https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64971
Jeffrey A. Law <law at redhat dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed|2015-02-09 00:00:00 |2016-03-23
CC| |rth at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #6 from Jeffrey A. Law <law at redhat dot com> ---
I've managed to reproduce this with a native aarch64 toolchain, but haven't
been able to trigger it with a cross.
The SImode address comes from expand_expr_addr_expr
Essentially address_mode is DImode and pointer_mode is SImode. tmode starts
off as VOIDmode, but is changed to SImode here:
7776 /* Target mode of VOIDmode says "whatever's natural". */
7777 if (tmode == VOIDmode)
7778 tmode = TYPE_MODE (TREE_TYPE (exp));
We get a DImode SYMBOL_REF back from expand_expr_addr_expr_1:
(symbol_ref/f:DI ("*.LC0") [flags 0x82] <var_decl 0x3ffb7fa1200 *.LC0>)
But we then try to convert that to tmode via:
7802 if (rmode != tmode)
7803 result = convert_memory_address_addr_space (tmode, result, as);
(symbol_ref/f:SI ("*.LC0") [flags 0x82] <var_decl 0x3ffb7fa1200 *.LC0>)
Which can't be recognized, as seen in c#0 -- because we expect the SYMBOL_REF
to have DImode.
Richard, want to take a peek? I've got an aarch64 native system in beaker.
Happy to pass along the hostname and passwd..
