> It would need strict_low_part unless we're dealing with subwords. I
> think the patch should maybe check for that in the !W_R_O case.
The code already does the check, it simply won't mess with strict_low_part.
> Does WORD_REGISTER_OPERATIONS really buy much on targets that use it?
Yes, it make
On 11/23/2016 03:32 AM, Paolo Bonzini wrote:
On 23/11/2016 11:26, Eric Botcazou wrote:
Does it really do that with a (set (reg1:QI)), as opposed to a
(set (strict_low_part (subreg:QI (reg1:DI)))?
That's the question (note that REE runs after register allocation).
IIRC, strict_low_part is r
On 11/23/2016 05:26 AM, Bernd Schmidt wrote:
On 11/23/2016 11:18 AM, Paolo Bonzini wrote:
On 23/11/2016 10:35, Eric Botcazou wrote:
I now wonder
whether this also matters for !WORD_REGISTER_OPERATIONS targets, e.g.
x86:
(set (reg1:DI) ...
...
(set (reg1:QI) (expression
On 11/23/2016 11:18 AM, Paolo Bonzini wrote:
On 23/11/2016 10:35, Eric Botcazou wrote:
I now wonder
whether this also matters for !WORD_REGISTER_OPERATIONS targets, e.g. x86:
(set (reg1:DI) ...
...
(set (reg1:QI) (expression:QI))
...
(set (reg2:DI)
> Please note that there is no guarantee that "(set (reg1:QI)
> (expression:QI))" will preserve high-order bits. This is guaranteed
> only when strict_low_part is used on the destination operand.
>
> C.f. *addqi_1 pattern, where QImode plus RTX can emit SImode ADDL insn
> vs. *addqi_1_slp, where Q
On 23/11/2016 11:26, Eric Botcazou wrote:
>> Does it really do that with a (set (reg1:QI)), as opposed to a
>> (set (strict_low_part (subreg:QI (reg1:DI)))?
>
> That's the question (note that REE runs after register allocation).
IIRC, strict_low_part is required even after register allocation,
> Does it really do that with a (set (reg1:QI)), as opposed to a
> (set (strict_low_part (subreg:QI (reg1:DI)))?
That's the question (note that REE runs after register allocation).
--
Eric Botcazou
Hello!
> I initially thought that this would only matter for WORD_REGISTER_OPERATIONS
> targets like Alpha, where the first set in QImode can implicitly set the whole
> DImode register so the use reads well-defined upper bits, but I now wonder
> whether this also matters for !WORD_REGISTER_OPERATI
On 23/11/2016 10:35, Eric Botcazou wrote:
> I now wonder
> whether this also matters for !WORD_REGISTER_OPERATIONS targets, e.g. x86:
>
> (set (reg1:DI) ...
> ...
>(set (reg1:QI) (expression:QI))
>...
>(set (reg2:DI) (any_extend:DI (reg1:QI)))
>
Hi,
this is a wrong code regression at -O2 on the mainline for Alpha coming from
the REE pass (Alpha is one of the 3 architectures enabling REE at -O2 but I'm
probably going to enable it for 64-bit SPARC too). The problem arises when a
copy is needed in combine_reaching_defs:
/* If the dest
10 matches
Mail list logo
