llvm dev - Feb 2020 - Eliminate some two entry PHI nodes - SimplifyCFG

Conditional on the target supporting cmov? Though that's probably not
optimal.



On Wed, Feb 5, 2020, 7:47 AM Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> Hi Ryan,
>
> On Mon, Feb 3, 2020 at 7:08 PM Ryan Taylor via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
> > SimplifyCFG FoldTwoEntryPhiNode looks to simplify all 2 entry phi
nodes
> in a block, if it can't do them all then it won't do any and
returns. There
> is a lot of code that is directly in this function geared toward this
> requirement. Is it possible currently to get this function (or pass) to
> simply fold "some" of the phis (without having to fold them
all?). I
> understand that "Simplify"CFG isn't really meant for this
type of thing but
> there is a lot of code that is re-usable to do this with some refactoring
> of this function and potentially some addition of target specifics.
> >
> > For some phis, it's beneficial to keep them as the generated COPYs
can
> be coalesced away later in the pipe. For other phis (for example, phis with
> all or potentially partial imm ops), it would be nice to opt these to
> selects (cmovs). It would also be nice to have this opt prior to the rest
> of SimplifyCFG as it might enable SimplifyCFG to do a better job of
> simplifying by reducing the number of phis, etc...
>
> I don't see a significant problem in adding this to
> FoldTwoEntryPhiNodes. The relevant changes would be isolated to the
> initial loop over the phis, and would be somewhat parallel to the use
> of simplifyInstruction.
>
> The main question is to how to make this conditional on a per-target
> basis, i.e. how to properly define any relevant heuristics via
> TargetTransformInfo.
>
> Cheers,
> Nicolai
>
>
>
> --
> Lerne, wie die Welt wirklich ist,
> aber vergiss niemals, wie sie sein sollte.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20200205/3e8b500b/attachment.html>

On Wed, Feb 5, 2020 at 1:53 PM Ryan Taylor <ryta1203 at gmail.com>
wrote:
> Conditional on the target supporting cmov? Though that's probably not
optimal.
No, it's not. Consider how the condition may be represented depending
on targets (e.g., flag registers....).

Cheers,
Nicolai

>
>
>
> On Wed, Feb 5, 2020, 7:47 AM Nicolai Hähnle <nhaehnle at gmail.com>
wrote:
>>
>> Hi Ryan,
>>
>> On Mon, Feb 3, 2020 at 7:08 PM Ryan Taylor via llvm-dev
>> <llvm-dev at lists.llvm.org> wrote:
>> > SimplifyCFG FoldTwoEntryPhiNode looks to simplify all 2 entry phi
nodes in a block, if it can't do them all then it won't do any and
returns. There is a lot of code that is directly in this function geared toward
this requirement. Is it possible currently to get this function (or pass) to
simply fold "some" of the phis (without having to fold them all?). I
understand that "Simplify"CFG isn't really meant for this type of
thing but there is a lot of code that is re-usable to do this with some
refactoring of this function and potentially some addition of target specifics.
>> >
>> > For some phis, it's beneficial to keep them as the generated
COPYs can be coalesced away later in the pipe. For other phis (for example, phis
with all or potentially partial imm ops), it would be nice to opt these to
selects (cmovs). It would also be nice to have this opt prior to the rest of
SimplifyCFG as it might enable SimplifyCFG to do a better job of simplifying by
reducing the number of phis, etc...
>>
>> I don't see a significant problem in adding this to
>> FoldTwoEntryPhiNodes. The relevant changes would be isolated to the
>> initial loop over the phis, and would be somewhat parallel to the use
>> of simplifyInstruction.
>>
>> The main question is to how to make this conditional on a per-target
>> basis, i.e. how to properly define any relevant heuristics via
>> TargetTransformInfo.
>>
>> Cheers,
>> Nicolai
>>
>>
>>
>> --
>> Lerne, wie die Welt wirklich ist,
>> aber vergiss niemals, wie sie sein sollte.


-- 
Lerne, wie die Welt wirklich ist,
aber vergiss niemals, wie sie sein sollte.

Just to be clear, I meant "cmov" as a generic definition of the target
supporting some type of conditional mov, not explicitly 'cmov', so it
would
be target dependent, of course.

What I meant is that just because a target supports "cmov" it still
might
not be beneficial to the target to do the transform, and determining that
could be quite complex depending on several things (the target info, target
specific backend passes, etc..).

On Wed, Feb 5, 2020, 7:58 AM Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> On Wed, Feb 5, 2020 at 1:53 PM Ryan Taylor <ryta1203 at gmail.com>
wrote:
> > Conditional on the target supporting cmov? Though that's probably
not
> optimal.
>
> No, it's not. Consider how the condition may be represented depending
> on targets (e.g., flag registers....).
>
> Cheers,
> Nicolai
>
>
> >
> >
> >
> > On Wed, Feb 5, 2020, 7:47 AM Nicolai Hähnle <nhaehnle at
gmail.com> wrote:
> >>
> >> Hi Ryan,
> >>
> >> On Mon, Feb 3, 2020 at 7:08 PM Ryan Taylor via llvm-dev
> >> <llvm-dev at lists.llvm.org> wrote:
> >> > SimplifyCFG FoldTwoEntryPhiNode looks to simplify all 2 entry
phi
> nodes in a block, if it can't do them all then it won't do any and
returns.
> There is a lot of code that is directly in this function geared toward this
> requirement. Is it possible currently to get this function (or pass) to
> simply fold "some" of the phis (without having to fold them
all?). I
> understand that "Simplify"CFG isn't really meant for this
type of thing but
> there is a lot of code that is re-usable to do this with some refactoring
> of this function and potentially some addition of target specifics.
> >> >
> >> > For some phis, it's beneficial to keep them as the
generated COPYs
> can be coalesced away later in the pipe. For other phis (for example, phis
> with all or potentially partial imm ops), it would be nice to opt these to
> selects (cmovs). It would also be nice to have this opt prior to the rest
> of SimplifyCFG as it might enable SimplifyCFG to do a better job of
> simplifying by reducing the number of phis, etc...
> >>
> >> I don't see a significant problem in adding this to
> >> FoldTwoEntryPhiNodes. The relevant changes would be isolated to
the
> >> initial loop over the phis, and would be somewhat parallel to the
use
> >> of simplifyInstruction.
> >>
> >> The main question is to how to make this conditional on a
per-target
> >> basis, i.e. how to properly define any relevant heuristics via
> >> TargetTransformInfo.
> >>
> >> Cheers,
> >> Nicolai
> >>
> >>
> >>
> >> --
> >> Lerne, wie die Welt wirklich ist,
> >> aber vergiss niemals, wie sie sein sollte.
>
>
>
> --
> Lerne, wie die Welt wirklich ist,
> aber vergiss niemals, wie sie sein sollte.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20200205/0eeb972c/attachment.html>