2008/5/29 Jeff Law <[EMAIL PROTECTED]>:
> Richard Sandiford wrote:
>>
>> Andy H <[EMAIL PROTECTED]> writes:
>>>>
>>>> If L_R_A does nothing with it,
>>>> the normal reload handling will first try:
>>>>
>>>> (const:HI (plus:HI (symbol_ref:HI ("chk_fail_buf") (const_int 2))))
>>>>
>>>
>>> This worked just as your described after I added test of
>>> reg_equiv_constant[] inside L_R_A .
>>>
>>> So I guess that looks like the fix for bug I posted.
>>>
>>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34641
>>>
>>> To summarize
>>>
>>> LEGITIMIZE_RELOAD_ADDRESS should now always check reg_equiv_constant
>>> before it trying to do any push_reload of register.
>>
>> TBH, I still think AVR is doing far too much in L_R_A. To quote
>> the current version:
>
> [...]
> Not only should there be some comments, those comments should clearly
> explain how L_R_A is improving the generated code. That is LRA's job, to
> implement target specific reload strategies which improve the generated
> code. If the AVR port is using L_R_A for *correctness*, then the AVR port
> is broken.
I wrote L_R_A for *correctness*, so in your terms AVR port is broken.
It's because AVR have only 3 pointer registers (X,Y,Z) and only 2 of
them Y and Z are base pointers.
Y is a frame pointer, so only one general base pointer - Z.
Offset also very limited -63 to +63.
In L_R_A I play with POINTER_REGS (X,Y,Z) and BASE_POINTER_REGS (Y,Z).
My experiments with reload show me that reload can't handle all
difficult situations.
Also, Jeff I know that you think that reload can.
GCC havn't something like (define_address ...) and reload can't handle
POINTER_REGS and
BASE_POINTER_REGS differently. In my version of L_R_A I tried to do it.
Denis.
