double precision float on ESP8266

[ThomasChr]

Phew, okay that sound resonable!

Just ask if you have any problems building the firmware!

[pythoncoder]

The Pyboard D SF6W has hardware double precision FP. As far as I know the other ports which implement it do it in software.

Astronomical work definitely needs DP, especially if you get into things like predicting solar eclipses...

[glubuluck]

The Pyboard D SF6W has hardware double precision FP. As far as I know the other ports which implement it do it in software.

Astronomical work definitely needs DP, especially if you get into things like predicting solar eclipses...

Thanks, good to know ! But not the same budget.. And actually i can manage with software implementation, since there is no time of computing issue.

[pidou46]

Hi,

Sometime ago I made some thought, about this subject to build an heliostat.

The best option to me would be:
- port a fixed point precision library to microptyhon( the original one is python 3.1 so it should be quite easy)
- code the astronomic algorithm using this library.

Here is the thread for reference:
viewtopic.php?f=15&t=3405&p=19923#p19923

Finaly, I gived up because single precision was enough for me.

[pidou46]

I have made some tests:

I got FixedPoint.py

To avoid an error I remove 'TypeError' from line 248 (I will check later thi point)

"class FXfamilyError(FXexception, TypeError): "

I upload it to my esp32.

import FixedPoint
x=FixedPoint.FXnum=(1)
y=FixedPoint.FXnum=(3)
print(x/y)
0.333333333333333333315

default family is 64bits

I try 128bit adn get an error

[jimmo]

default family is 64bits

I try 128bit adn get an error

The FixedPoint library is quite large -- it's going to use up a significant amount of the RAM on your ESP8266.

This is probably why it fails at 128bit?

I generated an mpy module from FixedPoint.py and was able to do at least basic stuff in 128-bit.

Code: Select all

>>> a = FixedPoint.FXnum(1)
>>> b = FixedPoint.FXnum(3)
>>> print(a/b)
0.33333333333333333331
>>> f = FixedPoint.FXfamily(128)
>>> a = FixedPoint.FXnum(1, f)
>>> b = FixedPoint.FXnum(3, f)
>>> print(a/b)
0.333333333333333333333333333333333333332

To do that I checked out the repo at v1.10 (to match the firmware version on my ESP8266), and ran

Code: Select all

git checkout v1.10
cd mpy-cross
make -j 16
./mpy-cross -X emit=native -O3 ~/src/github.com/rwpenney/spfpm/FixedPoint.py

This gave me ~/src/github.com/rwpenney/spfpm/FixedPoint.mpy, which I put on the ESP8266 instead of the .py file.

A better approach would be to make a frozen module out of it, so the bytecode will be stored in ROM (Flash) instead. You'll have to build your own firmware, after placing FixedPoint.py in the ports/esp8266/modules directory.

[pidou46]

Thanks,

I hope it could help ThomasChr

[apollo2mond]

hi,, if i follow the described way i get this error:
TypeError: multiple bases have instance lay-out conflict

any further hints regarding?

many thanks

[jimmo]

hi,, if i follow the described way i get this error:
TypeError: multiple bases have instance lay-out conflict

hi!
Sorry there have been lots of things described in this post - which part were you following?
It's possible though you might have run into to what @pidou46 said "To avoid an error I remove 'TypeError' from line 248 (I will check later thi point)"

[glubuluck]

Hello,
i just purchased an ESP32-WROOM-32, to try the Roberthh's solution : compile custom FW with double float on ESP32.
But actually, i need help for that...

What do i need ? (sources, compiler..)
How to get it work ?
sorry but my knowledge in micro-controller is just enough to flash regular FW
Thanks a lot

Matthieu