[Bug middle-end/41455] memcpy not tail called if it's a struct assignment

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=41455

Jakub Jelinek  changed:

   What|Removed |Added

 Status|NEW |ASSIGNED
   Assignee|unassigned at gcc dot gnu.org  |jakub at gcc dot gnu.org

--- Comment #6 from Jakub Jelinek  ---
Created attachment 42803
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42803=edit
gcc8-pr41445.patch

Untested fix.

[Bug middle-end/41455] memcpy not tail called if it's a struct assignment

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=41455

--- Comment #5 from Richard Biener  ---
We could do that I suppose.  Of course memcpy will expand to alias-set zero
block copies while aggregate assignment can use better info.

So in the end this means doing full block-copy expansion on GIMPLE rather than
on RTL.  Which sounds messy / impossible given stuff like rep; movb;

Another option is to not use __builtin_memcpy for aggregate assignments but
some internal function call that we can hand extra parameters (for aliasing).

[Bug middle-end/41455] memcpy not tail called if it's a struct assignment

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=41455

--- Comment #4 from Andrew Pinski  ---
I suspect what we can do is "lower" assignments to memcpy in a pass right
before tail call to optimize this case.