<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Apr 15, 2016 at 9:50 AM, Duncan P. N. Exon Smith <span dir="ltr"><<a href="mailto:dexonsmith@apple.com" target="_blank">dexonsmith@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Since I haven't heard any objections to the direction, I'm planning to<br>
commit this (step 4) when I find some cycles; likely over the weekend.<br>
<br>
To make this more concrete (in case someone is silently concerned) I've<br>
posted my WIP patches below. They apply cleanly to r266414. There are<br>
a few mechanical changes missing that are tracked in the commit<br>
messages (such as LangRef, or new tests in some cases). To get<br>
`ninja check` to pass after the final commit you'll need to run the<br>
attached upgrade script. The clang changes are truly uninteresting so<br>
I've left them out.<br>
<br>
- 0001/2: Prep commits I need to flush out (sorry for the noise).<br>
- 0003: Add an explicit type map for ODR type uniquing.<br>
- 0004: Prep.<br>
- 0005: Add ODR type uniquing of members.<br></blockquote><div><br></div><div>What's this ^ for?<br><br>In theory, the list of /members/ of a type shouldn't vary between instances. The special cases (implicit special members, member function template specializations, and nested types) don't appear in the member list - instead they just list the type as their scope (so it's one directional, instead of bidirectional)<br><br>Is this to handle ODR violations? We currently don't handle them, right - we just pick one instance of the type & the user gets to keep the pieces. Is there a particular motivation for changing that? Some place where it's more problematic, we have to support it, etc?<br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
- 0006: Remove DITypeRef (string-based references) and strip down<br>
'retainedTypes:'.<br>
<br>
Once this is done, I expect the bitcode block layout improvements for<br>
lazy-loading of subprograms and composite types (steps 3 and 5 from<br>
this RFC) to be fairly straightforward.<br>
<br>
<br><br>
<br>
> On 2016-Mar-29, at 19:11, Eric Christopher <<a href="mailto:echristo@gmail.com">echristo@gmail.com</a>> wrote:<br>
><br>
> I have no objections to any of this FWIW :)<br>
><br>
> -eric<br>
><br>
> On Tue, Mar 29, 2016 at 6:46 PM Duncan P. N. Exon Smith <<a href="mailto:dexonsmith@apple.com">dexonsmith@apple.com</a>> wrote:<br>
><br>
> > On 2016-Mar-22, at 20:11, Eric Christopher <<a href="mailto:echristo@gmail.com">echristo@gmail.com</a>> wrote:<br>
> ><br>
> >> On Tue, Mar 22, 2016 at 8:04 PM David Blaikie <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>> wrote:<br>
> >> +pcc, who had some other ideas/patch out for improving memory usage of debug info<br>
> >> +Reid, who's responsible for the windows/CodeView/PDB debug info which is motivating some of the ideas about changes to type emission<br>
> ><br>
> > So I discussed this with Adrian and Mehdi at the social last Thursday and I'm getting set to finish the write up. I think it'll have some bearing on this proposal as I think it'll change how we want to take a look at the format of DISubprogram metadata a bit more.<br>
><br>
> (The interesting bit here is to make a clearer split between<br>
> DISubprogram declarations (part of the type hierarchY) and<br>
> DISubprogram definitions (part of the code/line table/variable<br>
> locations). I think that'll end up being mostly orthogonal to what<br>
> I'm trying to do.)<br>
><br>
> > That said, most of it is orthogonal to the changes Duncan is talking about here. Just puts the pressure on to get the other proposal written up.<br>
><br>
> Which is now here:<br>
> <a href="http://lists.llvm.org/pipermail/llvm-dev/2016-March/097773.html" rel="noreferrer" target="_blank">http://lists.llvm.org/pipermail/llvm-dev/2016-March/097773.html</a><br>
><br>
> >> Baking into the IR more about types as units has pretty direct overlap with Reid/CodeView/etc - so, yeah, that'll takes ome discussion (but, as you say, it's not in your immediate plan anyway, so we can come back to that - but would be good for whoever gets there first to discuss it with the others)<br>
><br>
> After thinking for a few days, I don't think this will bake anything<br>
> new into the IR. If anything it removes a few special cases.<br>
><br>
> More at the bottom.<br>
><br>
> >>> Motivation<br>
> >>> ==========<br>
> >>><br>
> >>> Based on some analysis Mehdi ran (ping him for details), there are three<br>
> >>> (related) compile-time bottlenecks we're seeing with `-flto=thin -g`:<br>
> >>><br>
> >>> a) Reading the large number of Metadata bitcode records in the global<br>
> >>> metadata block. I'm talking about raw `BitStreamer` calls here.<br>
> >>><br>
> >>> b) Creating unnecessary `DI*` instances (that aren't relevant to code).<br>
> >>><br>
> >>> c) Emitting unnecessary `DI*` instances (that aren't relevant to code).<br>
> >>><br>
> >>> Here is my recollection of some peak memory stats on a small testcase<br>
> >>> during thin-LTO, which should be a decent indicator of (b):<br>
> >>><br>
> >>> - ~150MB: DILocation<br>
> >>> - ~100MB: DISubprogram<br>
> >>> - ~70MB: DILocalVariable<br>
> >>> - ~50MB: (cumulative) DIType descendents<br>
> >>><br>
> >>> It looks, suprisingly, like types are not the primary bottleneck.<br>
><br>
> (Probably wrong for (a), BTW. Caveats matter.)<br>
><br>
> >>> There are caveats:<br>
> >>><br>
> >>> - `DISubprogram` declarations -- member function descriptors -- are<br>
> >>> part of the type hierarchy.<br>
> >>> - Most of the type hierarchy gets uniqued at parse time.<br>
> >>> - As a result, these data are a poor indicator for (a).<br>
><br>
> ((a) is the main bottleneck for compile-time of -flto=thin (since it's<br>
> quadratic in the number of files). (b) only affects memory. Also<br>
> important, but at least it scales linearly.)<br>
><br>
> >>> Even so, non-types are substantial.<br>
> >>><br>
> >>> Proposal<br>
> >>> ========<br>
> >>><br>
> >>> Short version<br>
> >>> -------------<br>
> >>><br>
> >>> 4. Remove `DICompositeType`s from `retainedTypes:`, similar to (2).<br>
><br>
> This is the part that's relevant to the new RFC Eric just posted.<br>
><br>
> >>> Long version<br>
> >>> -------------<br>
> >>><br>
> >>> 4. Implement my proposal to remove the `DICompositeType` name map from<br>
> >>> `retainedTypes:`.<br>
> >>><br>
> >>> <a href="http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160125/327936.html" rel="noreferrer" target="_blank">http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160125/327936.html</a><br>
> >>><br>
> >>> Similar to (2) above, this will naturally filter the types that get<br>
> >>> linked in to the ones actually used by the code being linked.<br>
> >>><br>
> >>> It should also allow the reader to skip records for types that have<br>
> >>> already been loaded in the main module.<br>
><br>
> The essential things I want to accomplish with this part:<br>
><br>
> - Make `type:` operands less special: instead of referencing types<br>
> indirectly through MDString, point directly at the type node.<br>
><br>
> - Stop using `retainedTypes:` by default (only for -gfull, etc.).<br>
><br>
> - Avoid blowing up memory in -flto=full (which converting MDString<br>
> references back to pointers would do naively, through<br>
> re-introducing cycles). Note that this needs to be handled<br>
> somehow at BitcodeReader time.<br>
><br>
> After chatting with Eric, I don't think this conflicts at all with the<br>
> other RFC. Unifying the `type:` operands might actually help both.<br>
><br>
> One good point David mentioned last week was that we don't want to<br>
> teach the IR any more about types. Rather than inventing some new<br>
> context (as I suggested originally), I figure LTO plugins can just<br>
> pass a (StringRef -> DIType*) map to the BitcodeReader, and the Module<br>
> itself doesn't need to know anything about it.<br>
<br>
<br></blockquote></div><br></div></div>