[committed] CRIS: Fix PR middle-end/113109; "throw" failing
Checks
Commit Message
No test-case, but the regress-367 from r14-6674-g4759383245ac97 is
"back" to regress-10 for cris-elf+cris-sim with this patch applied
to gcc from that revision.
Also, I wonder why none of those other targets with a MEM for
EH_RETURN_HANDLER_RTX with an address relative to
frame_pointer_rtx (as opposed to hard_frame_pointer_rtx or
virtual_incoming_args_rtx) don't see the same problem.
Oh well. Merry Xmas.
brgds, H-P
-- >8 --
TL;DR: the "dse1" pass removed the eh-return-address store. The
PA also marks its EH_RETURN_HANDLER_RTX as volatile, for the same
reason, as does visum. See PR32769 - it's the same thing on PA.
Conceptually, it's logical that stores to incoming args are
optimized out on the return path or if no loads are seen -
at least before epilogue expansion, when the subsequent load
isn't seen in the RTL, as is the case for the "dse1" pass.
I haven't looked into why this problem, that appeared for the PA
already in 2007, was seen for CRIS only recently (with
r14-6674-g4759383245ac97).
PR middle-end/113109
* config/cris/cris.cc (cris_eh_return_handler_rtx): New function.
* config/cris/cris-protos.h (cris_eh_return_handler_rtx): Prototype.
* config/cris/cris.h (EH_RETURN_HANDLER_RTX): Redefine to call
cris_eh_return_handler_rtx.
---
gcc/config/cris/cris-protos.h | 1 +
gcc/config/cris/cris.cc | 16 ++++++++++++++++
gcc/config/cris/cris.h | 3 +--
3 files changed, 18 insertions(+), 2 deletions(-)
@@ -28,6 +28,7 @@ extern bool cris_reload_address_legitimized (rtx, machine_mode, int, int, int);
extern int cris_side_effect_mode_ok (enum rtx_code, rtx *, int, int,
int, int, int);
extern rtx cris_return_addr_rtx (int, rtx);
+extern rtx cris_eh_return_handler_rtx ();
extern rtx cris_split_movdx (rtx *);
extern bool cris_base_p (const_rtx, bool);
extern bool cris_base_or_autoincr_p (const_rtx, bool);
@@ -1382,6 +1382,22 @@ cris_return_addr_rtx (int count, rtx frameaddr ATTRIBUTE_UNUSED)
: NULL_RTX;
}
+/* Setting the EH return return address is done by a *store* to a memory
+ address expressed as relative to "*incoming* args". That store will
+ be optimized away, unless the MEM is marked as volatile. N.B.: no
+ optimization opportunities are expected to be lost due to this hack;
+ __builtin_eh_return isn't called from elsewhere than the EH machinery
+ in libgcc. */
+
+rtx
+cris_eh_return_handler_rtx ()
+{
+ rtx ret = cris_return_addr_rtx (0, NULL_RTX);
+ gcc_assert (MEM_P (ret));
+ MEM_VOLATILE_P (ret) = true;
+ return ret;
+}
+
/* Accessor used in cris.md:return because cfun->machine isn't available
there. */
@@ -551,8 +551,7 @@ enum reg_class
#define CRIS_STACKADJ_REG CRIS_STRUCT_VALUE_REGNUM
#define EH_RETURN_STACKADJ_RTX gen_rtx_REG (SImode, CRIS_STACKADJ_REG)
-#define EH_RETURN_HANDLER_RTX \
- cris_return_addr_rtx (0, NULL)
+#define EH_RETURN_HANDLER_RTX cris_eh_return_handler_rtx ()
#define INIT_EXPANDERS cris_init_expanders ()