How to add compiler options in nRF Connect SDK / implicit conversion from 'float' to 'double'

[puz_md]

Hi,

I just switched from nRF Connect SDK v2.5.3 to v2.8.0. With the new SDK, I get very many warnings called:
implicit conversion from 'float' to 'double' to match other operand of binary expression [-Wdouble-promotion]

The warning did not appear in v2.5.3.

I found out that there is a gcc compiler option called -fsingle-precision-constant which would fix the problem.

Question 1: Why does this warning appear in v2.8.0 but not in v2.5.3? Was the default interpretation for constants changed in the new SDK? (numbers without explicit type information interpreted as float in v2.5.3 and as double in v2.8.0?)

Then, I tried adding the option with the following line in CMakeLists.txt, which did not help:
add_compile_options(-fsingle-precision-constant)

Later, I found another instruction that worked:
target_compile_options(app PRIVATE -fsingle-precision-constant)

Question 2: What is the difference between add_compile_options() and target_compile_options()? Why did the first instruction not work?

Best regards,
Michael

[Håkon Alseth]

Hi,

 

With Zephyr v3.6.0 and newer, this was added as a default warning flag:

https://docs.zephyrproject.org/latest/releases/release-notes-3.6.html

Added -Wdouble-promotion as a default warning when compiling to warn developers with single-precision floats easily being promoted to double-precision.

 

A typical function that emits such a warning is printf (and variants):

https://stackoverflow.com/questions/28097564/why-does-printf-promote-a-float-to-a-double

Later, I found another instruction that worked:
target_compile_options(app PRIVATE -fsingle-precision-constant)

You can also append using CONFIG_COMPILER_OPT="-f..."

Question 2: What is the difference between add_compile_options() and target_compile_options()? Why did the first instruction not work?

To my understanding, add_compile_options() works on a directory level, and not a target level. Ordering comes into play, and thus using a command to set it to the target is often recommended.

 

Kind regards,

Håkon

[puz_md]

Oh... this means that our v2.5.3 based implementation runs inefficiently, even without the warnings. Good to know.

Thanks for the feedback!