Re: bug or feature? CSE optimizes away global, fixed register

[Andrew Haley <aph at redhat dot com>]

Georg-Johann Lay wrote:
> Andrew Haley schrieb:
>> Georg-Johann Lay wrote:
>>> David Hagood schrieb:
>>>> On Sun, 2009-06-07 at 12:00 +0200, Georg-Johann Lay wrote:
>>>>> in the test cases
>>>>>
>>>>> register unsigned char reg2 asm ("r2");
>>>>> As gcc 3.4.6 behaves as expected (e.g. in bar, the access to reg2
>>>>> remains in the loop), I wonder if this aggressive optimization is a
>>>>> bug?
>>>>> IMO that is a bug because reg2 is a fixed register and should
>>>>> therefore be treated just like volatile. Unfortunately, gcc has
>>>>> nothing like REG_VOLATILE_P to tag a reg being volatile, but as I
>>>>> said global fixed registers should be treated that way.
>>>> Try adding a volatile to the variable:
>>>>
>>>> register volatile unsigned char reg2 asm ("r2");
>>> That has no effect. Note that register is not a mem. Older versions of
>>> gcc reported a warning on volatile register like "effect is not what you
>>> might expect". Moreover, gcc cannot track volatile registers.
>>> rtl.h reads:
>>>
>>>   /* 1 in a MEM or ASM_OPERANDS expression if the memory reference is
>>> volatile.
>>>      1 in an INSN, CALL_INSN, JUMP_INSN, CODE_LABEL, BARRIER, or NOTE
>>>      if it has been deleted.
>>>      1 in a REG expression if corresponds to a variable declared by the
>>> user,
>>>      0 for an internally generated temporary.
>>>      1 in a SUBREG with a negative value.
>>>      1 in a LABEL_REF, REG_LABEL_TARGET or REG_LABEL_OPERAND note for a
>>>      non-local label.
>>>      In a SYMBOL_REF, this flag is used for machine-specific
>>> purposes.  */
>>>   unsigned int volatil : 1;
>>>
>>> What gcc knows is that some registers are or may be fixed, so it should
>>> care about that.
>>
>> Surely if you're using a global register as a mailbox, all you have to do
>> is read/write it with volatile asms.
> 
> That does not work. The generated code is /exactly/ the same.

Can you show a test case that use volatile asms, please?  I am very
surprised that does not work.

> I did not write "register volatile" in my original post because volatile
> is just syntactic sugar together with register. It is confusing and
> misleading to use volatile together with register because gcc ignores it
> and does not even report a warning.

Indeed not, no.  "register volatile" makes no sense.  I don't see how
volatile asm statements would fail, though.

Andrew.