Attached is Kaz's patch from the PR.
Tested against rev 185555 with
make -k check RUNTESTFLAGS="--target_board=sh-sim
It fixes a few "sometimes failing" libstdc++ failures.
No new failures otherwise.
OK to apply?
Maybe this one should be backported to 4.7.x, too?
* config/sh/sh.c (sh_expand_prologue): Emit blockage at the end
of prologue for unwinder and profiler.
--- gcc/config/sh/sh.c (revision 185554)
+++ gcc/config/sh/sh.c (working copy)
@@ -7239,6 +7239,13 @@
emit_insn (gen_shcompact_incoming_args ());
+ /* If we are profiling, make sure no instructions are scheduled before
+ the call to mcount. Similarly if some call instructions are swapped
+ before frame related insns, it'll confuse the unwinder because
+ currently SH has no unwind info for function epilogues. */
+ if (crtl->profile || flag_exceptions || flag_unwind_tables)
+ emit_insn (gen_blockage ());
current_function_static_stack_size = stack_usage;