diff options
author | James Y Knight <jyknight@google.com> | 2016-01-15 16:33:06 +0000 |
---|---|---|
committer | James Y Knight <jyknight@google.com> | 2016-01-15 16:33:06 +0000 |
commit | ac03dca41245fe4cf308e7405d6e947b45341547 (patch) | |
tree | 618e6891cf0bf5d04e375a798b99cd46c101dcbd /llvm/lib/Transforms/Utils/Local.cpp | |
parent | 0771b18257a6bd33ec9f9890d8cb0eed01e729d3 (diff) | |
download | bcm5719-llvm-ac03dca41245fe4cf308e7405d6e947b45341547.tar.gz bcm5719-llvm-ac03dca41245fe4cf308e7405d6e947b45341547.zip |
Stop increasing alignment of externally-visible globals on ELF
platforms.
With ELF, the alignment of a global variable in a shared library will
get copied into an executables linked against it, if the executable even
accesss the variable. So, it's not possible to implicitly increase
alignment based on access patterns, or you'll break existing binaries.
This happened to affect libc++'s std::cout symbol, for example. See
thread: http://thread.gmane.org/gmane.comp.compilers.clang.devel/45311
(This is a re-commit of r257719, without the bug reported in
PR26144. I've tweaked the code to not assert-fail in
enforceKnownAlignment when computeKnownBits doesn't recurse far enough
to find the underlying Alloca/GlobalObject value.)
Differential Revision: http://reviews.llvm.org/D16145
llvm-svn: 257902
Diffstat (limited to 'llvm/lib/Transforms/Utils/Local.cpp')
-rw-r--r-- | llvm/lib/Transforms/Utils/Local.cpp | 33 |
1 files changed, 20 insertions, 13 deletions
diff --git a/llvm/lib/Transforms/Utils/Local.cpp b/llvm/lib/Transforms/Utils/Local.cpp index 92e3a6b118b..3e695539f3e 100644 --- a/llvm/lib/Transforms/Utils/Local.cpp +++ b/llvm/lib/Transforms/Utils/Local.cpp @@ -944,37 +944,44 @@ bool llvm::EliminateDuplicatePHINodes(BasicBlock *BB) { static unsigned enforceKnownAlignment(Value *V, unsigned Align, unsigned PrefAlign, const DataLayout &DL) { + assert(PrefAlign > Align); + V = V->stripPointerCasts(); if (AllocaInst *AI = dyn_cast<AllocaInst>(V)) { + // TODO: ideally, computeKnownBits ought to have used + // AllocaInst::getAlignment() in its computation already, making + // the below max redundant. But, as it turns out, + // stripPointerCasts recurses through infinite layers of bitcasts, + // while computeKnownBits is not allowed to traverse more than 6 + // levels. + Align = std::max(AI->getAlignment(), Align); + if (PrefAlign <= Align) + return Align; + // If the preferred alignment is greater than the natural stack alignment // then don't round up. This avoids dynamic stack realignment. if (DL.exceedsNaturalStackAlignment(PrefAlign)) return Align; - // If there is a requested alignment and if this is an alloca, round up. - if (AI->getAlignment() >= PrefAlign) - return AI->getAlignment(); AI->setAlignment(PrefAlign); return PrefAlign; } if (auto *GO = dyn_cast<GlobalObject>(V)) { + // TODO: as above, this shouldn't be necessary. + Align = std::max(GO->getAlignment(), Align); + if (PrefAlign <= Align) + return Align; + // If there is a large requested alignment and we can, bump up the alignment // of the global. If the memory we set aside for the global may not be the // memory used by the final program then it is impossible for us to reliably // enforce the preferred alignment. - if (!GO->isStrongDefinitionForLinker()) + if (!GO->canIncreaseAlignment()) return Align; - if (GO->getAlignment() >= PrefAlign) - return GO->getAlignment(); - // We can only increase the alignment of the global if it has no alignment - // specified or if it is not assigned a section. If it is assigned a - // section, the global could be densely packed with other objects in the - // section, increasing the alignment could cause padding issues. - if (!GO->hasSection() || GO->getAlignment() == 0) - GO->setAlignment(PrefAlign); - return GO->getAlignment(); + GO->setAlignment(PrefAlign); + return PrefAlign; } return Align; |