[Bug target/69176] [6 Regression] ICE in in final_scan_insn, at final.c:2981 on aarch64-linux-gnu

[rth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69176

Richard Henderson <rth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #37267|0                           |1
        is obsolete|                            |

--- Comment #13 from Richard Henderson <rth at gcc dot gnu.org> ---
Created attachment 37281
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37281&action=edit
second patch

Ok, so the patch didn't survive overnight testing.

The primary issue being that stack pointer adjustments need to
be split immediately, since try_split will refuse to operate on
an instruction that modifies the stack pointer.

Thanks for the test cases in #c11, I'd forgotten about exposing
this to cse.  This adjusted patch handles them as expected.

The lack of condition on the peepholes was just a think-o.
Testing for single-insn move was always intended, just forgotten.

Full testing for this is still a couple hours off completion.