https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97673
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
So the issue is that we are inlining
# .MEM_5 = VDEF <.MEM_1(D)>
master.0.sub3noiso.constprop (1, z_2(D), y_3(D), x_4(D), 0B, 0B, 1, 1, 0, 0);
$12 = void
and duplicate the clones BLOCK tree via
if (DECL_INITIAL (fn))
{
if (gimple_block (stmt))
{
tree *var;
prepend_lexical_block (id->block,
remap_blocks (DECL_INITIAL (fn), id));
but the actual stmts we copy refer to the original function BLOCKs.
That's likely because we ask for the untransformed body but that process
does not "update" the scope tree? Adding
diff --git a/gcc/tree-inline.c b/gcc/tree-inline.c
index cd3fad1fa98..84f71d9c6cc 100644
--- a/gcc/tree-inline.c
+++ b/gcc/tree-inline.c
@@ -6513,6 +6513,7 @@ tree_function_versioning (tree old_decl, tree new_decl,
delete param_body_adjs;
free_dominance_info (CDI_DOMINATORS);
free_dominance_info (CDI_POST_DOMINATORS);
+ verify_gimple_in_cfg (cfun, false);
gcc_assert (!id.debug_stmts.exists ());
pop_cfun ();
yields
> ./f951 -quiet t.f90 -O3 -fno-early-inlining --param large-stack-frame=4000
t.f90:9:20:
9 | entry sub3noisoEntry(x,y,z)
| ^
Error: location references block not in block tree
# .MEM_5 = VDEF <.MEM_4(D)>
D.4037 = {};
during IPA pass: inline
t.f90:9:20: internal compiler error: verify_gimple failed
0x157e641 verify_gimple_in_cfg(function*, bool)
/home/rguenther/src/gcc2/gcc/tree-cfg.c:5467
0x15fd84b tree_function_versioning(tree_node*, tree_node*,
vec<ipa_replace_map*, va_gc, vl_embed>*, ipa_param_adjustments*, bool,
bitmap_head*, basic_block_def*)
/home/rguenther/src/gcc2/gcc/tree-inline.c:6516
0xd97875 cgraph_node::materialize_clone()
/home/rguenther/src/gcc2/gcc/cgraphclones.c:1131
0xd7cfb0 cgraph_node::get_untransformed_body()
/home/rguenther/src/gcc2/gcc/cgraph.c:3903
0x15f727c expand_call_inline
/home/rguenther/src/gcc2/gcc/tree-inline.c:4850
