Thanks for your reply @AldoG
We got the flags from the NXP example.
Looking more into the compiler flags and reference manual, we came to the conclusion that we should use the compiler flag -mfpu=fpv5-d16, since the IMX8MP has a Cortex-M7 with double precision. With that flag, all our target tests are working and math library calls etc works as expected.
However, should it work with the single precision flag?
With the single precision flag our FPU tests sometimes worked. Moving around the statements in the source code and/or adding debug code sometimes made it work. So it seems to be timing or code layout dependent.
It would be great if you added FPU examples.
Best regards,
Björn