question about float insns like ceil/floor on mips machine

question about float insns like ceil/floor on mips machine

[Amker.Cheng]

Hi:
  I found although there are standard pattern names such as "ceilm2/floorm2",
there is no insn pattern in mips.md for such float insns on mips target.
further more, there is no ceil/floor rtl code in rtl.def either.

based on these facts, I assuming those float insns are not supported by gcc,
but don't know why, seems not difficult to add such insns.

Did I miss anything important? please help, thanks.

-- 
Best Regards.

Re: question about float insns like ceil/floor on mips machine

[Ian Lance Taylor]

"Amker.Cheng" <amker.cheng@gmail.com> writes:

>   I found although there are standard pattern names such as "ceilm2/floorm2",
> there is no insn pattern in mips.md for such float insns on mips target.
> further more, there is no ceil/floor rtl code in rtl.def either.
>
> based on these facts, I assuming those float insns are not supported by gcc,
> but don't know why, seems not difficult to add such insns.

You're right that gcc does not currently have a way to represent those
instructions directly in RTL.  In current gcc it would have to be done
using an unspec.  E.g., see ceildf2 and ceildf2_fpr in rs6000.md.

You're also right that it would not be difficult to add direct RTL
support.  I'm not sure how useful that would be, since I don't see a lot
of additional optimization opportunity.  An unspec will handle
scheduling correctly, the only benefit I see of direct RTL exposure
would be that you could do optimizations like floor(ceil(x)) => ceil(x),
but that seems unlikely to occur in real code.

Ian

Re: question about float insns like ceil/floor on mips machine

[Richard Guenther]

On Mon, Jul 19, 2010 at 11:12 AM, Ian Lance Taylor <iant@google.com> wrote:
> "Amker.Cheng" <amker.cheng@gmail.com> writes:
>
>>   I found although there are standard pattern names such as "ceilm2/floorm2",
>> there is no insn pattern in mips.md for such float insns on mips target.
>> further more, there is no ceil/floor rtl code in rtl.def either.
>>
>> based on these facts, I assuming those float insns are not supported by gcc,
>> but don't know why, seems not difficult to add such insns.
>
> You're right that gcc does not currently have a way to represent those
> instructions directly in RTL.  In current gcc it would have to be done
> using an unspec.  E.g., see ceildf2 and ceildf2_fpr in rs6000.md.
>
> You're also right that it would not be difficult to add direct RTL
> support.  I'm not sure how useful that would be, since I don't see a lot
> of additional optimization opportunity.  An unspec will handle
> scheduling correctly, the only benefit I see of direct RTL exposure
> would be that you could do optimizations like floor(ceil(x)) => ceil(x),
> but that seems unlikely to occur in real code.

And we'd be doing that at the tree level anyway.

Richard.

> Ian
>