Jun Bum Lim [Mon, 1 Feb 2016 20:55:11 +0000 (20:55 +0000)]
Avoid inlining call sites in unreachable-terminated block
Summary:
If the normal destination of the invoke or the parent block of the call site is unreachable-terminated, there is little point in inlining the call site unless there is literally zero cost. Unlike my previous change (D15289), this change specifically handle the call sites followed by unreachable in the same basic block for call or in the normal destination for the invoke. This change could be a reasonable first step to conservatively inline call sites leading to an unreachable-terminated block while BFI / BPI is not yet available in inliner.
Reviewers: manmanren, majnemer, hfinkel, davidxl, mcrosier, dblaikie, eraman
Subscribers: dblaikie, davidxl, mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D16616
llvm-svn: 259403
Chad Rosier [Mon, 1 Feb 2016 20:54:36 +0000 (20:54 +0000)]
Remove extra semicolon. NFC.
llvm-svn: 259402
Sanjoy Das [Mon, 1 Feb 2016 20:48:14 +0000 (20:48 +0000)]
[SCEV] Clean up isKnownPredicateViaConstantRanges; NFCI
- ScalarEvolution::isKnownPredicateViaConstantRanges duplicates some
logic already present in ConstantRange, use ConstantRange for those
bits.
- In some cases ScalarEvolution::isKnownPredicateViaConstantRanges
returns `false` to mean "definitely false" (e.g. see the
`LHSRange.getSignedMin().sge(RHSRange.getSignedMax())` case for
`ICmpInst::ICMP_SLT`), but for `isKnownPredicateViaConstantRanges`,
`false` actually means "don't know". Get rid of this extra bit of
code to avoid confusion.
llvm-svn: 259401
Sanjoy Das [Mon, 1 Feb 2016 20:48:10 +0000 (20:48 +0000)]
[SCEV] Rename isKnownPredicateWithRanges; NFC
Make it obvious that it uses constant ranges, and use `Via` instead of
`With`, like other similar functions in SCEV.
llvm-svn: 259400
Rafael Espindola [Mon, 1 Feb 2016 20:36:49 +0000 (20:36 +0000)]
Fix infinite recursion in MCAsmStreamer::EmitValueImpl.
If a target can only emit 8-bits data, we would loop in EmitValueImpl
since it will try to split a 32-bits data in 1 chunk of 32-bits.
No test since all current targets can emit 32bits at a time.
Patch by Alexandru Guduleasa!
llvm-svn: 259399
Teresa Johnson [Mon, 1 Feb 2016 20:16:35 +0000 (20:16 +0000)]
[ThinLTO] Ensure function summary output order is stable
Iterate over the function list instead of a DenseMap of Function pointers
when emitting the function summary into the module.
This fixes PR26419.
llvm-svn: 259398
Jim Ingham [Mon, 1 Feb 2016 20:05:37 +0000 (20:05 +0000)]
Add a note on how to create an empty target and use it to attach to
processes whose executables you don't know up front.
llvm-svn: 259396
Nico Weber [Mon, 1 Feb 2016 20:03:53 +0000 (20:03 +0000)]
Revert r259143, it broke check-lld on Windows (see PR26388).
llvm-svn: 259395
Rafael Espindola [Mon, 1 Feb 2016 19:56:12 +0000 (19:56 +0000)]
Add a test for r258362.
Thanks to Mehdi for finding it.
llvm-svn: 259394
Eugene Zelenko [Mon, 1 Feb 2016 19:47:24 +0000 (19:47 +0000)]
Sort checks alphabetically in ReadabilityTidyModule.cpp.
llvm-svn: 259393
Sanjay Patel [Mon, 1 Feb 2016 19:39:52 +0000 (19:39 +0000)]
[InstCombine] simplify masked store intrinsics with all ones or zeros masks
A masked store with a zero mask means there's no store.
A masked store with an allOnes mask means it's a normal vector store.
This is a continuation of:
http://reviews.llvm.org/rL259369
llvm-svn: 259392
Davide Italiano [Mon, 1 Feb 2016 19:22:16 +0000 (19:22 +0000)]
[llvm-nm] Simplify the code a bit. NFCI.
Fix a style violation while I'm here.
llvm-svn: 259391
Reid Kleckner [Mon, 1 Feb 2016 19:13:28 +0000 (19:13 +0000)]
[ubsan] Stop thinking we can rely on C++ ABI knowledge on Windows
We already disabled various tests relying on C++ ABI knowledge, but we
still tried to build in this configuration on Windows which was a
mistake.
Fixes PR26415.
llvm-svn: 259388
Balaram Makam [Mon, 1 Feb 2016 19:13:07 +0000 (19:13 +0000)]
AArch64: Implement missed conditional compare sequences.
Summary:
This is an extension to the existing implementation of r242436 which
restricts to only select inputs. This version fixes missed opportunities
in pr26084 by attempting to lower conditional compare sequences of
and/or trees with setcc leafs. This will additionaly handle the case
when a tree with select input is not a conjunction-disjunction tree
but some of the sub trees are conjunction-disjunction trees.
Reviewers: jmolloy, t.p.northover, mcrosier, MatzeB
Subscribers: mcrosier, llvm-commits, junbuml, haicheng, mssimpso, gberry
Differential Revision: http://reviews.llvm.org/D16291
llvm-svn: 259387
Pete Cooper [Mon, 1 Feb 2016 19:10:10 +0000 (19:10 +0000)]
Atomize the ObjC category list section.
__DATA, __objc_catlist contains a list of pointers to categories.
We want to atomize it so that the ObjC pass can later optimize and remove
categories. That will be a later patch.
llvm-svn: 259386
Matthew Simpson [Mon, 1 Feb 2016 19:09:24 +0000 (19:09 +0000)]
Add test case missing from r259357 (NFC)
llvm-svn: 259385
Geoff Berry [Mon, 1 Feb 2016 19:07:06 +0000 (19:07 +0000)]
[AArch64] Simplify prolog/epilog callee save/restore. NFC.
Summary:
Factor out common code for callee-save register pair calculation. This
is intended to simplify follow-on changes that reduce the number of
registers saved/restored.
Depends on D16732
Reviewers: mcrosier, jmolloy, t.p.northover
Subscribers: aemerson, rengolin, mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D16734
llvm-svn: 259384
Reid Kleckner [Mon, 1 Feb 2016 18:58:24 +0000 (18:58 +0000)]
Fix attribute((mode([word|unwind_word]))) for x32
Patch by H.J. Lu
```
typedef unsigned int gcc_word __attribute__((mode(word)));
```
and
```
typedef unsigned int gcc_unwind_word __attribute__((mode(unwind_word)));
```
define the largest unsigned integer types which can be stored in a
general purpose register, which may not be the pointer type. For x32,
they aren't pointer nor unsigned long. We should
1. Make getUnwindWordWidth and getRegisterWidth virtual,
2. Override them for x32, similar to hasInt128Type.
3. Use getRegisterWidth for __attribute__((mode(word)));
This fixes PR 24706.
Reviewers: rnk
Differential Revision: http://reviews.llvm.org/D16779
llvm-svn: 259383
Peter Collingbourne [Mon, 1 Feb 2016 18:55:50 +0000 (18:55 +0000)]
docs: Clarify that cfi-unrelated-cast is based on lifetime.
Also restore Makefile.sphinx which is needed to build the documentation.
llvm-svn: 259382
Ulrich Weigand [Mon, 1 Feb 2016 18:31:19 +0000 (18:31 +0000)]
[SystemZ] Fix wrong-code generation for certain always-false conditions
We've found another bug in the code generation logic conditions for a
certain class of always-false conditions, those of the form
if ((a & 1) < 0)
These only reach the back end when compiling without optimization.
The bug was introduced by the choice of using TEST UNDER MASK
to implement a check for
if ((a & MASK) < VAL)
as
if ((a & MASK) == 0)
where VAL is less than the the lowest bit of MASK. This is correct
in all cases except for VAL == 0, in which case the original
condition is always false, but the replacement isn't.
Fixed by excluding that particular case.
llvm-svn: 259381
Colin LeMahieu [Mon, 1 Feb 2016 18:15:39 +0000 (18:15 +0000)]
[NFC] Referencing manual for reason why subregbit is checked
llvm-svn: 259380
Zachary Turner [Mon, 1 Feb 2016 18:12:59 +0000 (18:12 +0000)]
Always write the session file in UTF-8.
This patch attempts to solve the Python 2 / Python 3 incompatibilities by
introducing a new `encoded_file` abstraction that we use instead of
`io.open()`. The problem with the builtin implementation of `io.open` is
that `read` and `write` accept and return `unicode` objects, which are not
always convenient to work with in Python 2. We solve this by making
`encoded_file.open()` return the same object returned by `io.open()` but
with hooked `read()` and `write()` methods. These hooked methods will
accept binary or text data, and conditionally convert what it gets to a
`unicode` object using the correct encoding. When calling `read()` it
also does any conversion necessary to convert the output back into the
native `string` type of the running python version.
Differential Revision: http://reviews.llvm.org/D16736
llvm-svn: 259379
Zachary Turner [Mon, 1 Feb 2016 18:12:54 +0000 (18:12 +0000)]
Fix misnamed variable in TestProcessAPI.py
llvm-svn: 259378
Sanjay Patel [Mon, 1 Feb 2016 17:46:18 +0000 (17:46 +0000)]
fix broken check lines
Without the colon, it doesn't mean anything!
llvm-svn: 259377
Benjamin Kramer [Mon, 1 Feb 2016 17:42:01 +0000 (17:42 +0000)]
Move LocInfoType from Sema to AST.
While transient and only used during parsing, LocInfoTypes are still used
from ASTDumper and are part of the AST.
llvm-svn: 259376
David Majnemer [Mon, 1 Feb 2016 17:37:56 +0000 (17:37 +0000)]
[InstCombine] Don't transform (X+INT_MAX)>=(Y+INT_MAX) -> (X<=Y)
This miscompile came about because we tried to use a transform which was
only appropriate for xor operators when addition was present.
This fixes PR26407.
llvm-svn: 259375
Jun Bum Lim [Mon, 1 Feb 2016 17:03:07 +0000 (17:03 +0000)]
[ValueTracking] Improve isKnownNonZero for PHI of non-zero constants
It is clear that a PHI is a non-zero if all incoming values are non-zero constants.
llvm-svn: 259370
Sanjay Patel [Mon, 1 Feb 2016 17:00:10 +0000 (17:00 +0000)]
[InstCombine] simplify masked load intrinsics with all ones or zeros masks
A masked load with a zero mask means there's no load.
A masked load with an allOnes mask means it's a normal vector load.
Differential Revision: http://reviews.llvm.org/D16691
llvm-svn: 259369
Pavel Labath [Mon, 1 Feb 2016 16:50:28 +0000 (16:50 +0000)]
Remove flaky annotation for TestCallWithTimeout on linux
The test has passed last 100 runs of the buildbot.
llvm-svn: 259368
Geoff Berry [Mon, 1 Feb 2016 16:47:51 +0000 (16:47 +0000)]
[PrologEpilogInserter] Add some debug output for callee-save frame object allocation
Reviewers: mcrosier
Subscribers: mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D16733
llvm-svn: 259367
Arpith Chacko Jacob [Mon, 1 Feb 2016 16:32:47 +0000 (16:32 +0000)]
[OpenMP] Prevent nesting of target constructs within target code execution regions.
Summary:
This patch enhances Sema to check for the following restriction:
OpenMP 4.5 [2.17 Nesting of Regions]
If a target, target update, target data, target enter data, or
target exit data construct is encountered during execution of a
target region, the behavior is unspecified.
Reviewers: ABataev
Differential Revision: http://reviews.llvm.org/D16758
llvm-svn: 259366
Geoff Berry [Mon, 1 Feb 2016 16:29:19 +0000 (16:29 +0000)]
[AArch64] Simplify callee-save register save/restore. NFC.
Summary:
Simplify callee-save register save/restore code generation by
remembering the size of the callee-save area when it is computed so we
don't have to scan the prologue/epilogue instructions again later to
reconstruct it.
This is intended to simplify follow-on changes that reduce the number of
registers saved/restored.
Reviewers: mcrosier, jmolloy, t.p.northover
Subscribers: aemerson, rengolin, mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D16732
llvm-svn: 259365
Matthew Simpson [Mon, 1 Feb 2016 16:07:01 +0000 (16:07 +0000)]
[LV] Rename RdxPHIsToFix to PHIsToFix (NFC)
In the future, we will vectorize recurrences other than reductions. This patch
renames a few variables and updates their associated comments to enable them to
be reused for non-reduction PHI nodes.
This change was requested in the review for D16197.
llvm-svn: 259364
Asaf Badouh [Mon, 1 Feb 2016 15:48:21 +0000 (15:48 +0000)]
[X86][AVX512VBMI] add encoding and intrinsics for Multishift
Differential Revision: http://reviews.llvm.org/D16399
llvm-svn: 259363
Aaron Ballman [Mon, 1 Feb 2016 15:31:15 +0000 (15:31 +0000)]
Add a new check, readability-redundant-control-flow, that check for some forms of redundant control flow statements. Currently checks for return statements at the end of a function with a void return type and continue statements at the end of looping statements.
Patch by Richard Thomson.
llvm-svn: 259362
Vasileios Kalintiris [Mon, 1 Feb 2016 15:19:35 +0000 (15:19 +0000)]
[mips] Split large test file into 3 smaller ones.
Remove the old select.ll file and use select-int.ll, select-flt.ll,
select-dbl.ll for testing selects on integers, floats & doubles respectivelly.
llvm-svn: 259361
Daniel Sanders [Mon, 1 Feb 2016 15:13:31 +0000 (15:13 +0000)]
[mips] Range check uimm16 and fix several bugs this revealed.
Summary:
The bugs were:
* teq and similar take 4-bit unsigned immediates on microMIPS.
* teqi and similar have side-effects like teq do.
* shll_s.w and shra_r.w take 5-bit unsigned immediates.
* The various DSP ext* instructions take a 5-bit immediate.
* repl.qh takes an 8-bit unsigned immediate.
* repl.ph takes a 10-bit unsigned immediate.
* rddsp/wrdsp take a 10-bit unsigned immediate.
* teqi and similar take signed 16-bit immediates (10-bit for microMIPS).
* Out-of-range immediate macros for or/xor take a simm32/simm64 depending
on architecture. I'll fix the simm64 case properly when I reach simm32.
lui is a bit more lenient than GAS and accepts signed immediates in addition
to unsigned. This is because MipsMCExpr can produce signed values when
constant folding and it currently lacks a way of knowing it should fold to
an unsigned value.
Reviewers: vkalintiris
Subscribers: dsanders, llvm-commits
Differential Revision: http://reviews.llvm.org/D15446
llvm-svn: 259360
Aaron Ballman [Mon, 1 Feb 2016 14:11:47 +0000 (14:11 +0000)]
Reapply r259210 with a fix for RegistryTest.cpp.
Patch by Richard Thomson.
llvm-svn: 259359
Amjad Aboud [Mon, 1 Feb 2016 14:09:41 +0000 (14:09 +0000)]
Improved macro emission in dwarf.
Changed emitting offset of macinfo entry into compiler unit DIE to use "addSectionLabel" method rather than explicitly calculating size/offset of macro entry.
Differential Revision: http://reviews.llvm.org/D16292
llvm-svn: 259358
Matthew Simpson [Mon, 1 Feb 2016 13:38:29 +0000 (13:38 +0000)]
Reapply commit r258404 with fix.
The previous patch caused PR26364. The fix is to ensure that we don't enter a
cycle when iterating over use-def chains.
llvm-svn: 259357
Pavel Labath [Mon, 1 Feb 2016 13:29:41 +0000 (13:29 +0000)]
Remove Timer::Initialize routine
Summary:
I've run into an issue when running unit tests, where the underlying problem turned out to be
that we were creating Timer objects (through several layers of indirection) without calling
Timer::Initialize. Since Timer's thread-local storage was not properly initialized, we were
overwriting gtest's own thread-local storage, causing test failures.
Instead of requiring that every test calls Timer::Initialize(), I remove the function altogether:
The thread-local storage can be initialized on-demand, and the g_file variable initialized to
stdout and never changed, so I have simply removed it.
Reviewers: clayborg, zturner, tberghammer
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D16722
llvm-svn: 259356
Benjamin Kramer [Mon, 1 Feb 2016 13:22:39 +0000 (13:22 +0000)]
Remove the egregious PCHContainer layering hack that doesn't seem to be necessary anymore.
llvm-svn: 259355
Tobias Grosser [Mon, 1 Feb 2016 11:54:13 +0000 (11:54 +0000)]
ScopInfo: Correct schedule construction
For schedule generation we assumed that the reverse post order traversal used by
the domain generation is sufficient, however it is not. Once a loop is
discovered, we have to completely traverse it, before we can generate the
schedule for any block/region that is only reachable through a loop exiting
block.
To this end, we add a "loop stack" that will keep track of loops we
discovered during the traversal but have not yet traversed completely.
We will never visit a basic block (or region) outside the most recent
(thus smallest) loop in the loop stack but instead queue such blocks
(or regions) in a waiting list. If the waiting list is not empty and
(might) contain blocks from the most recent loop in the loop stack the
next block/region to visit is drawn from there, otherwise from the
reverse post order iterator.
We exploit the new property of loops being always completed before additional
loops are processed, by removing the LoopSchedules map and instead keep all
information in LoopStack. This clarifies that we indeed always only keep a
stack of in-process loops, but will never keep incomplete schedules for an
arbitrary set of loops. As a result, we can simplify some of the existing code.
This patch also adds some more documentation about how our schedule construction
works.
This fixes http://llvm.org/PR25879
This patch is an modified version of Johannes Doerfert's initial fix.
Differential Revision: http://reviews.llvm.org/D15679
llvm-svn: 259354
Daniel Jasper [Mon, 1 Feb 2016 11:21:07 +0000 (11:21 +0000)]
clang-format: Fix incorrect pointer detection in lambdas in constructor
initializers.
Before:
Constructor() : member([](A *a, B * b) {}) {}
After:
Constructor() : member([](A *a, B *b) {}) {}
llvm-svn: 259353
Daniel Jasper [Mon, 1 Feb 2016 11:21:02 +0000 (11:21 +0000)]
clang-format: Add option to disable string literal formatting.
llvm-svn: 259352
Daniel Jasper [Mon, 1 Feb 2016 11:20:55 +0000 (11:20 +0000)]
clang-format: Fix alignment of trailing multiline columns.
llvm-svn: 259351
Daniel Jasper [Mon, 1 Feb 2016 11:20:47 +0000 (11:20 +0000)]
clang-format: [JS] Treat "in" as a proper operator.
llvm-svn: 259350
JF Bastien [Mon, 1 Feb 2016 10:46:16 +0000 (10:46 +0000)]
WebAssembly NFC: simplify control flow
This should now be easier to read.
llvm-svn: 259349
Ewan Crawford [Mon, 1 Feb 2016 10:39:24 +0000 (10:39 +0000)]
DWARF RenderScript vendor extension
Patch adds a DWARF language vendor extension for RenderScript.
We are already using this identifier in LLDB with a hard coded value, so it's preferable to use a LLVM generated enum instead.
The language is intended to be added to the next version of the standard.
See http://www.dwarfstd.org/ShowIssue.php?issue=150331.1
Reviewers: dexonsmith, echristo
Subscribers: probinson domipheus, srhines, llvm-commits
Differential Revision: http://reviews.llvm.org/D16409
llvm-svn: 259348
Tobias Grosser [Mon, 1 Feb 2016 10:07:43 +0000 (10:07 +0000)]
ScopInfo: Drop dead code in schedule description
In https://llvm.org/svn/llvm-project/polly/trunk@251870 code was committed to
avoid a failure in the presence of infinite loops, but the test case committed
along with this change passes without the actual change. I looked back into the
code and also checked with the original committer (Johannes), but could not find
the reason why the code is needed. The introduction of LoopStacks for
buildSchedule in one of the next commits will make it even more clear that this
code is not needed, but I remove this ahead of time to facilitate bisecting in
case I missed something.
llvm-svn: 259347
Igor Breger [Mon, 1 Feb 2016 09:57:15 +0000 (09:57 +0000)]
AVX512: fix mask handling for gather/scatter/prefetch intrinsics.
Differential Revision: http://reviews.llvm.org/D16755
llvm-svn: 259346
Artem Dergachev [Mon, 1 Feb 2016 09:29:17 +0000 (09:29 +0000)]
[analyzer] Use a wider integer type for an array index.
Avoids unexpected overflows while performing pointer arithmetics in 64-bit code.
Moreover, neither PointerDiffType nor 'int' can be used as a common array index
type because arrays may have size (and indexes) more than PTRDIFF_MAX but less
than SIZE_MAX.
Patch by Aleksei Sidorin!
Differential Revision: http://reviews.llvm.org/D16063
llvm-svn: 259345
Abhishek Aggarwal [Mon, 1 Feb 2016 09:01:42 +0000 (09:01 +0000)]
Set correct ThreadStopInfo in case of trace event
Summary:
- The patch solves Bug 23478 and Bug 19311. Resolving
Bug 23478 also resolves Bug 23039.
Correct ThreadStopInfo is set for Linux and FreeBSD
platforms.
- Summary:
When a trace event is reported, we need to check
whether the trace event lands at a breakpoint site.
If it lands at a breakpoint site then set the thread's
StopInfo with the reason 'breakpoint'. Else, set the reason
to be 'Trace'.
Change-Id: I0af9765e782fd74bc0cead41548486009f8abb87
Signed-off-by: Abhishek Aggarwal <abhishek.a.aggarwal@intel.com>
Reviewers: jingham, emaste, lldb-commits, clayborg, ovyalov
Subscribers: emaste
Differential Revision: http://reviews.llvm.org/D16720
llvm-svn: 259344
Simon Pilgrim [Mon, 1 Feb 2016 08:59:30 +0000 (08:59 +0000)]
[X86][SSE] Find source of the inserted element of INSERTPS
Minor patch to trace back through target shuffles to the source of the inserted element in a (V)INSERTPS shuffle.
Differential Revision: http://reviews.llvm.org/D16652
llvm-svn: 259343
Igor Breger [Mon, 1 Feb 2016 07:56:09 +0000 (07:56 +0000)]
AVX512 : Fix SETCCE lowering for KNL 32 bit.
Differential Revision: http://reviews.llvm.org/D16752
llvm-svn: 259342
Frederic Riss [Mon, 1 Feb 2016 04:43:14 +0000 (04:43 +0000)]
[dsymutil] Skip mach-o paired relocations
Noticed while working on scattered relocations.
I do not think these relocs can actually happen in the debug_info section,
but if they happen the code would mishandle them. Explicitely skip them
and warn if we encounter one.
llvm-svn: 259341
David Majnemer [Mon, 1 Feb 2016 04:28:59 +0000 (04:28 +0000)]
[X86] Cleanup the WinEHState pass
Remove unnecessary includes and class state.
No functional change intended.
llvm-svn: 259340
Frederic Riss [Mon, 1 Feb 2016 03:44:22 +0000 (03:44 +0000)]
[dsymutil] Support scattered relocs.
Although it seems like clang will never emit scattered relocations in
the debug information (at least I couldn't find a way), we have too
support them for the benefit of other compilers.
As clang doesn't generate them, the included testcase was produced
from hacked up assembly.
llvm-svn: 259339
David Majnemer [Mon, 1 Feb 2016 03:29:38 +0000 (03:29 +0000)]
Revert r258580 and r258581.
Those commits created an artificial edge from a cleanup to a synthesized
catchswitch in order to get the MSVC personality routine to execute
cleanups which don't cleanupret and are not wrapped by a catchswitch.
This worked well enough but is not a complete solution in situations
where there the cleanup infinite loops.
However, the real deal breaker behind this approach comes about from a
degenerate case where the cleanup is post-dominated by unreachable *and*
throws an exception. This ends poorly because the catchswitch will
inadvertently catch the exception.
Because of this we should go back to our previous behavior of not
executing certain cleanups (identical behavior with the Itanium ABI
implementation in clang, GCC and ICC).
N.B. I think this could be salvaged by making the catchpad rethrow the
exception and properly transforming throwing calls in the cleanup into
invokes.
llvm-svn: 259338
Tobias Grosser [Mon, 1 Feb 2016 01:55:08 +0000 (01:55 +0000)]
ScopInfo: use std::distance to shorten code [NFC]
llvm-svn: 259337
Craig Topper [Mon, 1 Feb 2016 01:33:42 +0000 (01:33 +0000)]
[TableGen] Store result of getInstructionsByEnumValue in an ArrayRef instead of accidentally copying to a vector.
llvm-svn: 259336
David Majnemer [Mon, 1 Feb 2016 01:33:17 +0000 (01:33 +0000)]
Mark DR1250 as implemented
We implemented this DR back in r258768 but forgot to mark it as
implemented.
llvm-svn: 259335
Frederic Riss [Sun, 31 Jan 2016 22:06:35 +0000 (22:06 +0000)]
[MCDwarf] Fix encoding of line tables with weird custom parameters
With poorly chosen custom parameters, the line table encoding logic would
sometimes end up generating a special opcode bigger than 255, which is wrong.
The set of default parameters that LLVM uses isn't subject to this bug.
When carefully chosing the line table parameters, it's impossible to fall into the
corner case that this patch fixes. The standard however doesn't require that these
parameters be carefully chosen. And even if it did, we shouldn't generate broken
encoding.
Add a unittest for this specific encoding bug, and while at it, create some unit
tests for the encoding logic using different sets of parameters.
llvm-svn: 259334
Craig Topper [Sun, 31 Jan 2016 20:36:20 +0000 (20:36 +0000)]
Replace utostr_32 use with utostr to match removal from llvm.
llvm-svn: 259333
Tobias Grosser [Sun, 31 Jan 2016 20:25:46 +0000 (20:25 +0000)]
cmake: Add additional libraries to fix the darwin shared library build
darwin requires the additional linkages of...
LLVMBitReader
LLVMMCParser
LLVMObject
LLVMProfileData
LLVMTarget
LLVMVectorize
as the darwin requires all of the weak undefined symbols in a library to be
resolved when linking it against an executable (unless
-Wl,-undefined,dynamic_lookup is used to override the default behavior of
-Wl,-undefined,error).
Contributed-by: Jack Howarth
llvm-svn: 259332
Craig Topper [Sun, 31 Jan 2016 20:00:26 +0000 (20:00 +0000)]
Remove utostr_32 as it has no uses anymore.
llvm-svn: 259331
Craig Topper [Sun, 31 Jan 2016 20:00:24 +0000 (20:00 +0000)]
Replace usages of llvm::utostr_32 with just llvm::utostr. While this is less efficient, its unclear the few places that were using the _32 version were doing so for efficiency.
llvm-svn: 259330
Craig Topper [Sun, 31 Jan 2016 20:00:22 +0000 (20:00 +0000)]
Merge utohex_buffer into utohexstr, it's only caller. Also change utohexstr to use the std::string constructor that takes a start and end pointer. This saves a call to strlen. NFC
llvm-svn: 259329
Sanjay Patel [Sun, 31 Jan 2016 16:35:23 +0000 (16:35 +0000)]
add helper function for minnum/maxnum ; NFC
llvm-svn: 259326
Sanjay Patel [Sun, 31 Jan 2016 16:34:48 +0000 (16:34 +0000)]
use range-based for loop; NFC
llvm-svn: 259325
Sanjay Patel [Sun, 31 Jan 2016 16:34:11 +0000 (16:34 +0000)]
fix formatting; NFC
llvm-svn: 259324
Sanjay Patel [Sun, 31 Jan 2016 16:33:33 +0000 (16:33 +0000)]
simplify; NFC
llvm-svn: 259323
Sanjay Patel [Sun, 31 Jan 2016 16:32:23 +0000 (16:32 +0000)]
clean up; NFC
function names, comments, formatting, typos
llvm-svn: 259322
JF Bastien [Sun, 31 Jan 2016 08:19:40 +0000 (08:19 +0000)]
WebAssembly: more failures are gone
llvm-svn: 259321
JF Bastien [Sun, 31 Jan 2016 08:05:05 +0000 (08:05 +0000)]
WebAssembly: update expected failures
r259305 fixed a few assertions around FrameIndex, and I forgot to update these failures despite having run the torture tests.
llvm-svn: 259320
Frederic Riss [Sun, 31 Jan 2016 04:39:16 +0000 (04:39 +0000)]
[dsymutil] Fix FileCheck command.
Damn case-insensitive filesystem...
llvm-svn: 259319
Frederic Riss [Sun, 31 Jan 2016 04:29:34 +0000 (04:29 +0000)]
[dsymutil] Fix handling of common symbols.
llvm-dsymutil was misinterpreting the value of common symbols as their
address when it actually contains their size. This didn't impact
llvm-dsymutil's ability to link the debug information for common symbols
because these are always found by name and not by address. Things could
however go wrong when the size of a common object matched the object
file address of another symbol. Depending on the link order of the symbols
the common object might incorrectly evict this other object from the
address to symbol mapping, and then link the evicted symbol with a wrong
binary address.
Use the new ability to have symbols without an object file address to fix
this.
llvm-svn: 259318
Frederic Riss [Sun, 31 Jan 2016 04:29:22 +0000 (04:29 +0000)]
[dsymutil] Allow debug map mappings with no object file address. NFC
This change just changes the data structure that ties symbol names,
object file address and linked binary addresses to accept mappings
with no object file address. Such symbol mappings are not fed into
the debug map yet, so this patch is NFC.
A subsequent patch will make use of this functionality for common
symbols.
llvm-svn: 259317
Craig Topper [Sun, 31 Jan 2016 04:20:03 +0000 (04:20 +0000)]
Replace usage of llvm::utostr_32 with just llvm::utostr. While this is less efficient, its unclear that the one place using the _32 version was doing so for efficiency.
llvm-svn: 259316
Tim Shen [Sun, 31 Jan 2016 03:59:34 +0000 (03:59 +0000)]
[SelectionDAG] Eliminate exponential behavior in WalkChainUsers
llvm-svn: 259315
Craig Topper [Sun, 31 Jan 2016 01:55:15 +0000 (01:55 +0000)]
No need to use utostr/utohexstr when writing into a raw_ostream. NFC
llvm-svn: 259314
Craig Topper [Sun, 31 Jan 2016 01:12:38 +0000 (01:12 +0000)]
Shrink character buffer size in raw_ostream::write_hex to 16 characters intead of 20 as that's the largest string a 64-bit hex value can be.
llvm-svn: 259313
Craig Topper [Sun, 31 Jan 2016 01:12:35 +0000 (01:12 +0000)]
Use std::end instead of repeating buffer sizes.
llvm-svn: 259312
Bruno Cardoso Lopes [Sun, 31 Jan 2016 00:47:51 +0000 (00:47 +0000)]
[Parser] Update CachedTokens while parsing ObjectiveC template argument list
Consider the following ObjC++ snippet:
--
@protocol PA;
@protocol PB;
@class NSArray<ObjectType>;
typedef int some_t;
id<PA> FA(NSArray<id<PB>> *h, some_t group);
--
This would hit an assertion in the parser after generating an annotation token
while trying to update the token cache:
Assertion failed: (CachedTokens[CachedLexPos-1].getLastLoc() == Tok.getAnnotationEndLoc() && "The annotation should be until the most recent cached token")
...
7 clang::Preprocessor::AnnotatePreviousCachedTokens(clang::Token const&) + 494
8 clang::Parser::TryAnnotateTypeOrScopeTokenAfterScopeSpec(bool, bool, clang::CXXScopeSpec&, bool) + 1163
9 clang::Parser::TryAnnotateTypeOrScopeToken(bool, bool) + 361
10 clang::Parser::isCXXDeclarationSpecifier(clang::Parser::TPResult, bool*) + 598
...
The cached preprocessor token in this case is:
greatergreater '>>' Loc=<testcase.mm:7:24>
while the annotation ("NSArray<id<PB>>") ends at "testcase.mm:7:25", hence the
assertion.
Properly update the CachedTokens during template parsing to contain
two greater tokens instead of a greatergreater.
Differential Revision: http://reviews.llvm.org/D15173
rdar://problem/
23494277
llvm-svn: 259311
Craig Topper [Sun, 31 Jan 2016 00:20:26 +0000 (00:20 +0000)]
No need to use utostr when putting integers into a raw_ostream. NFC
llvm-svn: 259310
Craig Topper [Sun, 31 Jan 2016 00:20:24 +0000 (00:20 +0000)]
Convert an unsigned to Twine instead of using utostr since we're already building a Twine. NFC
llvm-svn: 259309
Craig Topper [Sun, 31 Jan 2016 00:15:35 +0000 (00:15 +0000)]
Convert int to Twine instead of using utostr since it was already being added to a Twine. NFC
llvm-svn: 259308
Jingyue Wu [Sat, 30 Jan 2016 23:48:47 +0000 (23:48 +0000)]
[doc] improve the doc for CUDA
1. Mentioned that CUDA support works best with trunk.
2. Simplified the example by removing its dependency on the CUDA samples.
3. Explain the --cuda-gpu-arch flag.
llvm-svn: 259307
Derek Schuff [Sat, 30 Jan 2016 21:43:08 +0000 (21:43 +0000)]
[WebAssembly] Fix uses of FrameIndex as store values
Previously the code assumed all uses of FI on loads and stores were as
addresses. This checks whether the use is the address or a value and
handles the latter case as it does for non-memory instructions.
llvm-svn: 259306
JF Bastien [Sat, 30 Jan 2016 14:11:26 +0000 (14:11 +0000)]
WebAssembly: don't optimize frameindex store
The previous code was incorrect (can't getReg a frameindex). We could instead optimize it to reduce tree height, but I'm not sure that's worthwhile yet because we then try to eliminate the frameindex.
This patch also fixes frame index elimination for operations which may load or store: it used to assume the base was operand 2 and immediate offset operand 1. That's not true for stores, where they're 4 and 3.
llvm-svn: 259305
Benjamin Kramer [Sat, 30 Jan 2016 12:16:23 +0000 (12:16 +0000)]
[AST] Pull simple method inline.
llvm-svn: 259304
JF Bastien [Sat, 30 Jan 2016 11:19:26 +0000 (11:19 +0000)]
WebAssembly NFC: fix build warning
WebAssemblyFrameLowering.cpp:158:44: warning: enumeral and non-enumeral type in conditional expression [enabled by default]
llvm-svn: 259303
Davide Italiano [Sat, 30 Jan 2016 08:03:54 +0000 (08:03 +0000)]
assert(false) -> llvm_unreachable().
llvm-svn: 259302
Ismail Donmez [Sat, 30 Jan 2016 07:14:31 +0000 (07:14 +0000)]
Fix build with VS2015
llvm-svn: 259301
Gerolf Hoflehner [Sat, 30 Jan 2016 05:58:38 +0000 (05:58 +0000)]
[BasicAA] NFC - revised comment for function adjustToPointerSize()
llvm-svn: 259300
Gerolf Hoflehner [Sat, 30 Jan 2016 05:52:53 +0000 (05:52 +0000)]
[BasicAA] Fix for missing must alias (D16343)
llvm-svn: 259299
Gerolf Hoflehner [Sat, 30 Jan 2016 05:35:09 +0000 (05:35 +0000)]
[BasicAA] Update on r259290 - added missing cast
llvm-svn: 259298
Matt Arsenault [Sat, 30 Jan 2016 05:19:45 +0000 (05:19 +0000)]
AMDGPU: Fix emitting invalid workitem intrinsics for HSA
The AMDGPUPromoteAlloca pass was emitting the read.local.size
calls, which with HSA was incorrectly selected to reading from
the offset mesa uses off of the kernarg pointer.
Error on intrinsics which aren't supported by HSA, and start
emitting the correct IR to read the workgroup size
out of the dispatch pointer.
Also initialize the pass so it can be tested with opt, and
start moving towards not depending on the subtarget as an
argument.
Start emitting errors for the intrinsics not handled with HSA.
llvm-svn: 259297
Matt Arsenault [Sat, 30 Jan 2016 05:10:59 +0000 (05:10 +0000)]
AMDGPU: Stop checking intrinsics not used by HSA for dispatch-ptr
Only the dispatch.ptr intrinsic is supposed to be used now to get
the workgroup size, and the read.local.size intrinsics do not
work correctly.
llvm-svn: 259296
Matt Arsenault [Sat, 30 Jan 2016 05:02:00 +0000 (05:02 +0000)]
InstCombine: fabs(x) * fabs(x) -> x * x
llvm-svn: 259295