable
   string a = "A"; string b= "B"; string c= "C"; a = a+b+c;
   this on the on the other hand goes wild. Its 3 string allocations initially. then a+b is another allocation, lets call that string ab, and then ab+c does another allocation. So over all you did 5 allocations.

   Ignoring memory fragmentation, this can actually slow down performance quite a bit, because memory allocations are painfully slow.

   A better way to construct strings is :
   string a = "A"; string b= "B"; string c= "C"; a += b; a += c;
   This does 3 initial allocations and resizes if nessecary, which you can prevent by guessing the size on initialization of the string.

5. The gameCleared method. This one is interesting. After every move you check for the state of the game, for that you iterate over the whole board and check for a field with more than 2048 points. Now in general that inefficient, but lets roll with that. If you use a multidimensional array, you can actually use a single non-nested loop because a multidimensional array is internally a one-dimensional one with length dimx*dimy. Nested loops are very inefficient in itself. Everytime you run the outer loop once a portion of the stack has to be allocated to run the inner loop(imagine the inner part of a loop as a function that gets called on every iteration). so what happens is that your nested loop created a lot of stack allocations for a simple if-statement. Using a non nested loop will already help.

   BUT there is an even better solution. Dont loop at all. You have a place where you calculate the results of a move, mainly the new field numbers. Add the if statement there and keep track of it in a bool tied to the Game-Object. And in gameCleared just return the value of said bool.

6. BONUS: The ESP32 is very powerfull in many regards. Especially since it has an FPU, but the caviat is that the FPU does not accelerate doubles only floats, im not sure if the AI Code can be converted to using floats, but if it can, that should give you a massive boost (the operations can be between 2 and 20x slower for doublen than for float)

7. BONUS2: I dont know if applicable here, but the ESP32 allows for multithreading, you can even use the usual std::thread for that.