platform/upstream/llvm.git
6 years agoAMDGPU: Use splat vectors for undefs when folding canonicalize
Matt Arsenault [Sun, 12 Aug 2018 08:42:54 +0000 (08:42 +0000)]
AMDGPU: Use splat vectors for undefs when folding canonicalize

If one of the elements is undef, use the canonicalized constant
from the other element instead of 0.

Splat vectors are more useful for other optimizations, such
as matching vector clamps. This was breaking on clamps
of half3 from the undef 4th component.

llvm-svn: 339512

6 years agoAMDGPU: Fix packing undef parts of build_vector
Matt Arsenault [Sun, 12 Aug 2018 08:42:46 +0000 (08:42 +0000)]
AMDGPU: Fix packing undef parts of build_vector

llvm-svn: 339511

6 years ago[LLF][ELF] - Simplify. NFC.
George Rimar [Sun, 12 Aug 2018 07:52:24 +0000 (07:52 +0000)]
[LLF][ELF] - Simplify. NFC.

llvm-svn: 339510

6 years ago[TargetLowering] Simplify one of the special cases in SimplifyDemandedBits for XOR...
Craig Topper [Sun, 12 Aug 2018 06:52:03 +0000 (06:52 +0000)]
[TargetLowering] Simplify one of the special cases in SimplifyDemandedBits for XOR. NFCI

We were checking for all bits being Known by checking Known.Zero|Known.One, but if all the bits are known then the value should be a Constant and we can just check for that instead.

llvm-svn: 339509

6 years ago[TargetLowering] Use APInt::isSubsetOf to simplify some code. NFC
Craig Topper [Sun, 12 Aug 2018 05:34:15 +0000 (05:34 +0000)]
[TargetLowering] Use APInt::isSubsetOf to simplify some code. NFC

llvm-svn: 339508

6 years ago[X86] Remove unnecessary AddedComplexity line. NFC
Craig Topper [Sun, 12 Aug 2018 03:22:18 +0000 (03:22 +0000)]
[X86] Remove unnecessary AddedComplexity line. NFC

The use of the or_is_add predicate already gives enough of a complexity boost to get the patterns ordered properly.

llvm-svn: 339507

6 years ago[ASTImporter] Added test case for opaque enums
Raphael Isemann [Sat, 11 Aug 2018 23:43:46 +0000 (23:43 +0000)]
[ASTImporter] Added test case for opaque enums

Reviewers: a.sidorin, a_sidorin

Reviewed By: a_sidorin

Subscribers: a_sidorin, martong, cfe-commits

Differential Revision: https://reviews.llvm.org/D50550

llvm-svn: 339506

6 years ago[ASTImporter] Added test case for CXXConversionDecl importing
Raphael Isemann [Sat, 11 Aug 2018 23:43:02 +0000 (23:43 +0000)]
[ASTImporter] Added test case for CXXConversionDecl importing

Reviewers: a.sidorin, a_sidorin

Reviewed By: a_sidorin

Subscribers: a_sidorin, martong, cfe-commits

Differential Revision: https://reviews.llvm.org/D50552

llvm-svn: 339505

6 years agoUse a DenseMap for looking up functions by UID in CompileUnit::FindFunctionByUID
Raphael Isemann [Sat, 11 Aug 2018 23:40:27 +0000 (23:40 +0000)]
Use a DenseMap for looking up functions by UID in CompileUnit::FindFunctionByUID

Summary:
Instead of iterating over our vector of functions, we might as well use a map here to
directly get the function we need.

Thanks to Vedant for pointing this out.

Reviewers: vsk

Reviewed By: vsk

Subscribers: mgrang, lldb-commits

Differential Revision: https://reviews.llvm.org/D50225

llvm-svn: 339504

6 years agoAdd missing _LIBCXXABI_FUNC_VIS to __gxx_personality_seh0
Martin Storsjo [Sat, 11 Aug 2018 19:36:06 +0000 (19:36 +0000)]
Add missing _LIBCXXABI_FUNC_VIS to __gxx_personality_seh0

This was missed in SVN r337754.

llvm-svn: 339503

6 years ago[Dominators] Remove the DeferredDominance class
Chijun Sima [Sat, 11 Aug 2018 08:12:07 +0000 (08:12 +0000)]
[Dominators] Remove the DeferredDominance class

Summary: After converting all existing passes to use the new DomTreeUpdater interface, there isn't any usage of the original DeferredDominance class. Thus, we can safely remove it from the codebase.

Reviewers: kuhar, brzycki, dmgreen, davide, grosser

Reviewed By: kuhar, brzycki

Subscribers: mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D49747

llvm-svn: 339502

6 years ago[UnJ] Improve explicit loop count checks
David Green [Sat, 11 Aug 2018 07:37:31 +0000 (07:37 +0000)]
[UnJ] Improve explicit loop count checks

Try to improve the computed counts when it has been explicitly set by a pragma
or command line option. This moves the code around, so that first call to
computeUnrollCount to get a sensible count and override that if explicit unroll
and jam counts are specified.

Also added some extra debug messages for when unroll and jamming is disabled.

Differential Revision: https://reviews.llvm.org/D50075

llvm-svn: 339501

6 years ago[UnJ] Create a hasInvariantIterationCount function. NFC
David Green [Sat, 11 Aug 2018 06:57:28 +0000 (06:57 +0000)]
[UnJ] Create a hasInvariantIterationCount function. NFC

Pulled out a separate function for some code that calculates
if an inner loop iteration count is invariant to it's outer
loop.

Differential Revision: https://reviews.llvm.org/D50063

llvm-svn: 339500

6 years ago[X86] Remove the AL/AX/EAX/RAX short immediate forms from the macro fusion shouldSche...
Craig Topper [Sat, 11 Aug 2018 06:42:51 +0000 (06:42 +0000)]
[X86] Remove the AL/AX/EAX/RAX short immediate forms from the macro fusion shouldScheduleAdjacent. NFC

These instructions are only created by the backend during MCInst lowering.

llvm-svn: 339499

6 years ago[X86] Add the mem-reg form of CMP to the macro fusion shouldScheduleAdjacent.
Craig Topper [Sat, 11 Aug 2018 06:42:50 +0000 (06:42 +0000)]
[X86] Add the mem-reg form of CMP to the macro fusion shouldScheduleAdjacent.

Unlike the other arithmetic instructions the mem-reg form of compare is just a load and not a RMW operation. According to the Intel optimization manual, this form is also supported by macro fusion.

llvm-svn: 339498

6 years ago[X86] Remove ADD8mi and ADDmr from the macro fusion shouldScheduleAdjacent.
Craig Topper [Sat, 11 Aug 2018 06:42:49 +0000 (06:42 +0000)]
[X86] Remove ADD8mi and ADDmr from the macro fusion shouldScheduleAdjacent.

The are RMW of memory operations. They aren't eligible for macro fusion.

llvm-svn: 339497

6 years ago[X86] Change the MOV32ri64 pseudo instruction to def a GR64 directly instead of wrapp...
Craig Topper [Sat, 11 Aug 2018 05:33:00 +0000 (05:33 +0000)]
[X86] Change the MOV32ri64 pseudo instruction to def a GR64 directly instead of wrapping it in a SUBREG_TO_REG.

Now we switch to the subregister in expandPostRAPseudos where we already switched the opcode.

This simplifies a few isel patterns that used the pseudo directly. And magically seems to have improved our ability to CSE it in the undef-label.ll test.

llvm-svn: 339496

6 years agoFix WebAssembly instruction printer after r339474
Richard Trieu [Sat, 11 Aug 2018 04:18:05 +0000 (04:18 +0000)]
Fix WebAssembly instruction printer after r339474

Treat the stack variants of control instructions the same as regular
instructions.  Otherwise, the vector ControlFlowStack will be the wrong
size and have out-of-bounds access.  This was detected by MemorySanitizer.

llvm-svn: 339495

6 years agoMake the section boundary checks on Windows not depend on the order as they are emitt...
Douglas Yung [Sat, 11 Aug 2018 02:46:47 +0000 (02:46 +0000)]
Make the section boundary checks on Windows not depend on the order as they are emitted in reverse when the compiler is built by Visual C++.

llvm-svn: 339494

6 years ago[analyzer] Fix keyboard navigation for .msgNote events
George Karpenkov [Sat, 11 Aug 2018 01:47:41 +0000 (01:47 +0000)]
[analyzer] Fix keyboard navigation for .msgNote events

Does not go to msgNote's.

Differential Revision: https://reviews.llvm.org/D50595

llvm-svn: 339493

6 years ago[gold] Fix Tests cases on i686
Tom Stellard [Sat, 11 Aug 2018 01:08:34 +0000 (01:08 +0000)]
[gold] Fix Tests cases on i686

Reviewers: tejohnson

Reviewed By: tejohnson

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D50583

llvm-svn: 339492

6 years agoAMDGPU/GlobalISel: Define instruction mapping for G_INSERT
Tom Stellard [Sat, 11 Aug 2018 00:51:54 +0000 (00:51 +0000)]
AMDGPU/GlobalISel: Define instruction mapping for G_INSERT

Reviewers: arsenm

Reviewed By: arsenm

Subscribers: kzhuravl, wdng, nhaehnle, yaxunl, rovka, kristof.beyls, dstuttard, tpr, t-tye, llvm-commits

Differential Revision: https://reviews.llvm.org/D49625

llvm-svn: 339491

6 years agoFix WebAssembly tests after r339474
Richard Trieu [Fri, 10 Aug 2018 23:58:11 +0000 (23:58 +0000)]
Fix WebAssembly tests after r339474

Add flags to llc RUN lines to keep tests passing.

llvm-svn: 339490

6 years ago[analyzer] InnerPointerChecker: improve warning messages and notes.
Reka Kovacs [Fri, 10 Aug 2018 23:56:57 +0000 (23:56 +0000)]
[analyzer] InnerPointerChecker: improve warning messages and notes.

Differential Revision: https://reviews.llvm.org/D49570

llvm-svn: 339489

6 years ago[analyzer] [NFC] [tests] Move plist-based diagnostics tests to separate files, use...
George Karpenkov [Fri, 10 Aug 2018 23:51:43 +0000 (23:51 +0000)]
[analyzer] [NFC] [tests] Move plist-based diagnostics tests to separate files, use diff instead of a FileCheck - II

Some files were missed by https://reviews.llvm.org/D50545

Differential Revision: https://reviews.llvm.org/D50590

llvm-svn: 339488

6 years ago[ELF][HEXAGON] Put test back in.
Sid Manning [Fri, 10 Aug 2018 23:46:11 +0000 (23:46 +0000)]
[ELF][HEXAGON] Put test back in.

The registers are tied.

llvm-svn: 339487

6 years ago[ELF][HEXAGON] Remove test that breaks assembler.
Sid Manning [Fri, 10 Aug 2018 22:54:54 +0000 (22:54 +0000)]
[ELF][HEXAGON] Remove test that breaks assembler.

Remove testcase the breaks the assembler parser.  Will fix llvm-mc
and put this back afterward.

llvm-svn: 339486

6 years agoRe-commit "[NFC] More ConstantMerge refactoring"
JF Bastien [Fri, 10 Aug 2018 22:41:09 +0000 (22:41 +0000)]
Re-commit "[NFC] More ConstantMerge refactoring"

My previous change moved some code upwards which caused an assert in debug mode
because the global value didn't necessarily have an initializer. Don't do that.

llvm-svn: 339485

6 years ago[DepInfo] Use isl++ in Dependences::isValidSchedule. NFC.
Michael Kruse [Fri, 10 Aug 2018 22:33:27 +0000 (22:33 +0000)]
[DepInfo] Use isl++ in Dependences::isValidSchedule. NFC.

Also change StatementToIslMapTy to hold isl::map, because it is used as a
parameter.

llvm-svn: 339484

6 years ago[clangd] Avoid duplicates in findDefinitions response
Simon Marchi [Fri, 10 Aug 2018 22:27:53 +0000 (22:27 +0000)]
[clangd] Avoid duplicates in findDefinitions response

Summary:
When compile_commands.json contains some source files expressed as
relative paths, we can get duplicate responses to findDefinitions.  The
responses only differ by the URI, which are different versions of the
same file:

    "result": [
        {
            ...
            "uri": "file:///home/emaisin/src/ls-interact/cpp-test/build/../src/first.h"
        },
        {
            ...
            "uri": "file:///home/emaisin/src/ls-interact/cpp-test/src/first.h"
        }
    ]

In getAbsoluteFilePath, we try to obtain the realpath of the FileEntry
by calling tryGetRealPathName.  However, this can fail and return an
empty string.  It may be bug a bug in clang, but in any case we should
fall back to computing it ourselves if it happens.

I changed getAbsoluteFilePath so that if tryGetRealPathName succeeds, we
return right away (a real path is always absolute).  Otherwise, we try
to build an absolute path, as we did before, but we also call
VFS->getRealPath to make sure to get the canonical path (e.g. without
any ".." in it).

Reviewers: malaperle

Subscribers: hokein, ilya-biryukov, ioeric, MaskRay, jkorous, cfe-commits

Differential Revision: https://reviews.llvm.org/D48687

llvm-svn: 339483

6 years ago[analyzer] Record nullability implications on getting items from NSDictionary
George Karpenkov [Fri, 10 Aug 2018 22:27:04 +0000 (22:27 +0000)]
[analyzer] Record nullability implications on getting items from NSDictionary

If we get an item from a dictionary, we know that the item is non-null
if and only if the key is non-null.

This patch is a rather hacky way to record this implication, because
some logic needs to be duplicated from the solver.
And yet, it's pretty simple, performant, and works.

Other possible approaches:

 - Record the implication, in future rely on Z3 to pick it up.
 - Generalize the current code and move it to the constraint manager.

rdar://34990742

Differential Revision: https://reviews.llvm.org/D50124

llvm-svn: 339482

6 years ago[LICM] Hoist assumes out of loops
Philip Reames [Fri, 10 Aug 2018 22:21:56 +0000 (22:21 +0000)]
[LICM] Hoist assumes out of loops

If we have an assume which is known to execute and whose operand is invariant, we can lift that into the pre-header. So long as we don't change which paths the assume executes on, this is a legal transformation. It's likely to be a useful canonicalization as other transforms only look for dominating assumes.

Differential Revision: https://reviews.llvm.org/D50364

llvm-svn: 339481

6 years agoRevert "[NFC] More ConstantMerge refactoring"
JF Bastien [Fri, 10 Aug 2018 22:10:20 +0000 (22:10 +0000)]
Revert "[NFC] More ConstantMerge refactoring"

Sanitizers seem unhappy.

llvm-svn: 339480

6 years agoFix unused lambda capture warning from r339472.
Eli Friedman [Fri, 10 Aug 2018 22:03:25 +0000 (22:03 +0000)]
Fix unused lambda capture warning from r339472.

llvm-svn: 339479

6 years ago[NFC] More ConstantMerge refactoring
JF Bastien [Fri, 10 Aug 2018 21:58:00 +0000 (21:58 +0000)]
[NFC] More ConstantMerge refactoring

This makes my upcoming patch much easier to read.

llvm-svn: 339478

6 years ago[ELF][HEXAGON] Add R_HEX_8_X relocation
Sid Manning [Fri, 10 Aug 2018 21:48:40 +0000 (21:48 +0000)]
[ELF][HEXAGON] Add R_HEX_8_X relocation

Differential Revision: https://reviews.llvm.org/D50577

llvm-svn: 339477

6 years ago[analyzer] Fix tracking expressions through negation operator
George Karpenkov [Fri, 10 Aug 2018 21:42:19 +0000 (21:42 +0000)]
[analyzer] Fix tracking expressions through negation operator

Differential Revision: https://reviews.llvm.org/D50537

llvm-svn: 339476

6 years ago[analyzer] [NFC] [tests] Move plist-based diagnostics tests to separate files, use...
George Karpenkov [Fri, 10 Aug 2018 21:36:45 +0000 (21:36 +0000)]
[analyzer] [NFC] [tests] Move plist-based diagnostics tests to separate files, use diff instead of a FileCheck

Some of the analyzer tests check the exact plist output, in order to
verify that the diagnostics produced is correct.
Current testing setup has many issues:

plist output clobbers tests, making them harder to read
it is impossible to debug test failures given error messages from FileCheck.
The only recourse is manually creating the files and using the diff
again, it is impossible to update the tests given the error message:
the only process is a tedious manual one,
going from a separate plist file to CHECK directives.

This patch offers a much better approach of using "diff" directly in place of FileCheck,
and moving tests to separate files.

Generated using the following script:

```
import os
import glob
import re
import subprocess

diagnostics_key = "// CHECK:  <key>diagnostics</key>"

def process_file(f, data):
    idx = data.index(diagnostics_key)
    plist_out_f = 'ExpectedOutputs/plists/%s.plist' % f
    plist_out_folder = os.path.join('ExpectedOutputs/plists/', os.path.dirname(f))
    plist_data = data[idx:]
    plist_data = plist_data.replace('// CHECK: ', '')
    plist_data = plist_data.replace('// CHECK-NEXT: ', '')
    plist_data += "</dict>\n</plist>\n"
    data = data[:idx]

    ptn = re.compile("FileCheck --?input-file(=| )(%t|%t\.plist) %s")

    if not ptn.findall(data):
        print "none found =/ skipping..."
        return

    data = ptn.sub(lambda m: "tail -n +11 %s | diff -u -w - %%S/../%s" % (m.group(2), plist_out_f), data)

    with open(f, 'w') as out_f:
        out_f.write(data)

    subprocess.check_call(["mkdir", "-p", plist_out_folder])
    with open(plist_out_f, 'w') as out_f:
        out_f.write(plist_data)

def main():
    files = glob.glob("**/*.*")
    for f in files:
        with open(f) as f_handler:
            data = f_handler.read()
        if diagnostics_key in data:
            print "Converting %s" %f
            process_file(f, data)

if __name__ == "__main__":
    main()
```

Differential Revision: https://reviews.llvm.org/D50545

llvm-svn: 339475

6 years ago[WebAssembly] Added default stack-only instruction mode for MC.
Wouter van Oortmerssen [Fri, 10 Aug 2018 21:32:47 +0000 (21:32 +0000)]
[WebAssembly] Added default stack-only instruction mode for MC.

Summary:
Moved Explicit Locals pass to last.
Made that pass obligatory.
Made it convert from register to stack based instructions, and removed the registers.
Fixes to related code that was expecting register based instructions.
Added the correct testing flag to all tests, depending on what the
format they were expecting so far.
Translated one test to stack format as example: reg-stackify-stack.ll

tested:
llvm-lit -v `find test -name WebAssembly`
unittests/MC/*

Reviewers: dschuff, sunfish

Subscribers: jfb, llvm-commits, aheejin, eraman, jgravelle-google, sbc100

Differential Revision: https://reviews.llvm.org/D50568

llvm-svn: 339474

6 years agoRemove copy-pasted and unrelated comment [NFC]
Raphael Isemann [Fri, 10 Aug 2018 21:31:44 +0000 (21:31 +0000)]
Remove copy-pasted and unrelated comment [NFC]

That comment was copied from the
CombineConsecutiveEntriesWithEqualData() implementation below,
and doesn't actually describe what's happening in the current
function.

llvm-svn: 339473

6 years ago[ARM] Adjust AND immediates to make them cheaper to select.
Eli Friedman [Fri, 10 Aug 2018 21:21:53 +0000 (21:21 +0000)]
[ARM] Adjust AND immediates to make them cheaper to select.

LLVM normally prefers to minimize the number of bits set in an AND
immediate, but that doesn't always match the available ARM instructions.
In Thumb1 mode, prefer uxtb or uxth where possible; otherwise, prefer
a two-instruction sequence movs+ands or movs+bics.

Some potential improvements outlined in
ARMTargetLowering::targetShrinkDemandedConstant, but seems to work
pretty well already.

The ARMISelDAGToDAG fix ensures we don't generate an invalid UBFX
instruction due to a larger-than-expected mask. (It's orthogonal, in
some sense, but as far as I can tell it's either impossible or nearly
impossible to reproduce the bug without this change.)

According to my testing, this seems to consistently improve codesize by
a small amount by forming bic more often for ISD::AND with an immediate.

Differential Revision: https://reviews.llvm.org/D50030

llvm-svn: 339472

6 years ago[MS Demangler] Support extern "C" functions.
Zachary Turner [Fri, 10 Aug 2018 21:09:05 +0000 (21:09 +0000)]
[MS Demangler] Support extern "C" functions.

There are two cases we need to support with extern "C"
functions.  The first is the case of a '9' indicating that
the function has no prototype.  This occurs when we mangle
a symbol inside of an extern "C" function, but not the
function itself.

The second case is when we have an overloaded extern "C"
functions.  In this case we emit $$J0 to indicate this.
This patch adds support for both of these cases.

llvm-svn: 339471

6 years ago[InstCombine] add tests for fsub factorization; NFC
Sanjay Patel [Fri, 10 Aug 2018 21:00:27 +0000 (21:00 +0000)]
[InstCombine] add tests for fsub factorization; NFC

The tests show that;
1. The fold doesn't fire for vectors, but it should.
2. The fold fires regardless of uses, but it shouldn't.

llvm-svn: 339470

6 years ago[InstCombine] rearrange code for foldSelectBinOpIdentity; NFCI
Sanjay Patel [Fri, 10 Aug 2018 20:30:35 +0000 (20:30 +0000)]
[InstCombine] rearrange code for foldSelectBinOpIdentity; NFCI

This is a retry of rL339439 with a fix for the problem that
caused the original commit to be reverted at rL339446.

That problem was that the compare can be integer while
the binop is FP or vice-versa, so we need to use the binop
type when we ask for the identity constant.

A test to guard against the problem was added at rL339453.

llvm-svn: 339469

6 years ago[llvm-mca] Make InstrBuilder::getOrCreateInstrDesc private. NFC.
Matt Davis [Fri, 10 Aug 2018 20:24:27 +0000 (20:24 +0000)]
[llvm-mca] Make InstrBuilder::getOrCreateInstrDesc private. NFC.

llvm-svn: 339468

6 years ago[InstCombine] add tests to show disabling of libcall/intrinsic shrinking; NFC
Sanjay Patel [Fri, 10 Aug 2018 20:12:36 +0000 (20:12 +0000)]
[InstCombine] add tests to show disabling of libcall/intrinsic shrinking; NFC

llvm-svn: 339467

6 years agoResubmit r339450 - [MS Demangler] Add conversion operator tests
Zachary Turner [Fri, 10 Aug 2018 20:08:46 +0000 (20:08 +0000)]
Resubmit r339450 - [MS Demangler] Add conversion operator tests

This was broken because of a malformed check line.  Incidentally,
this exposed a case where we crash when we should just be returning
an error, so we should fix that.  The demangler shouldn't crash due
to user input.

llvm-svn: 339466

6 years ago[MS Demangler] Demangle cv qualifiers on template args.
Zachary Turner [Fri, 10 Aug 2018 19:57:36 +0000 (19:57 +0000)]
[MS Demangler] Demangle cv qualifiers on template args.

Before we wouldn't properly demangle something like
Foo<const int>.  Template args have a special escape sequence
'$$C' that is optional, but if it is present contains
qualifiers.  So we need to check for this and only if it
present, demangle qualifiers before demangling the type.

With this fix, we re-enable some tests that were previously
marked FIXME.

llvm-svn: 339465

6 years agoAMDGPU: More canonicalized operations
Matt Arsenault [Fri, 10 Aug 2018 19:20:17 +0000 (19:20 +0000)]
AMDGPU: More canonicalized operations

llvm-svn: 339464

6 years agorevert r339450 - [MS Demangler] Add conversion operator tests
Sanjay Patel [Fri, 10 Aug 2018 19:20:16 +0000 (19:20 +0000)]
revert r339450 - [MS Demangler] Add conversion operator tests

Something here causes an assertion failure that killed a bunch of bots.
Example:
http://lab.llvm.org:8011/builders/reverse-iteration/builds/7021/steps/check_all/logs/stdio

llvm-svn: 339463

6 years agoAMDGPU: Combine and of seto/setuo and fp_class
Matt Arsenault [Fri, 10 Aug 2018 18:58:56 +0000 (18:58 +0000)]
AMDGPU: Combine and of seto/setuo and fp_class

Clear the nan (or non-nan) test bits from the mask.

llvm-svn: 339462

6 years agoAMDGPU: Turn class x, p_zero|n_zero into fcmp oeq x, 0
Matt Arsenault [Fri, 10 Aug 2018 18:58:49 +0000 (18:58 +0000)]
AMDGPU: Turn class x, p_zero|n_zero into fcmp oeq x, 0

The library does use this for some reason.

llvm-svn: 339461

6 years agoAMDGPU: Match isfinite pattern to class instructions
Matt Arsenault [Fri, 10 Aug 2018 18:58:41 +0000 (18:58 +0000)]
AMDGPU: Match isfinite pattern to class instructions

llvm-svn: 339460

6 years agoInvalidate static locals when escaping lambdas
George Karpenkov [Fri, 10 Aug 2018 18:28:04 +0000 (18:28 +0000)]
Invalidate static locals when escaping lambdas

Lambdas can affect static locals even without an explicit capture.

rdar://39537031

Differential Revision: https://reviews.llvm.org/D50368

llvm-svn: 339459

6 years agoAMDGPU: Add LLVM_FALLTHROUGH
Matt Arsenault [Fri, 10 Aug 2018 17:57:12 +0000 (17:57 +0000)]
AMDGPU: Add LLVM_FALLTHROUGH

llvm-svn: 339458

6 years ago[tests, libstdcxx] Add missing test category on the TestDataFormatterStdUniquePtr...
Stella Stamenova [Fri, 10 Aug 2018 17:52:45 +0000 (17:52 +0000)]
[tests, libstdcxx] Add missing test category on the TestDataFormatterStdUniquePtr tests

Each test needs to be marked with the add_test_categories decorator individually.

llvm-svn: 339457

6 years agoAllow relockable scopes with thread safety attributes.
Aaron Ballman [Fri, 10 Aug 2018 17:33:47 +0000 (17:33 +0000)]
Allow relockable scopes with thread safety attributes.

Patch by Aaron Puchert

llvm-svn: 339456

6 years agoUpdate the coding standards and developer policy documentation surrounding whitespace.
Aaron Ballman [Fri, 10 Aug 2018 17:26:07 +0000 (17:26 +0000)]
Update the coding standards and developer policy documentation surrounding whitespace.

Clarify that you should not introduce trailing whitespace when making a commit and that you should not remove trailing whitespace that's unrelated to code you are changing or are about to change. Then clarified the developer policy around what is considered an obvious whitespace commit.

llvm-svn: 339455

6 years ago[clangd] extend the publishDiagnostics response to send back fixits to the
Alex Lorenz [Fri, 10 Aug 2018 17:25:07 +0000 (17:25 +0000)]
[clangd] extend the publishDiagnostics response to send back fixits to the
client if the client supports this extension

This commit extends the 'textDocument/publishDiagnostics' notification sent from
Clangd to the client.  When it's enabled, Clangd sends out the fixits associated
with the appropriate diagnostic in the body of the 'publishDiagnostics'
notification. The client can enable this extension by setting 'clangdFixSupport'
to true in the textDocument capabilities during initialization.

Differential Revision: https://reviews.llvm.org/D50415

llvm-svn: 339454

6 years ago[InstCombine] add/update tests for selectBinOpIdentity; NFC
Sanjay Patel [Fri, 10 Aug 2018 17:20:24 +0000 (17:20 +0000)]
[InstCombine] add/update tests for selectBinOpIdentity; NFC

This includes a test that would have exposed the bug in rL339439
which was reverted at rL339446. The compare can be integer while
the binop is FP or vice-versa, so we need to use the binop type
when we ask for the identity constant.

llvm-svn: 339453

6 years agoMake changes to the check strings so that the test I modified in r339438
Akira Hatanaka [Fri, 10 Aug 2018 17:07:27 +0000 (17:07 +0000)]
Make changes to the check strings so that the test I modified in r339438
passes on 32-bit targets.

llvm-svn: 339452

6 years ago[libcxx] Mark charconv tests as failing for previous libcxx versions.
Volodymyr Sapsai [Fri, 10 Aug 2018 17:03:47 +0000 (17:03 +0000)]
[libcxx] Mark charconv tests as failing for previous libcxx versions.

<charconv> was added in r338479. Previous libcxx versions don't have
this functionality and corresponding tests should be failing.

Reviewers: mclow.lists, ldionne, EricWF

Reviewed By: ldionne

Subscribers: christof, dexonsmith, lichray, EricWF, cfe-commits

Differential Revision: https://reviews.llvm.org/D50543

llvm-svn: 339451

6 years ago[MS Demangler] Add conversion operator tests.
Zachary Turner [Fri, 10 Aug 2018 16:55:59 +0000 (16:55 +0000)]
[MS Demangler] Add conversion operator tests.

The mangled names were added in the original commit, but
the demangled equivalents weren't, so nothing was actually
being checked.

llvm-svn: 339450

6 years ago[hwasan] Remove liblog dependency.
Evgeniy Stepanov [Fri, 10 Aug 2018 16:38:57 +0000 (16:38 +0000)]
[hwasan] Remove liblog dependency.

HWASan will not run on older Android releases where we use
__android_log_write for logging.

This dependency is also harmful in the case when libc itself depends
on hwasan, because it creates a loop of
  libc -> hwasan -> liblog -> libc
which makes liblog vs libc initialization order undetermined.

Without liblog the loop is just
  libc -> hwasan -> libc
and any init order issues can be solved in hwasan.

llvm-svn: 339449

6 years ago[llvm-objcopy] NFC: consistently use typename ELFT::<X> definitions in headers
Jordan Rupprecht [Fri, 10 Aug 2018 16:25:58 +0000 (16:25 +0000)]
[llvm-objcopy] NFC: consistently use typename ELFT::<X> definitions in headers

llvm-svn: 339448

6 years ago[hwasan] Add -hwasan-with-ifunc flag.
Evgeniy Stepanov [Fri, 10 Aug 2018 16:21:37 +0000 (16:21 +0000)]
[hwasan] Add -hwasan-with-ifunc flag.

Summary: Similar to asan's flag, it can be used to disable the use of ifunc to access hwasan shadow address.

Reviewers: vitalybuka, kcc

Subscribers: srhines, hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D50544

llvm-svn: 339447

6 years ago[InstCombine] revert r339439 - rearrange code for foldSelectBinOpIdentity
Sanjay Patel [Fri, 10 Aug 2018 16:12:19 +0000 (16:12 +0000)]
[InstCombine] revert r339439 - rearrange code for foldSelectBinOpIdentity

That was supposed to be NFC, but it exposed a logic hole somewhere that
caused bots to fail.

llvm-svn: 339446

6 years ago[InstCombine][NFC] Added tests for select with binop fold
David Bolvansky [Fri, 10 Aug 2018 15:29:09 +0000 (15:29 +0000)]
[InstCombine][NFC] Added tests for select with binop fold

llvm-svn: 339441

6 years agoRichManglingContext: Make m_ipd_str_len a local variable and simplify processIPDStrRe...
Stefan Granitz [Fri, 10 Aug 2018 15:21:33 +0000 (15:21 +0000)]
RichManglingContext: Make m_ipd_str_len a local variable and simplify processIPDStrResult + polishing in test and Mangled

llvm-svn: 339440

6 years ago[InstCombine] rearrange code for foldSelectBinOpIdentity; NFCI
Sanjay Patel [Fri, 10 Aug 2018 15:11:26 +0000 (15:11 +0000)]
[InstCombine] rearrange code for foldSelectBinOpIdentity; NFCI

This should make it easier to folow and to add the planned enhancements
such as D50190.

llvm-svn: 339439

6 years ago[CodeGen] Merge equivalent block copy/helper functions.
Akira Hatanaka [Fri, 10 Aug 2018 15:09:24 +0000 (15:09 +0000)]
[CodeGen] Merge equivalent block copy/helper functions.

Clang generates copy and dispose helper functions for each block literal
on the stack. Often these functions are equivalent for different blocks.
This commit makes changes to merge equivalent copy and dispose helper
functions and reduce code size.

To enable merging equivalent copy/dispose functions, the captured object
infomation is encoded into the helper function name. This allows IRGen
to check whether an equivalent helper function has already been emitted
and reuse the function instead of generating a new helper function
whenever a block is defined. In addition, the helper functions are
marked as linkonce_odr to enable merging helper functions that have the
same name across translation units and marked as unnamed_addr to enable
the linker's deduplication pass to merge functions that have different
names but the same content.

rdar://problem/42640608

Differential Revision: https://reviews.llvm.org/D50152

llvm-svn: 339438

6 years ago[clang-tidy] check_clang_tidy.py: support CHECK-NOTES prefix
Roman Lebedev [Fri, 10 Aug 2018 15:05:46 +0000 (15:05 +0000)]
[clang-tidy] check_clang_tidy.py: support CHECK-NOTES prefix

Summary:
Currently, there is two configured prefixes: `CHECK-FIXES` and `CHECK-MESSAGES`
`CHECK-MESSAGES` checks that there are no test output lines with `warning:|error:`, which are not explicitly handled in lit tests.
However there does not seem to be a nice way to enforce for all the `note:` to be checked.
This was useful for me when developing D36836.

Reviewers: alexfh, klimek, aaron.ballman, hokein

Reviewed By: alexfh, aaron.ballman

Subscribers: JonasToth, JDevlieghere, xazax.hun, cfe-commits

Tags: #clang-tools-extra

Differential Revision: https://reviews.llvm.org/D36892

llvm-svn: 339437

6 years ago[MS Demangler] Properly demangle conversion operators.
Zachary Turner [Fri, 10 Aug 2018 15:04:56 +0000 (15:04 +0000)]
[MS Demangler] Properly demangle conversion operators.

These were completely broken before.  We need to handle
the 'B' operator tag.

llvm-svn: 339436

6 years ago[MS Demangler] Disable a couple of tests.
Zachary Turner [Fri, 10 Aug 2018 14:53:33 +0000 (14:53 +0000)]
[MS Demangler] Disable a couple of tests.

The check lines are marked FIXME but not the mangled names.
This is causing an error.

llvm-svn: 339435

6 years ago[MS Demangler] Fix several issues related to templates.
Zachary Turner [Fri, 10 Aug 2018 14:31:04 +0000 (14:31 +0000)]
[MS Demangler] Fix several issues related to templates.

These were uncovered when porting the mangling tests in
ms-templates.cpp from clang/CodeGenCXX over to demangling
tests.  The main issues fixed here are surrounding integer
literal signed and unsignedness, empty array dimensions,
and pointer and reference non-type template parameters.

Differential Revision: https://reviews.llvm.org/D50512

llvm-svn: 339434

6 years ago[clang-tidy: modernize] modernize-redundant-void-arg crashes when a function body...
Alexander Kornienko [Fri, 10 Aug 2018 13:59:33 +0000 (13:59 +0000)]
[clang-tidy: modernize] modernize-redundant-void-arg crashes when a function body is in a macro

Fixes https://bugs.llvm.org/show_bug.cgi?id=28406

Patch by IdrissRio.

Differential revision: https://reviews.llvm.org/D49800

llvm-svn: 339433

6 years ago[ARM] Disallow zexts in ARMCodeGenPrepare
Sam Parker [Fri, 10 Aug 2018 13:57:13 +0000 (13:57 +0000)]
[ARM] Disallow zexts in ARMCodeGenPrepare

Enabling ARMCodeGenPrepare by default caused a whole load of
failures. This is due to zexts and truncs not being handled properly.
ZExts are messy so it's just easier to disable for now and truncs
are allowed only as 'sinks'. I still need to figure out why allowing
them as 'sources' causes so many failures. The other main changes are
that we are explicit in the types that we converting to, it's now
always 'TypeSize'. Type support is also now performed while checking
for valid opcodes as it unnecessarily complicated having the checks
are different stages.

I've moved the tests around too, so we have the zext and truncs in
their own file as well as the overflowing opcode tests.

Differential Revision: https://reviews.llvm.org/D50518

llvm-svn: 339432

6 years ago[libc++] Enable aligned allocation based on feature test macro, irrespective of standard
Louis Dionne [Fri, 10 Aug 2018 13:24:56 +0000 (13:24 +0000)]
[libc++] Enable aligned allocation based on feature test macro, irrespective of standard

Summary:
The current code enables aligned allocation functions when compiling in C++17
and later. This is a problem because aligned allocation functions might not
be supported on the target platform, which leads to an error at link time.

Since r338934, Clang knows not to define __cpp_aligned_new when it's not
available on the target platform -- this commit takes advantage of that to
only use aligned allocation functions when they are available.

Reviewers: vsapsai, EricWF

Subscribers: christof, dexonsmith, cfe-commits, EricWF, mclow.lists

Differential Revision: https://reviews.llvm.org/D50344

llvm-svn: 339431

6 years agoAmend "Remove unused type Either from Utility library".
Tatyana Krasnukha [Fri, 10 Aug 2018 13:01:26 +0000 (13:01 +0000)]
Amend "Remove unused type Either from Utility library".

llvm-svn: 339430

6 years agoFix a deprecated warning in the last commit.
David Chisnall [Fri, 10 Aug 2018 12:53:18 +0000 (12:53 +0000)]
Fix a deprecated warning in the last commit.

Done as a separate commit to make it easier to cherry pick the changes
to the release branch.

llvm-svn: 339429

6 years agoAdd Windows support for the GNUstep Objective-C ABI V2.
David Chisnall [Fri, 10 Aug 2018 12:53:13 +0000 (12:53 +0000)]
Add Windows support for the GNUstep Objective-C ABI V2.

Summary:
Introduces funclet-based unwinding for Objective-C and fixes an issue
where global blocks can't have their isa pointers initialised on
Windows.

After discussion with Dustin, this changes the name mangling of
Objective-C types to prevent a C++ catch statement of type struct X*
from catching an Objective-C object of type X*.

Reviewers: rjmccall, DHowett-MSFT

Reviewed By: rjmccall, DHowett-MSFT

Subscribers: mgrang, mstorsjo, smeenai, cfe-commits

Differential Revision: https://reviews.llvm.org/D50144

llvm-svn: 339428

6 years ago[clang-tidy] run-clang-tidy.py - add synchronisation to the output
Andi-Bogdan Postelnicu [Fri, 10 Aug 2018 11:50:47 +0000 (11:50 +0000)]
[clang-tidy]  run-clang-tidy.py - add synchronisation to the output
Differential Revision: https://reviews.llvm.org/D49851

llvm-svn: 339427

6 years ago[clangd] Allow consuming limited number of items
Kirill Bobyrev [Fri, 10 Aug 2018 11:50:44 +0000 (11:50 +0000)]
[clangd] Allow consuming limited number of items

This patch modifies `consume` function to allow retrieval of limited
number of symbols. This is the "cheap" implementation of top-level
limiting iterator. In the future we would like to have a complete limit
iterator implementation to insert it into the query subtrees, but in the
meantime this version would be enough for a fully-functional
proof-of-concept Dex implementation.

Reviewers: ioeric, ilya-biryukov

Reviewed by: ioeric

Differential Revision: https://reviews.llvm.org/D50500

llvm-svn: 339426

6 years ago[X86][SSE] Pull out repeated shift getOpcode() calls. NFCI.
Simon Pilgrim [Fri, 10 Aug 2018 11:42:42 +0000 (11:42 +0000)]
[X86][SSE] Pull out repeated shift getOpcode() calls. NFCI.

llvm-svn: 339425

6 years agoclang-cl: accept -fcrash-diagnostics-dir=
Hans Wennborg [Fri, 10 Aug 2018 11:40:50 +0000 (11:40 +0000)]
clang-cl: accept -fcrash-diagnostics-dir=

llvm-svn: 339424

6 years agoFix a wrong type bug in ParsedAttr::TypeTagForDatatypeData
Bruno Ricci [Fri, 10 Aug 2018 11:20:20 +0000 (11:20 +0000)]
Fix a wrong type bug in ParsedAttr::TypeTagForDatatypeData

This patch fixes a wrong type bug inside ParsedAttr::TypeTagForDatatypeData.
The details to the best of my knowledge are as follow. The incredible thing
is that everything works out just fine by chance due to a sequence of lucky
coincidences in the layout of various types.

The struct ParsedAttr::TypeTagForDatatypeData contains among other things
a ParsedType *MatchingCType, where ParsedType is just OpaquePtr<QualType>.

However the member MatchingCType is initialized in the constructor for
type_tag_for_datatype attribute as follows:

new (&ExtraData.MatchingCType) ParsedType(matchingCType);

This results in the ParsedType being constructed in the location of the
ParsedType * Later ParsedAttr::getMatchingCType do return
*getTypeTagForDatatypeDataSlot().MatchingCType; which instead of
dereferencing the ParsedType * will dereference the QualType inside
the ParsedType. Now this QualType in this case contains no qualifiers
and therefore is a valid Type *. Therefore getMatchingCType returns a
Type or at least the stuff that is in the first sizeof(void*) bytes of it,
But it turns out that Type inherits from ExtQualsCommonBase and that the
first member of ExtQualsCommonBase is a const Type *const BaseType. This
Type * in this case points to the original Type pointed to by the
QualType and so everything works fine even though all the types were wrong.

This bug was only found because I changed the layout of Type,
which obviously broke all of this long chain of improbable events.

Reviewed By: erichkeane

Differential Revision: https://reviews.llvm.org/D50532

llvm-svn: 339423

6 years agoFix -Wimplicit-fallthrough warning introduced in rL339397.
Simon Pilgrim [Fri, 10 Aug 2018 11:02:44 +0000 (11:02 +0000)]
Fix -Wimplicit-fallthrough warning introduced in rL339397.

llvm-svn: 339422

6 years ago[Tablegen][SubtargetEmitter] refactor method `emitSchedModelHelpersImpl()`. NFCI
Andrea Di Biagio [Fri, 10 Aug 2018 10:43:43 +0000 (10:43 +0000)]
[Tablegen][SubtargetEmitter] refactor method `emitSchedModelHelpersImpl()`. NFCI

Part of the logic has been moved to helper functions to (hopefully) improve
readability.
Added a few code comments to better describe how the algorithm works.

No functional change intended.

llvm-svn: 339421

6 years agoclang-cl: Support /guard:cf,nochecks
Hans Wennborg [Fri, 10 Aug 2018 09:49:21 +0000 (09:49 +0000)]
clang-cl: Support /guard:cf,nochecks

This extension emits the guard cf table without inserting the
instrumentation. Currently that's what clang-cl does with /guard:cf
anyway, but this allows a user to request that explicitly.

Differential Revision: https://reviews.llvm.org/D50513

llvm-svn: 339420

6 years agoRename the cfguard module flag to cfguardtable
Hans Wennborg [Fri, 10 Aug 2018 09:48:53 +0000 (09:48 +0000)]
Rename the cfguard module flag to cfguardtable

The previous name sounds like it inserts cfguard implementation, but it
really just emits the table of address-taken functions. Change the name
to better reflect that.

Clang will be updated in the next commit.

llvm-svn: 339419

6 years ago[NFC] Add tests that demonstrate that MustExecute is fundamentally broken
Max Kazantsev [Fri, 10 Aug 2018 09:20:46 +0000 (09:20 +0000)]
[NFC] Add tests that demonstrate that MustExecute is fundamentally broken

llvm-svn: 339417

6 years ago[clangd] Fix a "-Wdangling-else" compiler warning in the test.
Haojian Wu [Fri, 10 Aug 2018 08:34:16 +0000 (08:34 +0000)]
[clangd] Fix a "-Wdangling-else" compiler warning in the test.

llvm-svn: 339416

6 years ago[clang-tidy] Omit cases where loop variable is not used in loop body in
Haojian Wu [Fri, 10 Aug 2018 08:25:51 +0000 (08:25 +0000)]
[clang-tidy] Omit cases where loop variable is not used in loop body in
performance-for-range-copy check.

Summary:
The upstream change r336737 make the check too smart to fix the case
where loop variable could be used as `const auto&`.

But for the case below, changing to `const auto _` will introduce
an unused complier warning.

```
for (auto _ : state) {
  // no references for _.
}
```

This patch omit this case, and it is safe to do it as the case is very rare.

Reviewers: ilya-biryukov, alexfh

Subscribers: xazax.hun, cfe-commits

Differential Revision: https://reviews.llvm.org/D50447

llvm-svn: 339415

6 years ago[MSan] Shrink the register save area for non-SSE builds
Alexander Potapenko [Fri, 10 Aug 2018 08:06:43 +0000 (08:06 +0000)]
[MSan] Shrink the register save area for non-SSE builds

If code is compiled for X86 without SSE support, the register save area
doesn't contain FPU registers, so `AMD64FpEndOffset` should be equal to
`AMD64GpEndOffset`.

llvm-svn: 339414

6 years ago[ELF] - Get rid of SyntheticSection::postThunkContents(). NFCI.
George Rimar [Fri, 10 Aug 2018 07:24:18 +0000 (07:24 +0000)]
[ELF] - Get rid of SyntheticSection::postThunkContents(). NFCI.

It turns out that postThunkContents() is only used for
sorting symbols in .symtab.

Though we can instead move the logic to SymbolTableBaseSection::finalizeContents(),
postpone calling it and then get rid of postThunkContents completely.

Differential revision: https://reviews.llvm.org/D49547

llvm-svn: 339413

6 years ago[LLD][ELF] - Fix crash when using empty --defsym.
George Rimar [Fri, 10 Aug 2018 06:32:39 +0000 (06:32 +0000)]
[LLD][ELF] - Fix crash when using empty --defsym.

We have a crash issue when handling the empty -defsym.

For parsing this option we are using ScriptParser class which is used
generally for reading the linker script. For empty defsym case, we
pass the empty memory buffer and crash in the place removed in https://reviews.llvm.org/rL336436.

But reverting of the above patch would not help here (we would still crash but a bit later). And
even after fixing the crash we would report something like
"lld.exe: error: -defsym:1: unexpected EOF"
It is probably not the appropriate message because mentions EOF.

I think the issue should be handled on a higher level like this patch does.
So we do not want to pass the empty memory buffer first of all I believe.

Differential revision: https://reviews.llvm.org/D50498

llvm-svn: 339412

6 years ago[MemorySSA] "Fix" lifetime intrinsic handling
George Burgess IV [Fri, 10 Aug 2018 05:14:43 +0000 (05:14 +0000)]
[MemorySSA] "Fix" lifetime intrinsic handling

MemorySSA currently creates MemoryAccesses for lifetime intrinsics, and
sometimes treats them as clobbers. This may/may not be the best way
forward, but while we're doing it, we should consider
MayAlias/PartialAlias to be clobbers.

The ideal fix here is probably to remove all of this reasoning about
lifetimes from MemorySSA + put it into the passes that need to care. But
that's a wayyy broader fix that needs some consensus, and we have
miscompiles + a release branch today, and this should solve the
miscompiles just as well.

differential revision is D43269. Landing without an explicit LGTM (and
without using the special please-autoclose-this syntax) so we can still
use that revision as a place to decide what the right fix here is.

llvm-svn: 339411

6 years ago[InstCombine] Transform str(n)cmp to memcmp
David Bolvansky [Fri, 10 Aug 2018 04:32:54 +0000 (04:32 +0000)]
[InstCombine] Transform str(n)cmp to memcmp

Summary:
Motivation examples:
int strcmp_memcmp() {
    char buf[12];
    return strcmp(buf, "key") == 0;
}

int strcmp_memcmp2() {
    char buf[12];
    return strcmp(buf, "key") != 0;
}

int strncmp_memcmp() {
    char buf[12];
    return strncmp(buf, "key", 3) == 0;
}

can be turned to memcmp.

See test file for more cases.

Reviewers: efriedma

Reviewed By: efriedma

Subscribers: spatel, llvm-commits

Differential Revision: https://reviews.llvm.org/D50233

llvm-svn: 339410

6 years agoFix size calculation from r339380
Richard Trieu [Fri, 10 Aug 2018 01:30:10 +0000 (01:30 +0000)]
Fix size calculation from r339380

r339380 changed the trailing types of ParsedAttr to use llvm::TrailingObjects.
However, it did not copy over one of the size attributes, causing a too
small allocation for this object.  The error was detected with
AddressSanitizer use-after-poison

llvm-svn: 339409

6 years agoRevert "SafeStack: Delay thread stack clean-up"
Vlad Tsyrklevich [Fri, 10 Aug 2018 00:36:04 +0000 (00:36 +0000)]
Revert "SafeStack: Delay thread stack clean-up"

This reverts commit r339405, it's failing on Darwin buildbots because
it doesn't seem to have a tgkill/thr_kill2 interface. It has a
__pthread_kill() syscall, but that relies on having a handle to the
thread's port which is not equivalent to it's tid.

llvm-svn: 339408