platform/upstream/llvm.git
5 years ago[analyzer] NFC: Move getStmt() and createEndOfPath() out of PathDiagnostic.
Artem Dergachev [Wed, 11 Sep 2019 20:54:21 +0000 (20:54 +0000)]
[analyzer] NFC: Move getStmt() and createEndOfPath() out of PathDiagnostic.

These static functions deal with ExplodedNodes which is something we don't want
the PathDiagnostic interface to know anything about, as it's planned to be
moved out of libStaticAnalyzerCore.

Differential Revision: https://reviews.llvm.org/D67382

llvm-svn: 371659

5 years ago[analyzer] NFC: Re-implement stack hints as a side map in BugReport.
Artem Dergachev [Wed, 11 Sep 2019 20:54:17 +0000 (20:54 +0000)]
[analyzer] NFC: Re-implement stack hints as a side map in BugReport.

That's one of the few random entities in the PathDiagnostic interface that
are specific to the Static Analyzer. By moving them out we could let
everybody use path diagnostics without linking against Static Analyzer.

Differential Revision: https://reviews.llvm.org/D67381

llvm-svn: 371658

5 years agoAdd a LineTable class to GSYM and test it.
Greg Clayton [Wed, 11 Sep 2019 20:51:03 +0000 (20:51 +0000)]
Add a LineTable class to GSYM and test it.

This patch adds the ability to create a gsym::LineTable object, populate it, encode and decode it and test all functionality.

The full format of the LineTable encoding is specified in the header file LineTable.h.

Differential Revision: https://reviews.llvm.org/D66602

llvm-svn: 371657

5 years ago[clang-scan-deps] add skip excluded conditional preprocessor block preprocessing...
Alex Lorenz [Wed, 11 Sep 2019 20:40:31 +0000 (20:40 +0000)]
[clang-scan-deps] add skip excluded conditional preprocessor block preprocessing optimization

This commit adds an optimization to clang-scan-deps and clang's preprocessor that skips excluded preprocessor
blocks by bumping the lexer pointer, and not lexing the tokens until reaching appropriate #else/#endif directive.
The skip positions and lexer offsets are computed when the file is minimized, directly from the minimized tokens.

On an 18-core iMacPro with macOS Catalina Beta I got 10-15% speed-up from this optimization when running clang-scan-deps on
the compilation database for a recent LLVM and Clang (3511 files).

Differential Revision: https://reviews.llvm.org/D67127

llvm-svn: 371656

5 years agoFix -Wnonportable-include-path suppression for header maps with absolute paths.
Volodymyr Sapsai [Wed, 11 Sep 2019 20:39:04 +0000 (20:39 +0000)]
Fix -Wnonportable-include-path suppression for header maps with absolute paths.

In `DirectoryLookup::LookupFile` parameter `HasBeenMapped` doesn't cover
the case when clang finds a file through a header map but doesn't remap
the lookup filename because the target path is an absolute path. As a
result, -Wnonportable-include-path suppression for header maps
introduced in r301592 wasn't triggered.

Change parameter `HasBeenMapped` to `IsInHeaderMap` and use parameter
`MappedName` to track the filename remapping. This way we can handle
both relative and absolute paths in header maps, and account for their
specific properties, like filename remapping being a property preserved
across lookups in multiple directories.

rdar://problem/39516483

Reviewers: dexonsmith, bruno

Reviewed By: dexonsmith

Subscribers: jkorous, cfe-commits, ributzka

Differential Revision: https://reviews.llvm.org/D58094

llvm-svn: 371655

5 years ago[Plugins/Process] Remove direct use of ClangASTContext from InferiorCallPOSIX
Alex Langford [Wed, 11 Sep 2019 20:36:28 +0000 (20:36 +0000)]
[Plugins/Process] Remove direct use of ClangASTContext from InferiorCallPOSIX

Summary:
InferiorCallPOSIX directly grabs a ClangASTContext from the Target it
has and does no error checking. I don't think these functions have a
reason to know about clang specifically. Additionally, using
`GetScratchTypeSystemForLanguage` forces us to do error checking since
it returns an Expected.

Differential Revision: https://reviews.llvm.org/D67427

llvm-svn: 371654

5 years ago[llvm-reduce] Fix a bug, improve error handling when running test
Reid Kleckner [Wed, 11 Sep 2019 20:29:22 +0000 (20:29 +0000)]
[llvm-reduce] Fix a bug, improve error handling when running test

llvm::sys::ExecuteAndWait can report errors, so let's make use of that.

Second, while iterating uses of functions to remove, a call can appear
multiple times. Use a SetVector so we don't attempt to erase such a call
twice.

llvm-svn: 371653

5 years agoUpdate link to the DWARF spec.
Adrian Prantl [Wed, 11 Sep 2019 19:57:29 +0000 (19:57 +0000)]
Update link to the DWARF spec.

llvm-svn: 371650

5 years ago[AMDGPU] Fix crash in phi-elimination hook.
Michael Liao [Wed, 11 Sep 2019 19:55:20 +0000 (19:55 +0000)]
[AMDGPU] Fix crash in phi-elimination hook.

Summary: - Pre-check in case there's just a single PHI insn.

Reviewers: alex-t, rampitec, arsenm

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, dstuttard, tpr, t-tye, hiraditya, llvm-commits, yaxunl

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67451

llvm-svn: 371649

5 years agoUpdate documentation.
Adrian Prantl [Wed, 11 Sep 2019 19:49:38 +0000 (19:49 +0000)]
Update documentation.

llvm-svn: 371648

5 years ago[libFuzzer] Remove hardcoded number of new features in merge_two_step.test.
Max Moroz [Wed, 11 Sep 2019 19:43:03 +0000 (19:43 +0000)]
[libFuzzer] Remove hardcoded number of new features in merge_two_step.test.

Summary:
The number of features can be different on different platforms.

This should fixed broken builders, e.g.
http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/7946

Reviewers: Dor1s

Reviewed By: Dor1s

Subscribers: kristof.beyls, delcypher, #sanitizers, llvm-commits

Tags: #llvm, #sanitizers

Differential Revision: https://reviews.llvm.org/D67458

llvm-svn: 371647

5 years ago[NFC] Added triple to test file to avoid arm buildbots failures
David Bolvansky [Wed, 11 Sep 2019 18:55:56 +0000 (18:55 +0000)]
[NFC] Added triple to test file to avoid arm buildbots failures

llvm-svn: 371646

5 years agoFix test failures after r371640
Matt Arsenault [Wed, 11 Sep 2019 18:55:20 +0000 (18:55 +0000)]
Fix test failures after r371640

r371640 evidently fixed bug 39481

llvm-svn: 371645

5 years ago[ConstantHoisting] Fix non-determinism.
Eli Friedman [Wed, 11 Sep 2019 18:55:00 +0000 (18:55 +0000)]
[ConstantHoisting] Fix non-determinism.

Differential Revision: https://reviews.llvm.org/D66114

llvm-svn: 371644

5 years ago[IRMover] Don't map globals if their types are the same
Pirama Arumuga Nainar [Wed, 11 Sep 2019 18:35:49 +0000 (18:35 +0000)]
[IRMover] Don't map globals if their types are the same

Summary:
During IR Linking, if the types of two globals in destination and source
modules are the same, it can only be because the global in the
destination module is originally from the source module and got added to
the destination module from a shared metadata.

We shouldn't map this type to itself in case the type's components get
remapped to a new type from the destination (for instance, during the
loop over SrcM->getIdentifiedStructTypes() further below in
IRLinker::computeTypeMapping()).

Fixes PR40312.

Reviewers: tejohnson, pcc, srhines

Subscribers: mehdi_amini, hiraditya, steven_wu, dexonsmith, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66814

llvm-svn: 371643

5 years ago[MS] Consder constexpr globals to be inline, as in C++17
Reid Kleckner [Wed, 11 Sep 2019 18:09:10 +0000 (18:09 +0000)]
[MS] Consder constexpr globals to be inline, as in C++17

Summary:
Microsoft seems to do this regardless of the language mode, so we must
also do it in order to be ABI compatible.

Fixes PR36125

Reviewers: thakis

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D47956

llvm-svn: 371642

5 years agoLiveIntervals: Split live intervals on multiple dead defs
Matt Arsenault [Wed, 11 Sep 2019 17:59:21 +0000 (17:59 +0000)]
LiveIntervals: Split live intervals on multiple dead defs

If there are multiple dead defs of the same virtual register, these
are required to be split into multiple virtual registers with separate
live intervals to avoid a verifier error.

llvm-svn: 371640

5 years agoConsolidate swap, swap_ranges, and iter_swap in <type_traits>.
Zoe Carver [Wed, 11 Sep 2019 17:39:24 +0000 (17:39 +0000)]
Consolidate swap, swap_ranges, and iter_swap in <type_traits>.

NFC. Thanks to @Quuxplusone (Arthur O'Dwyer) for this change.

llvm-svn: 371639

5 years ago[libc++] Add a CMake cache for Apple-specific configuration options
Louis Dionne [Wed, 11 Sep 2019 16:57:19 +0000 (16:57 +0000)]
[libc++] Add a CMake cache for Apple-specific configuration options

llvm-svn: 371638

5 years ago[Clang][Bundler] Replace std::vector by SmallVector [NFC]
Sergey Dmitriev [Wed, 11 Sep 2019 16:28:47 +0000 (16:28 +0000)]
[Clang][Bundler] Replace std::vector by SmallVector [NFC]

Differential Revision: https://reviews.llvm.org/D67413

llvm-svn: 371637

5 years agogn build: Merge r371635
Nico Weber [Wed, 11 Sep 2019 16:26:59 +0000 (16:26 +0000)]
gn build: Merge r371635

llvm-svn: 371636

5 years agoReland "clang-misexpect: Profile Guided Validation of Performance Annotations in...
Petr Hosek [Wed, 11 Sep 2019 16:19:50 +0000 (16:19 +0000)]
Reland "clang-misexpect: Profile Guided Validation of Performance Annotations in LLVM"

This patch contains the basic functionality for reporting potentially
incorrect usage of __builtin_expect() by comparing the developer's
annotation against a collected PGO profile. A more detailed proposal and
discussion appears on the CFE-dev mailing list
(http://lists.llvm.org/pipermail/cfe-dev/2019-July/062971.html) and a
prototype of the initial frontend changes appear here in D65300

We revised the work in D65300 by moving the misexpect check into the
LLVM backend, and adding support for IR and sampling based profiles, in
addition to frontend instrumentation.

We add new misexpect metadata tags to those instructions directly
influenced by the llvm.expect intrinsic (branch, switch, and select)
when lowering the intrinsics. The misexpect metadata contains
information about the expected target of the intrinsic so that we can
check against the correct PGO counter when emitting diagnostics, and the
compiler's values for the LikelyBranchWeight and UnlikelyBranchWeight.
We use these branch weight values to determine when to emit the
diagnostic to the user.

A future patch should address the comment at the top of
LowerExpectIntrisic.cpp to hoist the LikelyBranchWeight and
UnlikelyBranchWeight values into a shared space that can be accessed
outside of the LowerExpectIntrinsic pass. Once that is done, the
misexpect metadata can be updated to be smaller.

In the long term, it is possible to reconstruct portions of the
misexpect metadata from the existing profile data. However, we have
avoided this to keep the code simple, and because some kind of metadata
tag will be required to identify which branch/switch/select instructions
are influenced by the use of llvm.expect

Patch By: paulkirth
Differential Revision: https://reviews.llvm.org/D66324

llvm-svn: 371635

5 years agoRevert [InstCombine] Use SimplifyFMulInst to simplify multiply in fma.
Florian Hahn [Wed, 11 Sep 2019 16:17:03 +0000 (16:17 +0000)]
Revert [InstCombine] Use SimplifyFMulInst to simplify multiply in fma.

This introduces additional rounding error in some cases. See D67434.

This reverts r371518 (git commit 18a1f0818b659cee13865b4fad2648d85984a4ed)

llvm-svn: 371634

5 years ago[Clang][Bundler] Fix for a potential memory leak [NFC]
Sergey Dmitriev [Wed, 11 Sep 2019 16:03:21 +0000 (16:03 +0000)]
[Clang][Bundler] Fix for a potential memory leak [NFC]

Bundler leaks memory if it is called with -type=o but given input isn't an object file (though it has to have a known binary type like IR, archive, etc...). Memory leak is happening when binary object returned by the createBinary(...) call cannot be casted to an ObjectFile type. In this case returned BinaryOrErr object releases ownership of the binary, but no one is taking it (see line 626).

Differential Revision: https://reviews.llvm.org/D67416

llvm-svn: 371633

5 years ago[OPENMP] Update the diagnosis message for canonical loop form, by Chi
Alexey Bataev [Wed, 11 Sep 2019 15:44:06 +0000 (15:44 +0000)]
[OPENMP] Update the diagnosis message for canonical loop form, by Chi
Chun Chen.

The previous patch (https://reviews.llvm.org/D54441) support the
relational-op != very well for openmp canonical loop form, however,
it didn't update the diagnosis message. So this patch is simply
update the diagnosis message by adding !=, update the test
related to it, and update the section number for canonical loop
form for OpenMP 5.0 in comment.

Differential Revision: https://reviews.llvm.org/D66559

llvm-svn: 371631

5 years ago[InstSimplify] Pass SimplifyQuery into simplifyUnsignedRangeCheck() and use it for...
Roman Lebedev [Wed, 11 Sep 2019 15:32:46 +0000 (15:32 +0000)]
[InstSimplify] Pass SimplifyQuery into simplifyUnsignedRangeCheck() and use it for isKnownNonZero()

This was actually the original intention in D67332,
but i messed up and forgot about it.
This patch was originally part of D67411, but precommitting this.

llvm-svn: 371630

5 years ago[NFC][InstSimplify] Add extra test for D67411 with @llvm.assume
Roman Lebedev [Wed, 11 Sep 2019 15:28:03 +0000 (15:28 +0000)]
[NFC][InstSimplify] Add extra test for D67411 with @llvm.assume

llvm-svn: 371629

5 years ago[scudo][standalone] Android related improvements
Kostya Kortchinsky [Wed, 11 Sep 2019 14:48:41 +0000 (14:48 +0000)]
[scudo][standalone] Android related improvements

Summary:
This changes a few things to improve memory footprint and performances
on Android, and fixes a test compilation error:
- add `stdlib.h` to `wrappers_c_test.cc` to address
  https://bugs.llvm.org/show_bug.cgi?id=42810
- change Android size class maps, based on benchmarks, to improve
  performances and lower the Svelte memory footprint. Also change the
  32-bit region size for said configuration
- change the `reallocate` logic to reallocate in place for sizes larger
  than the original chunk size, when they still fit in the same block.
  This addresses patterns from `memory_replay` dumps like the following:
```
202: realloc 0xb48fd000 0xb4930650 12352
202: realloc 0xb48fd000 0xb48fd000 12420
202: realloc 0xb48fd000 0xb48fd000 12492
202: realloc 0xb48fd000 0xb48fd000 12564
202: realloc 0xb48fd000 0xb48fd000 12636
202: realloc 0xb48fd000 0xb48fd000 12708
202: realloc 0xb48fd000 0xb48fd000 12780
202: realloc 0xb48fd000 0xb48fd000 12852
202: realloc 0xb48fd000 0xb48fd000 12924
202: realloc 0xb48fd000 0xb48fd000 12996
202: realloc 0xb48fd000 0xb48fd000 13068
202: realloc 0xb48fd000 0xb48fd000 13140
202: realloc 0xb48fd000 0xb48fd000 13212
202: realloc 0xb48fd000 0xb48fd000 13284
202: realloc 0xb48fd000 0xb48fd000 13356
202: realloc 0xb48fd000 0xb48fd000 13428
202: realloc 0xb48fd000 0xb48fd000 13500
202: realloc 0xb48fd000 0xb48fd000 13572
202: realloc 0xb48fd000 0xb48fd000 13644
202: realloc 0xb48fd000 0xb48fd000 13716
202: realloc 0xb48fd000 0xb48fd000 13788
...
```
  In this situation we were deallocating the old chunk, and
  allocating a new one for every single one of those, but now we can
  keep the same chunk (we just updated the header), which saves some
  heap operations.

Reviewers: hctim, morehouse, vitalybuka, eugenis, cferris, rengolin

Reviewed By: morehouse

Subscribers: srhines, delcypher, #sanitizers, llvm-commits

Tags: #llvm, #sanitizers

Differential Revision: https://reviews.llvm.org/D67293

llvm-svn: 371628

5 years ago[OPENMP]Updated status page, NFC.
Alexey Bataev [Wed, 11 Sep 2019 14:44:30 +0000 (14:44 +0000)]
[OPENMP]Updated status page, NFC.

llvm-svn: 371627

5 years agogn build: Merge r371562
Nico Weber [Wed, 11 Sep 2019 14:40:16 +0000 (14:40 +0000)]
gn build: Merge r371562

llvm-svn: 371626

5 years agoRevert "[LLDB][ELF] Load both, .symtab and .dynsym sections"
Konrad Kleine [Wed, 11 Sep 2019 14:33:37 +0000 (14:33 +0000)]
Revert "[LLDB][ELF] Load both, .symtab and .dynsym sections"

This reverts commit 3a4781bbf4f39a25562b4c61c9a9ab2483a96b41.

llvm-svn: 371625

5 years agoRevert "[LLDB][ELF] Fixup for comments in D67390"
Konrad Kleine [Wed, 11 Sep 2019 14:33:21 +0000 (14:33 +0000)]
Revert "[LLDB][ELF] Fixup for comments in D67390"

This reverts commit 813f05915d29904878d926f9849ca3dbe78096af.

llvm-svn: 371624

5 years ago[lldb][NFC] Make include directories in Clang expression parser a std::string
Raphael Isemann [Wed, 11 Sep 2019 14:33:11 +0000 (14:33 +0000)]
[lldb][NFC] Make include directories in Clang expression parser a std::string

We never compare these directories (where ConstString would be good) and
essentially just convert this back to a normal string in the end. So we might
as well just use std::string. Also makes it easier to unittest this code
(which was the main motivation for this change).

llvm-svn: 371623

5 years agoLLVM: Optimization Pass: Remove conflicting attribute, if any, before
Whitney Tsang [Wed, 11 Sep 2019 14:26:22 +0000 (14:26 +0000)]
LLVM: Optimization Pass: Remove conflicting attribute, if any, before
adding new read attribute to an argument
Summary: Update optimization pass to prevent adding read-attribute to an
argument without removing its conflicting attribute.

A read attribute, based on the result of the attribute deduction
process, might be added to an argument. The attribute might be in
conflict with other read/write attribute currently associated with the
argument. To ensure the compatibility of attributes, conflicting
attribute, if any, must be removed before a new one is added.

The following snippet shows the current behavior of the compiler, where
the compilation process is aborted due to incompatible attributes.

$ cat x.ll
; ModuleID = 'x.bc'

%_type_of_d-ccc = type <{ i8*, i8, i8, i8, i8 }>

@d-ccc = internal global %_type_of_d-ccc <{ i8* null, i8 1, i8 13, i8 0,
i8 -127 }>, align 8

define void @foo(i32* writeonly %.aaa) {
foo_entry:
  %_param_.aaa = alloca i32*, align 8
  store i32* %.aaa, i32** %_param_.aaa, align 8
  store i8 0, i8* getelementptr inbounds (%_type_of_d-ccc,
%_type_of_d-ccc* @d-ccc, i32 0, i32 3)
  ret void
}

$ opt -O3 x.ll
Attributes 'readnone and writeonly' are incompatible!
void (i32*)* @foo
in function foo
LLVM ERROR: Broken function found, compilation aborted!
The purpose of this changeset is to fix the above error. This fix is
based on a suggestion from Johannes @jdoerfert (many thanks!!!)
Authored By: anhtuyen
Reviewer: nicholas, rnk, chandlerc, jdoerfert
Reviewed By: rnk
Subscribers: hiraditya, jdoerfert, llvm-commits, anhtuyen, LLVM
Tag: LLVM
Differential Revision: https://reviews.llvm.org/D58694

llvm-svn: 371622

5 years ago[ConstProp] add tests for fma that produce NaN; NFC
Sanjay Patel [Wed, 11 Sep 2019 14:18:48 +0000 (14:18 +0000)]
[ConstProp] add tests for fma that produce NaN; NFC

llvm-svn: 371621

5 years ago[libFuzzer] Make -merge=1 to reuse coverage information from the control file.
Max Moroz [Wed, 11 Sep 2019 14:11:08 +0000 (14:11 +0000)]
[libFuzzer] Make -merge=1 to reuse coverage information from the control file.

Summary:
This change allows to perform corpus merging in two steps. This is useful when
the user wants to address the following two points simultaneously:

1) Get trustworthy incremental stats for the coverage and corpus size changes
    when adding new corpus units.
2) Make sure the shorter units will be preferred when two or more units give the
    same unique signal (equivalent to the `REDUCE` logic).

This solution was brainstormed together with @kcc, hopefully it looks good to
the other people too. The proposed use case scenario:

1) We have a `fuzz_target` binary and `existing_corpus` directory.
2) We do fuzzing and write new units into the `new_corpus` directory.
3) We want to merge the new corpus into the existing corpus and satisfy the
    points mentioned above.
4) We create an empty directory `merged_corpus` and run the first merge step:

    `
    ./fuzz_target -merge=1 -merge_control_file=MCF ./merged_corpus ./existing_corpus
    `

    this provides the initial stats for `existing_corpus`, e.g. from the output:

    `
    MERGE-OUTER: 3 new files with 11 new features added; 11 new coverage edges
    `

5) We recreate `merged_corpus` directory and run the second merge step:

    `
    ./fuzz_target -merge=1 -merge_control_file=MCF ./merged_corpus ./existing_corpus ./new_corpus
    `

    this provides the final stats for the merged corpus, e.g. from the output:

    `
    MERGE-OUTER: 6 new files with 14 new features added; 14 new coverage edges
    `

Alternative solutions to this approach are:

A) Store precise coverage information for every unit (not only unique signal).
B) Execute the same two steps without reusing the control file.

Either of these would be suboptimal as it would impose an extra disk or CPU load
respectively, which is bad given the quadratic complexity in the worst case.

Tested on Linux, Mac, Windows.

Reviewers: morehouse, metzman, hctim, kcc

Reviewed By: morehouse

Subscribers: JDevlieghere, delcypher, mgrang, #sanitizers, llvm-commits, kcc

Tags: #llvm, #sanitizers

Differential Revision: https://reviews.llvm.org/D66107

llvm-svn: 371620

5 years ago[ConstProp] move test file from InstSimplify; NFC
Sanjay Patel [Wed, 11 Sep 2019 14:01:11 +0000 (14:01 +0000)]
[ConstProp] move test file from InstSimplify; NFC

These are constant folding tests; there is no code
directly in InstSimplify for this.

llvm-svn: 371619

5 years ago[lldb][NFC] Remove dead code in SBAddress::GetDescription
Raphael Isemann [Wed, 11 Sep 2019 13:57:41 +0000 (13:57 +0000)]
[lldb][NFC] Remove dead code in SBAddress::GetDescription

llvm-svn: 371618

5 years ago[InstSimplify] regenerate test CHECKs; NFC
Sanjay Patel [Wed, 11 Sep 2019 13:56:07 +0000 (13:56 +0000)]
[InstSimplify] regenerate test CHECKs; NFC

llvm-svn: 371617

5 years ago[Alignment][NFC] use llvm::Align for AsmPrinter::EmitAlignment
Guillaume Chatelet [Wed, 11 Sep 2019 13:37:35 +0000 (13:37 +0000)]
[Alignment][NFC] use llvm::Align for AsmPrinter::EmitAlignment

Summary:
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790

Reviewers: courbet

Subscribers: dschuff, sdardis, nemanjai, hiraditya, kbarton, jrtc27, MaskRay, atanasyan, jsji, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67443

llvm-svn: 371616

5 years ago[LangRef] add link for fma intrinsic
Sanjay Patel [Wed, 11 Sep 2019 13:25:32 +0000 (13:25 +0000)]
[LangRef] add link for fma intrinsic

llvm-svn: 371615

5 years ago[lldb][NFC] Sort files in unittests/Expression/CMakeLists.txt
Raphael Isemann [Wed, 11 Sep 2019 12:57:49 +0000 (12:57 +0000)]
[lldb][NFC] Sort files in unittests/Expression/CMakeLists.txt

llvm-svn: 371614

5 years ago[LangRef] fix punctuation; NFC
Sanjay Patel [Wed, 11 Sep 2019 12:22:24 +0000 (12:22 +0000)]
[LangRef] fix punctuation; NFC

llvm-svn: 371612

5 years agogn build: add include_dir that's necessary after r371564
Nico Weber [Wed, 11 Sep 2019 12:21:09 +0000 (12:21 +0000)]
gn build: add include_dir that's necessary after r371564

llvm-svn: 371611

5 years ago[InstCombine] fold sign-bit compares of srem
Sanjay Patel [Wed, 11 Sep 2019 12:04:26 +0000 (12:04 +0000)]
[InstCombine] fold sign-bit compares of srem

(srem X, pow2C) sgt/slt 0 can be reduced using bit hacks by masking
off the sign bit and the module (low) bits:
https://rise4fun.com/Alive/jSO
A '2' divisor allows slightly more folding:
https://rise4fun.com/Alive/tDBM

Any chance to remove an 'srem' use is probably worthwhile, but this is limited
to the one-use improvement case because doing more may expose other missing
folds. That means it does nothing for PR21929 yet:
https://bugs.llvm.org/show_bug.cgi?id=21929

Differential Revision: https://reviews.llvm.org/D67334

llvm-svn: 371610

5 years ago[Alignment] Use llvm::Align in MachineFunction and TargetLowering - fixes mir parsing
Guillaume Chatelet [Wed, 11 Sep 2019 11:16:48 +0000 (11:16 +0000)]
[Alignment] Use llvm::Align in MachineFunction and TargetLowering - fixes mir parsing

Summary:
This catches malformed mir files which specify alignment as log2 instead of pow2.
See https://reviews.llvm.org/D65945 for reference,

This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790

Reviewers: courbet

Subscribers: MatzeB, qcolombet, dschuff, arsenm, sdardis, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, s.egerton, pzheng, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67433

llvm-svn: 371608

5 years ago[mips][msa] Fix infinite loop for mips.nori.b intrinsic
Simon Atanasyan [Wed, 11 Sep 2019 11:16:06 +0000 (11:16 +0000)]
[mips][msa] Fix infinite loop for mips.nori.b intrinsic

When value of immediate in `mips.nori.b` is 255 (which has all ones in
binary form as 8bit integer) DAGCombiner and Legalizer would fall in an
infinite loop. DAGCombiner would try to simplify `or %value, -1` by
turning `%value` into UNDEF. Legalizer will turn it back into `Constant<0>`
which would then be again turned into UNDEF by DAGCombiner. To avoid this
loop we make UNDEF legal for MSA int types on Mips.

Patch by Mirko Brkusanin.

Differential Revision: https://reviews.llvm.org/D67280

llvm-svn: 371607

5 years agoFix -Wdocumentation warning - void function doesn't need a @returns. NFCI.
Simon Pilgrim [Wed, 11 Sep 2019 11:12:00 +0000 (11:12 +0000)]
Fix -Wdocumentation warning - void function doesn't need a @returns. NFCI.

llvm-svn: 371606

5 years ago[Diagnostics] Add -Wsizeof-array-div
David Bolvansky [Wed, 11 Sep 2019 10:59:47 +0000 (10:59 +0000)]
[Diagnostics] Add -Wsizeof-array-div

Summary: Clang version of https://www.viva64.com/en/examples/v706/

Reviewers: rsmith

Differential Revision: https://reviews.llvm.org/D67287

llvm-svn: 371605

5 years ago[NFC] Updated objsize-64.ll test
David Bolvansky [Wed, 11 Sep 2019 10:51:26 +0000 (10:51 +0000)]
[NFC] Updated objsize-64.ll test

llvm-svn: 371604

5 years ago[NFC] Fixed test
David Bolvansky [Wed, 11 Sep 2019 10:42:30 +0000 (10:42 +0000)]
[NFC] Fixed test

llvm-svn: 371603

5 years ago[InstCombine] Fixed handling of isOpNewLike (PR11748)
David Bolvansky [Wed, 11 Sep 2019 10:37:03 +0000 (10:37 +0000)]
[InstCombine] Fixed handling of isOpNewLike (PR11748)

llvm-svn: 371602

5 years ago[LoopInterchange] Drop unused splitInnerLoopHeader declaration.
Florian Hahn [Wed, 11 Sep 2019 10:32:15 +0000 (10:32 +0000)]
[LoopInterchange] Drop unused splitInnerLoopHeader declaration.

llvm-svn: 371601

5 years ago[LLDB][ELF] Fixup for comments in D67390
Konrad Kleine [Wed, 11 Sep 2019 10:12:36 +0000 (10:12 +0000)]
[LLDB][ELF] Fixup for comments in D67390

llvm-svn: 371600

5 years ago[LLDB][ELF] Load both, .symtab and .dynsym sections
Konrad Kleine [Wed, 11 Sep 2019 10:00:30 +0000 (10:00 +0000)]
[LLDB][ELF] Load both, .symtab and .dynsym sections

Summary:
This change ensures that the .dynsym section will be parsed even when there's already is a .symtab.

It is motivated because of minidebuginfo (https://sourceware.org/gdb/current/onlinedocs/gdb/MiniDebugInfo.html#MiniDebugInfo).

There it says:

    Keep all the function symbols not already in the dynamic symbol table.

That means the .symtab embedded inside the .gnu_debugdata does NOT contain the symbols from .dynsym. But in order to put a breakpoint on all symbols we need to load both. I hope this makes sense.

My other patch D66791 implements support for minidebuginfo, that's why I need this change.

Reviewers: labath, espindola, alexshap

Subscribers: JDevlieghere, emaste, arichardson, MaskRay, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D67390

llvm-svn: 371599

5 years agoRevert "clang-misexpect: Profile Guided Validation of Performance Annotations in...
Dmitri Gribenko [Wed, 11 Sep 2019 09:16:17 +0000 (09:16 +0000)]
Revert "clang-misexpect: Profile Guided Validation of Performance Annotations in LLVM"

This reverts commit r371584. It introduced a dependency from compiler-rt
to llvm/include/ADT, which is problematic for multiple reasons.

One is that it is a novel dependency edge, which needs cross-compliation
machinery for llvm/include/ADT (yes, it is true that right now
compiler-rt included only header-only libraries, however, if we allow
compiler-rt to depend on anything from ADT, other libraries will
eventually get used).

Secondly, depending on ADT from compiler-rt exposes ADT symbols from
compiler-rt, which would cause ODR violations when Clang is built with
the profile library.

llvm-svn: 371598

5 years ago[ARM] Take into account -mcpu and -mfpu options while handling 'crypto' feature
Diogo N. Sampaio [Wed, 11 Sep 2019 09:06:17 +0000 (09:06 +0000)]
[ARM] Take into account -mcpu and -mfpu options while handling 'crypto' feature

Submittin in behalf of krisb (Kristina Bessonova) <ch.bessonova@gmail.com>

Summary:
'+crypto' means '+aes' and '+sha2' for arch >= ARMv8 when they were
not disabled explicitly. But this is correctly handled only in case of
'-march' option, though the feature may also be specified through
the '-mcpu' or '-mfpu' options. In the following example:

  $ clang -mcpu=cortex-a57 -mfpu=crypto-neon-fp-armv8

'aes' and 'sha2' are disabled that is quite unexpected:

  $ clang -cc1 -triple armv8--- -target-cpu cortex-a57
    <...> -target-feature -sha2 -target-feature -aes -target-feature +crypto

This exposed by https://reviews.llvm.org/D63936 that makes
the 'aes' and 'sha2' features disabled by default.

So, while handling the 'crypto' feature we need to take into account:
  - a CPU name, as it provides the information about architecture
    (if no '-march' option specified),
  - features, specified by the '-mcpu' and '-mfpu' options.

Reviewers: SjoerdMeijer, ostannard, labrinea, dnsampaio

Reviewed By: dnsampaio

Subscribers: ikudrin, javed.absar, kristof.beyls, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D66018

Author: krisb
llvm-svn: 371597

5 years ago[LLDB] Do not try to canonicalize gethostname() result
David Zarzycki [Wed, 11 Sep 2019 08:32:37 +0000 (08:32 +0000)]
[LLDB] Do not try to canonicalize gethostname() result

This code is trying too hard and failing. Either the result of
gethostname() is canonical or it is not. If it is not, then trying to
canonicalize it is – for various reasons – a lost cause. For example, a
given machine might have multiple network interfaces with multiple
addresses per interface, each with a different canonical name.
Separably, the result of HostInfoPosix::GetHostname() and latency
thereof shouldn't depend on whether networking is up or down or what
network the machine happened to be attached to at any given moment (like
a laptop that travels between work and home).

https://reviews.llvm.org/D67230

llvm-svn: 371596

5 years ago[LoopInterchange] Properly move condition, induction increment and ops to latch.
Florian Hahn [Wed, 11 Sep 2019 08:23:23 +0000 (08:23 +0000)]
[LoopInterchange] Properly move condition, induction increment and ops to latch.

Currently we only rely on the induction increment to come before the
condition to ensure the required instructions get moved to the new
latch.

This patch duplicates and moves the required instructions to the
newly created latch. We move the condition to the end of the new block,
then process its operands. We stop at operands that are defined
outside the loop, or are the induction PHI.

We duplicate the instructions and update the uses in the moved
instructions, to ensure other users remain intact. See the added
test2 for such an example.

Reviewers: efriedma, mcrosier

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D67367

llvm-svn: 371595

5 years ago[NFC][ARM] Add and modify tests
Sam Parker [Wed, 11 Sep 2019 08:17:48 +0000 (08:17 +0000)]
[NFC][ARM] Add and modify tests

Add test for ParallelDSP.

llvm-svn: 371594

5 years ago[LTO] Avoid calling GlobalValue::getGUID (MD5) twice
Fangrui Song [Wed, 11 Sep 2019 07:38:21 +0000 (07:38 +0000)]
[LTO] Avoid calling GlobalValue::getGUID (MD5) twice

llvm-svn: 371593

5 years ago[TLI][AMDGPU] AMDPAL does not have library functions
Tim Renouf [Wed, 11 Sep 2019 07:26:39 +0000 (07:26 +0000)]
[TLI][AMDGPU] AMDPAL does not have library functions

Configure TLI to say that r600/amdgpu does not have any library
functions, such that InstCombine does not do anything like turn sin/cos
into the library function @tan with sufficient fast math flags.

Differential Revision: https://reviews.llvm.org/D67406

Change-Id: I02f907d3e64832117ea9800e9f9285282856e5df
llvm-svn: 371592

5 years ago[llvm-objcopy] Simplify --prefix-alloc-sections
Fangrui Song [Wed, 11 Sep 2019 07:23:35 +0000 (07:23 +0000)]
[llvm-objcopy] Simplify --prefix-alloc-sections

Handle --prefix-alloc-sections after --rename-sections so that --prefix-alloc-sections code
does not have to check if renaming has been performed.

Reviewed By: jhenderson

Differential Revision: https://reviews.llvm.org/D66931

llvm-svn: 371591

5 years ago[Attributor] Implement "noalias" callsite argument deduction
Hideto Ueno [Wed, 11 Sep 2019 07:00:33 +0000 (07:00 +0000)]
[Attributor] Implement "noalias" callsite argument deduction

Summary: Now, `nocapture` is deduced in Attributor therefore, this patch introduces deduction for `noalias` callsite argument using `nocapture`.

Reviewers: jdoerfert, sstefan1

Reviewed By: jdoerfert

Subscribers: lebedev.ri, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67286

llvm-svn: 371590

5 years ago[Attributor][Fix] Manifest nocapture only in CSArgument or Argument
Hideto Ueno [Wed, 11 Sep 2019 06:52:11 +0000 (06:52 +0000)]
[Attributor][Fix] Manifest nocapture only in CSArgument or Argument

Summary:
We can query to Attributor whether the value is captured in the scope or not on the following way:

```
    const auto & NoCapAA = A.getAAFor<AANoCapture>(*this, IRPosition::value(V));
```
And if V is CallSiteReturned then `getDeducedAttribute` will add `nocatpure` to the callsite returned value. It is not valid.
This patch checks the position is an argument or call site argument.

This is tested in D67286.

Reviewers: jdoerfert, sstefan1

Reviewed By: jdoerfert

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67342

llvm-svn: 371589

5 years ago[ELF][test] Make tests more tolerant to exact symbol addresses
Fangrui Song [Wed, 11 Sep 2019 06:20:14 +0000 (06:20 +0000)]
[ELF][test] Make tests more tolerant to exact symbol addresses

llvm-svn: 371588

5 years ago[Debuginfo][Instcombiner] Do not clone dbg.declare.
Alexey Lapshin [Wed, 11 Sep 2019 06:07:16 +0000 (06:07 +0000)]
[Debuginfo][Instcombiner] Do not clone dbg.declare.

TryToSinkInstruction() has a bug: While updating debug info for
sunk instruction, it could clone dbg.declare intrinsic.
That is wrong. There could be only one dbg.declare.
The fix is to not clone dbg.declare intrinsic and to update
it`s arguments, to not to point to sunk instruction.

Differential Revision: https://reviews.llvm.org/D67217

llvm-svn: 371587

5 years ago[CodeGen] Add alias for cpu_dispatch function with IFunc & Fix resolver linkage type
Fangrui Song [Wed, 11 Sep 2019 01:54:48 +0000 (01:54 +0000)]
[CodeGen] Add alias for cpu_dispatch function with IFunc & Fix resolver linkage type

Multi-versioned functions defined by cpu_dispatch and implemented with IFunc
can not be called outside the translation units where they are defined due to
lack of symbols. This patch add function aliases for these functions and thus
make them visible outside.

Differential Revision: https://reviews.llvm.org/D67058
Patch by Senran Zhang

llvm-svn: 371586

5 years ago[llvm-nm] Fix -DBUILD_SHARED_LIBS=ON builds after D66160/r371576
Fangrui Song [Wed, 11 Sep 2019 01:49:00 +0000 (01:49 +0000)]
[llvm-nm] Fix -DBUILD_SHARED_LIBS=ON builds after D66160/r371576

llvm-svn: 371585

5 years agoclang-misexpect: Profile Guided Validation of Performance Annotations in LLVM
Petr Hosek [Wed, 11 Sep 2019 01:09:16 +0000 (01:09 +0000)]
clang-misexpect: Profile Guided Validation of Performance Annotations in LLVM

This patch contains the basic functionality for reporting potentially
incorrect usage of __builtin_expect() by comparing the developer's
annotation against a collected PGO profile. A more detailed proposal and
discussion appears on the CFE-dev mailing list
(http://lists.llvm.org/pipermail/cfe-dev/2019-July/062971.html) and a
prototype of the initial frontend changes appear here in D65300

We revised the work in D65300 by moving the misexpect check into the
LLVM backend, and adding support for IR and sampling based profiles, in
addition to frontend instrumentation.

We add new misexpect metadata tags to those instructions directly
influenced by the llvm.expect intrinsic (branch, switch, and select)
when lowering the intrinsics. The misexpect metadata contains
information about the expected target of the intrinsic so that we can
check against the correct PGO counter when emitting diagnostics, and the
compiler's values for the LikelyBranchWeight and UnlikelyBranchWeight.
We use these branch weight values to determine when to emit the
diagnostic to the user.

A future patch should address the comment at the top of
LowerExpectIntrisic.cpp to hoist the LikelyBranchWeight and
UnlikelyBranchWeight values into a shared space that can be accessed
outside of the LowerExpectIntrinsic pass. Once that is done, the
misexpect metadata can be updated to be smaller.

In the long term, it is possible to reconstruct portions of the
misexpect metadata from the existing profile data. However, we have
avoided this to keep the code simple, and because some kind of metadata
tag will be required to identify which branch/switch/select instructions
are influenced by the use of llvm.expect

Patch By: paulkirth
Differential Revision: https://reviews.llvm.org/D66324

llvm-svn: 371584

5 years agoAh, only skip this for embedded darwin targets.
Jason Molenda [Wed, 11 Sep 2019 01:03:24 +0000 (01:03 +0000)]
Ah, only skip this for embedded darwin targets.

llvm-svn: 371583

5 years agoSkip a test in TestProcessIO.py when running against ios devices.
Jason Molenda [Wed, 11 Sep 2019 01:02:30 +0000 (01:02 +0000)]
Skip a test in TestProcessIO.py when running against ios devices.

llvm-svn: 371582

5 years agoEmit -Wmicrosoft-enum-value warning instead of error in MS ABI
Reid Kleckner [Wed, 11 Sep 2019 01:01:06 +0000 (01:01 +0000)]
Emit -Wmicrosoft-enum-value warning instead of error in MS ABI

Summary:
The first NFC change is to replace a getCXXABI().isMicrosoft() check
with getTriple().isWindowsMSVCEnvironment(). This code takes effect in
non-C++ compilations, so it doesn't make sense to check the C++ ABI. In
the MS ABI, enums are always considered to be "complete" because the
underlying type of an unfixed enum will always be 'int'. This behavior
was moved from -fms-compatibility to MS ABI back in r249656.

The second change is functional, and it downgrades an error to a warning
when the MS ABI is used rather than only under -fms-compatibility. The
reasoning is that it's unreasonable for the following code to reject the
following code for all MS ABI targets with -fno-ms-compatibility:
  enum Foo { Foo_Val = 0xDEADBEEF };
This is valid code for any other target, but in the MS ABI, Foo_Val just
happens to be negative. With this change, clang emits a
-Wmicrosoft-enum-value warning on this code, but compiles it without
error.

Fixes PR38478

Reviewers: hans, rsmith, STL_MSFT

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D67304

llvm-svn: 371581

5 years agoRevert "llvm-reduce: Add pass to reduce parameters"
David Blaikie [Wed, 11 Sep 2019 00:44:17 +0000 (00:44 +0000)]
Revert "llvm-reduce: Add pass to reduce parameters"

Looks to be failing on asan buildbots

This reverts commit r371567.

llvm-svn: 371580

5 years agoRevert "Adding support for overriding LLVM_ENABLE_RUNTIMES for runtimes builds."
Petr Hosek [Wed, 11 Sep 2019 00:43:35 +0000 (00:43 +0000)]
Revert "Adding support for overriding LLVM_ENABLE_RUNTIMES for runtimes builds."

This reverts commit r371566: this doesn't correctly handle the case
where LLVM_ENABLE_RUNTIMES contains a list of values.

llvm-svn: 371579

5 years ago[clang-scan-deps][NFC] Fix tests - prevent FileCheck matching test dir path
Jan Korous [Wed, 11 Sep 2019 00:30:26 +0000 (00:30 +0000)]
[clang-scan-deps][NFC] Fix tests - prevent FileCheck matching test dir path

Differential Revision: https://reviews.llvm.org/D67379

llvm-svn: 371578

5 years ago[Object][TextAPI] NFC, fix tapi lit tests
Cyndy Ishida [Wed, 11 Sep 2019 00:13:01 +0000 (00:13 +0000)]
[Object][TextAPI] NFC, fix tapi lit tests

FileCheck command had wrong casing, which works fine on macOS, repairs the linux build bots

llvm-svn: 371577

5 years ago[llvm-nm] Add tapi file support
Cyndy Ishida [Wed, 11 Sep 2019 00:00:53 +0000 (00:00 +0000)]
[llvm-nm] Add tapi file support

Summary:
This commit is the final one for adding tapi support to the llvm-nm implementation.
This commit also has accompanying tests the additions to lib/Object

Reviewers: ributzka, steven_wu

Reviewed By: ributzka

Subscribers: hiraditya, plotfi, dexonsmith, rupprecht, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66160

llvm-svn: 371576

5 years agoRemove xfail i386 NetBSD mark in vptr-non-unique-typeinfo.cpp
Kamil Rytarowski [Tue, 10 Sep 2019 23:55:03 +0000 (23:55 +0000)]
Remove xfail i386 NetBSD mark in vptr-non-unique-typeinfo.cpp

This test passes now.

llvm-svn: 371575

5 years agoRemove xfail NetBSD mark from ignored-interceptors-mmap.cpp
Kamil Rytarowski [Tue, 10 Sep 2019 23:42:16 +0000 (23:42 +0000)]
Remove xfail NetBSD mark from ignored-interceptors-mmap.cpp

This test now passes.

llvm-svn: 371574

5 years ago[MemorySSA] MemorySSA should not model debuginfo, and need not update it.
Alina Sbirlea [Tue, 10 Sep 2019 23:36:43 +0000 (23:36 +0000)]
[MemorySSA] MemorySSA should not model debuginfo, and need not update it.

Reverts the change in r371084, but keeps the test.
After r371565, debuginfo cannot be modelled in MemorySSA, even with a
non-standard AA pipeline.

llvm-svn: 371573

5 years ago[GlobalISel] When a tail call is emitted in a block, stop translating it
Jessica Paquette [Tue, 10 Sep 2019 23:34:45 +0000 (23:34 +0000)]
[GlobalISel] When a tail call is emitted in a block, stop translating it

This fixes a crash in tail call translation caused by assume and lifetime_end
intrinsics.

It's possible to have instructions other than a return after a tail call which
will still have `Analysis::isInTailCallPosition` return true. (Namely,
lifetime_end and assume intrinsics.)

If we emit a tail call, we should stop translating instructions in the block.
Otherwise, we can end up emitting an extra return, or dead instructions in
general. This makes the verifier unhappy, and is generally unfortunate for
codegen.

This also removes the code from AArch64CallLowering that checks if we have a
tail call when lowering a return. This is covered by the new code now.

Also update call-translator-tail-call.ll to show that we now properly tail call
in the presence of lifetime_end and assume.

Differential Revision: https://reviews.llvm.org/D67415

llvm-svn: 371572

5 years ago[clang-format] Apply BAS_AlwaysBreak to C++11 braced lists
Owen Pan [Tue, 10 Sep 2019 23:26:45 +0000 (23:26 +0000)]
[clang-format] Apply BAS_AlwaysBreak to C++11 braced lists

See PR18455.

Differential Revision: https://reviews.llvm.org/D67395

llvm-svn: 371571

5 years ago[AArch64][GlobalISel] Support sibling calls with mismatched calling conventions
Jessica Paquette [Tue, 10 Sep 2019 23:25:12 +0000 (23:25 +0000)]
[AArch64][GlobalISel] Support sibling calls with mismatched calling conventions

Add support for sibcalling calls whose calling convention differs from the
caller's.

- Port over `CCState::resultsCombatible` from CallingConvLower.cpp into
  CallLowering. This is used to verify that the way the caller and callee CC
  handle incoming arguments matches up.

- Add `CallLowering::analyzeCallResult`. This is basically a port of
  `CCState::AnalyzeCallResult`, but using `ArgInfo` rather than `ISD::InputArg`.

- Add `AArch64CallLowering::doCallerAndCalleePassArgsTheSameWay`. This checks
  that the calling conventions are compatible, and that the caller and callee
  preserve the same registers.

For testing:

- Update call-translator-tail-call.ll to show that we can now handle this.

- Add a GISel line to tailcall-ccmismatch.ll to show that we will not tail call
  when the regmasks don't line up.

Differential Revision: https://reviews.llvm.org/D67361

llvm-svn: 371570

5 years agoUpdate ReleaseNotes: add enabling of MemorySSA.
Alina Sbirlea [Tue, 10 Sep 2019 23:22:37 +0000 (23:22 +0000)]
Update ReleaseNotes: add enabling of MemorySSA.

llvm-svn: 371569

5 years agoReland "Change the X86 datalayout to add three address spaces
Amy Huang [Tue, 10 Sep 2019 23:15:38 +0000 (23:15 +0000)]
Reland "Change the X86 datalayout to add three address spaces
 for 32 bit signed, 32 bit unsigned, and 64 bit pointers."
This reverts 57076d3199fc2b0af4a3736b7749dd5462cacda5.

Original review at https://reviews.llvm.org/D64931.
Review for added fix at https://reviews.llvm.org/D66843.

llvm-svn: 371568

5 years agollvm-reduce: Add pass to reduce parameters
David Blaikie [Tue, 10 Sep 2019 23:10:10 +0000 (23:10 +0000)]
llvm-reduce: Add pass to reduce parameters

Patch by Diego Treviño!

Differential Revision: https://reviews.llvm.org/D65479

llvm-svn: 371567

5 years agoAdding support for overriding LLVM_ENABLE_RUNTIMES for runtimes builds.
Puyan Lotfi [Tue, 10 Sep 2019 22:55:47 +0000 (22:55 +0000)]
Adding support for overriding LLVM_ENABLE_RUNTIMES for runtimes builds.

On some platforms, certain runtimes are not supported. For runtimes builds of
those platforms it would be nice if we could disable certain runtimes (ie
libunwind on Windows).

Differential Revision: https://reviews.llvm.org/D67195

llvm-svn: 371566

5 years ago[MemorySSA] Do not create memoryaccesses for debug info intrinsics.
Alina Sbirlea [Tue, 10 Sep 2019 22:35:27 +0000 (22:35 +0000)]
[MemorySSA] Do not create memoryaccesses for debug info intrinsics.

Summary:
Do not model debuginfo intrinsics in MemorySSA.
Regularly these are non-memory modifying instructions. With -disable-basicaa, they were being modelled as Defs.

Reviewers: george.burgess.iv

Subscribers: aprantl, Prazek, sanjoy.google, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67307

llvm-svn: 371565

5 years agollvm-reduce: Remove some unused headers/more narrowly include them
David Blaikie [Tue, 10 Sep 2019 22:31:35 +0000 (22:31 +0000)]
llvm-reduce: Remove some unused headers/more narrowly include them

llvm-svn: 371564

5 years agollvm-remove: Remove "using namespace" in header.
David Blaikie [Tue, 10 Sep 2019 22:10:00 +0000 (22:10 +0000)]
llvm-remove: Remove "using namespace" in header.

llvm-svn: 371563

5 years agollvm-reduce: Add pass to reduce Metadata
David Blaikie [Tue, 10 Sep 2019 22:09:58 +0000 (22:09 +0000)]
llvm-reduce: Add pass to reduce Metadata

Patch by Diego Treviño!

Differential Revision: https://reviews.llvm.org/D65026

llvm-svn: 371562

5 years agoMove LLVM_ENABLE_ABI_BREAKING_CHECKS variables to their own file
Eric Christopher [Tue, 10 Sep 2019 22:05:01 +0000 (22:05 +0000)]
Move LLVM_ENABLE_ABI_BREAKING_CHECKS variables to their own file
so that you don't have to link Error.o and all of its dependencies.

In more detail: global initializers in Error.o can't be elided with
-ffunction-sections/-gc-sections since they always need to be run
causing a fairly significant binary bloat if all you want is the
ABI breaking checks code.

Differential Revision: https://reviews.llvm.org/D67387

llvm-svn: 371561

5 years agoFix a thinko in handling the QSetLogging packet.
Jim Ingham [Tue, 10 Sep 2019 21:58:22 +0000 (21:58 +0000)]
Fix a thinko in handling the QSetLogging packet.

The comparison against LOG_MEMORY shortcut all the LOG_MEMORY_*
log channels.  It has to come last.

llvm-svn: 371560

5 years agoActually reorder not and env in crash-recovery-modules.m
Reid Kleckner [Tue, 10 Sep 2019 21:54:16 +0000 (21:54 +0000)]
Actually reorder not and env in crash-recovery-modules.m

llvm-svn: 371559

5 years ago[Loads] Move generic code out of vectorizer into a location it might be reused [NFC]
Philip Reames [Tue, 10 Sep 2019 21:33:53 +0000 (21:33 +0000)]
[Loads] Move generic code out of vectorizer into a location it might be reused [NFC]

llvm-svn: 371558

5 years agoWhen evaluating a __builtin_constant_p conditional, always enter
Richard Smith [Tue, 10 Sep 2019 21:24:09 +0000 (21:24 +0000)]
When evaluating a __builtin_constant_p conditional, always enter
constant-folding mode regardless of the original evaluation mode.

In order for this to be correct, we need to track whether we're checking
for a potential constant expression or checking for undefined behavior
separately from the evaluation mode enum, since we don't want to clobber
those states when entering constant-folding mode.

llvm-svn: 371557

5 years ago[ValueTracking] Factor our common speculation suppression logic [NFC]
Philip Reames [Tue, 10 Sep 2019 21:12:29 +0000 (21:12 +0000)]
[ValueTracking] Factor our common speculation suppression logic [NFC]

Expose a utility function so that all places which want to suppress speculation (when otherwise legal) due to ordering and/or sanitizer interaction can do so.

llvm-svn: 371556

5 years agollvm-reduce: Simplify testing using -implicit-check-not
David Blaikie [Tue, 10 Sep 2019 20:52:14 +0000 (20:52 +0000)]
llvm-reduce: Simplify testing using -implicit-check-not

Also fix llvm-reduce to use the specified output file name directly,
without appending '.ll' to the name.

llvm-svn: 371555

5 years ago[mips] Allow PT_LOAD to have overlapping p_offset ranges on EM_MIPS
Simon Atanasyan [Tue, 10 Sep 2019 20:19:59 +0000 (20:19 +0000)]
[mips] Allow PT_LOAD to have overlapping p_offset ranges on EM_MIPS

Port the D64906 <https://reviews.llvm.org/D64906> technique to MIPS.

Fix PR33131

llvm-svn: 371554