> I have backported it onto the 8 branch, where it fixes the failure (timeout)
> of gnat.dg/opt73.adb for PowerPC/Linux, after testing it on this platform.
And now onto the 7 branch, after testing on x86-64/Linux and PowerPC/Linux.
--
Eric Botcazou
> * rtlanal.c (get_initial_register_offset): Fall back to the raw estimate
> as long as the epilogue isn't completed.
I have backported it onto the 8 branch, where it fixes the failure (timeout)
of gnat.dg/opt73.adb for PowerPC/Linux, after testing it on this platform.
--
Eric
> Doesn't the frame have to be laid out correctly during the final reload
> subpass anyway, in order to get the right elimination choices and
> elimination offsets? If so, I'm not sure what the " &&
> info->reload_completed" check in rs6000_stack_info achieves. It seems like
> that's at least
Eric Botcazou writes:
> Hi,
>
> this is a regression present on all active branches since the controversial
> get_initial_register_offset stuff was added to rtlanal.c some time ago, and
> visible in the testsuite on PowerPC/Linux under the form of gnat.dg/opt73.adb
> timing out at run time.
>
> So, barring the removal of the get_initial_register_offset stuff, the only
> simple fix is probably to prevent it from calling into the back-end too
> early, for example with the attached fixlet. Tested on x86-64 and
> PowerPC/Linux.
>
> Thoughts? Where do we want to fix this?
>
>
> *
Hi,
this is a regression present on all active branches since the controversial
get_initial_register_offset stuff was added to rtlanal.c some time ago, and
visible in the testsuite on PowerPC/Linux under the form of gnat.dg/opt73.adb
timing out at run time.
The problem is that the compiler
