This is an archive of the discontinued LLVM Phabricator instance.

Remove the restriction of ten types on AligedCharArrayUnion
ClosedPublic

Authored by spyffe on Jan 6 2017, 6:24 PM.

Details

Summary

This patch uses C++11 parameter packs and constexpr functions to allow AlignedCharArrayUnion to hold an arbitrary number of types.

Diff Detail

Repository
rL LLVM

Event Timeline

spyffe updated this revision to Diff 83483.Jan 6 2017, 6:24 PM
spyffe retitled this revision from to Remove the restriction of ten types on AligedCharArrayUnion.
spyffe updated this object.
spyffe added reviewers: lhames, dblaikie.
spyffe set the repository for this revision to rL LLVM.
spyffe added a subscriber: llvm-commits.
davide requested changes to this revision.Jan 6 2017, 6:38 PM
davide added a reviewer: davide.
davide added a subscriber: davide.

Can you add an unittest to make sure we handle correctly > 10 arguments?

This revision now requires changes to proceed.Jan 6 2017, 6:38 PM
spyffe updated this revision to Diff 83515.Jan 6 2017, 10:20 PM
spyffe edited edge metadata.
spyffe removed rL LLVM as the repository for this revision.

Updated to add static and runtime unit tests.

dblaikie edited edge metadata.Jan 7 2017, 9:13 AM

Hmm - did have some comments, but I think, maybe, all the supported compilers support std::aligned_union now, so switching to that might be an/the right option now? *hopeful*

include/llvm/Support/AlignOf.h
131 ↗(On Diff #83515)

Leave a non-doxy comment about how std::aligned_union could be used once LLVM's supported compilers support it. (GCC 4.8 is what we'd need, I think - oh, 4.8 is our minimum)

unittests/ADT/AlignOfTest.cpp
31–34 ↗(On Diff #83515)

These shouldn't need to be in a class like this, should they?

36–41 ↗(On Diff #83515)

Not sure this test is needed/covers anything - tests that memset works, really? But, yeah... not sure what the right/appropriate testing would be for this.

38 ↗(On Diff #83515)

Should that be sizeof(MyU.buffer) rather than sizeof(U)? (it could have extra padding, etc - I guess not with this specific set of types)

I agree with David here. Can you check if std::aligned_union actually works with 4.8 and the minimum version of MSVC we support (I think 2015) and switch to it instead?

According to the GCC 5 changelist, support for std::aligned_union was introduced with GCC 5.
I was able to verify with Compiler Explorer that std::aligned_union is unsupported (as in, not found) by 4.9.5. It is supported by 5.1.

According to Microsoft, Visual Studio 2015 supports std::aligned_union.
I was able to verify with "Compile Visual Studio C++ Online" that Visual C++ 2015 supports it.

davide accepted this revision.Jan 9 2017, 11:24 AM
davide edited edge metadata.

According to the GCC 5 changelist, support for std::aligned_union was introduced with GCC 5.
I was able to verify with Compiler Explorer that std::aligned_union is unsupported (as in, not found) by 4.9.5. It is supported by 5.1.

Oh, too bad.

According to Microsoft, Visual Studio 2015 supports std::aligned_union.
I was able to verify with "Compile Visual Studio C++ Online" that Visual C++ 2015 supports it.

Thanks for checking.

I'm fine with this going in as long as David is happy.

This revision is now accepted and ready to land.Jan 9 2017, 11:24 AM

Where do you actually plan to use this? In LLVM proper or in some downstream project?

dblaikie accepted this revision.Jan 9 2017, 11:34 AM
dblaikie edited edge metadata.

Please include a comment in the header/near/in the definition of AlignedCharArrayUnion about the minimum compiler version before we can switch to std::aligned_union.

Where do you actually plan to use this? In LLVM proper or in some downstream project?

My plan was to use this to implement std::variant (a C++17 feature) on the cheap inside LLVM. I mentioned wanting to implement a tagged union and Lang Hames said it would be useful in several places. Hence the review.

spyffe updated this revision to Diff 83679.Jan 9 2017, 1:39 PM
spyffe edited edge metadata.
spyffe set the repository for this revision to rL LLVM.

Added a comment at David's request, and moved the test into the Support/AlignOf unit tests. Also removed the memcpy test.

This revision was automatically updated to reflect the committed changes.
spyffe added a comment.Jan 9 2017, 3:52 PM

This failed to build under Visual Studio:

C:\PROGRA~2\MICROS~1.0\VC\bin\AMD64_~2\cl.exe   /nologo /TP -DGTEST_HAS_RTTI=0 -DUNICODE -D_CRT_NONSTDC_NO_DEPRECATE -D_CRT_NONSTDC_NO_WARNINGS -D_CRT_SECURE_NO_DEPRECATE -D_CRT_SECURE_NO_WARNINGS -D_HAS_EXCEPTIONS=0 -D_SCL_SECURE_NO_DEPRECATE -D_SCL_SECURE_NO_WARNINGS -D_UNICODE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -Ilib\Support -IC:\b\slave\sanitizer-windows\llvm\lib\Support -Iinclude -IC:\b\slave\sanitizer-windows\llvm\include /DWIN32 /D_WINDOWS   /W4 -wd4141 -wd4146 -wd4180 -wd4244 -wd4258 -wd4267 -wd4291 -wd4345 -wd4351 -wd4355 -wd4456 -wd4457 -wd4458 -wd4459 -wd4503 -wd4624 -wd4722 -wd4800 -wd4100 -wd4127 -wd4512 -wd4505 -wd4610 -wd4510 -wd4702 -wd4245 -wd4706 -wd4310 -wd4701 -wd4703 -wd4389 -wd4611 -wd4805 -wd4204 -wd4577 -wd4091 -wd4592 -wd4319 -wd4324 -w14062 -we4238 /Zc:inline /Zc:strictStrings /Oi /Zc:rvalueCast /MD /O2 /Ob2   -UNDEBUG  /EHs-c- /GR- /showIncludes /Folib\Support\CMakeFiles\LLVMSupport.dir\CommandLine.cpp.obj /Fdlib\Support\CMakeFiles\LLVMSupport.dir\ /FS -c C:\b\slave\sanitizer-windows\llvm\lib\Support\CommandLine.cpp
C:\b\slave\sanitizer-windows\llvm\include\llvm/Support/AlignOf.h(135): error C2027: use of undefined type 'llvm::detail::AlignerImpl<Ts...>'
C:\b\slave\sanitizer-windows\llvm\include\llvm/Support/AlignOf.h(135): note: see declaration of 'llvm::detail::AlignerImpl<Ts...>'
C:\b\slave\sanitizer-windows\llvm\include\llvm/Support/AlignOf.h(136): note: see reference to class template instantiation 'llvm::AlignedCharArrayUnion<Ts...>' being compiled
ninja: build stopped: subcommand failed.

I did some restricted testing with http://rextester.com/l/cpp_online_compiler_visual but it looks like that wasn't sufficient.

mgorny added a subscriber: rnk.Jan 10 2017, 8:36 AM

This seems to be the source of new test failure on 32-bit x86:

[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from AlignOfTest
[ RUN      ] AlignOfTest.BasicAlignedArray
/var/tmp/portage/sys-devel/llvm-9999/work/llvm-9999/unittests/Support/AlignOfTest.cpp:151: Failure
      Expected: alignof(T<long long>)
      Which is: 4
To be equal to: alignof(AlignedCharArrayUnion<long long>)
      Which is: 8
/var/tmp/portage/sys-devel/llvm-9999/work/llvm-9999/unittests/Support/AlignOfTest.cpp:153: Failure
      Expected: alignof(T<double>)
      Which is: 4
To be equal to: alignof(AlignedCharArrayUnion<double>)
      Which is: 8
[  FAILED  ] AlignOfTest.BasicAlignedArray (1 ms)

It should be noted that this specific commit fails due to assertion:

/home/mgorny/llvm/unittests/Support/AlignOfTest.cpp:110:1: error: static assertion failed: Statically-computed alignment must be right
 static_assert(alignof(U) == alignof(uint64_t), "Statically-computed alignment must be right");
 ^

However, rL291515 (@rnk) following it builds fine and has this test failure.