Mehdi AMINI via llvm-dev
2017-Jun-07 18:00 UTC
[llvm-dev] [RFC][ThinLTO] llvm-dis ThinLTO summary dump format
2017-06-07 10:19 GMT-07:00 David Blaikie <dblaikie at gmail.com>:> > > On Wed, Jun 7, 2017 at 10:01 AM Mehdi AMINI <joker.eph at gmail.com> wrote: > >> 2017-06-07 9:44 GMT-07:00 David Blaikie <dblaikie at gmail.com>: >> >>> >>> >>> On Tue, Jun 6, 2017 at 2:21 PM Mehdi AMINI <joker.eph at gmail.com> wrote: >>> >>>> 2017-06-06 13:38 GMT-07:00 David Blaikie <dblaikie at gmail.com>: >>>> >>>>> >>>>> >>>>> On Tue, Jun 6, 2017 at 1:26 PM Mehdi AMINI <joker.eph at gmail.com> >>>>> wrote: >>>>> >>>>>> 2017-06-05 14:27 GMT-07:00 David Blaikie via llvm-dev < >>>>>> llvm-dev at lists.llvm.org>: >>>>>> >>>>>>> I know there's been a bunch of discussion here already, but I was >>>>>>> wondering if perhaps someone (probably Teresa? Peter?) could: >>>>>>> >>>>>>> 1) summarize the current state >>>>>>> 2) describe the end-goal >>>>>>> 3) describe what steps (& how this patch relates) are planned to get >>>>>>> to (2) >>>>>>> >>>>>>> My naive thoughts, not being intimately familiar with any of this: >>>>>>> Usually bitcode and textual IR support go in together or around the same >>>>>>> time, and designed that way from the start (take r211920 for examaple, >>>>>>> which added an explicit representation of COMDATs to the IR). This seems to >>>>>>> have been an oversight in the implementation of IR summaries (is that an >>>>>>> accurate representation/statement?) >>>>>>> >>>>>> >>>>>> More or less: it was not an oversight. >>>>>> The summaries are not really part of the IR, it is more like an >>>>>> "analysis result" that is serialized. It can always be recomputed from the >>>>>> IR. This aspect makes it quite "special", it is the only analysis result >>>>>> that I know of that we serialize. >>>>>> >>>>> >>>>> The use list work seems pretty similar in some ways (granted, can't be >>>>> recomputed to match, hence the desire to serialize it for test case >>>>> implementation). >>>>> >>>> >>>> I see use-list as a leaky implementation detail of the IR that we >>>> serialized because it impact the processing of the IR. >>>> >>>> Summaries are more like serializing the CFG for example. >>>> >>>> >>>>> But it looks like the same is true here to a degree - there are test >>>>> cases that exercise the summary handling, so they want summaries for input >>>>> (for now, I think, I've seen test cases that run another LLVM tool to >>>>> insert/create a summary to then feed that back in for a test), or to test >>>>> that the resulting summary is correct. >>>>> >>>> >>>> We have cases were we want summaries as an input and check a combined >>>> summary as an output, and for these having the YAML representation will be >>>> useful (we didn't have it before). >>>> >>> >>> What I'm suggesting is that this is an (optional) IR feature as much as >>> any other >>> >> >> Well I disagree with this at this point, because I haven't read anything >> that would support it. >> I'd be happy to revise my position if you were providing any argument >> that would make this holds in face of any other analysis result. >> > > I'm not intending to make an argument that would generalize over other > analysis results, nor following why there should be one. The bitcode > support for this doesn't generalize over other analysis results so I'm not > sure why the textual IR would. > > >> - so it seems slightly odd that it'd be YAML rather than something that >>> looked more like the rest of the IR. Though I'm not outright opposed to >>> YAML here - just want to make sure this information is being treated as a >>> first class IR construct (as much as use order, comdats, etc are for rough >>> examples) >>> >> >> YAML was pushed forward as an easy way to get there IIRC. It wasn't set >> in stone and it was clearly open to change it to a more integrate format. >> So I'm supportive of anyone who would replace this with a more >> "textual-IR integrated" format, I haven't proposed this in this thread >> because Teresa is interested in getting something readable "quickly". >> > > & I'm interested in pushing back on that a bit to discuss what the > end/desired/ideal goal is here & to understand whether such an intermediate > state is beneficial or whether it'd be better to look at the end goal & > increments to approach that. > > >> My point was more that as an intermediate step, I rather reuse the >> existing YAML serialization than creating yet another dump. >> >> >> >>> >>> Can summaries be standalone? I thought they could (that'd be ideal for >>>>> the distributed situation - only the summary needs to go to the 'thin link' >>>>> step, I think? (currently maybe only the debug info is stripped for that - >>>>> but ideally other unused IR wouldn't be shipped there as well, I would >>>>> think) >>>>> >>>> >>>> Yes conceptually they can be standalone. >>>> >>> >>> This seems to provide the strongest/clear motivation for having >>> summaries as a first class (though optional) IR construct. >>> >> >> No, this provide a strong motivation to have a proper serialization, I >> don't see how you connect this to the rest of the IR. >> > > Because it can appear both with the rest of the IR and without it? So "it > can be derived from the IR" doesn't seem to hold as a generalization - > you're right that there could be no way to write it beside/with IR, and a > non-IR-like syntax to write it separately. >I haven't advocated for this. But the fact that this is an analysis result that we serialize alongside the IR, whether in bitcode form or textual form, does not make it automatically part of the IR to me. We just bundle the IR with one of the analysis result because there exist a consumer for the analysis result that benefit from not having to read all the IR to get the information. The only reason for the summaries to exist is a performance one. ThinLTO could be implemented without summaries, reconstructing them on the fly when needed.> But that doesn't seem like what's needed/desired for testing - since > current tests work around this by generating bitcode with summaries to test > - so bitcode with summary seems like something that's desired as an input > to tests. Feels like IR to me. >No it isn't IR, unless you consider that any other analysis result is "IR". Another analogy is that if you have an analysis that is trying to find all the SCCs on the callgraph, its input is the result of the callgraphy analysis, and its output is a list of SCCs. You could write an input test as IR and opt would run the callgraph analysis and feed it into the SCC analysis and you would get the list of SCC as an output for your test. But you could also have a textual format to write a callgraph and feed it to your test. It wouldn't come to my mind to write IR *and* the callgraph that corresponds to this IR together as an input to the SCC analysis test.> > >> & now there's an effort to correct that. >>>>>>> >>>>>> >>>>>> The main motivation here, I believe, is more to help dev to have >>>>>> human readable/understandable dump for ThinLTO bitcodes. Having to inspect >>>>>> separately summaries is a pain. >>>>>> >>>>> >>>>> Not sure I quite follow - inspect separately? >>>>> >>>> >>>> llvm-dis does not display summaries today, so you can't just use >>>> llvm-dis like a "regular" flow. >>>> >>>> >>>>> How are they inspected today? >>>>> >>>> >>>> llvm-bcanalyzer? And now the YAML dump as well. >>>> >>>> >>>>> & also, I think there are test cases that want to/are currently >>>>> testing summary input but do so somewhat awkwardly by using another tool to >>>>> produce the summary first. Ideally the test case would have the summary >>>>> written in to start, I would think, if that's a codepath worth testing? >>>>> >>>> >>>> The IR already contains all the information, so why repeating it? >>>> >>> >>> For the same reason that it's relevant to test cases which way it's >>> encoded, etc (in the same way that the LLVM IR repeats types of uses, for >>> example - even though they're totally redundant from a "does this have all >>> the semantic information required) & because it can be standalone. >>> >> >>> >>>> This makes the test case harder to maintain, in the vast majority, I >>>> expect that if a test needs IR then it shouldn't need to include a summary >>>> as well (and vice-versa). >>>> >>> >>> Ah, sorry, I'm not suggesting it should be required - in the same way >>> it's not required in the bitcode. But if you want a summary in the bitcode >>> when assembling a .ll file it seems OK To say you write it in the IR, >>> >> >> No it does not seem OK to me to write summaries alongside the IR in tests >> in general (outside of specific need like testing the round-trip of >> course). >> > > I'm not sure I understand "in tests in general" - the only things that > would include summaries are tests testing summary handling, in which case > it seems sort of useful to have that be explicit in the IR >Why? If you want to test summary handling, why would you need the IR? (cf above).> - at the moment that's being worked around in tests by running IR through > LLVM to generate a summary, then feeding that back in to test the summary. > That doesn't seem to be ideal testing (in the same way that LLVM isn't > tested by running source code through clang and testing the resulting IR - > the two features are separated and tested independently) >This is what we do for every analysis (cf callgraph analogy above).> > >> It is entirely redundant and I don't perceive any benefit, I don't see >> why you would want to do that? >> >> >> >>> and equally if there is a summary in the bitcode it seems reasonable >>> that it be printed in the .ll file by llvm-dis. >>> >> >> I agree and I advocated for this earlier. >> > > That sounds like IR to me - it's a feature that appears in bitcode and IR > files, roundtrips through either. >Let's define "IR" then. That does not pass my bar for IR: serializing an analysis result is not enough to change its fundamental properties that it isn't part of the IR per-se.> > >> In the majority of test we have we want to check if the importing does >>>> what it is supposed to do, and if the linkage are correctly adjusted. With >>>> a YAML (or other) serialization for the summaries this could indeed been >>>> done purely with summaries, without any IR involved. >>>> >>> >>> I'm not sure I understand - you mean for executions of tools that don't >>> need the rest of the IR, there could be a different/separate tool that >>> consumes YAML summaries and produces YAML >>> >> >> It does not have to be a separate tool: a tool that is looking to operate >> purely on summary should just ask to get the summaries out of the input >> file. The input being textual or bitcode shouldn't matter much at this >> point. >> > > Agreed, but I'm not understanding how this isn't IR if it can be textual > or bitcode and roundtrip through both. >My best guess is that we don't have the same definition for "IR".> > >> This is exactly how 'opt' and 'llc' operate. >> >> summaries and that would be tested - but the "consuming a summary in a >>> bitcode file" would not be? >>> >> >> This is exactly what we're doing with (almost) *all* of the .ll test: we >> write them as textual, and read them back as textual, and not as bitcode. >> >> >>> I'm not sure I understand the benefit of this separation and asymmetry >>> with the bitcode form of the same data. >>> >> >> Have you tried to write a test directly in bitcode? ;) >> > > I didn't mean to suggest writing tests in bitcode - I meant I didn't > understand the asymmetry that it sounded like you were suggesting - where a > summary could be explicitly read from a textual file, but not from a .ll > file (that a .ll file would only contain an implicit summary, I guess, > never an explicit one - unlike bitcode which could contain an explicit > summary, standalone or alongside the rest of the IR). >I don't think this represent my position. I mentioned earlier in the thread that we should have "llvm-dis | llvm-as" being able to round-trip bitcode, including summaries when present.>This does not mean that testing how we handle summaries in ThinLTO requires to write *both* IR and summary in the textual input. I'm advocating that this would b a terrible way of writing test, which I believe is an orthogonal concern to *supporting* round-tripping bitcode files that contain summaries. -- Mehdi -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170607/3cdb74fd/attachment.html>
David Blaikie via llvm-dev
2017-Jun-07 18:06 UTC
[llvm-dev] [RFC][ThinLTO] llvm-dis ThinLTO summary dump format
> > I don't think this represent my position. I mentioned earlier in the > thread that we should have "llvm-dis | llvm-as" being able to round-trip > bitcode, including summaries when present. >OK, then I think there's been a lot of confusion, so maybe focussing on these comments might be more useful. How do you envision that happening? That the summary would be written in the textual representation? (or it could be reconstructed, as you mention, by having a single bit/flag/something in the textual representation that says "this has a summary" & then a summary could be generated from the Module and written out to the bitcode) That's all I'm really interested in in this discussion. That summaries seem like they should be as well represented in .ll files as they are in .bc files. And if that's the case, it seems weird for them to be YAML - YAML in a .ll file would seem out of place/inconsistent with the rest of what's in a .ll file, to me. Not impossible, but a bit 'off'.> This does not mean that testing how we handle summaries in ThinLTO > requires to write *both* IR and summary in the textual input. > I'm advocating that this would b a terrible way of writing test, which I > believe is an orthogonal concern to *supporting* round-tripping bitcode > files that contain summaries. > > -- > Mehdi > > >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170607/d511566b/attachment-0001.html>
Mehdi AMINI via llvm-dev
2017-Jun-07 19:09 UTC
[llvm-dev] [RFC][ThinLTO] llvm-dis ThinLTO summary dump format
2017-06-07 11:06 GMT-07:00 David Blaikie <dblaikie at gmail.com>:> I don't think this represent my position. I mentioned earlier in the >> thread that we should have "llvm-dis | llvm-as" being able to round-trip >> bitcode, including summaries when present. >> > > OK, then I think there's been a lot of confusion, so maybe focussing on > these comments might be more useful. > > How do you envision that happening? That the summary would be written in > the textual representation? >Yes, I don't think anyone in the thread argued against this. The original proposal was not accounting for reading them back though.> (or it could be reconstructed, as you mention, by having a single > bit/flag/something in the textual representation that says "this has a > summary" & then a summary could be generated from the Module and written > out to the bitcode) >> That's all I'm really interested in in this discussion. That summaries > seem like they should be as well represented in .ll files as they are in > .bc files. >Right.> > And if that's the case, it seems weird for them to be YAML - YAML in a .ll > file would seem out of place/inconsistent with the rest of what's in a .ll > file, to me. Not impossible, but a bit 'off'. >I agree, and this was raised as a concern in the first place when the YAML format was introduced. But IIRC Peter pushed for the YAML as an intermediate easy step to get results, and advocated for punting nailing down an IR format to later. Now the proposal in this thread is to dump summaries with llvm-dis, which I agree with, but what I raised in this thread is that we should be able to read them back. I also mentioned that we shouldn't keep *two* separate textual formats. As a first step, having `llvm-dis | llvm-as` using the YAML format seems like a pragmatic way of getting it done. I'd be happy if anyone wants to replace the YAML format with a more "integrated way" in the textual IR format, I didn't feel like it was worth blocking getting the feature in `llvm-dis` though. -- Mehdi -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170607/e4d94b2e/attachment.html>