Daniel Berlin [Sun, 21 May 2017 23:41:51 +0000 (23:41 +0000)]
SmallPtrSetImpl/SmallPtrSet: Add a public value_type and key_type
llvm-svn: 303518
Michael Kruse [Sun, 21 May 2017 22:46:57 +0000 (22:46 +0000)]
[CodeGen] Support partial write accesses.
Allow the BlockGenerator to generate memory writes that are not defined
over the complete statement domain, but only over a subset of it. It
generates a condition that evaluates to 1 if executing the subdomain,
and only then execute the access.
Only write accesses are supported. Read accesses would require a PHINode
which has a value if the access is not executed.
Partial write makes DeLICM able to apply mappings that are not defined
over the entire domain (for instance, a branch that leaves a loop with
a PHINode in its header; a MemoryKind::PHI write when leaving is never
read by its PHI read).
Differential Revision: https://reviews.llvm.org/D33255
llvm-svn: 303517
Davide Italiano [Sun, 21 May 2017 20:55:53 +0000 (20:55 +0000)]
[NewGVN] Actually check the NewGVN output.
Apparently I messed up squashing two consecutive commits.
llvm-svn: 303516
Davide Italiano [Sun, 21 May 2017 20:50:16 +0000 (20:50 +0000)]
[NewGVN] Add a test for non most dominating leader.
Taken from PR32845. Dan removed the most dominating leader check
in r303443, but we check this test anyway to make sure things
don't regress.
llvm-svn: 303515
Tobias Grosser [Sun, 21 May 2017 20:38:33 +0000 (20:38 +0000)]
[ScopInfo] Translate updateDimensionality to isl C++ [NFC]
llvm-svn: 303514
Davide Italiano [Sun, 21 May 2017 20:30:27 +0000 (20:30 +0000)]
[InstCombine] Take in account the size in sext->lshr->trunc patterns.
Otherwise we end up miscompiling, transforming:
define i8 @tinky() {
%sext = sext i1 1 to i16
%hibit = lshr i16 %sext, 15
%tr = trunc i16 %hibit to i8
ret i8 %tr
}
into:
%sext = sext i1 1 to i8
ret i8 %sext
and the first get folded to ret i8 1, while the second gets folded
to ret i8 -1.
Eventually we should get rid of this transform entirely, but for now,
this at least fixes a know correctness bug.
Differential Revision: https://reviews.llvm.org/D33338
llvm-svn: 303513
Tobias Grosser [Sun, 21 May 2017 20:23:26 +0000 (20:23 +0000)]
[isl++] add isl_constraint to C++ bindings [NFC]
llvm-svn: 303512
Tobias Grosser [Sun, 21 May 2017 20:23:23 +0000 (20:23 +0000)]
[ScopInfo] Translate wrapConstantDimensions to isl C++ [NFC]
llvm-svn: 303511
Tobias Grosser [Sun, 21 May 2017 20:23:20 +0000 (20:23 +0000)]
[ScopInfo] Translate addRangeBoundsToSet to isl C++ [NFC]
llvm-svn: 303510
Tobias Grosser [Sun, 21 May 2017 16:34:09 +0000 (16:34 +0000)]
Fix include order to stop clang-format complains
llvm-svn: 303509
Tobias Grosser [Sun, 21 May 2017 16:21:33 +0000 (16:21 +0000)]
[ScheduleOptimizer] Move schedule construction to isl C++ [NFC]
llvm-svn: 303508
Tobias Grosser [Sun, 21 May 2017 16:12:21 +0000 (16:12 +0000)]
[Simplify] Move to isl C++
llvm-svn: 303507
Tobias Grosser [Sun, 21 May 2017 16:00:32 +0000 (16:00 +0000)]
[isl++] Export (almost) all functions from isl
This commit exports the majority of the isl functions to the isl C++ interface.
The official isl C++ bindings still require discussions to define the set of
functions that are officially supported. As a result, the officially exported
functionality will be rather limited until these discussions conclude and a
non-trivial set of isl functions is officially supported through the isl C++
bindings. Starting from this commit we ship with Polly an extended version of
the official isl C++ bindings to ensure sufficient functionality is available
such that LLVM developers can make efficient use of isl through C++. The
practical experience Polly gathers with its bindings will then be used to
gradually upstream patches to isl to extend the official bindings.
llvm-svn: 303506
Tobias Grosser [Sun, 21 May 2017 15:59:15 +0000 (15:59 +0000)]
[isl++] Rebase isl C++ bindings on top of
29aee98ce
This reduces the diff to the official isl C++ bindings and solves a correctness
issue with isl::booleans, where isl_bool_error results were accidentally
converted to isl::boolean::true.
llvm-svn: 303505
Sanjay Patel [Sun, 21 May 2017 15:18:52 +0000 (15:18 +0000)]
[InstCombine] add tests for potential (lshr(sext X), C) folds; NFC
As discussed in:
https://reviews.llvm.org/D33338
...we may be able to remove a wider pattern match by doing these more
basic canonicalizations.
llvm-svn: 303504
Tobias Grosser [Sun, 21 May 2017 13:16:05 +0000 (13:16 +0000)]
[isl++] Move isl raw_ostream printers into separate header
Instead of relying on these functions to be part of the isl C++ bindings, we
just define this functionality independently. This allows us to use isl C++
bindings that do not contain LLVM specific functionality.
llvm-svn: 303503
Igor Breger [Sun, 21 May 2017 11:13:56 +0000 (11:13 +0000)]
[GlobalISel][X86] Fix G_TRUNC instruction selection.
Updated tests with -verify-machineinstrs flag.
It fixes 3 tests failed with machine verifier enabled and listed
in PR27481
llvm-svn: 303502
Craig Topper [Sun, 21 May 2017 07:29:07 +0000 (07:29 +0000)]
[Format] Add curly braces to suppress a -Wmisleading-indentation warning from gcc.
llvm-svn: 303501
Hiroshi Inoue [Sun, 21 May 2017 06:00:05 +0000 (06:00 +0000)]
Summary
PPC backend eliminates compare instructions by using record-form instructions in PPCInstrInfo::optimizeCompareInstr, which is called from peephole optimization pass.
This patch improves this optimization to eliminate more compare instructions in two types of common case.
- comparison against a constant 1 or -1
The record-form instructions set CR bit based on signed comparison against 0. So, the current implementation does not exploit the record-form instruction for comparison against a non-zero constant.
This patch enables record-form optimization for constant of 1 or -1 if possible; it changes the condition "greater than -1" into "greater than or equal to 0" and "less than 1" into "less than or equal to 0".
With this patch, compare can be eliminated in the following code sequence, as an example.
uint64_t a, b;
if ((a | b) & 0x8000000000000000ull) { ... }
else { ... }
- andi for 32-bit comparison on PPC64
Since record-form instructions execute 64-bit signed comparison and so we have limitation in eliminating 32-bit comparison, i.e. with cmplwi, using the record-form. The original implementation already has such checks but andi. is not recognized as an instruction which executes implicit zero extension and hence safe to convert into record-form if used for equality check.
%1 = and i32 %a, 10
%2 = icmp ne i32 %1, 0
br i1 %2, label %foo, label %bar
In this simple example, LLVM generates andi. + cmplwi + beq on PPC64.
This patch make it possible to eliminate the cmplwi for this case.
I added andi. for optimization targets if it is safe to do so.
Differential Revision: https://reviews.llvm.org/D30081
llvm-svn: 303500
George Burgess IV [Sun, 21 May 2017 05:31:29 +0000 (05:31 +0000)]
[Docs] Fix LangRef links referred in GetElementPtr.rst
Patch by chenwj!
Differential Revision: https://reviews.llvm.org/D32929
llvm-svn: 303499
Sanjoy Das [Sun, 21 May 2017 05:02:12 +0000 (05:02 +0000)]
Revert "[SCEV] Clarify behavior around max backedge taken count"
This reverts commit r303497 since it breaks the msan bootstrap bot:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-bootstrap/builds/1379/
llvm-svn: 303498
Sanjoy Das [Sun, 21 May 2017 01:47:50 +0000 (01:47 +0000)]
[SCEV] Clarify behavior around max backedge taken count
This change makes the split between the "exact" backedge taken count
and the "maximum" backedge taken count a bit more obvious. Both of
these are upper bounds on the number of times the loop header
executes (since SCEV does not account for most kinds of abnormal
control flow), but the latter is guaranteed to be a constant.
There were a few places where the max backedge taken count *was* a
non-constant; I've changed those to compute constants instead.
At this point, I'm not sure if the constant max backedge count can be
computed by calling `getUnsignedRange(Exact).getUnsignedMax()` without
losing precision. If it can, we can simplify even further by making
`getMaxBackedgeTakenCount` a thin wrapper around
`getBackedgeTakenCount` and `getUnsignedRange`.
llvm-svn: 303497
Xin Tong [Sun, 21 May 2017 00:37:55 +0000 (00:37 +0000)]
Revert "Add pthread_self function prototype and make it speculatable."
This reverts commit
143d7445b5dfa2f6d6c45bdbe0433d9fc531be21.
Build breaking
llvm-svn: 303496
Xin Tong [Sat, 20 May 2017 22:40:25 +0000 (22:40 +0000)]
Add pthread_self function prototype and make it speculatable.
Summary: This allows pthread_self to be pulled out of a loop by LICM.
Reviewers: hfinkel, arsenm, davide
Reviewed By: davide
Subscribers: davide, wdng, llvm-commits
Differential Revision: https://reviews.llvm.org/D32782
llvm-svn: 303495
Martell Malone [Sat, 20 May 2017 21:54:15 +0000 (21:54 +0000)]
COFF: Fix another StringRef return error
This should appease the lld build bot regression
Following up on rL303493
llvm-svn: 303494
Martell Malone [Sat, 20 May 2017 21:00:36 +0000 (21:00 +0000)]
COFF: Fix single StringRef return error
This should appease the lld build bot regression
Intrroduced by rL303490
llvm-svn: 303493
Faisal Vali [Sat, 20 May 2017 19:58:04 +0000 (19:58 +0000)]
Fix PR25627: constant expressions being odr-used in template arguments.
This patch ensures that clang processes the expression-nodes that are generated when disambiguating between types and expressions within template arguments as constant-expressions by installing the ConstantEvaluated ExpressionEvaluationContext just before attempting the disambiguation - and then making sure that Context carries through into ParseConstantExpression (by refactoring it out into a function that does not create its own EvaluationContext: ParseConstantExpressionInExprEvalContext)
Note, prior to this patch, trunk would correctly disambiguate and identify the expression as an expression - and while it would annotate the token with the expression - it would fail to complete the odr-use processing (specifically, failing to trigger Sema::UpdateMarkingForLValueToRValue as is done for all Constant Expressions, which would remove it from being considered odr-used). By installing the ConstantExpression Evaluation Context prior to disambiguation, and making sure it carries though, we ensure correct processing of the expression-node.
For e.g:
template<int> struct X { };
void f() {
const int N = 10;
X<N> x; // should be OK.
[] { return X<N>{}; }; // Should be OK - no capture - but clang errors!
}
See a related bug: https://bugs.llvm.org//show_bug.cgi?id=25627
In summary (and reiteration), the fix is as follows:
- Remove the EnteredConstantEvaluatedContext action from ParseTemplateArgumentList (relying on ParseTemplateArgument getting it right)
- Add the EnteredConstantEvaluatedContext action just prior to undergoing the disambiguating parse, and if the parse succeeds for an expression, carry the context though into a refactored version of ParseConstantExpression that does not create its own ExpressionEvaluationContext.
See https://reviews.llvm.org/D31588 for additional context regarding some of the more fragile and complicated approaches attempted, and Richard's feedback that eventually shaped the simpler and more robust rendition that is being committed.
Thanks Richard!
llvm-svn: 303492
Martell Malone [Sat, 20 May 2017 19:56:44 +0000 (19:56 +0000)]
COFF: migrate def parser from LLD to LLVM [2/2]
This is split up into two commits.
This commit removes the DEF parser from LLD
See the previous commit for the creation in LLVM.
Reviewers: ruiu
Differential Revision: https://reviews.llvm.org/D32689
llvm-svn: 303491
Martell Malone [Sat, 20 May 2017 19:56:29 +0000 (19:56 +0000)]
COFF: migrate def parser from LLD to LLVM [1/2]
This is split up into two commits.
The will create the DEF parser in LLVM.
Check the following commit to see the removal from LLD
Reviewers: ruiu
Differential Revision: https://reviews.llvm.org/D32689
llvm-svn: 303490
Xin Tong [Sat, 20 May 2017 19:40:24 +0000 (19:40 +0000)]
Fix test typo. NFC
llvm-svn: 303489
Piotr Padlewski [Sat, 20 May 2017 08:56:18 +0000 (08:56 +0000)]
[Devirtualization] insert placement new barrier with -O0
Summary:
To not break LTO with different optimizations levels, we should insert
the barrier regardles of optimization level.
Reviewers: rjmccall, rsmith, mehdi_amini
Reviewed By: mehdi_amini
Subscribers: mehdi_amini, cfe-commits
Differential Revision: https://reviews.llvm.org/D32401
llvm-svn: 303488
Duncan P. N. Exon Smith [Sat, 20 May 2017 05:25:19 +0000 (05:25 +0000)]
Docs: Fix pluralization in CMake docs
Fixed "most platform" to read "most platforms", and changed "are not
using the same" to "use different" as a drive-by.
llvm-svn: 303487
George Burgess IV [Sat, 20 May 2017 04:52:29 +0000 (04:52 +0000)]
[docs] Small style nits.
llvm-svn: 303486
Tobias Grosser [Sat, 20 May 2017 04:22:26 +0000 (04:22 +0000)]
Fix buildbots after r303429
A test case with a GPU runline was added without setting 'REQUIRES=pollyacc'. We
drop the GPU run line, as the basic functionality can already be tested with
the normal code generation.
llvm-svn: 303485
Argyrios Kyrtzidis [Sat, 20 May 2017 04:11:33 +0000 (04:11 +0000)]
[index] Fix forward declarations interfering with USR generation of external source symbols
Patch by Nathan Hawes.
https://reviews.llvm.org/D33346
llvm-svn: 303484
David Blaikie [Sat, 20 May 2017 03:32:51 +0000 (03:32 +0000)]
Fix -Wunneeded-internal-declaration by removing constant arrays only used in sizeof expressions, in favor of constants containing the size directly
llvm-svn: 303483
David Blaikie [Sat, 20 May 2017 03:32:49 +0000 (03:32 +0000)]
llvm-symbolizer: Support multiple CUs in a single DWO file
llvm-svn: 303482
Vitaly Buka [Sat, 20 May 2017 02:09:25 +0000 (02:09 +0000)]
Revert "[compiler-rt] Change default of allow_user_segv_handler to true"
Failed libFuzzer tests on Windows.
This reverts commit r303476.
llvm-svn: 303481
Eric Beckmann [Sat, 20 May 2017 01:49:19 +0000 (01:49 +0000)]
Add functionality to cvtres to parse all entries in res file.
Summary: Added the new modules in the Object/ folder. Updated the
llvm-cvtres interface as well, and added additional tests.
Subscribers: llvm-commits, mgorny
Differential Revision: https://reviews.llvm.org/D33180
llvm-svn: 303480
Richard Smith [Sat, 20 May 2017 01:36:41 +0000 (01:36 +0000)]
Rename RAII objects for performing eager instantiation to have names that
describe what they're for, not how they do it, and factor out a bit more
common code into them.
llvm-svn: 303479
Matthias Braun [Sat, 20 May 2017 01:29:55 +0000 (01:29 +0000)]
CodeGenModule: Always output wchar_size, check LLVM assumptions.
Re-commit r303463 now that LLVM is fixed and adjust some lit tests.
llvm::TargetLibraryInfo needs to know the size of wchar_t to work on
functions like `wcslen`. This patch changes clang to always emit the
wchar_size module flag (it would only do so for ARM previously).
This also adds an `assert()` to ensure the LLVM defaults based on the
target triple are in sync with clang.
Differential Revision: https://reviews.llvm.org/D32982
llvm-svn: 303478
Matthias Braun [Sat, 20 May 2017 01:28:52 +0000 (01:28 +0000)]
Fix breakage after r303461
- Improve wchar_t size predicitions based on target triple.
- Be less strict in wchar_t size verifier.
llvm-svn: 303477
Vitaly Buka [Sat, 20 May 2017 01:04:59 +0000 (01:04 +0000)]
[compiler-rt] Change default of allow_user_segv_handler to true
Reviewers: eugenis
Subscribers: srhines, kubamracek, llvm-commits
Differential Revision: https://reviews.llvm.org/D32443
llvm-svn: 303476
Davide Italiano [Sat, 20 May 2017 00:46:54 +0000 (00:46 +0000)]
[NewGVN] Create a StoreExpression instead of a VariableExpression.
In the case where we have an operand defined by a lod of the
same memory location. Historically this was a VariableExpression
because we wanted to make sure they ended up in the same class,
but if we create the right expression, they end up in the same
class anyway.
Fixes PR32897. Thanks to Dan for the detailed discussion and the
fix suggestion.
llvm-svn: 303475
Matthias Braun [Sat, 20 May 2017 00:38:27 +0000 (00:38 +0000)]
Revert "CodeGenModule: Always output wchar_size, check LLVM assumptions."
Let's revert this for now (and with it the assert()) to get the bots
back to green until I have LLVM synced up properly.
This reverts commit r303463.
llvm-svn: 303474
Davide Italiano [Sat, 20 May 2017 00:24:04 +0000 (00:24 +0000)]
[NewGVN] Get rid of an assertion.
This was here because we don't want to switch leaders too much,
in order to avoid fixpoint(ing) issue, but it's not sure if it
matters in practice.
A first step towards fixing PR32897.
llvm-svn: 303473
Hubert Tong [Sat, 20 May 2017 00:21:55 +0000 (00:21 +0000)]
Fix valid-for-expr ellipses eaten as invalid decl
Summary:
The trial parse for declarative syntax accepts an invalid pack
declaration syntax, which is ambiguous with valid pack expansions of
expressions. This commit removes the invalid pack declaration syntax to
avoid mistaking valid pack expansions as invalid declarator components.
Additionally, the trial parse of a //template-argument-list// then needs
to handle the optional ellipsis that is part of that grammar, as opposed
to relying on the trial parse for declarators accepting stray ellipses.
Reviewers: rsmith, rcraik, aaron.ballman
Reviewed By: rsmith
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D33339
llvm-svn: 303472
Galina Kistanova [Sat, 20 May 2017 00:02:08 +0000 (00:02 +0000)]
Cosmetic. Added braces to address gcc warning: suggest explicit braces to avoid ambiguous 'else' [-Wdangling-else].
llvm-svn: 303471
Adrian Prantl [Sat, 20 May 2017 00:00:08 +0000 (00:00 +0000)]
Revert "Revert "ThinLTO: Verify bitcode before lauching the ThinLTOCodeGenerator.""
This reapplies commit r303438 modified to not verify cross-imported
bitcode in FunctionImporter.
rdar://problem/
31233625
Differential Revision: https://reviews.llvm.org/D33370
llvm-svn: 303470
Richard Smith [Fri, 19 May 2017 23:49:00 +0000 (23:49 +0000)]
Remove last (unnecessary) use of mapping from SourceLocation to Module and
remove the mechanism for doing so.
This mechanism was incorrect in the presence of preprocessed modules (and
#pragma clang module begin/end).
llvm-svn: 303469
Richard Smith [Fri, 19 May 2017 23:32:38 +0000 (23:32 +0000)]
[modules] Simplify module macro handling in non-local-submodule-visibility mode.
When reaching the end of a module, we used to convert its macros to
ModuleMacros but also leave them in the MacroDirective chain for the
identifier. This meant that every lookup of such a macro would find two
(identical) definitions. It also made it difficult to determine the correct
owner for a macro when reaching the end of a module: the most recent
MacroDirective in the chain could be from an #included submodule rather than
the current module.
Simplify this: whenever we convert a MacroDirective to a ModuleMacro when
leaving a module, clear out the MacroDirective chain for that identifier, and
just rely on the ModuleMacro to provide the macro definition information.
(We don't want to do this for local submodule visibility mode, because in that
mode we maintain a distinct MacroDirective chain for each submodule, and we
need to keep around the prior MacroDirective in case we re-enter the submodule
-- for instance, if its header is #included more than once in a module build,
we need the include guard directive to stick around. But the problem doesn't
arise in this case for the same reason: each submodule has its own
MacroDirective chain, so the macros don't leak out of submodules in the first
place.)
This reinstates r302932, reverted in r302947, with a fix for a bug that
resulted in us sometimes losing macro definitions due to failing to clear out
the overridden module macro list when promoting a directive to a module macro.
llvm-svn: 303468
Adrian Prantl [Fri, 19 May 2017 23:32:21 +0000 (23:32 +0000)]
Revert "ThinLTO: Verify bitcode before lauching the ThinLTOCodeGenerator."
This reverts commit r303438 while deliberating buildbot breakage.
llvm-svn: 303467
Erik Pilkington [Fri, 19 May 2017 23:02:49 +0000 (23:02 +0000)]
[test] Add support for Objective-C++ tests
Differential revision: https://reviews.llvm.org/D33049
llvm-svn: 303466
Jason Molenda [Fri, 19 May 2017 23:00:42 +0000 (23:00 +0000)]
Don't override the SDKs and force the use of /System/Library/Frameworks to
find the python headers when buiding crashinfo.c
<rdar://problem/
32288076>
llvm-svn: 303465
Vitaly Buka [Fri, 19 May 2017 22:37:16 +0000 (22:37 +0000)]
[compiler-rt] Switch handle_<signal> flags from bool to enum.
Summary: We are going to make it tri-state and remove allow_user_segv_handler.
Reviewers: eugenis, alekseys, kcc
Subscribers: kubamracek, dberris, llvm-commits
Differential Revision: https://reviews.llvm.org/D33159
llvm-svn: 303464
Matthias Braun [Fri, 19 May 2017 22:37:15 +0000 (22:37 +0000)]
CodeGenModule: Always output wchar_size, check LLVM assumptions.
llvm::TargetLibraryInfo needs to know the size of wchar_t to work on
functions like `wcslen`. This patch changes clang to always emit the
wchar_size module flag (it would only do so for ARM previously).
This also adds an `assert()` to ensure the LLVM defaults based on the
target triple are in sync with clang.
Differential Revision: https://reviews.llvm.org/D32982
llvm-svn: 303463
Vitaly Buka [Fri, 19 May 2017 22:37:13 +0000 (22:37 +0000)]
[compiler-rt] Add negative test for boolean flags.
Reviewers: eugenis, alekseyshl
Subscribers: kubamracek, dberris, llvm-commits
Differential Revision: https://reviews.llvm.org/D33344
llvm-svn: 303462
Matthias Braun [Fri, 19 May 2017 22:37:09 +0000 (22:37 +0000)]
SimplifyLibCalls: Optimize wcslen
Refactor the strlen optimization code to work for both strlen and wcslen.
This especially helps with programs in the wild where people pass
L"string"s to const std::wstring& function parameters and the wstring
constructor gets inlined.
This also fixes a lingerind API problem/bug in getConstantStringInfo()
where zeroinitializers would always give you an empty string (without a
length) back regardless of the actual length of the initializer which
did not work well in the TrimAtNul==false causing the PR mentioned
below.
Note that the fixed getConstantStringInfo() needed fixes to SelectionDAG
memcpy lowering and may lead to some cases for out-of-bounds
zeroinitializer accesses not getting optimized anymore. So some code
with UB may produce out of bound memory reads now instead of just
producing zeros.
The refactoring "accidentally" fixes http://llvm.org/PR32124
Differential Revision: https://reviews.llvm.org/D32839
llvm-svn: 303461
Matthias Braun [Fri, 19 May 2017 22:37:01 +0000 (22:37 +0000)]
Verifier: Check wchar_size module flag.
Differential Revision: https://reviews.llvm.org/D32974
llvm-svn: 303460
Richard Trieu [Fri, 19 May 2017 22:35:41 +0000 (22:35 +0000)]
[ODRHash] Revert r303450 to fix buildbot
r303450
[ODRHash] Support TemplateName and TemplateArgument
llvm-svn: 303459
Reid Kleckner [Fri, 19 May 2017 22:23:47 +0000 (22:23 +0000)]
Fix off-by-one bug in AttributeList::addAttributes index handling
getParamAlignment expects an argument number, not an AttributeList
index.
Johan Englan, who works on LDC, found this bug and told me about it off
list.
llvm-svn: 303458
Galina Kistanova [Fri, 19 May 2017 21:08:28 +0000 (21:08 +0000)]
Added LLVM_FALLTHROUGH to address gcc warning: this statement may fall through.
llvm-svn: 303457
Evgeniy Stepanov [Fri, 19 May 2017 20:58:48 +0000 (20:58 +0000)]
[safestack] Disable stack coloring by default.
Workaround for apparent miscompilation of PR32143.
llvm-svn: 303456
Galina Kistanova [Fri, 19 May 2017 20:31:51 +0000 (20:31 +0000)]
Added missing break.
llvm-svn: 303454
Daniel Berlin [Fri, 19 May 2017 20:22:20 +0000 (20:22 +0000)]
NewGVN: Fix PR32838.
This is a complicated bug involving two issues:
1. What do we do with phi nodes when we prove all arguments are not
live?
2. When is it safe to use value leaders to determine if we can ignore
an argumnet?
llvm-svn: 303453
Daniel Berlin [Fri, 19 May 2017 20:22:14 +0000 (20:22 +0000)]
NewGVN: Print out the StoredValue of a StoreExpression
llvm-svn: 303452
Richard Smith [Fri, 19 May 2017 20:20:13 +0000 (20:20 +0000)]
Do not issue -Wnullability-completeness for dependent types that are not written as pointer types.
llvm-svn: 303451
Richard Trieu [Fri, 19 May 2017 20:05:16 +0000 (20:05 +0000)]
[ODRHash] Support TemplateName and TemplateArgument
llvm-svn: 303450
Simon Pilgrim [Fri, 19 May 2017 19:47:29 +0000 (19:47 +0000)]
Fix line-endings.
llvm-svn: 303448
Davide Italiano [Fri, 19 May 2017 19:41:11 +0000 (19:41 +0000)]
[InstCombine] *Actually* commit the test showing the miscompile.
Clarify a comment while I'm here.
llvm-svn: 303447
Zachary Turner [Fri, 19 May 2017 19:26:58 +0000 (19:26 +0000)]
Resubmit "[CodeView] Provide a common interface for type collections."
This was originally reverted because it was a breaking a bunch
of bots and the breakage was not surfacing on Windows. After much
head-scratching this was ultimately traced back to a bug in the
lit test runner related to its pipe handling. Now that the bug
in lit is fixed, Windows correctly reports these test failures,
and as such I have finally (hopefully) fixed all of them in this
patch.
llvm-svn: 303446
Davide Italiano [Fri, 19 May 2017 19:23:24 +0000 (19:23 +0000)]
[InstCombine] Add tests to demonstrate the miscompile in PR33078.
llvm-svn: 303445
Daniel Berlin [Fri, 19 May 2017 19:01:27 +0000 (19:01 +0000)]
Last of the major pieces to NewGVN - yay!
Summary:
NewGVN: Handle equivalence between phi of ops and op of phis.
This makes our GVN mostly-complete. It would be complete, modulo some
deliberate choices we make. This means it detects roughly all herband
equivalences in polynomial time, including cases notoriously hard for
other GVN's to detect. It also detects a very large swath of the
cases we currently rely on instcombine to detect that involve folding
upwards through phis.
Fixes PR 31125, 31463, PR 31868
Reviewers: davide
Subscribers: Prazek, llvm-commits
Differential Revision: https://reviews.llvm.org/D32151
llvm-svn: 303444
Daniel Berlin [Fri, 19 May 2017 19:01:24 +0000 (19:01 +0000)]
NewGVN: Get rid of most dominating leader check
llvm-svn: 303443
Daniel Berlin [Fri, 19 May 2017 19:01:21 +0000 (19:01 +0000)]
BasicAA: Uninserted instructions have no parent, and notDifferentParent explicitly allows for this case, but getParent crashes when handed one.
llvm-svn: 303442
Amaury Sechet [Fri, 19 May 2017 18:20:44 +0000 (18:20 +0000)]
[DAGCombine] (addcarry 0, 0, X) -> (ext/trunc X)
Summary:
While this makes some case better and some case worse - so it's unclear if it is a worthy combine just by itself - this is a useful canonicalisation.
As per discussion in D32756 .
Reviewers: jyknight, nemanjai, mkuper, spatel, RKSimon, zvi, bkramer
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D32916
llvm-svn: 303441
Zachary Turner [Fri, 19 May 2017 18:12:07 +0000 (18:12 +0000)]
[lit] Take the last error when executing pipelines.
This seems to have been present since the beginning of time,
which is quite surprising. The symptom was this: Suppose you
have a test with a run line that looks like this:
RUN: foo | FileCheck %s
foo prints some output and then due to a bug in the program it
asserts. On Windows this results in the program returning a
negative exit code. But if enough output had been printed
already by the tool so that the FileCheck match would succeed
then FileCheck would return 0, and because of bad logic in
lit this 0 return value would overwrite the failed return
value from previous items in the pipeline. This only happened
with negative exit codes.
The most sensible behavior is to just take whatever the first
exit code is. There is no logical ordering defined on exit
codes, so comparing with < and > does not make a lot of sense.
Instead, as soon as we find the first non-successful return
value, that should be the result of the entire expression.
This fixes the issue, as now tests which fail on non-Windows
platforms also fail for me on Windows as well.
llvm-svn: 303440
Anna Thomas [Fri, 19 May 2017 18:00:30 +0000 (18:00 +0000)]
[NFC][loopIdiom] Clang format change rL303434
llvm-svn: 303439
Adrian Prantl [Fri, 19 May 2017 17:55:02 +0000 (17:55 +0000)]
ThinLTO: Verify bitcode before lauching the ThinLTOCodeGenerator.
rdar://problem/
31233625
Differential Revision: https://reviews.llvm.org/D33151
llvm-svn: 303438
Adrian Prantl [Fri, 19 May 2017 17:54:58 +0000 (17:54 +0000)]
Rewrite llvm-lto's codegen() using ThinCodeGenerator::run(). NFC-ish.
Differential Revision: https://reviews.llvm.org/D33360
llvm-svn: 303437
Matthias Braun [Fri, 19 May 2017 17:25:20 +0000 (17:25 +0000)]
Fix typo in test
llvm-svn: 303436
Simon Pilgrim [Fri, 19 May 2017 17:19:26 +0000 (17:19 +0000)]
[X86][FMA] Tests showing missed fmsubadd opportunities (PR30633)
llvm-svn: 303435
Anna Thomas [Fri, 19 May 2017 17:05:36 +0000 (17:05 +0000)]
[LoopIdiom] Refactor return value of isLegalStore [NFC]
Summary:
This NFC simply refactors the return value of LoopIdiomRecognize::isLegalStore() from bool to an enumeration, and
removes the return-through-parameter mechanism that the function was using. This function is constructed such that it will
only ever recognize a single store idiom (memset, memset_pattern, or memcpy), and never a combination of these. As such it
makes much more sense for the return value to be the single idiom that the store matches, rather than
having a separate argument-return for each idiom -- it's cleaner, and makes it clearer that
only a single idiom can be matched.
Patch by Daniel Neilson!
Reviewers: anna, sanjoy, davide, haicheng
Reviewed By: anna, haicheng
Subscribers: haicheng, mzolotukhin, llvm-commits
Differential Revision: https://reviews.llvm.org/D33359
llvm-svn: 303434
Craig Topper [Fri, 19 May 2017 16:56:53 +0000 (16:56 +0000)]
[InstSimplify] Fix 80 column violation. NFC
llvm-svn: 303433
Vassil Vassilev [Fri, 19 May 2017 16:46:06 +0000 (16:46 +0000)]
[modules] Further delay calling DeclMustBeEmitted until it's safe.
As discussed in D30793, we have some unsafe calls to isConsumerInterestedIn().
This patch implements Richard's suggestion (from the inline comment) that we
should track if we just deserialized an declaration. If we just deserialized,
we can skip the unsafe call because we know it's interesting. If we didn't just
deserialize the declaration, calling isConsumerInterestedIn() should be safe.
We tried to create a test case for this but we were not successful.
Patch by Raphael Isemann (D32499)!
llvm-svn: 303432
Craig Topper [Fri, 19 May 2017 16:43:54 +0000 (16:43 +0000)]
[APInt] Add support for dividing or remainder by a uint64_t or int64_t.
Summary:
This patch adds udiv/sdiv/urem/srem/udivrem/sdivrem methods that can divide by a uint64_t. This makes division consistent with all the other arithmetic operations.
This modifies the interface of the divide helper method to work on raw arrays instead of APInts. This way we can pass the uint64_t in for the RHS without wrapping it in an APInt. This required moving all the Quotient and Remainder allocation handling up to the callers. For udiv/urem this was as simple as just creating the Quotient/Remainder with the right size when they were declared. For udivrem we have to rely on reallocate not changing the contents of the variable LHS or RHS is aliased with the Quotient or Remainder APInts. We also have to zero the upper bits of Remainder and Quotient that divide doesn't write to if lhsWords/rhsWords is smaller than the width.
I've update the toString method to use the new udivrem.
Reviewers: hans, dblaikie, RKSimon
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D33310
llvm-svn: 303431
Siddharth Bhat [Fri, 19 May 2017 15:07:45 +0000 (15:07 +0000)]
[Fortran Support] Materialize outermost dimension for Fortran array.
- We use the outermost dimension of arrays since we need this
information to generate GPU transfers.
- In general, if we do not know the outermost dimension of the array
(because the indexing expression is non-affine, for example) then we
simply cannot generate transfer code.
- However, for Fortran arrays, we can use the Fortran array
representation which stores the dimensions of all arrays.
- This patch uses the Fortran array representation to generate code that
computes the outermost dimension size.
Differential Revision: https://reviews.llvm.org/D32967
llvm-svn: 303429
Dmitry Preobrazhensky [Fri, 19 May 2017 14:27:52 +0000 (14:27 +0000)]
[AMDGPU][MC] Corrected disassembler to decode instructions with 2 literals
See bug 32922: https://bugs.llvm.org//show_bug.cgi?id=32922
Reviewers: artem.tamazov, vpykhtin
Differential Revision: https://reviews.llvm.org/D32912
llvm-svn: 303428
Artur Pilipenko [Fri, 19 May 2017 14:02:46 +0000 (14:02 +0000)]
[LoopPredication] NFC. Extract LoopICmp struct and parseLoopICmp helper
llvm-svn: 303427
Artur Pilipenko [Fri, 19 May 2017 14:00:58 +0000 (14:00 +0000)]
[LoopPredication] NFC. Extract LoopPredication::expandCheck helper
llvm-svn: 303426
Artur Pilipenko [Fri, 19 May 2017 14:00:04 +0000 (14:00 +0000)]
[LoopPredication] NFC. Extract CanExpand helper lambda
llvm-svn: 303425
Artur Pilipenko [Fri, 19 May 2017 13:59:34 +0000 (13:59 +0000)]
[LoopPredication] NFC. Add an early exit if there is no guards in the loop
llvm-svn: 303424
Dmitry Preobrazhensky [Fri, 19 May 2017 13:36:09 +0000 (13:36 +0000)]
[AMDGPU][MC] Fixed bugs in export instruction
See Bugs 33019, 33056:
https://bugs.llvm.org//show_bug.cgi?id=33019
https://bugs.llvm.org//show_bug.cgi?id=33056
Reviewers: artem.tamazov, vpykhtin
Differential Revision: https://reviews.llvm.org/D33288
llvm-svn: 303423
Francis Ricci [Fri, 19 May 2017 13:34:02 +0000 (13:34 +0000)]
Use write instead of read permissions to check for global sections on mac
Summary:
The LINKEDIT section is very large and is read-only. Scanning this
section caused LSan on darwin to be very slow. When only writable sections
are scanned for global pointers, performance improved by a factor of about 25x.
Reviewers: alekseyshl, kubamracek
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D33322
llvm-svn: 303422
Guy Blank [Fri, 19 May 2017 12:35:15 +0000 (12:35 +0000)]
[X86][AVX512] Make i1 illegal in the CodeGen
This patch defines the i1 type as illegal in the X86 backend for AVX512.
For DAG operations on <N x i1> types (build vector, extract vector element, ...) i8 is used, and should be truncated/extended.
This should produce better scalar code for i1 types since GPRs will be used instead of mask registers.
Differential Revision: https://reviews.llvm.org/D32273
llvm-svn: 303421
Tobias Grosser [Fri, 19 May 2017 12:13:02 +0000 (12:13 +0000)]
[ScopDetection] Allow detection of full functions
This is useful when only analyzing functions.
llvm-svn: 303420
Krzysztof Parzyszek [Fri, 19 May 2017 12:03:34 +0000 (12:03 +0000)]
Restore and update documentation comment for EmitPointerWithAlignment
llvm-svn: 303419
Daniel Sanders [Fri, 19 May 2017 11:08:33 +0000 (11:08 +0000)]
[globalisel][tablegen] Demote OptForSize/OptForMinSize/ForCodeSize to per-function predicates.
Summary:
This causes them to be re-computed more often than necessary but resolves
objections that were raised post-commit on r301750.
Reviewers: qcolombet, ab, t.p.northover, rovka, kristof.beyls
Reviewed By: qcolombet
Subscribers: igorb, llvm-commits
Differential Revision: https://reviews.llvm.org/D32861
llvm-svn: 303418
Amara Emerson [Fri, 19 May 2017 10:40:18 +0000 (10:40 +0000)]
Fix vector pass-through value being unused in IRBuilder::CreateMaskedGather
Also s/0/nullptr in the call site in LV.
llvm-svn: 303416
Krasimir Georgiev [Fri, 19 May 2017 10:34:57 +0000 (10:34 +0000)]
[clang-format] Handle trailing comment sections in import statement lines
Summary:
This patch updates the handling of multiline trailing comment sections in
import statement lines to make it more consistent with the case in general.
This includes updating the parsing logic to collect the trailing comment
sections and the formatting logic to not insert escaped newlines at the end of
comment lines in import statement lines.
Specifically, before this patch this code:
```
#include <a> // line 1
// line 2
```
will be turned into two unwrapped lines, whereas this code:
```
int i; // line 1
// line 2
```
is turned into a single unwrapped line, enabling reflowing across comments.
An example where the old behaviour is bad is when partially formatting the lines
3 to 4 of this code:
```
#include <a> // line 1
// line 2
int i;
```
which gets turned into:
```
#include <a> // line 1
// line 2
int i;
```
because the two comment lines were independent and the indent was copied.
Reviewers: djasper
Reviewed By: djasper
Subscribers: cfe-commits, klimek
Differential Revision: https://reviews.llvm.org/D33351
llvm-svn: 303415