platform/upstream/llvm.git
4 years ago[Driver][Bundler] Improve bundling of object files.
Alexey Bataev [Thu, 15 Aug 2019 17:15:35 +0000 (17:15 +0000)]
[Driver][Bundler] Improve bundling of object files.

Summary:
Previously, object files were bundled using partial linking. It resulted
in the following structure of the bundled objects:
```
<host_code>
clang-offload-bundle
__CLANG_OFFLOAD_BUNDLE__<target>
<target_code>
```
But when we tried to unbundle object files, it worked correctly only for
the target objects. The host object remains bundled. It produced a lot of
junk sections in the host object files and in some cases may caused
incorrect linking.

Patch improves bundling of the object files. After this patch the
bundled object looks like this:

```
<host_code>
clang-offload-bundle
__CLANG_OFFLOAD_BUNDLE__<target>
<target_code>
__CLANG_OFFLOAD_BUNDLE__<host>
<host_code>
```

With this structure we are able to unbundle the host object files too so
that after unbundling they are the same as were before.
The host section is bundled twice. The bundled section is used to
unbundle the original host section.

Reviewers: yaxunl, tra, jlebar, hfinkel, jdoerfert

Subscribers: caomhin, kkwli0, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D65819

llvm-svn: 369019

4 years ago[Support] Base RWMutex on std::shared_timed_mutex (C++14)
Benjamin Kramer [Thu, 15 Aug 2019 16:55:23 +0000 (16:55 +0000)]
[Support] Base RWMutex on std::shared_timed_mutex (C++14)

This should have the same semantics. We use std::shared_mutex instead on
MSVC and C++17, std::shared_timed_mutex is less efficient than our
custom implementation on Windows, std::shared_mutex should be faster.

llvm-svn: 369018

4 years ago[LLVM][Alignment] Fix MSVC potential division by 0 warning (PR42911)
Simon Pilgrim [Thu, 15 Aug 2019 16:34:23 +0000 (16:34 +0000)]
[LLVM][Alignment] Fix MSVC potential division by 0 warning (PR42911)

Original Patch by @gchatelet (Guillaume Chatelet)

Differential Revision: https://reviews.llvm.org/D65859

llvm-svn: 369016

4 years ago[BUNDLER]Improve the test, NFC.
Alexey Bataev [Thu, 15 Aug 2019 16:28:24 +0000 (16:28 +0000)]
[BUNDLER]Improve the test, NFC.

Summary:
Make the test more portable and do not rely on the pre-bundled object
file.

Reviewers: Hahnfeld, hfinkel, jdoerfert

Subscribers: caomhin, kkwli0, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D66296

llvm-svn: 369015

4 years ago[Hexagon] Generate vector min/max for HVX
Krzysztof Parzyszek [Thu, 15 Aug 2019 16:13:17 +0000 (16:13 +0000)]
[Hexagon] Generate vector min/max for HVX

llvm-svn: 369014

4 years ago[llvm] Migrate llvm::make_unique to std::make_unique
Jonas Devlieghere [Thu, 15 Aug 2019 15:54:37 +0000 (15:54 +0000)]
[llvm] Migrate llvm::make_unique to std::make_unique

Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.

llvm-svn: 369013

4 years agoAdd a proposal for a libc project under the LLVM umbrella.
Siva Chandra [Thu, 15 Aug 2019 15:50:42 +0000 (15:50 +0000)]
Add a proposal for a libc project under the LLVM umbrella.

Reviewers: chandlerc, dlj, echristo, hfinkel, jfb, zturner

Subscribers: dexonsmith, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D64939

llvm-svn: 369012

4 years ago[cmake] install_symlink should obey DESTDIR unconditionally
Justin Bogner [Thu, 15 Aug 2019 15:36:13 +0000 (15:36 +0000)]
[cmake] install_symlink should obey DESTDIR unconditionally

Setting DESTDIR was erroneously buried under a condition here - if
it's set it should always be used.

llvm-svn: 369011

4 years ago[MCA] Slightly refactor class RetireControlUnit, and add the ability to override...
Andrea Di Biagio [Thu, 15 Aug 2019 15:27:40 +0000 (15:27 +0000)]
[MCA] Slightly refactor class RetireControlUnit, and add the ability to override the mask of used buffered resources in class mca::Instruction. NFCI

This patch teaches the RCU how to peek 'next' RCUTokens. A new method has been
added to the RetireControlUnit class with the goal of minimizing the complexity
of follow-up patches that will enable macro-fusion support in mca.

This patch also adds method Instruction::getNumMicroOpcodes() to simplify common
interactions with the instruction descriptor (a pattern quite common in some
pipeline stages).

Added the ability to override the default set of consumed scheduler resources
(this -again- is to simplify future patches that add support for macro-op fusion).

No functional change intended.

llvm-svn: 369010

4 years agoRe-land "[compiler-rt] Migrate llvm::make_unique to std::make_unique"
Jonas Devlieghere [Thu, 15 Aug 2019 14:57:44 +0000 (14:57 +0000)]
Re-land "[compiler-rt] Migrate llvm::make_unique to std::make_unique"

With the compiler-rt check for C++14 updated in r368960, this should now
be fine to land.

llvm-svn: 369009

4 years agoRemove SmallBitVector.h include. NFCI.
Simon Pilgrim [Thu, 15 Aug 2019 14:40:37 +0000 (14:40 +0000)]
Remove SmallBitVector.h include. NFCI.

SmallBitVector/BitVector types aren't used at all in the cpp file.

llvm-svn: 369008

4 years agoRemove BitVector.h include. NFCI.
Simon Pilgrim [Thu, 15 Aug 2019 14:39:28 +0000 (14:39 +0000)]
Remove BitVector.h include. NFCI.

BitVector type isn't used at all in the cpp file.

llvm-svn: 369007

4 years ago[PowerPC] Use xxleqv to set all one vector IMM(-1).
Jinsong Ji [Thu, 15 Aug 2019 14:32:51 +0000 (14:32 +0000)]
[PowerPC] Use xxleqv to set all one vector IMM(-1).

Summary:
xxspltib/vspltisb are 3 cycle PM instructions,
xxleqv is 2 cycle ALU instruction.

We should use xxleqv to set all one vectors.

Reviewers: hfinkel, nemanjai, steven.zhang

Subscribers: hiraditya, kbarton, MaskRay, shchenz, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65529

llvm-svn: 369006

4 years ago[clangd] Don't use Bind() where C++14 move capture works
Benjamin Kramer [Thu, 15 Aug 2019 14:16:06 +0000 (14:16 +0000)]
[clangd] Don't use Bind() where C++14 move capture works

llvm-svn: 369005

4 years ago[lib/Object] - Remove objdump-file-header.test
George Rimar [Thu, 15 Aug 2019 13:39:58 +0000 (13:39 +0000)]
[lib/Object] - Remove objdump-file-header.test

objdump-file-header.test is placed in the wrong folder.
I removed it and updated the existent llvm-objdump test cases with
the updated content of the file removed.

Differential revision: https://reviews.llvm.org/D66288

llvm-svn: 369004

4 years ago[OpenMP] Enable warning about "implicit fallthrough"
Jonas Hahnfeld [Thu, 15 Aug 2019 13:26:55 +0000 (13:26 +0000)]
[OpenMP] Enable warning about "implicit fallthrough"

Fix last warned location in ittnotify_static.cpp using the defined
macro KMP_FALLTHROUGH().

Differential Revision: https://reviews.llvm.org/D65871

llvm-svn: 369003

4 years ago[OpenMP] Remove 'unnecessary parentheses'
Jonas Hahnfeld [Thu, 15 Aug 2019 13:26:41 +0000 (13:26 +0000)]
[OpenMP] Remove 'unnecessary parentheses'

The variables in kmp_lock.cpp are really arrays of function pointers
that return void or int, not pointers to functions that return void*
or int*. The other changes are only cosmetic.

Differential Revision: https://reviews.llvm.org/D65870

llvm-svn: 369002

4 years ago[OMPT] Resolve warnings because of ints in if conditions
Jonas Hahnfeld [Thu, 15 Aug 2019 13:26:29 +0000 (13:26 +0000)]
[OMPT] Resolve warnings because of ints in if conditions

The implementation status can only be one of
ompt_event_UNIMPLEMENTED = ompt_set_never = 1
ompt_event_MAY_ALWAYS = ompt_set_always = 5

In both cases, the condition was already true, so just remove
the check.

Differential Revision: https://reviews.llvm.org/D65869

llvm-svn: 369001

4 years ago[lldb][NFC] Refactor remaining completion logic to use CompletionRequests
Raphael Isemann [Thu, 15 Aug 2019 13:14:10 +0000 (13:14 +0000)]
[lldb][NFC] Refactor remaining completion logic to use CompletionRequests

This patch moves the remaining completion functions from the
old completion API (that used several variables) to just
passing a single CompletionRequest.

This is for the most part a simple change as we just replace
the old arguments with a single CompletionRequest argument.

There are a few places where I had to create new CompletionRequests
in the called functions as CompletionRequests itself are immutable
and don't expose their internal match list anymore. This means that
if a function wanted to change the CompletionRequest or directly
access the result list, we need to work around this by creating
a new CompletionRequest and a temporary match/description list.

Preparation work for rdar://53769355

llvm-svn: 369000

4 years ago[OpenMP] Turn on -Wall compiler warnings by default
Jonas Hahnfeld [Thu, 15 Aug 2019 13:11:50 +0000 (13:11 +0000)]
[OpenMP] Turn on -Wall compiler warnings by default

Instead, maintain a list of disabled options to still build libomp and
libomptarget without warnings. This includes -Wno-error and -Wno-pedantic
to silence warnings that LLVM enables when building in-tree.

I tested the following compilers:
 * Clang 6.0, 7.0, 8.0
 * GCC 4.8.5 (CentOS 7), GCC 6, 7, 8, 9
 * Intel Compiler 16, 17, 18, 19

RFC thread on openmp-dev mailing list:
http://lists.llvm.org/pipermail/openmp-dev/2019-August/002668.html

Differential Revision: https://reviews.llvm.org/D65867

llvm-svn: 368999

4 years ago[DAGCombine] MergeConsecutiveStores - fix cppcheck/MSVC extension warning. NFCI.
Simon Pilgrim [Thu, 15 Aug 2019 13:07:14 +0000 (13:07 +0000)]
[DAGCombine] MergeConsecutiveStores - fix cppcheck/MSVC extension warning. NFCI.

Set the StartIdx type to size_t so that it matches the StoreNodes SmallVector size() and index types.

Silences the MSVC analyzer warning that unsigned increment might overflow before exceeding size_t on 64-bit targets - this isn't likely to happen but it means we use consistent types and reduces the warning "noise" a little.

llvm-svn: 368998

4 years ago[CodeGen] Do the Simple Early Return in block-placement pass to optimize the blocks
Kang Zhang [Thu, 15 Aug 2019 13:05:16 +0000 (13:05 +0000)]
[CodeGen] Do the Simple Early Return in block-placement pass to optimize the blocks

Summary:

This patch has trigger a bug of r368339, and the r368339 has been reverted, So upstream this patch again.

In `block-placement` pass, it will create some patterns for unconditional we can do the simple early retrun.
But the `early-ret` pass is before `block-placement`, we don't want to run it again.
This patch is to do the simple early return to optimize the blocks at the last of `block-placement`.

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D63972

llvm-svn: 368997

4 years ago[ARM] Fix alignment checks for BE VLDRH
David Green [Thu, 15 Aug 2019 12:54:47 +0000 (12:54 +0000)]
[ARM] Fix alignment checks for BE VLDRH

We need to allow any alignment at least 2, not just exactly 2, so that the big
endian loads and stores can be selected successfully. I've also added extra BE
testing for the load and store tests.

Thanks to Oliver for the report.

Differential Revision: https://reviews.llvm.org/D66222

llvm-svn: 368996

4 years ago[SDAG][x86] check for relaxed math when matching an FP reduction
Sanjay Patel [Thu, 15 Aug 2019 12:43:15 +0000 (12:43 +0000)]
[SDAG][x86] check for relaxed math when matching an FP reduction

If the last step in an FP add reduction allows reassociation and doesn't care
about -0.0, then we are free to recognize that computation as a reduction
that may reorder the intermediate steps.

This is requested directly by PR42705:
https://bugs.llvm.org/show_bug.cgi?id=42705
and solves PR42947 (if horizontal math instructions are actually faster than
the alternative):
https://bugs.llvm.org/show_bug.cgi?id=42947

Differential Revision: https://reviews.llvm.org/D66236

llvm-svn: 368995

4 years ago[MCA] Slightly refactor the logic in ResourceManager. NFCI
Andrea Di Biagio [Thu, 15 Aug 2019 12:39:55 +0000 (12:39 +0000)]
[MCA] Slightly refactor the logic in ResourceManager. NFCI

This patch slightly changes the API in the attempt to simplify resource buffer
queries. It is done in preparation for a patch that will enable support for
macro fusion.

llvm-svn: 368994

4 years ago[ValueTracking] Add MustPreserveNullness arg to functions analyzing calls. (NFC)
Florian Hahn [Thu, 15 Aug 2019 12:13:02 +0000 (12:13 +0000)]
[ValueTracking] Add MustPreserveNullness arg to functions analyzing calls. (NFC)

Some uses of getArgumentAliasingToReturnedPointer and
isIntrinsicReturningPointerAliasingArgumentWithoutCapturing require the
calls/intrinsics to preserve the nullness of the argument.

For alias analysis, the nullness property does not really come into
play.

This patch explicitly sets it to true. In D61669, the alias analysis
uses will be switched to not require preserving nullness.

Reviewers: nlopes, efriedma, hfinkel, sanjoy, aqjune, jdoerfert

Reviewed By: jdoerfert

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D64150

llvm-svn: 368993

4 years ago[ADT] PointerUnion: Use C++14 constexpr std::min
Benjamin Kramer [Thu, 15 Aug 2019 11:49:00 +0000 (11:49 +0000)]
[ADT] PointerUnion: Use C++14 constexpr std::min

llvm-svn: 368992

4 years ago[clangd] llvm::integer_sequence -> std::integer_sequence.
Benjamin Kramer [Thu, 15 Aug 2019 11:09:22 +0000 (11:09 +0000)]
[clangd] llvm::integer_sequence -> std::integer_sequence.

llvm-svn: 368991

4 years agoReplace llvm::integer_sequence and friends with the C++14 standard version
Benjamin Kramer [Thu, 15 Aug 2019 10:56:05 +0000 (10:56 +0000)]
Replace llvm::integer_sequence and friends with the C++14 standard version

The implementation in libc++ takes O(1) compile time, ours was O(n).

llvm-svn: 368990

4 years agoBump llvm-go to C++14
Benjamin Kramer [Thu, 15 Aug 2019 10:55:25 +0000 (10:55 +0000)]
Bump llvm-go to C++14

llvm-svn: 368989

4 years ago[ARM] MVE predicate store patterns
David Green [Thu, 15 Aug 2019 10:41:42 +0000 (10:41 +0000)]
[ARM] MVE predicate store patterns

Stack loads and stores were already working, but direct stores were not. This
adds the patterns for them, same as predicate loads.

Differential Revision: https://reviews.llvm.org/D66213

llvm-svn: 368988

4 years ago[AArch64] Change location of frame-record within callee-save area.
Sander de Smalen [Thu, 15 Aug 2019 10:34:16 +0000 (10:34 +0000)]
[AArch64] Change location of frame-record within callee-save area.

This patch changes the location of the frame-record (FP, LR) to the
bottom of the callee-saved area. According to the AAPCS the location of
the frame-record within the stackframe is unspecified (section 5.2.3 The
Frame Pointer), so the compiler should be free to choose a different
location.

The reason for changing the location of the frame-record is to prepare
the frame for allocating an SVE area below the callee-saves. This way the
compiler can use the VL-scaled addressing modes to directly access SVE
objects from the frame-pointer.

            :                :
        | stack |        | stack |
        |  args |        |  args |
        +-------+        +-------+
        |  x30  |        |  x19  |
        |  x29  |        |  x20  |
  FP -> |- - - -|        |  x21  |
        |  x19  |   ==>  |  x22  |
        |  x20  |        |- - - -|
        |  x21  |        |  x30  |
        |  x22  |        |  x29  |
        +-------+        +-------+ <- FP
        |///////|        |///////|         // realignment gap
        |- - - -|        |- - - -|
        |spills/|        |spills/|
        | locals|        | locals|
  SP -> +-------+        +-------+ <- SP

Things to point out:
- The algorithm to find a paired register should be prevented from
  accidentally pairing some callee-saved register with LR that is not
  FP, since they should always be paired together when the frame
  has a frame-record.
- For Darwin platforms the location of the frame-record is unchanged,
  since the unwind encoding does not allow for encoding this position
  dynamically and other tools currently depend on the former layout.

Reviewers: efriedma, rovka, rengolin, thegameg, greened, t.p.northover

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D65653

llvm-svn: 368987

4 years agoAdd ptrmask intrinsic
Florian Hahn [Thu, 15 Aug 2019 10:12:26 +0000 (10:12 +0000)]
Add ptrmask intrinsic

This patch adds a ptrmask intrinsic which allows masking out bits of a
pointer that must be zero when accessing it, because of ABI alignment
requirements or a restriction of the meaningful bits of a pointer
through the data layout.

This avoids doing a ptrtoint/inttoptr round trip in some cases (e.g. tagged
pointers) and allows us to not lose information about the underlying
object.

Reviewers: nlopes, efriedma, hfinkel, sanjoy, jdoerfert, aqjune

Reviewed by: sanjoy, jdoerfert

Differential Revision: https://reviews.llvm.org/D59065

llvm-svn: 368986

4 years agoRevert rL368939 "Remove LVALUE / RVALUE workarounds"
Russell Gallop [Thu, 15 Aug 2019 10:12:11 +0000 (10:12 +0000)]
Revert rL368939 "Remove LVALUE / RVALUE workarounds"

This reverts commit cad8356d699b36c73abb267f65db575ddacbd652.

To unbreak Windows bots

llvm-svn: 368985

4 years ago[Support] Fix Wundef warning
Sven van Haastregt [Thu, 15 Aug 2019 10:05:22 +0000 (10:05 +0000)]
[Support] Fix Wundef warning

llvm-svn: 368984

4 years ago[ELF][AArch64] Improve error message for unknown relocations
Fangrui Song [Thu, 15 Aug 2019 10:02:54 +0000 (10:02 +0000)]
[ELF][AArch64] Improve error message for unknown relocations

Like rLLD354040.

Previously, for unrecognized relocation types, in -no-pie/-pie mode, we got something like:

  foo.o: unrecognized relocation ...

In -shared mode:

  error: can't create dynamic relocation ... against symbol: yyy in readonly segment

Delete the default case from AArch64::getRelExpr and add the error there.

Reviewed By: grimar

Differential Revision: https://reviews.llvm.org/D66277

llvm-svn: 368983

4 years ago[llvm-objcopy] Allow 'protected' visibility to be set when using
Chris Jackson [Thu, 15 Aug 2019 09:45:09 +0000 (09:45 +0000)]
[llvm-objcopy] Allow 'protected' visibility to be set when using
add-symbol

Reviewers: Maskray, rupprecht

Differential Revision: https://reviews.llvm.org/D65891

llvm-svn: 368982

4 years ago[ARM] MVE trunc to i1 vectors
David Green [Thu, 15 Aug 2019 09:26:51 +0000 (09:26 +0000)]
[ARM] MVE trunc to i1 vectors

This adds patterns for selecting trunc instructions from full vectors to i1's
vectors.

Differential Revision: https://reviews.llvm.org/D66201

llvm-svn: 368981

4 years ago[analyzer] Warn about -analyzer-configs being meant for development purposes only
Kristof Umann [Thu, 15 Aug 2019 08:53:16 +0000 (08:53 +0000)]
[analyzer] Warn about -analyzer-configs being meant for development purposes only

This is more of a temporary fix, long term, we should convert AnalyzerOptions.def
into the universally beloved (*coughs*) TableGen format, where they can more
easily be separated into developer-only, alpha, and user-facing configs.

Differential Revision: https://reviews.llvm.org/D66261

llvm-svn: 368980

4 years ago[analyzer] Add docs for cplusplus.InnerPointer
Kristof Umann [Thu, 15 Aug 2019 08:52:10 +0000 (08:52 +0000)]
[analyzer] Add docs for cplusplus.InnerPointer

Differential Revision: https://reviews.llvm.org/D60281

llvm-svn: 368979

4 years ago[InstCombine] Precommit test case for D66216
Florian Hahn [Thu, 15 Aug 2019 08:42:12 +0000 (08:42 +0000)]
[InstCombine] Precommit test case for D66216

llvm-svn: 368978

4 years agoMemoryBuffer: Add a missing error-check to getOpenFileImpl
Pavel Labath [Thu, 15 Aug 2019 08:20:15 +0000 (08:20 +0000)]
MemoryBuffer: Add a missing error-check to getOpenFileImpl

Summary:
In case the function was called with a desired read size *and* the file
was not an "mmap()" candidate, the function was falling back to a
"pread()", but it was failing to check the result of that system call.
This meant that the function would return "success" even though the read
operation failed, and it returned a buffer full of uninitialized memory.

Reviewers: rnk, dblaikie

Subscribers: kristina, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66224

llvm-svn: 368977

4 years ago[clang] Loop pragma parsing. NFC.
Sjoerd Meijer [Thu, 15 Aug 2019 07:39:05 +0000 (07:39 +0000)]
[clang] Loop pragma parsing. NFC.

Just a refactoring and a tidy up.

Differential Revision: https://reviews.llvm.org/D64564

llvm-svn: 368976

4 years ago[lldb] Let LLDB depend on intrinsics_gen to fix non-deterministic module build failures
Raphael Isemann [Thu, 15 Aug 2019 07:29:53 +0000 (07:29 +0000)]
[lldb] Let LLDB depend on intrinsics_gen to fix non-deterministic module build failures

Summary:
When building with modules we currently fail randomly to build LLDB's modules and get error messages like below:
```
In file included from <module-includes>:1:
In file included from llvm-project/llvm/include/llvm/IR/Argument.h:18:
llvm-project/llvm/include/llvm/IR/Attributes.h:74:14: fatal error: 'llvm/IR/Attributes.inc' file not found
    #include "llvm/IR/Attributes.inc"
             ^~~~~~~~~~~~~~~~~~~~~~~~
...
In file included from llvm-project/lldb/source/Plugins/ABI/SysV-arm/ABISysV_arm.cpp:9:
llvm-project/lldb/source/Plugins/ABI/SysV-arm/ABISysV_arm.h:12:10: fatal error: could not build module 'lldb_Wrapper'
```

The reason for this is that our source files can include Clang headers which in turn include files like Attributes.inc which
are generated by intrinsics_gen. However, it seems Clang can't express this dependency in its CMake code, so intrinsics_gen
is actually not guaranteed to be built before we start parsing LLDB source files and Clang headers. Clang worked around this
by letting all libraries depend on intrinsics_gen, which at least lets their builds pass.

As I haven't figured out how to solve these dependencies properly and I want to get the LLDB+Modules build green,
I suggest we copy Clang's hack until we figure out how to get Clang's dependencies right.

Reviewers: sgraenitz, aprantl, JDevlieghere

Reviewed By: JDevlieghere

Subscribers: mgorny, javed.absar, kristof.beyls, mgrang, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D66208

llvm-svn: 368975

4 years ago[llvm-readobj][MachO] Fix section type printing
Seiya Nuta [Thu, 15 Aug 2019 07:22:04 +0000 (07:22 +0000)]
[llvm-readobj][MachO] Fix section type printing

Summary:
Currently, llvm-readobj mistakenly decodes section type as section attribute.

This patch fixes the bug and affected tests.

Reviewers: JDevlieghere, jhenderson, rupprecht, alexshap, echristo

Reviewed By: jhenderson, rupprecht, alexshap, echristo

Subscribers: javed.absar, rupprecht, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66075

llvm-svn: 368974

4 years ago[LV] fold-tail predication should be respected even with assume_safety
Dorit Nuzman [Thu, 15 Aug 2019 07:12:14 +0000 (07:12 +0000)]
[LV] fold-tail predication should be respected even with assume_safety

assume_safety implies that loads under "if's" can be safely executed
speculatively (unguarded, unmasked). However this assumption holds only for the
original user "if's", not those introduced by the compiler, such as the
fold-tail "if" that guards us from loading beyond the original loop trip-count.
Currently the combination of fold-tail and assume-safety pragmas results in
ignoring the fold-tail predicate that guards the loads, generating unmasked
loads. This patch fixes this behavior.

Differential Revision: https://reviews.llvm.org/D66106

Reviewers: Ayal, hsaito, fhahn
llvm-svn: 368973

4 years ago[X86] Add isel pattern to match VZEXT_MOVL and a v2i64 scalar_to_vector bitcasted...
Craig Topper [Thu, 15 Aug 2019 06:46:30 +0000 (06:46 +0000)]
[X86] Add isel pattern to match VZEXT_MOVL and a v2i64 scalar_to_vector bitcasted from x86mmx to MOVQ2DQ.

We already had the pattern for just the scalar to vector and bitcast,
but not the case where we wanted zeroes in the high half of the xmm.

llvm-svn: 368972

4 years ago[X86] Make sure load is non-volatile in the MMX_X86movdq2q (loadv2i64) isel pattern.
Craig Topper [Thu, 15 Aug 2019 06:46:26 +0000 (06:46 +0000)]
[X86] Make sure load is non-volatile in the MMX_X86movdq2q (loadv2i64) isel pattern.

This pattern will narrow the load so we should make sure its not
volatile.

llvm-svn: 368971

4 years ago[Clang] Pragma vectorize_predicate implies vectorize
Sjoerd Meijer [Thu, 15 Aug 2019 06:24:40 +0000 (06:24 +0000)]
[Clang] Pragma vectorize_predicate implies vectorize

New pragma "vectorize_predicate(enable)" now implies "vectorize(enable)",
and it is ignored when vectorization is disabled with e.g.
"vectorize(disable) vectorize_predicate(enable)".

Differential Revision: https://reviews.llvm.org/D65776

llvm-svn: 368970

4 years ago[X86] Add test cases for _mm_movepi64_pi64 and _mm_movpi64_epi64.
Craig Topper [Thu, 15 Aug 2019 06:20:33 +0000 (06:20 +0000)]
[X86] Add test cases for _mm_movepi64_pi64 and _mm_movpi64_epi64.

llvm-svn: 368969

4 years ago[X86] Remove unneeded isel pattern for v4f32->v4i32 fp_to_sint and conversion to...
Craig Topper [Thu, 15 Aug 2019 05:52:02 +0000 (05:52 +0000)]
[X86] Remove unneeded isel pattern for v4f32->v4i32 fp_to_sint and conversion to MMX.

fp_to_sint is turned into X86cvttp2si during isel preprocessing.
The other redundant isel patterns were removed previously, but I
missed this one because its in the MMX td file.

llvm-svn: 368968

4 years ago[X86] Disable custom type legalization for v2i32/v4i16/v8i8->i64.
Craig Topper [Thu, 15 Aug 2019 05:51:58 +0000 (05:51 +0000)]
[X86] Disable custom type legalization for v2i32/v4i16/v8i8->i64.

The default legalization can take care of this.

llvm-svn: 368967

4 years ago[X86] Disable custom type legalization for v2i32/v4i16/v8i8->f64 bitcast.
Craig Topper [Thu, 15 Aug 2019 05:51:54 +0000 (05:51 +0000)]
[X86] Disable custom type legalization for v2i32/v4i16/v8i8->f64 bitcast.

The generic legalization handles this in the same way so just use
that.

llvm-svn: 368966

4 years ago[X86] Remove some unreachable code from LowerBITCAST.
Craig Topper [Thu, 15 Aug 2019 05:51:50 +0000 (05:51 +0000)]
[X86] Remove some unreachable code from LowerBITCAST.

llvm-svn: 368965

4 years ago[ELF][PPC] Improve error message for unknown relocations
Fangrui Song [Thu, 15 Aug 2019 05:22:23 +0000 (05:22 +0000)]
[ELF][PPC] Improve error message for unknown relocations

Like rLLD354040.

Previously, for unrecognized relocation types, in -no-pie mode:

  foo.o: unrecognized reloc 256

In -pie/-shared mode:

  error: can't create dynamic relocation R_PPC_xxx against symbol: yyy in readonly segment

llvm-svn: 368964

4 years ago[llvm-objdump] Add warning messages if disassembly + source for problematic inputs
Michael Pozulp [Thu, 15 Aug 2019 05:15:22 +0000 (05:15 +0000)]
[llvm-objdump] Add warning messages if disassembly + source for problematic inputs

Summary: Addresses https://bugs.llvm.org/show_bug.cgi?id=41905

Reviewers: jhenderson, rupprecht, grimar

Reviewed By: jhenderson, grimar

Subscribers: RKSimon, MaskRay, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62462

llvm-svn: 368963

4 years agoFix variable mismatch between signature and body
Jonas Devlieghere [Thu, 15 Aug 2019 05:09:09 +0000 (05:09 +0000)]
Fix variable mismatch between signature and body

I updated the signature to conform to the LLDB coding style but
accidentally forgot to update the function body.

llvm-svn: 368962

4 years ago[llvm-objcopy] Move duplicate tablegen from objcopy and strip into one file
Michael Pozulp [Thu, 15 Aug 2019 04:45:41 +0000 (04:45 +0000)]
[llvm-objcopy] Move duplicate tablegen from objcopy and strip into one file

Summary: This avoids maintaining the same options in two different places.

Reviewers: jhenderson, alexshap, rupprecht, MaskRay

Reviewed By: jhenderson, rupprecht, MaskRay

Subscribers: MaskRay, wolfgangp, jakehehrlich, abrachet, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65991

llvm-svn: 368961

4 years ago[CMake] Check for C++14 instead of C++11
Jonas Devlieghere [Thu, 15 Aug 2019 04:42:15 +0000 (04:42 +0000)]
[CMake] Check for C++14 instead of C++11

Now that LLVM moved to C++14, `COMPILER_RT_HAS_STD_CXX11_FLAG` should
become `COMPILER_RT_HAS_STD_CXX14_FLAG`.

I ran into this issue when replacing llvm::make_unique with
std::make_unique in an X-ray unit test. We are correctly passing
`-std=c++14`, but this got overwritten further down the invocation by
the compiler-rt flags. Given that this unit test is using LLVM headers,
this is bound to break sooner than later, regardless of my change.

Differential revision: https://reviews.llvm.org/D66271

llvm-svn: 368960

4 years ago[NFC] Fix documentation for some utility classes.
Jonas Devlieghere [Thu, 15 Aug 2019 04:35:46 +0000 (04:35 +0000)]
[NFC] Fix documentation for some utility classes.

This fixes a few warnings emitted when compiling with -Wdocumentation.

llvm-svn: 368959

4 years ago[Tooling] Add a hack to work around issues with matcher binding in r368681.
David L. Jones [Thu, 15 Aug 2019 04:10:11 +0000 (04:10 +0000)]
[Tooling] Add a hack to work around issues with matcher binding in r368681.

The change in r368681 contains a (probably unintentional) behavioral change for
rewrite rules with a single matcher. Previously, the single matcher would not
need to be bound (`joinCaseMatchers` returned it directly), even though a final
DynTypeMatcher was created and bound by `buildMatcher`. With the new change, a
single matcher will be bound, in addition to the final binding (which is now in
`buildMatchers`, but happens roughly at the same point in the overall flow).

This patch simply duplicates the "final matcher" trick: it creates an extra
DynTypedMatcher for each rewrite rule case matcher, and unconditionally makes it
bindable. This is probably not the right long-term fix, but it does allow
existing code to continue to work with this interface.

Subscribers: cfe-commits, gribozavr, ymandel

Tags: #clang

Differential Revision: https://reviews.llvm.org/D66273

llvm-svn: 368958

4 years ago[NFC] Remove documentation comment to fix warning.
Jonas Devlieghere [Thu, 15 Aug 2019 04:09:00 +0000 (04:09 +0000)]
[NFC] Remove documentation comment to fix warning.

Remove the @return to fix the warning: '@returns' command used in a
comment that is attached to a function returning void [-Wdocumentation]

llvm-svn: 368957

4 years ago[X86] Remove some dead code and combine some repeated code that's left.
Craig Topper [Thu, 15 Aug 2019 04:07:43 +0000 (04:07 +0000)]
[X86] Remove some dead code and combine some repeated code that's left.

If the width is 256 bits, then we must have AVX so the else here
was unnecessary. Once that's removed then the >= 256 bit code is
identical to the 128 bit code with a different VT so combine them.

llvm-svn: 368956

4 years ago[NFC] Update doc comment to fix warning.
Jonas Devlieghere [Thu, 15 Aug 2019 03:59:30 +0000 (03:59 +0000)]
[NFC] Update doc comment to fix warning.

This fixes the warning: parameter 'EnableNullFPSuppression' not found in
the function declaration [-Wdocumentation]

llvm-svn: 368954

4 years agoRevert "Expose TailCallKind via the LLVM C API"
Jonas Devlieghere [Thu, 15 Aug 2019 03:49:51 +0000 (03:49 +0000)]
Revert "Expose TailCallKind via the LLVM C API"

This is failing on several build bots. Reverting as discussed in
https://reviews.llvm.org/D66061.

llvm-svn: 368953

4 years agoRevert "[compiler-rt] Migrate llvm::make_unique to std::make_unique"
Jonas Devlieghere [Thu, 15 Aug 2019 03:02:57 +0000 (03:02 +0000)]
Revert "[compiler-rt] Migrate llvm::make_unique to std::make_unique"

The X-ray unit tests in compiler-rt are overriding the C++ version by
explicitly passing -std=c++11 in the compiler invocation. This poses a
problem as these tests are including LLVM headers that can now use C++14
features. I'm temporarily reverting this as I investigate the correct
solution.

llvm-svn: 368952

4 years agogn build: Merge r368918
Vitaly Buka [Thu, 15 Aug 2019 01:38:30 +0000 (01:38 +0000)]
gn build: Merge r368918

llvm-svn: 368951

4 years ago[NFCI] Always initialize BugReport const fields
Alex Langford [Thu, 15 Aug 2019 00:58:51 +0000 (00:58 +0000)]
[NFCI] Always initialize BugReport const fields

Summary:
Some compilers require that const fields of an object must be explicitly
initialized by the constructor. I ran into this issue building with clang
3.8 on Ubuntu 16.04.

Reviewers: compnerd, Szelethus, NoQ

Subscribers: cfe-commits, llvm-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D66265

llvm-svn: 368950

4 years ago[coroutine] Fixes "cannot move instruction since its users are not dominated by CoroB...
Gor Nishanov [Thu, 15 Aug 2019 00:48:51 +0000 (00:48 +0000)]
[coroutine] Fixes "cannot move instruction since its users are not dominated by CoroBegin" problem.

Summary:
Fixes https://bugs.llvm.org/show_bug.cgi?id=36578 and https://bugs.llvm.org/show_bug.cgi?id=36296.
Supersedes: https://reviews.llvm.org/D55966

One of the fundamental transformation that CoroSplit pass performs before splitting the coroutine is to find which values need to survive between suspend and resume and provide a slot for them in the coroutine frame to spill and restore the value as needed.

Coroutine frame becomes available once the storage for it was allocated and that point is marked in the pre-split coroutine with a llvm.coro.begin intrinsic.

FE normally puts all of the user-authored code that would be accessing those values after llvm.coro.begin, however, sometimes instructions accessing those values would end up prior to coro.begin. For example, writing out a value of the parameter into the alloca done by the FE or instructions that are added by the optimization passes such as SROA when it rewrites allocas.

Prior to this change, CoroSplit pass would try to move instructions that may end up accessing the values in the coroutine frame after CoroBegin. However it would run into problems (report_fatal_error) if some of the values would be used both in the allocation function (for example allocator is passed as a parameter to a coroutine) and in the use-authored body of the coroutine.

To handle this case and to simplify the instruction moving logic, this change removes all of the instruction moving. Instead, we only change the uses of the spilled values that are dominated by coro.begin and leave other instructions intact.

Before:

```
%var = alloca i32
%1 = getelementptr .. %var; ; will move this one after coro.begin
%f = call i8* @llvm.coro.begin(
```

After:

```
%var = alloca i32
%1 = getelementptr .. %var; stays put
%f = call i8* @llvm.coro.begin(
```
If we discover that there is a potential write into an alloca, prior to coro.begin we would copy its value from the alloca into the spill slot in the coroutine frame.

Before:

```
%var = alloca i32
store .. %var ; will move this one after coro.begin
%f = call i8* @llvm.coro.begin(
```

After:

```
%var = alloca i32
store .. %var ;stays put
%f = call i8* @llvm.coro.begin(
%tmp = load %var
store %tmp, %spill.slot.for.var
```

Note: This change does not handle array allocas as that is something that C++ FE does not produce, but, it can be added in the future if need arises

Reviewers: llvm-commits, modocache, ben-clayton, tks2103, rjmccall

Reviewed By: modocache

Subscribers: bartdesmet

Differential Revision: https://reviews.llvm.org/D66230

llvm-svn: 368949

4 years ago[sanitizer_common] Replace forkpty with posix_spawn on Darwin
Julian Lettner [Thu, 15 Aug 2019 00:18:55 +0000 (00:18 +0000)]
[sanitizer_common] Replace forkpty with posix_spawn on Darwin

On Darwin, we currently use forkpty to communicate with the "atos"
symbolizer. There are several problems that fork[pty] has, e.g. that
after fork, interceptors are still active and this sometimes causes
crashes or hangs. This is especially problematic for TSan, which uses
interceptors for OS-provided locks and mutexes, and even Libc functions
use those.

This patch replaces forkpty with posix_spawn on Darwin. Since
posix_spawn doesn't fork (at least on Darwin), the interceptors are not
a problem. Another benefit is that we'll handle post-fork failures (e.g.
sandbox disallows "exec") gracefully now.

Related revisions and previous attempts that were blocked by or had to
be revered due to test failures:
https://reviews.llvm.org/D48451
https://reviews.llvm.org/D40032

Reviewed By: kubamracek

Differential Revision: https://reviews.llvm.org/D65253

llvm-svn: 368947

4 years ago[compiler-rt] Migrate llvm::make_unique to std::make_unique
Jonas Devlieghere [Thu, 15 Aug 2019 00:06:49 +0000 (00:06 +0000)]
[compiler-rt] Migrate llvm::make_unique to std::make_unique

Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.

Differential revision: https://reviews.llvm.org/D66259

llvm-svn: 368946

4 years agoExpose TailCallKind via the LLVM C API
Robert Widmann [Wed, 14 Aug 2019 23:54:35 +0000 (23:54 +0000)]
Expose TailCallKind via the LLVM C API

Summary: This exposes `CallInst`'s tail call kind via new `LLVMGetTailCallKind` and `LLVMSetTailCallKind` functions. The motivation for this is to be able to see `musttail` for languages that require mandatory tail calls for correctness. Today only the weaker `LLVMSetTail` is exposed and there is no way to set `GuaranteedTailCallOpt` via the C API.

Reviewers: CodaFi, jyknight, deadalnix, rnk

Reviewed By: CodaFi

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66061

llvm-svn: 368945

4 years ago[clang-tools-extra] Migrate llvm::make_unique to std::make_unique
Jonas Devlieghere [Wed, 14 Aug 2019 23:52:23 +0000 (23:52 +0000)]
[clang-tools-extra] Migrate llvm::make_unique to std::make_unique

Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.

Differential revision: https://reviews.llvm.org/D66259

llvm-svn: 368944

4 years ago[Clang] Migrate llvm::make_unique to std::make_unique
Jonas Devlieghere [Wed, 14 Aug 2019 23:04:18 +0000 (23:04 +0000)]
[Clang] Migrate llvm::make_unique to std::make_unique

Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.

Differential revision: https://reviews.llvm.org/D66259

llvm-svn: 368942

4 years ago[www] Update DR status page to match latest version of CWG issues list.
Richard Smith [Wed, 14 Aug 2019 22:57:51 +0000 (22:57 +0000)]
[www] Update DR status page to match latest version of CWG issues list.

llvm-svn: 368941

4 years agoFix handling of class member access into a vector type.
Richard Smith [Wed, 14 Aug 2019 22:57:50 +0000 (22:57 +0000)]
Fix handling of class member access into a vector type.

When handling a member access into a non-class, non-ObjC-object type, we
would perform a lookup into the surrounding scope as if for an
unqualified lookup. If the member access was followed by a '<' and this
lookup (or the typo-correction for it) found a template name, we'd treat
the member access as naming that template.

Now we treat such accesses as never naming a template if the type of the
object expression is of vector type, so that vector component accesses
are never misinterpreted as naming something else. This is not entirely
correct, since it is in fact valid to name a template from the enclosing
scope in this context, when invoking a pseudo-destructor for the vector
type via an alias template, but that's very much a corner case, and this
change leaves that case only as broken as the corresponding case for
Objective-C types is.

This incidentally adds support for dr2292, which permits a 'template'
keyword at the start of a member access naming a pseudo-destructor.

llvm-svn: 368940

4 years agoRemove LVALUE / RVALUE workarounds
JF Bastien [Wed, 14 Aug 2019 22:48:12 +0000 (22:48 +0000)]
Remove LVALUE / RVALUE workarounds

Summary: LLVM_HAS_RVALUE_REFERENCE_THIS and LLVM_LVALUE_FUNCTION shouldn't be needed anymore because the minimum compiler versions support them.

Subscribers: jkorous, dexonsmith, cfe-commits, llvm-commits, hans, thakis, chandlerc, rnk

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D66240

llvm-svn: 368939

4 years ago[Attributor] Try to fix "missing field 'RetInsts' initializer" warning
Johannes Doerfert [Wed, 14 Aug 2019 22:32:29 +0000 (22:32 +0000)]
[Attributor] Try to fix "missing field 'RetInsts' initializer" warning

http://lab.llvm.org:8011/builders/lld-x86_64-darwin13/builds/35674/steps/build_Lld/logs/stdio

llvm-svn: 368938

4 years agoImprove anonymous class heuristic in ClangASTContext::CreateRecordType
Shafik Yaghmour [Wed, 14 Aug 2019 22:30:29 +0000 (22:30 +0000)]
Improve anonymous class heuristic in ClangASTContext::CreateRecordType

Summary:
Currently the heuristic used in ClangASTContext::CreateRecordType to identify an anonymous class is that there is that name is a nullptr or simply a null terminator. This heuristic is not accurate since it will also sweep up unnamed classes and lambdas. The improved heuristic relies on the requirement that an anonymous class must be contained within a class.

Differential Revision: https://reviews.llvm.org/D66175

llvm-svn: 368937

4 years ago[LLD] Migrate llvm::make_unique to std::make_unique
Jonas Devlieghere [Wed, 14 Aug 2019 22:28:17 +0000 (22:28 +0000)]
[LLD] Migrate llvm::make_unique to std::make_unique

Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.

Differential revision: https://reviews.llvm.org/D66259

llvm-svn: 368936

4 years ago[Polly] Migrate llvm::make_unique to std::make_unique
Jonas Devlieghere [Wed, 14 Aug 2019 22:28:12 +0000 (22:28 +0000)]
[Polly] Migrate llvm::make_unique to std::make_unique

Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.

Differential revision: https://reviews.llvm.org/D66259

llvm-svn: 368935

4 years ago[Bugpoint redesign] Modified Functions pass to consider declarations
Diego Trevino Ferrer [Wed, 14 Aug 2019 22:22:37 +0000 (22:22 +0000)]
[Bugpoint redesign] Modified Functions pass to consider declarations

Summary: This modification was put in place so the `ReduceMetadata` pass doesn't have to consider debug functions

Reviewers: dblaikie

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66257

llvm-svn: 368934

4 years ago[LLDB] Migrate llvm::make_unique to std::make_unique
Jonas Devlieghere [Wed, 14 Aug 2019 22:19:23 +0000 (22:19 +0000)]
[LLDB] Migrate llvm::make_unique to std::make_unique

Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.

Differential revision: https://reviews.llvm.org/D66259

llvm-svn: 368933

4 years ago[docs] Fix sphinx doc generation errors
Jordan Rupprecht [Wed, 14 Aug 2019 22:18:01 +0000 (22:18 +0000)]
[docs] Fix sphinx doc generation errors

Summary:
Errors fixed:
 - GettingStarted: Duplicate explicit target name: "cmake"
 - GlobalISel: Unexpected indentation
 - LoopTerminology: Explicit markup ends without a blank line; unexpected unindent
 - ORCv2: Definition list ends without a blank line; unexpected unindent
 - Misc: document isn't included in any toctree

Verified that a clean docs build (`rm -rf docs/ && ninja docs-llvm-html`) passes with no errors. Spot checked the individual pages to make sure they look OK.

Reviewers: thakis, dsanders

Reviewed By: dsanders

Subscribers: arphaman, llvm-commits, lhames, rovka, dsanders, reames

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66183

llvm-svn: 368932

4 years ago[Attributor][NFC] Make debug output consistent
Johannes Doerfert [Wed, 14 Aug 2019 22:04:28 +0000 (22:04 +0000)]
[Attributor][NFC] Make debug output consistent

llvm-svn: 368931

4 years ago[SCEV] Rename getMaxBackedgeTakenCount to getConstantMaxBackedgeTakenCount [NFC]
Philip Reames [Wed, 14 Aug 2019 21:58:13 +0000 (21:58 +0000)]
[SCEV] Rename getMaxBackedgeTakenCount to getConstantMaxBackedgeTakenCount [NFC]

llvm-svn: 368930

4 years ago[LifetimeAnalysis] Support std::stack::top() and std::optional::value()
Matthias Gehre [Wed, 14 Aug 2019 21:55:57 +0000 (21:55 +0000)]
[LifetimeAnalysis] Support std::stack::top() and std::optional::value()

Summary: Diagnose dangling pointers that come from std::stack::top() and std::optional::value().

Reviewers: gribozavr

Subscribers: cfe-commits, xazax.hun

Tags: #clang

Differential Revision: https://reviews.llvm.org/D66164

llvm-svn: 368929

4 years ago[Attributor][NFC] Try to eliminate warnings (debug build + fall through)
Johannes Doerfert [Wed, 14 Aug 2019 21:46:28 +0000 (21:46 +0000)]
[Attributor][NFC] Try to eliminate warnings (debug build + fall through)

llvm-svn: 368928

4 years ago[Attributor][NFC] Introduce statistics macros for new positions
Johannes Doerfert [Wed, 14 Aug 2019 21:46:25 +0000 (21:46 +0000)]
[Attributor][NFC] Introduce statistics macros for new positions

llvm-svn: 368927

4 years ago[SelectionDAGBuilder] Teach gather/scatter getUniformBase to look through vector...
Craig Topper [Wed, 14 Aug 2019 21:38:56 +0000 (21:38 +0000)]
[SelectionDAGBuilder] Teach gather/scatter getUniformBase to look through vector zeroinitializer indices in addition to scalar zeroes.

llvm-svn: 368926

4 years ago[Attributor][NFC] Add merge/join/clamp operators to the IntegerState
Johannes Doerfert [Wed, 14 Aug 2019 21:35:20 +0000 (21:35 +0000)]
[Attributor][NFC] Add merge/join/clamp operators to the IntegerState

Differential Revision: https://reviews.llvm.org/D66146

llvm-svn: 368925

4 years ago[Attributor] Use the AANoNull attribute directly in AADereferenceable
Johannes Doerfert [Wed, 14 Aug 2019 21:31:32 +0000 (21:31 +0000)]
[Attributor] Use the AANoNull attribute directly in AADereferenceable

Summary:
Instead of constantly keeping track of the nonnull status with the
dereferenceable information we can simply query the nonnull attribute
whenever we need the information (debug + manifest).

Reviewers: sstefan1, uenoku

Subscribers: hiraditya, bollu, jfb, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66113

llvm-svn: 368924

4 years ago[AArch64][GlobalISel] Custom selection for s8 load acquire.
Amara Emerson [Wed, 14 Aug 2019 21:30:30 +0000 (21:30 +0000)]
[AArch64][GlobalISel] Custom selection for s8 load acquire.

Implement this single atomic load instruction so that we can compile stack
protector code.

Differential Revision: https://reviews.llvm.org/D66245

llvm-svn: 368923

4 years ago[Attributor] Use liveness during the creation of AAReturnedValues
Johannes Doerfert [Wed, 14 Aug 2019 21:29:37 +0000 (21:29 +0000)]
[Attributor] Use liveness during the creation of AAReturnedValues

Summary:
As one of the first attributes, and one of the complex ones,
AAReturnedValues was not using liveness but we filtered the result after
the fact. This change adds liveness usage during the creation. The
algorithm is also improved and shorter.

The new algorithm will collect returned values over time using the
generic facilities that work with liveness already, e.g.,
genericValueTraversal which does not look at dead PHI node predecessors.
A test to show how this leads to better results is included.

Note: Unresolved calls and resolved calls are now tracked explicitly.

Reviewers: uenoku, sstefan1

Subscribers: hiraditya, bollu, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66120

llvm-svn: 368922

4 years ago[Attributor] Do not update or manifest dead attributes
Johannes Doerfert [Wed, 14 Aug 2019 21:25:08 +0000 (21:25 +0000)]
[Attributor] Do not update or manifest dead attributes

Summary:
If the associated context instruction is assumed dead we do not need to
update or manifest the state.

Reviewers: sstefan1, uenoku

Subscribers: hiraditya, bollu, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66116

llvm-svn: 368921

4 years agoRevert "[lldb] Reinstate original guard variable check"
Raphael Isemann [Wed, 14 Aug 2019 21:21:14 +0000 (21:21 +0000)]
Revert "[lldb] Reinstate original guard variable check"

It seems this breaks the following tests:
    lldb-Suite :: expression_command/call-function/TestCallUserDefinedFunction.py
    lldb-Suite :: expression_command/rdar42038760/TestScalarURem.py

Let's revert this patch and wait until we find an actual issue that could be
fixed by also doing the guard variable check on Windows.

llvm-svn: 368920

4 years ago[Attributor] Use IRPosition consistently
Johannes Doerfert [Wed, 14 Aug 2019 21:18:01 +0000 (21:18 +0000)]
[Attributor] Use IRPosition consistently

Summary:
The next attempt to clean up the Attributor interface before we grow it
further.

Before, we used a combination of two values (associated + anchor) and an
argument number (or -1) to determine a location. This was very fragile.
The new system uses exclusively IR positions and we restrict the
generation of IR positions to special constructor methods that verify
internal constraints we have. This will catch misuse early.

The auto-conversion, e.g., in getAAFor, is now performed through the
SubsumingPositionIterator. This iterator takes an IR position and allows
to visit all IR positions that "subsume" the given one, e.g., function
attributes "subsume" argument attributes of that function. For a
detailed breakdown see the class comment of SubsumingPositionIterator.

This patch also introduces the IRPosition::getAttrs() to extract IR
attributes at a certain position. The method knows how to look up in
different positions that are equivalent, e.g., the argument position for
call site arguments. We also introduce three new positions kinds such
that we have all IR positions where attributes can be placed and one for
"floating" values.

Reviewers: sstefan1, uenoku

Subscribers: hiraditya, bollu, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65977

llvm-svn: 368919

4 years ago[Bugpoint redesign] Added Pass to Remove Global Variables
Diego Trevino Ferrer [Wed, 14 Aug 2019 21:01:19 +0000 (21:01 +0000)]
[Bugpoint redesign] Added Pass to Remove Global Variables

Summary:
This pass tries to remove Global Variables, as well as their derived uses. For example if a variable `@x` is used by `%call1` and `%call2`, both these uses and the definition of `@x` are deleted. Moreover if `%call1` or `%call2` are used elsewhere those uses are also deleted, and so on recursively.

I'm still uncertain if this pass should remove derived uses, I'm open to suggestions.

Subscribers: mgorny, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D64176

llvm-svn: 368918

4 years ago[AMDGPU] Do not assume a default GCN target
Stanislav Mekhanoshin [Wed, 14 Aug 2019 20:55:15 +0000 (20:55 +0000)]
[AMDGPU] Do not assume a default GCN target

Differential Revision: https://reviews.llvm.org/D66246

llvm-svn: 368917

4 years agoFix thread comparison by making sure we never pass our special 'not a thread' value...
Marshall Clow [Wed, 14 Aug 2019 20:54:56 +0000 (20:54 +0000)]
Fix thread comparison by making sure we never pass our special 'not a thread' value to the underlying implementation. Fixes PR#42918.

llvm-svn: 368916