[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [avr-gcc-list] error with optimized boolean logic in gcc
From: |
Georg-Johann Lay |
Subject: |
Re: [avr-gcc-list] error with optimized boolean logic in gcc |
Date: |
Fri, 25 Nov 2011 16:02:12 +0100 |
User-agent: |
Thunderbird 2.0.0.24 (X11/20100302) |
Georg-Johann Lay wrote:
> Sean D'Epagnier wrote:
>> On Fri, Nov 25, 2011 at 11:18:17AM +0100, Georg-Johann Lay wrote:
>>>> In avr.md I changed:
>>>> return "sbi %i0,%2";
>>>> to
>>>> return "sbi %i0-0x20,%2";
>>>>
>>>> It fixed the problem. I think this needs to be done all over the place
>>> This is an incorrect fix, the problem must be somewhere else.
>>>
>> Indeed, I didn't like it much, but it helped me understand the problem.
>>
>>> %i shall subtract avr_current_arch->sfr_offset which is 0x20 for all
>>> architectures. The reason to use %i and not %m is to avoid magic
>>> numbers 0x20 all over the place, see top of following changeset and
>>> %i implementation in avr.c:print_operand()
>>>
>>> http://gcc.gnu.org/viewcvs?view=revision&revision=181552
>>>
>>> + else if (code == 'i')
>>> + {
>>> + if (!io_address_operand (addr, GET_MODE (x)))
>>> + fatal_insn ("bad address, not an I/O address:", addr);
>>> +
>>> + switch (INTVAL (addr))
>>> + {
>>> + case RAMPZ_ADDR: fprintf (file, "__RAMPZ__"); break;
>>> + case SREG_ADDR: fprintf (file, "__SREG__"); break;
>>> + case SP_ADDR: fprintf (file, "__SP_L__"); break;
>>> + case SP_ADDR+1: fprintf (file, "__SP_H__"); break;
>>> +
>>> + default:
>>> + fprintf (file, HOST_WIDE_INT_PRINT_HEX,
>>> + UINTVAL (addr) - avr_current_arch->sfr_offset);
>>> + break;
>>> + }
>>> + }
>>>
>> This makes a lot more sense now, but the above code is never reached.
>> If you go a bit higher up in the function you will see:
>>
>> else if (GET_CODE (x) == CONST_INT)
>> -> fprintf (file, HOST_WIDE_INT_PRINT_DEC, INTVAL (x) + abcd);
>> else if (GET_CODE (x) == MEM)
>> {
>> ...
>> bit of code from above
>> ...
>> }
>>
>> code is indeed 'i', but GET_CODE(x) is CONST_INT not MEM, so the conversion
>> never takes place.
>>
>> We could of course perform the conversion in both places, but I don't
>> really like that either. Maybe perform the subraction earlier as for abcd?
>>
>> Sean
>
> Please f'up
>
> http://gcc.gnu.org/ml/gcc-patches/2011-11/msg02400.html
It's upstream in 181722:
http://gcc.gnu.org/viewcvs?view=revision&revision=181722
Johann