Interesting. What compiler options would you have used? Do you know if the options are applicable for ARM as well?

"-march=...", e.g. "-march=skylake" or "-march=znver3" or whatever CPU you are using.

When you do not know the correct type, and you do not cross-compile, you can just use "-march=native".

I recommend to never use either gcc or clang without giving an explicit "-march=..." option. Otherwise you do not know which is the default compilation target and it is almost certain that the quality of the generated code will be bad. Even for code that will be distributed in binary form for multiple computers, you must choose consciously a compilation target that is the minimum that should be supported and you must not depend on a random default choice that may change at each compiler update.

For ARM, there are similar options for specifying the CPU model, e.g. "-march=armv8.2-a". However there are much fewer models and counting leading zeroes has been supported for about 20 years, so even the default ARM CPU model must include it.

On ARM only the compiler options for specifying the floating-point instruction subset can be tricky (an example of such an option: "-mfpu=fpv4-sp-d16"), when you are targetting embedded computers or microcontrollers, because there you can still encounter a lot of older ARM cores that have a much more restricted floating-point support than in more recent models. Moreover, some vendors of ARM-based devices may choose to reduce the cost by disabling some floating-point features, so knowing the name of an ARM core may be not enough for the selection of the compiler options, you must also know whether the optional core synthesis options have been enabled or disabled.