https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89139
David Malcolm <dmalcolm at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2019-02-01
CC| |dmalcolm at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #2 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Indeed, the last elimination is happening too late.
At -O3:
.040t.tailr1 converts the tail-recursion in "g" into a loop
.076i.inline inlines "g" into "foo", but an initialization of a temporary with
"h" remains in "foo"
.103t.dce2 eliminates that last use of "h".
With the "a * 2" -> "a" variant at -O3,
.025t.einline inlines "g" into "foo", converting the tail recursion into a
loop
.033t.fre1 eliminates the usage of "h"
cgraphunit.c:expand_all_functions happens at .090t ("fixup_cfg" and
siblings/descendants), so the elimination of the last "h" in .103t.dce2 happens
when we're already running cgraph_node::expand on "h" (whereas in the einline
variant, "h" has already been eliminated before
cgraphunit.c:expand_all_functions).
