llvm dev - Jan 2014 - [LLVMdev] Why should we have the LoopPass and LoopPassManager? Can we get rid of this complexity?

On Wed, Jan 22, 2014 at 12:33 AM, Andrew Trick <atrick at apple.com>
wrote:
> > There appear to be two chunks of "functionality" provided by
loop passes:
> >
> > 1) A worklist of loops to process. This is very rarely used:
> > 1.1) LoopSimplify and LoopUnswitch add loops to the queue.
>
> I’m making this up without much thought, but we may benefit from iterative
> loop transforms like Rotate -> LICM -> Unswitch -> Rotate ->
LICM. We might
> need to come up with a preferred alternative: we can either continue to
> convert transforms into a utilities, or we can invent new pass manager
> tricks. e.g. iterate over a set of function passes or selectively run a
> pass on “dirty” regions.

This is a really good point. Owen pointed it out to me as well in another
guise: when we unroll a loop, we need to re-run a bunch of the passes, but
re-running them when we *haven't* successfully unrolled a loop is a total
waste.

I'd like to think more about this, so a simpler option: what if we *just*
extract LoopSimplify and LCSSA? If all the LoopPasses preserve these, then
them being function passes would be fine. This would allow us to at least
*start* writing function passes over loops (like the LoopVectorizer) rather
than being forced to phrase them as LoopPasses.

I think I could implement this simpler option right away, and that would
unblock a lot of our efforts around unrolling, vectorizing, and PGO.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140122/93f989ae/attachment.html>

On Jan 22, 2014, at 12:44 AM, Chandler Carruth <chandlerc at gmail.com>
wrote:
> 
> On Wed, Jan 22, 2014 at 12:33 AM, Andrew Trick <atrick at apple.com>
wrote:
> > There appear to be two chunks of "functionality" provided by
loop passes:
> >
> > 1) A worklist of loops to process. This is very rarely used:
> > 1.1) LoopSimplify and LoopUnswitch add loops to the queue.
> 
> I’m making this up without much thought, but we may benefit from iterative
loop transforms like Rotate -> LICM -> Unswitch -> Rotate -> LICM.
We might need to come up with a preferred alternative: we can either continue to
convert transforms into a utilities, or we can invent new pass manager tricks.
e.g. iterate over a set of function passes or selectively run a pass on “dirty”
regions.
> 
> This is a really good point. Owen pointed it out to me as well in another
guise: when we unroll a loop, we need to re-run a bunch of the passes, but
re-running them when we *haven't* successfully unrolled a loop is a total
waste.
> 
> I'd like to think more about this, so a simpler option: what if we
*just* extract LoopSimplify and LCSSA? If all the LoopPasses preserve these,
then them being function passes would be fine. This would allow us to at least
*start* writing function passes over loops (like the LoopVectorizer) rather than
being forced to phrase them as LoopPasses.
> 
> I think I could implement this simpler option right away, and that would
unblock a lot of our efforts around unrolling, vectorizing, and PGO.
Great. LoopSimplify can be used as a utility to cleanup transformed loops if
needed.

LCSSA could eventually be eliminated as prerequisites for some loop passes that
should just use the SSA updater instead.

SCEV based passes would be next in line to convert to function passes.
Reproducing and debugging issues related to SCEV invalidation and loop pass
order is terrifying.

-Andy
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140122/e672cec8/attachment.html>

On Wed, Jan 22, 2014 at 1:01 AM, Andrew Trick <atrick at apple.com> wrote:
> On Jan 22, 2014, at 12:44 AM, Chandler Carruth <chandlerc at
gmail.com>
> wrote:
>
>
> On Wed, Jan 22, 2014 at 12:33 AM, Andrew Trick <atrick at apple.com>
wrote:
>
>> > There appear to be two chunks of "functionality"
provided by loop
>> passes:
>> >
>> > 1) A worklist of loops to process. This is very rarely used:
>> > 1.1) LoopSimplify and LoopUnswitch add loops to the queue.
>>
>> I’m making this up without much thought, but we may benefit from
>> iterative loop transforms like Rotate -> LICM -> Unswitch ->
Rotate ->
>> LICM. We might need to come up with a preferred alternative: we can
either
>> continue to convert transforms into a utilities, or we can invent new
pass
>> manager tricks. e.g. iterate over a set of function passes or
selectively
>> run a pass on “dirty” regions.
>
>
> This is a really good point. Owen pointed it out to me as well in another
> guise: when we unroll a loop, we need to re-run a bunch of the passes, but
> re-running them when we *haven't* successfully unrolled a loop is a
total
> waste.
>
> I'd like to think more about this, so a simpler option: what if we
*just*
> extract LoopSimplify and LCSSA? If all the LoopPasses preserve these, then
> them being function passes would be fine. This would allow us to at least
> *start* writing function passes over loops (like the LoopVectorizer) rather
> than being forced to phrase them as LoopPasses.
>
> I think I could implement this simpler option right away, and that would
> unblock a lot of our efforts around unrolling, vectorizing, and PGO.
>
>
> Great. LoopSimplify can be used as a utility to cleanup transformed loops
> if needed.
>
This would really only work with pretty specialized logic in the
LoopVectorizer and LSR to select only the cleanup aspects of LoopSimplify
that they need. And it wouldn't even have any utility in LSR.

Unfortunately, the loop pass manager makes doing this incrementally hard.
The loop vectorizer runs on the inner most loop, makes in not-simplified.
We then pop out to the outer loop and verify everything. The
now-function-pass LoopSimplify fails its verification because it verifies
*all* of the loops in the function, even though the inner loops aren't
going to be revisited.

My suggestion is to disable the whole-function verification in LoopSimplify
while we're fixing this. We can instead have LCSSA verify the specific loop
it is processing which will give us the same safety guarantees. Does that
sound OK?

That will let me do the following:

1) Hoist LoopSimplify to a function pass
2) Hoist LCSSA to a function pass
3) Make the LoopVectorizer a function pass
4) Make IVUsers a utility of LSR, run on-demand for the loop that LSR is
processing (no pass manager involvement)
5) Make LSR a function pass

As a consequence of this, we will actually fix the long-standing issue of
LoopSimplify vs. ScalarEvolution, and cut the number of runs of
LoopSimplify and LCSSA by half (!!!) so I think this is worth doing.

I have a patch that does #1 already, but wanted to check that you're OK
weakening the verification. Otherwise, I have to do 1, 2, 3, and 5 in a
single commit, or teach the LoopVectorizer and LSR to preserve
LoopSimplify... Yuck.

-Chandler
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140122/19ec6b23/attachment.html>