Closed escherstair closed 3 years ago
@escherstair Thanks for reporting it. It should have been caught. It looks like -Wall -Wextra -Werror is not enough to detect it.
I'll correct it.
@escherstair I have started to work on those issues of float promotion. It will take more time than I had anticipated because I want to correct all issues and the float16 code has a lot more because of ambiguity with types __fp16 and _Float16.
You have not seen those issues because float16 is disabled when building CMSIS-DSP with ARM Compiler 5.
Thanks for reporting this problem which is not detected with -Wall -Wextra on gcc and clang. I have added the missing option to track this issue in the future.
@christophe0606 changing the compiler from ARM Compiler 5 to 6 is one of the next tasks on my roadmap. So, thank you for your job. I hope that the release with these fixes will be released soon.
It should be corrected (among other related issues) in commit https://github.com/ARM-software/CMSIS_5/commit/347ca08b52808350c6328cd6a28f42f75dcddaad
I've just upgraded CMSIS-DSP from 1.8.0 to 1.9.0 and this brings to new compiler warnings. File arm_rotation2quaternion_f32.c gives several "single-precision operand implicitly converted to double-precision". The reason is that single-precision constant numbers should be written with 'f' suffix (i.e.
1.0f
). Here are the lines with the warning:Can you fix these warnings, please?