[Bug tree-optimization/90671] [10 Regression] ICE on valid code at -Os and above with -g enabled in gsi_split_seq_after, at gimple-iterator.c:345

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90671

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 46431
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46431&action=edit
gcc10-pr90671.patch

Untested fix.  If during the first call the template block is empty, neither
gsi_split_seq_after nor gsi_insert_seq_after can actually work.  My
understanding is in that case we want to hide the whole bb_seq from the
create_block_for_threading because it contains only debug stmts we do not want
to copy there.