Use precalculated factors

[Jozo132]

Replacing floating point division with floating point multiplication for faster data processing in main loop.

[gabriel-milan]

Hi @Jozo132, thanks for the collaboration.

Although I enjoy the idea of this PR, if I do merge this I think I'd have an issue:

• On 18bd718 you set 1 / 16384.0 as 0.00006103515 and that's accurate, but
• on 87c8f1c you change it from division to multiplication.

As the values are numerically equal, if you change the operation to the inverse one, results will change.

For that to work, you have either to leave the *_TRANSFORMATION_NUMBER as they are and change the operation to multiplication, or the opposite.

Since the whole idea is to increase data processing speed, the best thing to do is the first one. Do you wish to fix that and open a new PR?

[Jozo132]

Hi @Jozo132, thanks for the collaboration.

Although I enjoy the idea of this PR, if I do merge this I think I'd have an issue:

• On 18bd718 you set 1 / 16384.0 as 0.00006103515 and that's accurate, but
• on 87c8f1c you change it from division to multiplication.

As the values are numerically equal, if you change the operation to the inverse one, results will change.

For that to work, you have either to leave the *_TRANSFORMATION_NUMBER as they are and change the operation to multiplication, or the opposite.

Since the whole idea is to increase data processing speed, the best thing to do is the first one. Do you wish to fix that and open a new PR?

As you can see, the idea was to replace divisions with multiplication.
So if this abstraction holds true:
y = k / x <==> y = k * (1/x)
that means that a constant z = 1/x simplifies all repeating divisions to multiplications
y = k * z

My 'GIT' skills are terrible and I probably messed up somewhere.
Could you please edit the changes to make it work out?

[Jozo132]

Oops, sorry ... I thought that was the 'Close' window for the comment input

Anyway, now that I think about what you asked me. No.
It is correct. The division has to be replaced in the cpp file.
The explanation just above should be enough.

[gabriel-milan]

Omg my mistake.

When I read your PR's code, I looked on the commentaries right after your code line:

// (1 / 16384) precalculated

and assumed that 1 / 16384 was the previous value set. In that case, my first comment would make sense but yeah, you're right.

I'll merge this PR and publish a new release right after, thanks again and sorry for the misunderstanding.

[Jozo132]

@gabriel-milan I might have spotted a major issue. I have requested this pull request for each file sperately but we only merged ONE!!
When I just scrolled through your code I noticed the divisions were not replaced with multiplications but the constants were changed!