On Thu, Mar 24, 2022 at 2:57 PM Willy Tarreau <[email protected]> wrote:
> On Thu, Mar 24, 2022 at 02:30:32PM +0700, Ammar Faizi wrote:
> > Use appropriate register constraints if exist. Don't use register
> > variables for all inputs.
> >
> > Register variables with "r" constraint should be used when we need to
> > pass data through a specific register to extended inline assembly that
> > doesn't have a specific register constraint associated with it (anything
> > outside %rax, %rbx, %rcx, %rdx, %rsi, %rdi).
> >
> > It also simplifies the macro definition.
>
> I'm a bit bothered by this one because I went the exact opposite route
> in the early design precisely because I found that the current one was
> simpler. [...]
[...]
> I'd say that if there is any technical benefit in doing this (occasional
> code improvement or better support for older or exotic compilers), I'd say
> "ok go for it", but if it's only a matter of taste, I'm not convinced at
> all and am rather seeing this as a regression. Now if there's rough
> consensus around this approach I'll abide, but then I'd request that other
> archs are adapted as well so that we don't keep a different approach only
> for these two ones.
I don't see any technical benefit for x86-64, so I don't think there
is a need in doing this. Though I personally prefer to use register
constraints if they exist instead of register variables for everything
(oh yeah, matter of taste since I don't have any technical argument to
say it's better respecting the resulting codegen). The only real issue
is for the syscall6() implementation on i386 as we've been bitten by a
real compiler issue. In short, I am neutral on this change.
Regards~~
-- Viro
On Thu, Mar 24, 2022 at 03:33:57PM +0700, Alviro Iskandar Setiawan wrote:
> On Thu, Mar 24, 2022 at 2:57 PM Willy Tarreau <[email protected]> wrote:
> > On Thu, Mar 24, 2022 at 02:30:32PM +0700, Ammar Faizi wrote:
> > > Use appropriate register constraints if exist. Don't use register
> > > variables for all inputs.
> > >
> > > Register variables with "r" constraint should be used when we need to
> > > pass data through a specific register to extended inline assembly that
> > > doesn't have a specific register constraint associated with it (anything
> > > outside %rax, %rbx, %rcx, %rdx, %rsi, %rdi).
> > >
> > > It also simplifies the macro definition.
> >
> > I'm a bit bothered by this one because I went the exact opposite route
> > in the early design precisely because I found that the current one was
> > simpler. [...]
> [...]
> > I'd say that if there is any technical benefit in doing this (occasional
> > code improvement or better support for older or exotic compilers), I'd say
> > "ok go for it", but if it's only a matter of taste, I'm not convinced at
> > all and am rather seeing this as a regression. Now if there's rough
> > consensus around this approach I'll abide, but then I'd request that other
> > archs are adapted as well so that we don't keep a different approach only
> > for these two ones.
>
> I don't see any technical benefit for x86-64, so I don't think there
> is a need in doing this. Though I personally prefer to use register
> constraints if they exist instead of register variables for everything
> (oh yeah, matter of taste since I don't have any technical argument to
> say it's better respecting the resulting codegen). The only real issue
> is for the syscall6() implementation on i386 as we've been bitten by a
> real compiler issue. In short, I am neutral on this change.
Just to be clear, I usually only use register constraints as well but I
changed this for the syscalls since they were not sufficient, and found
that the mix of the two was really not great to deal with.
Thanks,
Willy
On 3/24/22 3:33 PM, Alviro Iskandar Setiawan wrote:
> On Thu, Mar 24, 2022 at 2:57 PM Willy Tarreau <[email protected]> wrote:
>> On Thu, Mar 24, 2022 at 02:30:32PM +0700, Ammar Faizi wrote:
>>> Use appropriate register constraints if exist. Don't use register
>>> variables for all inputs.
>>>
>>> Register variables with "r" constraint should be used when we need to
>>> pass data through a specific register to extended inline assembly that
>>> doesn't have a specific register constraint associated with it (anything
>>> outside %rax, %rbx, %rcx, %rdx, %rsi, %rdi).
>>>
>>> It also simplifies the macro definition.
>>
>> I'm a bit bothered by this one because I went the exact opposite route
>> in the early design precisely because I found that the current one was
>> simpler. [...]
> [...]
>> I'd say that if there is any technical benefit in doing this (occasional
>> code improvement or better support for older or exotic compilers), I'd say
>> "ok go for it", but if it's only a matter of taste, I'm not convinced at
>> all and am rather seeing this as a regression. Now if there's rough
>> consensus around this approach I'll abide, but then I'd request that other
>> archs are adapted as well so that we don't keep a different approach only
>> for these two ones.
>
> I don't see any technical benefit for x86-64, so I don't think there
> is a need in doing this. Though I personally prefer to use register
> constraints if they exist instead of register variables for everything
> (oh yeah, matter of taste since I don't have any technical argument to
> say it's better respecting the resulting codegen). The only real issue
> is for the syscall6() implementation on i386 as we've been bitten by a
> real compiler issue. In short, I am neutral on this change.
OK then, I will drop this patch in the next version. I agree that it
doesn't really show any technical benefit and there is no danger in
doing the current implementation.
And yes, the syscall6() for i386 is somewhat problematic and we've a
confirmed bug that lives in many versions of GCC and it's not even fixed
in the current trunk. It's proven that using register constraints can
be a valid workaround to deal with this bug.
2022-03-23 13:50:18 UTC, Jakub Jelinek wrote:
> Anyway, with the "b" etc. constraints (which is a good idea to use on
> x86 when it has single register constraints for those but can't be used
> on other arches which do not have such constraints) you just trigger
> slightly different path in the RA, [...]
See the discussion here:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105032#c7
^^^ That is only for syscall6() on i386.
As such, I will drop this patch and another one that does this on i386.
Thanks!
--
Ammar Faizi