According to the people at Ada Core Technologies, that develop gnat, the GNU Ada compiler, all compliance tests has been done using the -O2 flag with the compiler. Thus that mode should be the most tested and reliable option for generating code.
I have no problem believing that aerospace people distrust optimizers. They are in fact unlikely to trust anything before they have inspected the assembler output from the compiler.
It's been my experience that the aerospace industry distrust anything that's less than 30 years old. A lot of people have died from trying to introduce new technology, so they do kind of have a point. However, I think they have mostly learned the wrong lessons from the past.
Wrong is almost always contextual. If the cost of a false positive is that someone dies, and the cost of a false negative is that development is slower, guess which one is chosen?
Slow development also costs lives. One of the major causes of accidents in small plains is simply running out of fuel because the pilot forgot to fill up on the ground and or did not check the fuel gauge in the air. Now, there are a lot of ways to prevent these accidents, but at some point you need to change basic design elements and not just make a longer check list.
I have no problem believing that aerospace people distrust optimizers. They are in fact unlikely to trust anything before they have inspected the assembler output from the compiler.