[Bug tree-optimization/89209] [9 Regression] ICE in build_ref_for_model, at tree-sra.c:1791

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89209

--- Comment #5 from Martin Jambor <jamborm at gcc dot gnu.org> ---
HWell, no.  We create a special default-def SSA to insert into the IL
the information that an uninitialized value is being loaded but now
the SSA has aggregate type, which should not happen, I guess (even
though it surprisingly works).  We try to create only one such SSA
name for each part of an aggregate. A simple solution would be to
create a new one if V_C_E with an aggregate argument is on the RHS.
I'll look into the code a bit more to see if I can find other options.