[cfe-dev] [RFC] Clang SourceLocation overflow
Keane, Erich via cfe-dev
cfe-dev at lists.llvm.org
Wed Feb 3 12:10:43 PST 2021
Thanks for the response Richard! I’m working with my QA team to get a better reproducer to see if we can figure out what is the root-cause. It DOES include a couple of other files so I’m not sure all of what is entailed.
>There are also some relatively cheap things we could do to expand our capacity: we could remove the 'is macro location' bit without much effort (though this would slow down the current places that check the flag), which would double our available source location capacity.
This seems like a useful and cheap thing to do, if find this is a legitimate issue, I’ll see if I can put a patch together to do this one.
>And we could allow the division between local and imported locations to be determined dynamically rather than fixing a 50/50 split, which would in practice be likely to double the available capacity again. But those would only be useful if we're just a little over the limit; they wouldn't help if there's an asymptotic problem in our source location usage.
I don’t have a great idea on how to do this, or how we use this, but based on your description it seems worth-while. At least in non-module TUs I’d assume that the “imported-from-ast-file” is relatively rare. Also, I would think these imports happen ‘first’, right? So we could figure out the split as soon as we’re done with imports and optimize our space.
From: Richard Smith <richard at metafoo.co.uk>
Sent: Wednesday, February 3, 2021 11:54 AM
To: Keane, Erich <erich.keane at intel.com>
Cc: David Rector <davrecthreads at gmail.com>; clang developer list <cfe-dev at lists.llvm.org>
Subject: Re: [cfe-dev] [RFC] Clang SourceLocation overflow
On Tue, 2 Feb 2021 at 10:41, Keane, Erich via cfe-dev <cfe-dev at lists.llvm.org<mailto:cfe-dev at lists.llvm.org>> wrote:
Presumably, yes, we could hit that limit, and that limit is closer than we’d think.
Based on my understanding though, we currently stores the ‘offset’ into the buffer. If we were to go with your plan, we’d reduce the ‘SourceLocation’ space significantly, since we’d only have 1 value taken up by “ALongFunctionName” instead of ~15? Or, at least, only 1 instead of 3 in ‘int’.
I presume that anything short of expanding the size of SourceLocation is going to be a temporary measure at best.
One consideration: What do we think about making SourceLocation have an underlying type of ‘bitfield’ that is a smaller size than 64 bits? Then at least we could make SourceRange be a packed value? So make SourceLocation be 40 bits (obviously would still be 64 if stored directly), but at least we could pack 2 of them into 80 bits for any type that does some sort of bit-packing?
I doubt this would help in practice -- most of the storage in AST nodes tends to be SourceLocations and pointers.
Alternatively, what about making SourceRange (instead of 2 SourceLocation objects) be a SourceLocation + unsigned? It would make it a little smaller? Then we could do similar optimizations over time to the ASTNodes. That is, all of the TypeLoc types could just store things like open/close parens as offsets-from-the-original rather than SourceLocations, then calculate them when needed? I would assume anything that required re-calculation would be acceptable, since SourceLocations are seemingly quite rarely used (except for the few cases that Richard mentioned below).
That doesn't seem likely to work: we can't assume the beginning and end of a source range are close together in source location space (they could be in different SLocEntrys, which could be a long way apart from each other).
My preference is obviously to just make SourceLocation be uint64_t instead, but the impact on AST size is going to be significant. So I guess I’m hoping that Richard Smith can comment and help us figure out how much pain we are willing to go through for this?
We could do that, and having a build-time selector for 32/64-bit SourceLocations seems like it might not impose a huge cost. But I think we should first try to gain some confidence that we're addressing the right problem -- running out of source locations seems likely to indicate that there's something more fundamental wrong with the compilation. Currently we reserve one bit for an 'is macro location' flag, and divide the remaining addressable 2GB into a 1GB local region and a 1GB imported-from-AST-file region. The resulting "1GB of preprocessed source (including all intermediate stages of macro expansion)" implementation limit does not seem especially restrictive to me, so the first thing I think we should find out is how we're actually hitting that limit in the boost example. We can currently handle huge compilations without hitting the limit, so if a single header file can hit it, that seems indicative of a bug that's not just caused by the limit being too low.
One possible cause would be that a large header is included a *lot* and doesn't have a proper include guard. If that's the case, that seems like a problem that we should get fixed in boost -- or in Clang if it's a bug in include guard detection -- because that will contribute to long compile times too. I expect there are other cases where source location address space gets wasted that we could drill down into.
If we find the problem is our location tracking for macro expansions (eg, pushing a large volume of tokens through deeply nested macro expansions), we could probably find a way to turn that off; there may even be ways we can intelligently turn it off selectively in cases where we think the intermediary information is unlikely to be useful. We could also look into throwing away source location address space for macro expansions that ended up producing no tokens. But we need to understand the nature of the problem first.
There are also some relatively cheap things we could do to expand our capacity: we could remove the 'is macro location' bit without much effort (though this would slow down the current places that check the flag), which would double our available source location capacity. And we could allow the division between local and imported locations to be determined dynamically rather than fixing a 50/50 split, which would in practice be likely to double the available capacity again. But those would only be useful if we're just a little over the limit; they wouldn't help if there's an asymptotic problem in our source location usage.
From: David Rector <davrecthreads at gmail.com<mailto:davrecthreads at gmail.com>>
Sent: Tuesday, February 2, 2021 10:28 AM
To: Keane, Erich <erich.keane at intel.com<mailto:erich.keane at intel.com>>
Cc: clang developer list <cfe-dev at lists.llvm.org<mailto:cfe-dev at lists.llvm.org>>
Subject: Re: [cfe-dev] [RFC] Clang SourceLocation overflow
An additional consideration: if the files/buffers are so large that each SourceLocation cannot fit in 30 bits, does that imply that so many SourceLocations will be generated that the indices to them won’t be able to fit in 30 bits? That may be the case now that I think of it, in which case the only solution really would be 64 bit SourceLocations, or something still more creative/costly.
On Feb 2, 2021, at 11:53 AM, Keane, Erich <erich.keane at intel.com<mailto:erich.keane at intel.com>> wrote:
Huh, that is an interesting idea! The issue ends up being the callers of getOffset though, since it is a previate method.
That said, I wonder if the extra bit-use is just putting off the inevitable, and we should just use the vector in SourceManager for everything. My initial thought is that we could do away with the ‘IsMacro’ bit as well, and just encode that in the vector too. It makes isFileID and isMacroID require the SourceManager as well, but I wonder if that is OK?
From: David Rector <davrecthreads at gmail.com<mailto:davrecthreads at gmail.com>>
Sent: Tuesday, February 2, 2021 8:46 AM
To: Keane, Erich <erich.keane at intel.com<mailto:erich.keane at intel.com>>
Cc: clang developer list <cfe-dev at lists.llvm.org<mailto:cfe-dev at lists.llvm.org>>
Subject: Re: [cfe-dev] [RFC] Clang SourceLocation overflow
A possible solution not mentioned: reserve a bit in SourceLocation for "IsBigLoc". When IsBigLoc = true, the remaining 30 bits of the "ID" field is not an offset but an index into a vector of uintptr_ts in SourceManager, each holding 64-bit offset data.
Only a few changes to private methods and isolated details would be needed I think: e.g. method `unsigned SourceLocation::getOffset()` would become `uintptr_t SourceLocation::getOffset(const SourceManager &SM)`, and would dereference and return the larger data when IsBigLoc. And more generally `unsigned` types should be changed to `uintptr_t` everywhere 32-bit encodings are currently assumed (mostly in SourceManager).
This might be easier and less intrusive than allowing 64-bit SourceLocations depending on a build flag, if I understand that proposal correctly, since that would require templating virtually everything with the SourceLocation type.
On Feb 2, 2021, at 9:21 AM, Keane, Erich via cfe-dev <cfe-dev at lists.llvm.org<mailto:cfe-dev at lists.llvm.org>> wrote:
Original thread here: https://lists.llvm.org/pipermail/cfe-dev/2019-October/063459.html
I’m bringing this back up since we have a reproduction of this in Boost now. We haven’t finished analyzing what boost is doing, but simply doing an include of:
#include <libs/vmd/test/test_doc_modifiers_return_type.cxx>
Now causes us to run out of source locations, hitting:
/llvm/clang/lib/Basic/SourceManager.cpp:680: clang::SourceLocation clang::SourceManager::createExpansionLocImpl(const clang::SrcMgr::ExpansionInfo &, unsigned int, int, unsigned int): Assertion `NextLocalOffset + TokLength + 1 > NextLocalOffset && NextLocalOffset + TokLength + 1 <= CurrentLoadedOffset && "Ran out of source locations!"' failed.
From the last discussion, it seems that increasing our source-location size isn’t really acceptable due to how much it is stored in the AST( Multiple times per node), and giving up on location isn’t viable either. Additionally, the source location/source manager layout is quite complex and I don’t quite understand it yet, so I don’t have a good way of suggesting an alternative.
SO, I’d like to re-start the discussion into this, we need to find a way to make our compiler able to support more source-locations, as I can’t imagine this is going to be the only time we run into this.
-Erihc
>>>>>>>>>>>>>>>>>>>>
Hi Matt.
Thanks for the offer. Whenever you’re back and have a moment is fine by me, I don’t think we’re in a massive hurry.
Thanks,
Christof
From: Matt Asplund <mwasplund at gmail.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>
Sent: 10 October 2019 14:09
To: Christof Douma <Christof.Douma at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>
Cc: Richard Smith <richard at metafoo.co.uk<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>; Mikhail Maltsev <Mikhail.Maltsev at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>; Clang Dev <cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>; nd <nd at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>
Subject: Re: [cfe-dev] [RFC] Clang SourceLocation overflow
Sure, I can share my changes. In general my changes are very localized to the code paths I was hitting oveflow issues and tried to keep as much as possible using 32bit encodings. Is there an ETA for when you will start investigating, I am out of town for the next week but if needed can get access to my desktop.
-Matt
On Thu, Oct 10, 2019, 2:13 AM Christof Douma <Christof.Douma at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:Christof.Douma at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>> wrote:
Hi Richard.
Thanks for the clarification, I certainly had not realized that location tracking was needed for correctness of clang. Decoupling this sounds like a painful processes, and the only thing we get is errors without any location information. Sounds like not a great trade-off. We’ll go experiment a bit with the size impact of using 64-bits and will attempt to take the route of a cmake configuration option.
If there are people that have already done some experiments on the size impact of 64-bits location pointers, we’re welcome your insights. @Matt Asplund<mailto:mwasplund at gmail.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>> I believe you said that you’ve done some prototyping, is there something you can share?
Thanks,
Christof
From: Richard Smith <richard at metafoo.co.uk<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:richard at metafoo.co.uk<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>>
Sent: 08 October 2019 19:53
To: Christof Douma <Christof.Douma at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:Christof.Douma at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>>
Cc: Mikhail Maltsev <Mikhail.Maltsev at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:Mikhail.Maltsev at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>>; Clang Dev <cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>>; nd <nd at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:nd at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>>
Subject: Re: [cfe-dev] [RFC] Clang SourceLocation overflow
On Tue, 8 Oct 2019, 10:42 Christof Douma via cfe-dev, <cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>> wrote:
Hi Richard, Paul and other.
Thanks for the input so far. I wanted to point out that it’s not our code-base. Rather, we’re seeing more use of the LLVM technology in the automotive market and as usual we’re faced with existing code bases that are tried and tested with other toolchains (gcc or others) and when LLVM comes along things don’t always work directly.
We’ve suggested better ways of structuring their code and your suggestions are certainly good input. However, legacy code is especially sticky in any market that has to handle ‘safety’ concerns, like automotive, aerospace and medical markets. Code changes are pretty expensive in those fields. So while I hope that over time we see more sensible coding structures, I don’t expect that to happen any time soon. In the mean time, we’re searching for a solution for this coding pattern that doesn’t play well with clang.
Hope that gave some more background of where this question comes from.
Do all options that were suggested by Mikhail really require fundamental restructuring of major parts of clang? This surprised me, I had expected that the option 2 to be possible without a complete overhaul. (2 is “Track until an overflow occurs after that make the lexer output the <invalid location> special value for all subsequent tokens.”)
Clang uses source locations as part of the semantic representation of the AST in some cases (simple example: some forms of initialization might use parens or not, with different semantics, and we distinguish between them based on whether we have paren locations; there are also some checks that look at which of two source locations came first when determining which warnings or errors to produce, and so on). Maybe we could go through and fix all of those, but that's still a very large task and it'd be hard to check we got them all.
Not nice user experience but maybe doable? I was hoping there was something slightly better that still works without a major restructuring (maybe something that at least gives a rough location or something that only gives the location of the error and not the include stack under an option or using some kind of heuristic to detect that things go haywire).
As an alternative, I was curious if it would be possible and acceptable to make the switch between 32-bit and 64-bit location tracking a build-time/cmake decision? I’ve not done any estimation on the memory size growth, so maybe this is a dead end.
If someone is prepared to do the work to add and maintain this build mode, I think this might be a feasible option.
Thanks,
Christof
From: cfe-dev <cfe-dev-bounces at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:cfe-dev-bounces at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>> On Behalf Of Richard Smith via cfe-dev
Sent: 07 October 2019 20:36
To: Mikhail Maltsev <Mikhail.Maltsev at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:Mikhail.Maltsev at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>>
Cc: nd <nd at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:nd at arm.com<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>>; cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>
Subject: Re: [cfe-dev] [RFC] Clang SourceLocation overflow
On Wed, 2 Oct 2019 at 09:26, Mikhail Maltsev via cfe-dev <cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>> wrote:
Hi all,
We are experiencing a problem with Clang SourceLocation overflow.
Currently source locations are 32-bit values, one bit is a flag, which gives
a source location space of 2^31 characters.
When the Clang lexer processes an #include directive it reserves the total size
of the file being included in the source location space. An overflow can occur
if a large file (which does not have include guards by design) is included many
times into a single TU.
The pattern of including a file multiple times is for example required by
the AUTOSAR standard [1], which is widely used in the automotive industry.
Specifically the pattern is described in the Specification of Memory Mapping [2]:
Section 8.2.1, MEMMAP003:
"The start and stop symbols for section control are configured with section
identifiers defined in MemMap.h [...] For instance:
#define EEP_START_SEC_VAR_16BIT
#include "MemMap.h"
static uint16 EepTimer;
static uint16 EepRemainingBytes;
#define EEP_STOP_SEC_VAR_16BIT
#include "MemMap.h""
Section 8.2.2, MEMMAP005:
"The file MemMap.h shall provide a mechanism to select different code, variable
or constant sections by checking the definition of the module specific memory
allocation key words for starting a section [...]"
In practice MemMap.h can reach several MBs and can be included several thousand
times causing an overflow in the source location space.
The problem does not occur with GCC because it tracks line numbers rather than
file offsets. Column numbers are tracked separately and are optional. I.e., in
GCC a source location can be either a (line+column) tuple packed into 32 bits or
(when the line number exceeds a certain threshold) a 32-bit line number.
We are looking for an acceptable way of resolving the problem and propose the
following approaches for discussion:
1. Use 64 bits for source location tracking.
2. Track until an overflow occurs after that make the lexer output
the <invalid location> special value for all subsequent tokens.
3. Implement an approach similar to the one used by GCC and start tracking line
numbers instead of file offsets after a certain threshold. Resort to (2)
when even line numbers overflow.
4. (?) Detect the multiple inclusion pattern and track it differently (for now
we don't have specific ideas on how to implement this)
Is any of these approaches viable? What caveats should we expect? (we already
know about static_asserts guarding the sizes of certain class fields which start
failing in the first approach).
Other suggestions are welcome.
I don't think any of the above approaches are reasonable; they would all require fundamental restructuring of major parts of Clang, an efficiency or memory size hit for all other users of Clang, or some combination of those.
Your code pattern seems unreasonable; including a multi-megabyte file thousands of times is not a good idea. Can you split out parts of MemMap.h into a separate header that is only included once, and keep only the parts that actually change on repeated inclusion in MemMap.h itself?
_______________________________________________
cfe-dev mailing list
cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev><mailto:cfe-dev at lists.llvm.org<https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev>>
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20191010/bc5ff8cd/attachment.html>
_______________________________________________
cfe-dev mailing list
cfe-dev at lists.llvm.org<mailto:cfe-dev at lists.llvm.org>
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
_______________________________________________
cfe-dev mailing list
cfe-dev at lists.llvm.org<mailto:cfe-dev at lists.llvm.org>
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20210203/61868372/attachment-0001.html>
More information about the cfe-dev
mailing list