I read from an IMU values in the format FP16.
I store them as 16bit integer.
There is a practical and useful component to convert types.
Maybe you can add an input to read FP16 and get it as a floating point from Flowcode (is a FP32 ? )
2024-11-01_15-06-37.PNG (21.02 KiB) Viewed 6359 times
2024-11-01_15-08-43.PNG (11.07 KiB) Viewed 6359 times
Half precision floating-point
2024-11-01_15-16-06.PNG (33.54 KiB) Viewed 6359 times
That looks very good.
Now I still have a small error. I don't know if it comes from the STM32 board or from where.
I have heard that FP16 has advantages in many applications, especially with microprocessors and AI or Motion, because it is faster and needs less memory.
In my application I want to send the values from the IMU to the Flowcode App still FB16 and only convert them to FB32 in the app
But I can also send FP32 if necessary.
Project fp16 (1)
Device STM32F469xx
In file included from fp16 (1).c:20:0:
C:\c#_sourcecode\_#FC10\FP16toFP32\fp16 (1).h:246:19: error: conflicting types for 'ushort'
typedef MX_UINT32 ushort;
^
In file included from c:\program files (x86)\flowcode\compilers\starm\gcc\arm-none-eabi\include\stdio.h:61:0,
from c:\program files (x86)\flowcode\compilers\starm\boards\f469discovery.h:19,
from C:\c#_sourcecode\_#FC10\FP16toFP32\fp16 (1).h:32,
from fp16 (1).c:20:
c:\program files (x86)\flowcode\compilers\starm\gcc\arm-none-eabi\include\sys\types.h:107:24: note: previous declaration of 'ushort' was here
typedef unsigned short ushort; /* System V compatibility */
^
In file included from fp16 (1).c:20:0:
C:\c#_sourcecode\_#FC10\FP16toFP32\fp16 (1).h:247:19: error: conflicting types for 'uint'
typedef MX_UINT32 uint;
^
In file included from c:\program files (x86)\flowcode\compilers\starm\gcc\arm-none-eabi\include\stdio.h:61:0,
from c:\program files (x86)\flowcode\compilers\starm\boards\f469discovery.h:19,
from C:\c#_sourcecode\_#FC10\FP16toFP32\fp16 (1).h:32,
from fp16 (1).c:20:
c:\program files (x86)\flowcode\compilers\starm\gcc\arm-none-eabi\include\sys\types.h:108:22: note: previous declaration of 'uint' was here
typedef unsigned int uint; /* System V compatibility */
^
Error(s) in build
So either remove the #define (in supplementary code) or
Change the ushorts on the code to MX_UINT32 (and remove the #define)
Try option 1 first and check the results - ushort seems to need to be 32 bit (and the results are wrong if it is a 16 bit value) so I assume that 64 bit is now common as I'd assumed it would be 16...
That doesn't make the whole thing any easier..64Bit or 32Bit
In Flowcode:
Floating point
A floating point value can represent a much wider range of values than an integer can, but at a loss of accuracy over large ranges. Floating point values when downloaded will be 64-bit if the target supports them, or 32-bit if it does not.
FLOAT
Icon: Image
Range: -Infinite to +Infinite
Bit Depth: 32-bit signed
Float is a keyword in Flowcode and so cannot be used for variable, property or component names.
Mathematical operations such as a = b + c will be processed as an integer calculation if a, b or c are integer values.
for example you would expect 2.5 * 2 to equal 5 but in the world of integers the calculation only sees 2.5 * 2 = 4, the real portion of the number is lost.
The keyword float can be used to force a calculation to be done using floating point maths instead of integer maths.
e.g. FloatVar1 = FloatVar2 + float IntVar1
Did you know numeric constants such as 35 are also considered as integer values.
Using the float keyword or changing the constant value to for example 35.0 would force the value to be recognised as a floating point type variable
MX_UINT32 is 32 bit and so are floats - so the 'trick' to load a value as a float from it's memory address works okay..
The C compiler used by the person who wrote the code - I'm guessing that had 64 bit ints? But I might be wrong - it's far easier to have a fixed set of types rather than them being define (or redefined) at the start of every program...
It is simply converting and saving to an array, perfect!
but it is not possible to send as Float_Array.
Receiving in the app is also not possible, only as INT_Array.
And if I send the values as a Float instead of FB16, I send twice as many bytes.
Question: can I also do the conversion from FP16 to Float in the app?
and also add a send/receive Float_Array?
no send Float Array available:
2024-11-05_10-06-00.PNG (22.76 KiB) Viewed 6106 times
simply converting and saving to an Float_Array:
2024-11-05_10-18-44.PNG (154.6 KiB) Viewed 6106 times
I have tried to make a macro for the conversion FP16 to FP32 that works in emdeded and in the App.
I tested different values and it works. But three values don't work. Do you have an idea?
)