Use InlineArray for cache to match C

[soraros]

I think it's a good example of progressive disclosure of complexity.

• Make type (defaults to UInt32) an alias
• Use InlineArray instead of List for cache, remove bound check
• Format code using mblack (official Mojo code formatter with default settings)

[Yona5]

Running this version of code on the mojo playground is slower than main_v2.mojo.

[dorjeduck]

On my Mac 2 Pro laptop, i get nearly identical performance results:

List / UInt32 version:

Benchmark 1: ./main2
  Time (mean ± σ):      1.623 s ±  0.001 s    [User: 1.603 s, System: 0.020 s]
  Range (min … max):    1.622 s …  1.624 s    2 runs

InlineArray / UInt32 version (this PR)

Benchmark 1: ./main4
 Time (mean ± σ):      1.678 s ±  0.016 s    [User: 1.652 s, System: 0.021 s]
 Range (min … max):    1.666 s …  1.689 s    2 runs

Would love to see benchmark results of these two on other devices.

[soraros]

I'm not that concerned with performance (I know they should be roughly identical, for the generated assembly says so). This PR is more to showcase how to write such a program using arrays.

BTW, try running this benchmark script.

Assembly

[dorjeduck]

Thanks for the script: I get

--------------------------------------------------------------------------------
Benchmark results
--------------------------------------------------------------------------------
name,met (ms),iters
"list",2486.6900000000001,3
"array",1636.8683333333331,3

;-)

[martinvuyk]

see my comment about the benchmark, it is not so straightforward a change

[jabbalaci]

Why is it necessary to modify v1?

[martinvuyk]

@jabbalaci it's just the code formatter that mojo uses. If you run magic run mojo format . it will format all mojo files in that directory. The extension should do it automatically every time you save the file actually.

[soraros]

@jabbalaci I just wanted to minimise unrelated differences so that comparing the two files in a diff view would be clearer. And in order to avoid my own bias in coding style, I used the default code formatter. If you prefer not to include the cosmetic changes to v1, I can revert them.

[jabbalaci]

A @parameter line is also removed from v1. I don't know what it meant, I've never used Mojo.

[soraros]

@jabbalaci Good catch. Fixed.

[martinvuyk]

In my machine it seems like the InlineArray is a bit faster than SIMD, but this will likely change with server grade CPUs

Doing a global var cache and passing it as a function argument to is_muchausen():

"InlineArray[uint32]",3449.9183376666665,3
"SIMD[uint32]",3860.0218596666668,3

Doing a local alias cache inside the function is_muchausen()

"InlineArray[uint32]",3235.4407679999999,3
"SIMD[uint32]",3797.3388070000001,3

See PR #68 for more benchmark results in my machine