[Bug middle-end/90345] too pessimistic check whether pointer may alias a local variable

[vanyacpp at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90345

--- Comment #4 from Ivan Sorokin <vanyacpp at gmail dot com> ---
Making points-to analysis aware of SESE regions will definitely help here and
is a nice thing to have.

There is one more option. In my reduced test case the body of 'push_back' is
unavailable, but when it is it can be analysed and an attribute can be added
that 'push_back' only uses the received reference internally and does not
escape it.

>From my experiments this is what clang does: even when the body of 'push_back'
is not inlined it generates different code for 'operator*=' depending on
whether push_back escapes the received reference or not:

void push_back(uint32_t const&) __attribute__((noinline));

void big_integer::push_back(uint32_t const& a)
{
    __asm__("" : : : "memory");
    //__asm__("" : : "g"(&a) : "memory");
}

I guess with LTO enabled this type of analysis is quite powerful, as many
'const&' and 'this' parameters in C++ don't really escape.