Achieving O0 performance by disabling O1 level optimizations

[Eric Botcazou]

> I would greatly appreciate any guidance on the following:
> 
> 
>    - Are there optimization-related behaviors at -O1 that are controlled
>    implicitly (e.g., via internal optimize > 0 gating or other non-flag
>    mechanisms)?
>    - Is it expected that -O0 behavior cannot be fully reproduced by
>    disabling flags starting from -O1?
>    - Are there additional factors that significantly differentiate -O1 from
>    -O0, and can I redirect those properties while transitioning from O1 to
> O0?
>    - Can I implement any flag that will redirect to O0 properties even if
>      I run O1 with all defined disable flags?

There are global variables (optimize, optimize_debug, optimize_fast and 
optimize_size) directly set by the -Ox switches, in combination with others.
So no -Ox switch can be reduced to a simple combination of -f switches, and 
that's indeed particularly visible for -O1 vs -O0.  We already have way too 
many -f switches in GCC, so I'm not sure that more of them would be welcome.

-- 
Eric Botcazou