float math

[kmp84]

Hello Friends,
I know that floating point are so hard for PCM,PCH micro. I want to use example "ex_tank.c" for simple level tank volume measurement system.
Is it possible without float math ? Any suggestion and advice for math and which MCU for this project ?

Best Regards!

[drolleman]

It's more what other things the mcu is doing, to decide what mcu to use. All will do the job. Math operations are hard on cpu cycles. Using an int32 or int64 will reduce cpu cycles to get the job done depending on the accuracy needed. But for basic use and convenience just use the float. 16f part will be the slowest, 18f will be faster, 24fj will be faster yet and the fastest will be the 24E, that is if you have the pcd compiler.

drolleman

[kmp84]

Thanks for your answer mr. "drolleman".
In "ex_tank.c" has math with trig function.

[asmboy]

it all depends on the precision you need .

you can do this with scaled 32 bit integers where lets say
your dimensions are in centimeters.
* multiply all you dimensions by 100 or 1000
* define PI as say PIX at 314 or 3142
then doing the same math with 32 bit integers will leave lots more prog space free.

you can use printf with %w formatting to insert the decimal place where you want it for output

[temtronic]

also..

don't do 'PI/2'. Instead define a variable say halfPI = 1570. division takes a LOT of memory and CPU time.

if you want to get real clever precompute(use a PC or calculator) so the math uses binary divisors or multipliers( /8 instead of /10, /32 instead of /100.It is far easier and faster for the PIC to work in binary.The 'math' then becomes simple for the PIC, though harder for you .
when you 'think like a PIC' you'll be able to do more complicated math faster taking less memory.

Jay

[PCM programmer]

temtronic,
PI is a constant in math.h, and the compiler does the division at compile
time, not at run-time. See the ASM code below from the .LST file.
The compiler is loading PI/2 (1.57079637051 in Microchip 4-byte
floating point format) into temp1. No division routine is called:

[Ttelmah]

The example here is just that. An example. It is trying to show that the PIC can do some quite complex maths if you are not worried about size or speed.

Now the powers can more efficiently be done by simple multiplication. It is much more efficient (and more accurate...), to generate x*x*x than to generate x^3!. So a lump of the code can be shrunk straight away. You may be surprised at how much this alone can save!... Then the accuracy is in part 'fallacious'. The actual sin functions for instance do not give the full accuracy of the maths. Basically it takes too long to solve things really well. You can do a remarkably good 'integer cos' (giving the results in something like 1/100000th parts), that is quicker and smaller than the standard functions. I've done fast LCD libraries like this, using just a look-up table, and interpolation.

Now, I'd say any of the reasonably experienced programmers here could re-write 'ex_tank', in an afternoon, to give results accurate to a small fraction of a mm^3, and using probably 1/4 the time and space. However it'd involve some thought, and so if you want to do this, the _you_ are going to have to do this....

However just try getting rid of the pow terms and using *. It might save all the space you need. Just tried it using the example code for the 452, and generating this function:

[kmp84]

Thanks for all advice and suggestion. It will be very helpful for me. I did not understand how to replace trig function without float. Any example ?

[drolleman]

1st why are you using a 16f part?

Is this a large scale product, is the cost of the mcu important in the calculation? Can you add a couple of $$. if so go to a 18f or 24f part.


If you are at 80% already you will run out very quickly. The way 16f parts uses memory should have been retired a long time ago. I don't know why it still is used. When they retired the 17f they should have done the 16f as well.

1st guess of how much program memory requirements are. Then get a device that has double that or more. Start here then find the part to fit the job.

Trying to stuff in a shoebox in harder than placing the item in a large box.

Get it functionally working first, even if it is bloated. Then trim it down like the previous posts suggest.


drolleman

[asmboy]

come clean. WHAT are you trying to accomplish ?

[kmp84]

Thanks for all suggestion developers! All they are Welcome!
I'll use 18F it's no problem , but I was wondering that above simple math calc take so much ROM memory.
Because "asmboy" ask I will tell the project requirements:
Horizontal cylinder tank volume meausurment system (max 6 tank). In all tanks there are level meausure device which return level in "mm".
Returning level, volume over serial protocol like modbus rtu.

Thanks and Have a Nice day!

[Ttelmah]

OK.

If you look at the example, the complexity is because the tank here has elliptical ends. A simple tubular tank doesn't.

The area is just PI*R^2, and the volume, this * height.

Your actual maths size will be tiny compared to the example.

[kmp84]

[Ttelmah]

That doesn't make things much worse. Look for the solutions for the area of a segment of a circle. It's just the area of a sector of a circle, with the area of a triangle removed. Obviously, the volume is just this area times the length of the tank. Still doesn't get involved in the complexities of the elliptical ends involved in the example given.

[kmp84]

Hi again mr."Ttelmah",
In "ex tank.c" the volume of ellipsoid ends are calc with :