[llvm-dev] Aliasing rules for const type pointer

Nema, Ashutosh via llvm-dev llvm-dev at lists.llvm.org
Tue Jun 11 08:10:51 PDT 2019

Thanks Tim, I understood the point.


-----Original Message-----
From: Tim Northover <t.p.northover at gmail.com> 
Sent: Tuesday, June 11, 2019 1:57 PM
To: Nema, Ashutosh <Ashutosh.Nema at amd.com>
Cc: llvm-dev <llvm-dev at lists.llvm.org>
Subject: Re: [llvm-dev] Aliasing rules for const type pointer

[CAUTION: External Email]

Hi Ashutosh,

On Tue, 11 Jun 2019 at 06:32, Nema, Ashutosh via llvm-dev <llvm-dev at lists.llvm.org> wrote:
> In this test case the type for “C” is const int * type, it means the pointer is pointing to a memory which is of type constant.
> My understanding is value at *C can’t be modified in the program, so its invariant and possibly can be hoisted.

This isn't quite right. In C and C++, const applies to a specific pointer, and only rarely to the underlying memory[*]. A given object is often accessed through both const and non-const pointers througout its lifetime.

What const really means on a function like your foo is that the writer of foo has promised not to modify C through that pointer (and even that might only be an approximation since you can cast the const away). It's still possible for the caller to pass the same object in through A or B (as you say later), and depending on foo's contract that may be perfectly legitimate or an API violation. The compiler doesn't know which, so it can't assume they don't alias.

Think about memmove for a concrete example which is very close to your
foo: the source is a const pointer, but it's explicitly allowed to overlap with the dest.



[*] The memory itself is usually const for literal strings, and sometimes global variables declared const. Situations where the compiler knows that an object is immutable where it's actually allocated.

More information about the llvm-dev mailing list