[Bug target/59549] [SH] __builtin_return_address causes push/pop of PR

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59549

--- Comment #2 from Oleg Endo  ---
Ran over the following in sh.c, maybe it's related:

  /* For Shcompact, if not optimizing, we end up with a memory reference
 using the return address pointer for __builtin_return_address even
 though there is no actual need to put the PR register on the stack. 
*/
  pr_live |= df_regs_ever_live_p (RETURN_ADDRESS_POINTER_REGNUM);
}
  /* Force PR to be live if the prologue has to call the SHmedia
 argument decoder or register saver.  */
  has_call = pr_live;

[Bug target/59549] [SH] __builtin_return_address causes push/pop of PR

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59549

--- Comment #1 from Oleg Endo olegendo at gcc dot gnu.org ---
I haven't checked, but it could be related to this.  When shifts are
expanded/combined on SH2 (no dynamic shifts), library function calls might be
intially inserted but then removed again during combine/split1.  If that
happens, PR push/pop insns might appear, although a function doesn't contain
any function call.