This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [Patch,AVR,4.6]: Fix PR50063 GCC does not support FP = SP
Richard Guenther wrote:
> On Wed, Feb 22, 2012 at 11:39 AM, Georg-Johann Lay <avr@gjlay.de> wrote:
>> Denis Chertykov wrote:
>>> 2012/2/21 Richard Henderson:
>>>> On 02/21/12 09:08, Georg-Johann Lay wrote:
>>>>> PR rtl-optimization/50063
>>>>> * config/avr/avr.md (movhi_sp_r): Handle -1 (unknown IRQ state)
>>>>> and 2 (8-bit SP) in operand 2.
>>>>> * config/avr/avr.c (avr_prologue_setup_frame): Adjust prologue
>>>>> setup to use movhi_sp_r instead of vanilla move to write SP.
>>>>> Adjust REG_CFA notes to superseed unspec.
>>>>> (expand_epilogue): Adjust epilogue setup to use read_sp instead
>>>>> of vanilla move.
>>>>> As function body might contain CLI or SEI: Use irq_state 0 (IRQ
>>>>> known to be off) only with TARGET_NO_INTERRUPTS. Never use
>>>>> irq_state 1 (IRQ known to be on) here.
>>>> The CFA bits in avr_prologue_setup_frame look good.
>>>> I'll let Denis or Eric review the movhi_sp_r change.
>>> Approved.
>>>
>>> Denis.
>> Here is a patchlet for 4.6. It just sets -fno-dse to work around the problem
>> because I think back-porting all what's needed is too much change.
>>
>> Compiling avr-libc without DSE gives the same sizes for all objects which shows
>> that this is not a crucial optimization for avr.
>>
>> Ok for 4.6 branch?
>
> I don't think this kind of fixes are wanted. The patch misses a
> testcase to backport
> and the issue surely not only affects DSE.
>
> Richard.
Actually, hacking the backend to work around assumptions in rtl-optimizers by
hiding information in unspec_volatile is not wanted, too.
If I understand correctly, no one is inclined to fix the root cause of the bug,
i.e. don't let RTL optimizers make assumptions that shred some backends.
Johann