Sjoerd Meijer via llvm-dev
llvm-dev at lists.llvm.org
Mon Oct 7 02:08:49 PDT 2019
> The problem I see is that the warning isn't very actionable.
> Good warnings are supposed to be actionable, but what is the developer supposed to do in this case?
This diagnostic is unclear. But to be more precise, the first part says the optimisation could not be performed. This is spot on, and an improvement of what we had before because that didn't issue any diagnostic at all, so one could falsely be under the impression the transformation had actually been applied.
The second part of the diagnostic, the suggestion is unclear / is not applicable. This needs fixing.
The "workaround" is that extra optimisation remarks can be requested, that will tell in more detail what and what didn't happen.
> The code looks perfectly set up for vectorization, but the compiler doesn't vectorize. What's the fix for the code?
Not a strong argument but this is perhaps a bit of an edge case because this is vectorisation with -Oz; this problem won't appear (or it will less) with higher opt levels. In this particular case, as Florian mentiond, it's looprotate not doing something. But forgetting this for a moment, as I also mentioned earlier, I think there are a few things that are orthogonal to this particular change:
* diagnostics can be improved (what Hal and Michael brought up and discussed)
* the vectoriser can be made a bit smarter (or looprotate, as Florian mentioned, and someone is working on that)
> Another question is what we should do at -O0, because there I think we still don't warn.
I will look into this. I was hoping we would issue exactly the same warning, which in this case would be correct: the transformation didn't run.
From: Hans Wennborg <hans at chromium.org>
Sent: 07 October 2019 09:13
To: Sjoerd Meijer <Sjoerd.Meijer at arm.com>
Cc: Michael Kruse <llvmdev at meinersbur.de>; Finkel, Hal J. <hfinkel at anl.gov>; llvm-dev at lists.llvm.org <llvm-dev at lists.llvm.org>; Kruse, Michael <michael.kruse at anl.gov>; hans at hanshq.net <hans at hanshq.net>
Subject: Re: [llvm-dev] vectorize.enable
Sorry for coming late to this thread, I was out a few days.
I wasn't familiar with this pragma before, but if the intended meaning
is indeed "vectorize this" and not "if vectorizing, use this width",
then the change makes sense.
The problem I see is that the warning isn't very actionable. Good
warnings are supposed to be actionable, but what is the developer
supposed to do in this case? The code looks perfectly set up for
vectorization, but the compiler doesn't vectorize. What's the fix for
I think the thread mentioned that someone was working on fixing the
vectorization for this case. Maybe we should wait for that?
(Another question is what we should do at -O0, because there I think
we still don't warn. Does that mean the pragma doesn't have the same
meaning at -O0?)
On Fri, Oct 4, 2019 at 4:28 PM Sjoerd Meijer via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> Thanks for your replies. That was a very useful discussion.
> I won't recommit on a Friday afternoon, but will do on Monday, as it looks like we agreed again on the direction and the change.
> Orthogonal to this change, the interesting topics brought up are improved diagnostics, and the cases the vectoriser misses. I will briefly look why this particular case isn't vectorised, but I suspect that it's a simple case of some prep / clean-up passes not running at Oz.
> From: llvm-dev <llvm-dev-bounces at lists.llvm.org> on behalf of Finkel, Hal J. via llvm-dev <llvm-dev at lists.llvm.org>
> Sent: 03 October 2019 00:18
> To: Michael Kruse <llvmdev at meinersbur.de>
> Cc: llvm-dev at lists.llvm.org <llvm-dev at lists.llvm.org>; Kruse, Michael <michael.kruse at anl.gov>; hans at hanshq.net <hans at hanshq.net>
> Subject: Re: [llvm-dev] vectorize.enable
> On 10/2/19 5:31 PM, Michael Kruse wrote:
> > Am Mi., 2. Okt. 2019 um 15:56 Uhr schrieb Finkel, Hal J. <hfinkel at anl.gov>:
> >>> It's done by the WarnMissedTransformation and just looks for
> >>> transformation metadata that is still in the IR after all passes that
> >>> should have transformed them have ran. That is, it does not know why
> >>> it is still there -- it could be because the LoopVectorize pass is not
> >>> even in the pipeline -- and we cannot be more specific in the message.
> >>> However, -Rpass-missed=loop-vectorize may give more information.
> >> As I recall, there is some trade-off here because it's hard for a
> >> transformation to know that it's last - either the last run of that
> >> particular transformation in the pipeline or the last transformation in
> >> the pipeline that can service a particular transformation request (and
> >> this is especially true if there are multiple, separated pipelines
> >> involved, such as in LTO). This is why we did not have transformations
> >> warn if they can't perform the requested transformation for structural
> >> reasons - maybe they will be able to later. However, we also should
> >> improve the diagnostics in these cases.
> > This was not the only consideration. With ordered transformations,
> > such as vectorize after unroll-and-jam, the LoopVectorize does not
> > even have a chance to analyze the code since it is located after the
> > LoopUnrollAndJam pass. We would still warn that vectorization has not
> > been performed.
> >> I recommend that we consider taking a kind of delayed-diagnostic
> >> approach. When a pass cannot perform the requested transformation, it
> >> records some rationale into the metadata. That rationale can be reported
> >> by WarnMissedTransformation, if available, to make the diagnostic more
> >> helpful. If the transformation is later actually performed, then the
> >> extra information is discarded along with the transformation metadata
> >> itself.
> > I like the idea, but I am not sure how helpful are messages such as
> > "The exiting block is not the loop latch" or "Cannot identify array
> > bounds" are to the end user. It would still be an improvement.
> It probably wouldn't be, but might encourage some more-useful bug reports.
> > If there is no diagnostic metadata, do we keep emitting the current
> > message?
> That was my thought. If we don't know anything else, just keep doing
> what we're doing.
> > If there already is an explanation metadata, does the new one
> > override the old one or is it appended?
> Lacking data, I'll say append (because why throw away information?), but
> only print the last one by default (because, for example, if a pass runs
> multiple times, what I probably care about it why it couldn't perform
> the transformation during the last run, not the previous times when the
> IR might have been in a less-optimizable state) - that's probably less
> > We could also just a hint to the diagnostic such as
> > "-Rpass=loop-vectorize may provide more information".
> Agreed. That is probably the easiest thing, and might be the most useful
> as well. We just need a bit in the metadata to provide some information
> on whether the transformation was attempted but failed.
> > Michael
> Hal Finkel
> Lead, Compiler Technology and Programming Languages
> Leadership Computing Facility
> Argonne National Laboratory
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-dev