[LLVMdev] RFC: Loop distribution/Partial vectorization

[Adam Nemet]

Hi,

We'd like to propose new Loop Distribution pass.  The main motivation is to

allow partial vectorization of loops.  One such example is the main loop of

456.hmmer in SpecINT_2006.  The current version of the patch improves hmmer by

24% on ARM64 and 18% on X86.

The goal of the pass is to distribute a loop that can't be vectorized because of

memory dependence cycles.  The pass splits the part with cycles into a new loop

making the remainder of the loop a candidate for vectorization.  E.g.:

        for (k = 0; k < M; k++) {

         S1:   MC[k+1] = …

         // Cycle in S2 due to DC[k+1] -> DC[k] loop-carried dependence

         S2:   DC[k+1] = DC[k], MC[k]                                      

        }

  

      => (Loop Distribute)

  

        for (k = 0; k < M; k++) {

         S1:   MC[k+1] = ...

        }

        for (k = 0; k < M; k++) {

         S2:   DC[k+1] = DC[k], MC[k]

        }

  

      => (Loop Vectorize S1)

  

        for (k = 0; k < M; k += 4) {

         S1:   MC[k+1:k+5] = ...

        }

        for (k = 0; k < M; k++) {

         S2:   DC[k+1] = DC[k], MC[k]

        }

I'd like to collect feedback on the design decisions made so far.  These are

implemented in a proof-of-concept patch (http://reviews.llvm.org/D6930).

Here is the list of design choices:

- Loop Distribution is implemented as a separate pass to be run before the Loop

  Vectorizer.

- The pass reuses the Memory Dependence Checker framework from the Loop

  Vectorizer.  This along with the AccessAnalysis class is split out into a new

  LoopAccessAnalysis class.  We may want to turn this into an analysis pass on its own.

- It also reuses the Run-time Memory Check code from the Loop Vectorizer.  The

  hmmer loop requires memchecks.  This is again captured by the same

  LoopAccessAnalysis class.

- The actual loop distribution is implemented as follows:

  - The list of unsafe memory dependencies is computed for the loop.  Unsafe

    means that the dependence may be part of a cycle (this is what the current

    framework provides).

  - Partitions are created for each set of unsafe dependences.

  - Partitions are created for each of the remaining stores not yet encountered.

    The order of the partitions preserve the original order of the dependent

    memory accesses.

  - Simple partition merging is performed to minimize the number of new loops.

  - Partitions are populated with the other dependent instructions by following

    the SSA use-def chains and control dependence edges.

  - Finally, the actual distribution is performed by creating a loop for each

    partition.  For each partition we clone the loop and remove all the

    instructions that don't belong to the partition.

  - Also, if run-time memory checks are necessary, these are emitted.  We keep

    an original version of the loop around to branch too if the checks fail.

My plan is to proceed with the following steps:

- Bring the current functionality to trunk by splitting off smaller patches from

  the current patch and completing them.  The final commit will enable loop

  distribution with a command-line flag or a loop hint.

- Explore and fine-tune the proper cost model for loop distribution to allow

  partial vectorization.  This is essentially whether to partition and what

  these partitions should be.  Currently instructions are mapped to partitions

  using a simple heuristics to create a vectorizable partitions.  We may need to

  interact with the vectorizer to make sure the vectorization will actually

  happen and it will be overall profitable.

- Explore other potentials for loop distribution, e.g.:

  - Partial vectorization of loops that can't be if-converted

  - Classic loop distribution to improve spatial locality

  - Compute the Program Dependence Graph rather than the list of unsafe memory

    accesses and allow reordering of memory operations

  - Distribute a loop in order to recognize parts as loop idioms

    Long term, loop distribution could also become a transformation utility

    (Transform/Util).  That way, the loop transformation passes could use it to

    strip the loop from parts that inhibits the given optimization.

Please let me know if you have feedback either on the design or on the next

steps.

Thanks,

Adam

[Krzysztof Parzyszek]

Thanks for doing this. I really like the idea of having loop
distribution as a separate pass (and having dependence analysis as an
analysis pass).

A couple of comments that I have are below.

1. Handle situations like this:

for (k = 0; k < M; k++) {

for (i = 0; i < N; ++i) {
S1: MC[i][k+1] = …

}
S2: DC[k+1] = DC[k], MC[…][k]
}

Basically, recognize and handle dependencies between differently nested
expressions.


2. Make it general so that it can serve any purpose (not only
vectorization). Various targets may want to do different things and
distribute loops for various reasons that don't apply universally.


-Krzysztof

> _______________________________________________
> LLVM Developers mailing list
> LLV...@cs.uiuc.edu http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>


--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation

_______________________________________________
LLVM Developers mailing list
LLV...@cs.uiuc.edu http://llvm.cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

[Das, Dibyendu]

As far as 456.hmmer is concerned it may gain a bit more ( at least on x86) by peeling off the last part of the loop. This will (afaik) create three vectorizable loops instead of 2 after loop distribution.

-----Original Message-----
From: llvmdev...@cs.uiuc.edu [mailto:llvmdev...@cs.uiuc.edu] On Behalf Of Krzysztof Parzyszek
Sent: Tuesday, January 13, 2015 9:57 PM
To: llv...@cs.uiuc.edu
Subject: Re: [LLVMdev] RFC: Loop distribution/Partial vectorization

Thanks for doing this. I really like the idea of having loop distribution as a separate pass (and having dependence analysis as an analysis pass).

A couple of comments that I have are below.

1. Handle situations like this:

for (k = 0; k < M; k++) {
for (i = 0; i < N; ++i) {

S1: MC[i][k+1] = ...
}
S2: DC[k+1] = DC[k], MC[...][k]

}

Basically, recognize and handle dependencies between differently nested expressions.


2. Make it general so that it can serve any purpose (not only
vectorization). Various targets may want to do different things and
distribute loops for various reasons that don't apply universally.


-Krzysztof



On 1/12/2015 12:42 PM, Adam Nemet wrote:
> Hi,
>
> We'd like to propose new Loop Distribution pass. The main motivation is to
> allow partial vectorization of loops. One such example is the main loop of
> 456.hmmer in SpecINT_2006. The current version of the patch improves
> hmmer by
> 24% on ARM64 and 18% on X86.
>
> The goal of the pass is to distribute a loop that can't be vectorized
> because of
> memory dependence cycles. The pass splits the part with cycles into a
> new loop
> making the remainder of the loop a candidate for vectorization. E.g.:
>
> for (k = 0; k < M; k++) {

> S1: MC[k+1] = ...

[Adam Nemet]

> On Jan 13, 2015, at 8:27 AM, Krzysztof Parzyszek <kpar...@codeaurora.org> wrote:
>
> Thanks for doing this. I really like the idea of having loop distribution as a separate pass (and having dependence analysis as an analysis pass).

Great, thanks for your comments.

> A couple of comments that I have are below.
>
> 1. Handle situations like this:
>
> for (k = 0; k < M; k++) {
> for (i = 0; i < N; ++i) {
> S1: MC[i][k+1] = …
> }
> S2: DC[k+1] = DC[k], MC[…][k]
> }
>
> Basically, recognize and handle dependencies between differently nested expressions.

This would take some improvements to the current memory dependence checker which handles a single loop right now. Do you have some standard algorithms in mind that would benefit from this?

> 2. Make it general so that it can serve any purpose (not only vectorization). Various targets may want to do different things and distribute loops for various reasons that don't apply universally.

Agreed. As I alluded to in the original post there could be multiple optimization benefiting from this transformation. I’ll make sure that other heuristics are pluggable besides partial vectorization.

Adam

[Adam Nemet]

> On Jan 13, 2015, at 9:29 AM, Das, Dibyendu <Dibyen...@amd.com> wrote:
>
> As far as 456.hmmer is concerned it may gain a bit more ( at least on x86) by peeling off the last part of the loop. This will (afaik) create three vectorizable loops instead of 2 after loop distribution.

Agreed, that is the next step for hmmer. (I think you must mean two vectorizable loops out of the three distributed loops.)

My hope is that with some tuning, this pass and Sanjoy’s new range-check elimination pass should take us there.

Adam

[Philip Reames]

I like the general direction.  One potential concern I have is regards to distributing a loop which we turn out not to vectorize and potentially creating larger code for no clear benefit.  We'll have to see how this works in practice. 

My plan is to proceed with the following steps:

- Bring the current functionality to trunk by splitting off smaller patches from

  the current patch and completing them.  The final commit will enable loop

  distribution with a command-line flag or a loop hint.

I look forward to seeing your patches.  Getting this in incrementally will take some work on all sides, but is definitely better than trying to land one large patch.

- Explore and fine-tune the proper cost model for loop distribution to allow

  partial vectorization.  This is essentially whether to partition and what

  these partitions should be.  Currently instructions are mapped to partitions

  using a simple heuristics to create a vectorizable partitions.  We may need to

  interact with the vectorizer to make sure the vectorization will actually

  happen and it will be overall profitable.

As I said above, this is my biggest area of concern.  It'll be interesting to see where you end up.

- Explore other potentials for loop distribution, e.g.:

  - Partial vectorization of loops that can't be if-converted

  - Classic loop distribution to improve spatial locality

  - Compute the Program Dependence Graph rather than the list of unsafe memory

    accesses and allow reordering of memory operations

  - Distribute a loop in order to recognize parts as loop idioms

    Long term, loop distribution could also become a transformation utility

    (Transform/Util).  That way, the loop transformation passes could use it to

    strip the loop from parts that inhibits the given optimization.

Please let me know if you have feedback either on the design or on the next

steps.

Thanks,

Adam

[Krzysztof Parzyszek]

On 1/13/2015 1:05 PM, Adam Nemet wrote:
>
> This would take some improvements to the current memory dependence checker which handles a single loop right now. Do you have some standard algorithms in mind that would benefit from this?

The first candidate would be the loop distribution itself---it would be
able to handle more cases, even if its purpose is vectorization.

Overall, any loop nest transformation would be a consumer of such
information, but that probably wouldn't get past polly advocates. :)

If I come up with another motivating example other than the above, I'll
let you know.

-Krzysztof

[Hal Finkel]

----- Original Message -----
> From: "Adam Nemet" <ane...@apple.com>
> To: "LLVM Developers Mailing List" <llv...@cs.uiuc.edu>
> Sent: Monday, January 12, 2015 12:42:36 PM
> Subject: [LLVMdev] RFC: Loop distribution/Partial vectorization
>
>
>
> Hi,
>
>
> We'd like to propose new Loop Distribution pass. The main motivation
> is to
> allow partial vectorization of loops. One such example is the main
> loop of
> 456.hmmer in SpecINT_2006. The current version of the patch improves
> hmmer by
> 24% on ARM64 and 18% on X86.

Thanks for working on this! We definitely need this capability in LLVM (and for more than just enabling vectorization).

This is good. It would be nice if this new analysis could have the same interface as the DependenceAnalysis pass, so that it will be easy to switch between them. I think that, eventually, we'll want to switch everything to use something like DependenceAnalysis, at least at higher optimization levels.

>
>
> - It also reuses the Run-time Memory Check code from the Loop
> Vectorizer. The
> hmmer loop requires memchecks. This is again captured by the same
> LoopAccessAnalysis class.

I think this is also reasonable; we just want to make sure that we don't end up with double memory checks. I've seen cases in the past where the vectorizer has inserted checks that should have been eliminated as duplicates with other loop guards, SE guard domination checking may need to be improved.

This sounds reasonable.

>
>
> My plan is to proceed with the following steps:
>
>
> - Bring the current functionality to trunk by splitting off smaller
> patches from
> the current patch and completing them. The final commit will enable
> loop
> distribution with a command-line flag or a loop hint.

Okay, please do.

>
>
> - Explore and fine-tune the proper cost model for loop distribution
> to allow
> partial vectorization. This is essentially whether to partition and
> what
> these partitions should be. Currently instructions are mapped to
> partitions
> using a simple heuristics to create a vectorizable partitions. We may
> need to
> interact with the vectorizer to make sure the vectorization will
> actually
> happen and it will be overall profitable.

I think this sounds reasonable. Splitting to enable vectorization is important; one reason to have this process tightly integrated with vectorization is so that it can properly integrate with the vectorizers register pressure checking (we might split to reduce register pressure, thus enabling more interleaving, at least when doing so does not decrease spatial locality).

Independent of vectorization, loop splitting is important to reduce register pressure within loops (i.e. loops with too many phis, but that are splittable, could be split to prevent intra-iteration spilling). Also very important is splitting to reduce the number of hardware prefetching streams used by the loop. In every system on which I've worked, the hardware prefetchers have a finite set of resources to sustain prefetching streams (5-10 per thread, depending on the architecture). When a loop would require more streams than this then performance will greatly suffer, and splitting it highly profitable. I'd definitely like us to hit these two use cases too.

>
>
> - Explore other potentials for loop distribution, e.g.:
> - Partial vectorization of loops that can't be if-converted
> - Classic loop distribution to improve spatial locality
> - Compute the Program Dependence Graph rather than the list of unsafe
> memory
> accesses and allow reordering of memory operations

This would also be quite nice to have.

> - Distribute a loop in order to recognize parts as loop idioms

Indeed, once you have the partitions, splitting out a memcpy, etc. should not be hard; this is not always profitable, however.

>
>
> Long term, loop distribution could also become a transformation
> utility
> (Transform/Util). That way, the loop transformation passes could use
> it to
> strip the loop from parts that inhibits the given optimization.

This sounds good, but we still may want to schedule the transformation itself (late in the pipeline). We don't want to limit register-pressure-induced loop splitting, for example, to vectorizable loops.

Thanks again,
Hal

>
>
> Please let me know if you have feedback either on the design or on
> the next
> steps.
>
>
> Thanks,
> Adam
>
>

> _______________________________________________
> LLVM Developers mailing list
> LLV...@cs.uiuc.edu http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>

--
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory

[Adam Nemet]

Yes, that is precisely what Arnold and I discussed a few weeks ago.  We want to reuse something that's known to work initially to get us off the ground but then we want to be to swap in the DependenceAnalysis.  The idea was exactly as you describe it: to try to change the interface of the Memory Dependence Checker to match the interface of the DependenceAnalysis pass.

- It also reuses the Run-time Memory Check code from the Loop
Vectorizer. The
hmmer loop requires memchecks. This is again captured by the same
LoopAccessAnalysis class.

I think this is also reasonable; we just want to make sure that we don't end up with double memory checks. I've seen cases in the past where the vectorizer has inserted checks that should have been eliminated as duplicates with other loop guards, SE guard domination checking may need to be improved.

Yes, this was also on my list.  (Sorry.  I didn’t include everything in the original post because it would have been way too long).

I didn’t know we already have code that tries to deal with this.  Can you please point me to it?

OK, I haven’t thought of splitting due to register pressure.  I guess this makes sense both in vectorizable and non-vectorizable loops.

Do you have an example for the part in parentheses?  Do you mean that spatial locality would be decreased by interleaving?

Independent of vectorization, loop splitting is important to reduce register pressure within loops (i.e. loops with too many phis, but that are splittable, could be split to prevent intra-iteration spilling). Also very important is splitting to reduce the number of hardware prefetching streams used by the loop. In every system on which I've worked, the hardware prefetchers have a finite set of resources to sustain prefetching streams (5-10 per thread, depending on the architecture). When a loop would require more streams than this then performance will greatly suffer, and splitting it highly profitable. I'd definitely like us to hit these two use cases too.

Sure.  I think that this is essentially what I meant by loop distribution to improve spatial locality.   Exposing the target’s parameters for the HW prefetcher sounds like a nice way to model this.

- Explore other potentials for loop distribution, e.g.:
- Partial vectorization of loops that can't be if-converted
- Classic loop distribution to improve spatial locality
- Compute the Program Dependence Graph rather than the list of unsafe
memory
accesses and allow reordering of memory operations

This would also be quite nice to have.

- Distribute a loop in order to recognize parts as loop idioms

Indeed, once you have the partitions, splitting out a memcpy, etc. should not be hard; this is not always profitable, however.

Long term, loop distribution could also become a transformation
utility
(Transform/Util). That way, the loop transformation passes could use
it to
strip the loop from parts that inhibits the given optimization.

This sounds good, but we still may want to schedule the transformation itself (late in the pipeline). We don't want to limit register-pressure-induced loop splitting, for example, to vectorizable loops.

Sure.  I meant that there would still be a “stand-alone” loop distribution pass which would be another user of this transformation utility.

Thanks very much for your feedback, Hal!

Adam

[Renato Golin]

On 12 January 2015 at 18:42, Adam Nemet <ane...@apple.com> wrote:
> - Loop Distribution is implemented as a separate pass to be run before the
> Loop Vectorizer.

Agree, this would be the best place to put it. I'm wondering whether
it would be valid (or desirable) to put any metadata stopping the
vectorizers from looking at the loops that you have marked as
non-vectorizable.

cheers,
--renato

[Adam Nemet]

> On Jan 26, 2015, at 3:26 AM, Renato Golin <renato...@linaro.org> wrote:
>
> On 12 January 2015 at 18:42, Adam Nemet <ane...@apple.com> wrote:
>> - Loop Distribution is implemented as a separate pass to be run before the
>> Loop Vectorizer.
>
> Agree, this would be the best place to put it. I'm wondering whether
> it would be valid (or desirable) to put any metadata stopping the
> vectorizers from looking at the loops that you have marked as
> non-vectorizable.

This sounds like a good idea, thank you. I’ll experiment with it when I get this far.

Adam

[Hal Finkel]

We don't exactly. JumpThreading could handle the constant-range cases, but that also leaves a lot on the table (and we don't run it after loop vectorization regardless). What we do have is SE's isLoopEntryGuardedByCond, which I thought was used by the loop vectorizer to avoid adding unnecessary guards, but I don't see that now, so I might be wrong.

Not off hand. The interleaving factor is limited by the number of available registers. I recall running across loops that have a lot of phi values, and those limit the number of registers available for the intermediate values required by the interleaving. If the loop can be split, reducing the number of registers needed for phis, that can increase the number of interleavings possible.

Thanks again,
Hal