Assert failure with compile optimizations

[mathisloge]

If you compile the tests with gcc -g -Wall -Wextra -pedantic-errors tests/tests.c -o bin/tests -O3
you get the error message

===========================================================
ENCODE DECODE MOTOROLA 32 bits double negative value
===========================================================

step 9.2

value to be encoded: -50.6164129

encoded frame: 00 00 00 00 00 00 00 00

decoded value:       0.0000000
tests: tests/tests.c:736: main: Assertion `physical_value == decode_double(frame, startbit, length, MOTOROLA, factor, offset)' failed.
Aborted (core dumped)

The error occurs at optimization levels 1-3 and s (size optimization).

[nisennenmondai]

Hello,
Thank you for your input, I'll investigate this issue when I have a bit of time

nisen

[nisennenmondai]

Hello,
Bug is fixed, I don't know why but GCC is doing some weird optimization when the value past in encode functions is negative, resulting in a 0 being written in the frame. Physical value should be declared as volatile to force compiler not to optimize this variable, let me know if this works for you so I can close this issue.

3f2104a

nisen

[mathisloge]

Hey sorry,
the volatile fix works. Thanks

[nisennenmondai]

Hey sorry,
the volatile fix works. Thanks

I believe I found the real problem and it is quite interesting see this link bellow if you want more information

https://www.embeddeduse.com/2013/08/25/casting-a-negative-float-to-an-unsigned-int/

So basically I think the compiler is doing an implicit double cast float -> int -> unsigned int. But with optimization activated it is doing float -> unsigned int directly thus skipping the int cast resulting in an 0 being inserted when value is inferior to -1 as stated in the article. I am not 100% sure but this is my best guess. So volatile keyword is now removed and instead I put a cast to int before uint64_t.

[mathisloge]

Thanks for sharing the article. It is really interesting, Sometimes casting can be a real thieve of time.

At the moment I have no time to look at the problem in the code, but it sounds like a possible cause.