Question

Right now I’m doing as follow:

uint8_t ManualFlow = 40; // 0 -> 255     Unsigned Char

uint24_t ME; // 0 -> 16777215 Unsigned Short Long
ME = (uint24_t) ManualFlow*10; // Have to make this hack otherwise malfunction in calculation
ME /= 6;
ME *= (80 - 60);
ME /= 100;
ME *= 414;

The end result:

40*10 = 400
400/6 = 66
66*20 = 1320
1320/100 = 13
13*414 = 5382

What I would Love is similar to this:

4/60 = 0,0667 * 20 * 4188 * 0,998 = 5576 (more accurate).

How can I do this more accurate without using floats or doubles, and most important
not increase my code size too much.

Kind Regards
Sonite

Answer 1

conclusion:

My initial code had not so good accuracy and was “this big”.

By doing this below I increased the code with “38 byte” and gained better accuracy

ME = (uint24_t) ManualFlow*100;
ME /= 6;
ME *= (Port[2].AD - Port[3].AD);
ME /= 100;
ME *= 414;
ME /= 10;

The best accuracy I got by doing fixed-point but it increased the code to much “1148 bytes ->

// Utility macros for dealing with 16:16 fixed-point numbers
#define I2X(v) ((int32_t) ((v) * 65536.0 + 0.5))    // Int to Fix32
#define X2I(v) ((int16_t) ((v) + 0x8000 >> 16))     // Fix to Int

ME = I2X(ManualFlow*10); //400 * 65536.0 + 0.5 =   26214400
ME = I2X(ME/6); // 26214400 / 6 = 4369066
ME = I2X(ME * 20); // = 87381320
ME = I2X(ME / 100); // = 873813
ME = I2X(ME * 414); // 361758582
ME = X2I(ME); // 158FFF76 + 8000 >> 16 15907F76 >> 16 = 5520

Hope it can help some one else!

Kind Regards
Sonite