platform/upstream/llvm.git
4 years agoRevert "[InstCombine] Fold PHIs with equal incoming pointers"
Daniil Suchkov [Thu, 14 Nov 2019 10:42:01 +0000 (17:42 +0700)]
Revert "[InstCombine] Fold PHIs with equal incoming pointers"

This reverts commit a2f6ae9abffcba260c22bb235879f0576bf3b783.
It is reverted due to clang-cmake-armv7-selfhost buildbot failure.

4 years ago[lldb] Fix that trailing backslashes in source lines break the Clang highlighter
Raphael Isemann [Thu, 14 Nov 2019 10:11:20 +0000 (11:11 +0100)]
[lldb] Fix that trailing backslashes in source lines break the Clang highlighter

Summary:
Clang's raw Lexer doesn't produce any tokens for trailing backslashes in a line. This doesn't work with
LLDB's Clang highlighter which builds the source code to display from the list of tokens the Lexer returns.
This causes that lines with trailing backslashes are lacking the backslash and the following newline when
rendering source code in LLDB.

This patch removes the trailing newline from the current line we are highlighting. This way Clang doesn't
drop the backslash token and we just restore the newline after tokenising.

Fixes rdar://57091487

Reviewers: JDevlieghere, labath

Reviewed By: JDevlieghere, labath

Subscribers: labath, lldb-commits

Tags: #lldb

Differential Revision: https://reviews.llvm.org/D70177

4 years ago[InstCombine] Fold PHIs with equal incoming pointers
Daniil Suchkov [Thu, 14 Nov 2019 08:25:31 +0000 (15:25 +0700)]
[InstCombine] Fold PHIs with equal incoming pointers

This is a resubmission of bbb29738b58aaf6f6518269abdcf8f64131665a9 that
was reverted due to clang tests failures. It includes the fix and
additional IR tests for the missed case.

Summary:
In case when all incoming values of a PHI are equal pointers, this
transformation inserts a definition of such a pointer right after
definition of the base pointer and replaces with this value both PHI and
all it's incoming pointers. Primary goal of this transformation is
canonicalization of this pattern in order to enable optimizations that
can't handle PHIs. Non-inbounds pointers aren't currently supported.

Reviewers: spatel, RKSimon, lebedev.ri, apilipenko

Reviewed By: apilipenko

Tags: #llvm

Subscribers: hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D68128

4 years ago[AArch64][DebugInfo] Fix incorrect call site param value produced by MOVZXi
Djordje Todorovic [Thu, 14 Nov 2019 07:32:03 +0000 (08:32 +0100)]
[AArch64][DebugInfo] Fix incorrect call site param value produced by MOVZXi

This resolves the problem with the truncation of the immediate operand.

Differential Revision: https://reviews.llvm.org/D70168

4 years ago[NFC] Fix typo in ARMBaseRegisterInfo
Anna Welker [Thu, 14 Nov 2019 09:43:26 +0000 (09:43 +0000)]
[NFC] Fix typo in ARMBaseRegisterInfo

4 years ago[ThinLTO] Add correctness check for RO/WO variable import
evgeny [Thu, 14 Nov 2019 09:24:05 +0000 (12:24 +0300)]
[ThinLTO] Add correctness check for RO/WO variable import

This patch adds an assertion check for exported read/write-only
variables to be also in import list for module. If they aren't
we may face linker errors, because read/write-only variables are
internalized in their source modules. The patch also changes
export lists to store ValueInfo instead of GUID for performance
considerations.

Differential revision: https://reviews.llvm.org/D70128

4 years agoDWARFDebugLoclists: stricter base address handling
Pavel Labath [Tue, 12 Nov 2019 10:36:19 +0000 (11:36 +0100)]
DWARFDebugLoclists: stricter base address handling

Summary:
This removes the use of zero as a base address in section-based dumping.
Although this will often be true for (unlinked) object files with a
single compile unit, it is not true in general. This means that
section-based dumping will not be able to resolve entries referencing
the base address (DW_LLE_offset_pair) -- it wasn't able to do that
correctly before either, but now it will be more explicit about it. One
exception to that is if the location list contains an explicit
DW_LLE_base_address entry -- in this case the dumper will pick it up,
and resolve subsequent entries normally.

The patch also removes the fallback to zero in the "inline" dumping in
case the compile unit does not contain a base address.

Reviewers: dblaikie, probinson, JDevlieghere, aprantl, SouraVX

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70115

4 years ago[lldb][NFC] Simplify IOHandler constructor/destructor setup
Raphael Isemann [Thu, 14 Nov 2019 08:47:10 +0000 (09:47 +0100)]
[lldb][NFC] Simplify IOHandler constructor/destructor setup

We only need a default constructor because of DISALLOW_COPY_AND_ASSIGN,
but the non-virtual destructor isn't needed.

4 years ago[LLDB] Fix whitespace/tabs mismatch in lldbsuite Makefile.rules
Muhammad Omair Javaid [Thu, 14 Nov 2019 08:53:51 +0000 (13:53 +0500)]
[LLDB] Fix whitespace/tabs mismatch in lldbsuite Makefile.rules

This patch fixes whitespace/tabs mismatch in
lldb/packages/Python/lldbsuite/test/make/Makefile.rules

Legacy make files always used tabs though modern make version can
work with white-spaces I have chosen the legacy just to be safe.

Signed-off-by: Muhammad Omair Javaid <omair.javaid@linaro.org>
Differential Revision: https://reviews.llvm.org/D70154

4 years ago[clang][IFS][test] Reverting driver-test.c to what it was prior to e782192d5e6
Puyan Lotfi [Thu, 14 Nov 2019 08:39:16 +0000 (03:39 -0500)]
[clang][IFS][test] Reverting driver-test.c to what it was prior to e782192d5e6

My mistake. Changes I had in this test were for code changes that are
not landed yet. I am reverting driver-test.c back to what it was
originally.

4 years ago[clang][IFS][test] Removing -shared, fixing triple: driver-test.c/windows.cpp.
Puyan Lotfi [Thu, 14 Nov 2019 07:41:28 +0000 (02:41 -0500)]
[clang][IFS][test] Removing -shared, fixing triple: driver-test.c/windows.cpp.

Removing -shared as it is not used on a lot of targets in order to green failing
bots with this change. Also, tiding up the windows.cpp test as the
triple compile out can look slightly different that what you specified
on a windows bot.

4 years ago[clang][IFS][test][NFC] Tightening up clang-ifs tests to use -cc1 more often.
Puyan Lotfi [Thu, 14 Nov 2019 04:58:09 +0000 (23:58 -0500)]
[clang][IFS][test][NFC] Tightening up clang-ifs tests to use -cc1 more often.

Unless the test is explicitly testing a driver feature if clang
interface stubs I have changed the tests to use %clang_cc1. This should
make some changes I plan to make to the driver job pipeline cause fewer
test changes and breakages.

4 years agoCheck result of emitStrLen before passing it to CreateGEP
Dimitry Andric [Wed, 13 Nov 2019 21:15:30 +0000 (22:15 +0100)]
Check result of emitStrLen before passing it to CreateGEP

Summary:
This fixes PR43081, where the transformation of `strchr(p, 0) -> p +
strlen(p)` can cause a segfault, if `-fno-builtin-strlen` is used.  In
that case, `emitStrLen` returns nullptr, which CreateGEP is not designed
to handle.  Also add the minimized code from the PR as a test case.

Reviewers: xbolva00, spatel, jdoerfert, efriedma

Reviewed By: efriedma

Subscribers: lebedev.ri, hiraditya, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D70143

4 years ago[mips] Use local variable to escape repetitive calls of `getOpcode`. NFC
Simon Atanasyan [Thu, 14 Nov 2019 06:05:24 +0000 (09:05 +0300)]
[mips] Use local variable to escape repetitive calls of `getOpcode`. NFC

4 years ago[mips] Use isInt<> call instead of explicit range checking. NFC
Simon Atanasyan [Thu, 14 Nov 2019 05:45:37 +0000 (08:45 +0300)]
[mips] Use isInt<> call instead of explicit range checking. NFC

4 years ago[cmake] Prevent building with BUILD_SHARED_LIBS and LLVM_LINK_LLVM_DYLIB
Tom Stellard [Thu, 14 Nov 2019 04:29:19 +0000 (20:29 -0800)]
[cmake] Prevent building with BUILD_SHARED_LIBS and LLVM_LINK_LLVM_DYLIB

Summary: This doesn't work, so error early if someone tries it.

Reviewers: beanz, smeenai, compnerd, phosek

Reviewed By: beanz

Subscribers: mgorny, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70160

4 years ago[ModuleUtils] Clean up header file. [NFC]
Francesco Petrogalli [Thu, 14 Nov 2019 04:29:07 +0000 (04:29 +0000)]
[ModuleUtils] Clean up header file. [NFC]

Summary:
1. Remove unused forward declaration.
2. Remove unnecessary `\defgroup` in docygen documentation.

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70218

4 years agoWarn on /align if used without /driver
Rui Ueyama [Wed, 13 Nov 2019 04:43:25 +0000 (13:43 +0900)]
Warn on /align if used without /driver

/align is not supposed to be used without /driver, so it makes sense
to warn if only /align is passed. MSVC link.exe warns on this too.

Differential Revision: https://reviews.llvm.org/D70163

4 years agoImplement /driver, /driver:wdm and /driver:uponly
Rui Ueyama [Wed, 13 Nov 2019 04:53:15 +0000 (13:53 +0900)]
Implement /driver, /driver:wdm and /driver:uponly

This patch implements /driver, /driver:wdm and /driver:uponly as
described in
https://docs.microsoft.com/en-us/cpp/build/reference/driver-windows-nt-kernel-mode-driver?view=vs-2019.

Differential Revision: https://reviews.llvm.org/D70162

4 years ago[WebAssembly] -fwasm-exceptions enables reference-types
Heejin Ahn [Tue, 5 Nov 2019 00:04:15 +0000 (16:04 -0800)]
[WebAssembly] -fwasm-exceptions enables reference-types

Summary:
This adds `-mreference-types` and `-mno-reference-types` flags to clang
and make `-fwasm-exceptions` enables reference types feature in clang
and the backend.

Reviewers: tlively

Subscribers: dschuff, sbc100, jgravelle-google, sunfish, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D69832

4 years ago[unittests] Add InitializePasses.h includes
Heejin Ahn [Thu, 14 Nov 2019 02:19:54 +0000 (18:19 -0800)]
[unittests] Add InitializePasses.h includes

Summary:
After D70211, Pass.h does not include InitializePasses.h anymore, so
these files need to include InitializePasses.h directly.

Reviewers: rnk

Subscribers: MatzeB, mehdi_amini, zzheng, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70217

4 years ago[LLDB] Don't install the pretty stack trace handler twice.
Jonas Devlieghere [Thu, 14 Nov 2019 01:23:21 +0000 (17:23 -0800)]
[LLDB] Don't install the pretty stack trace handler twice.

I noticed that currently we are printing LLVM's pretty stack trace
twice. The reason is that we're calling PrintStackTraceOnErrorSignal in
addition to InitLLVM, which besides some other useful things, also
register LLVM's pretty stack trace handler.

Differential revision: https://reviews.llvm.org/D70216

4 years agoMove CodeGenFileType enum to Support/CodeGen.h
Reid Kleckner [Wed, 13 Nov 2019 23:17:46 +0000 (15:17 -0800)]
Move CodeGenFileType enum to Support/CodeGen.h

Avoids the need to include TargetMachine.h from various places just for
an enum. Various other enums live here, such as the optimization level,
TLS model, etc. Data suggests that this change probably doesn't matter,
but it seems nice to have anyway.

4 years ago[AMDGPU] Fixed dpp test. NFC.
Stanislav Mekhanoshin [Thu, 14 Nov 2019 00:36:01 +0000 (16:36 -0800)]
[AMDGPU] Fixed dpp test. NFC.

4 years ago[cmake] Remove confusing condition argument from else() NFC
Tom Stellard [Thu, 14 Nov 2019 00:01:33 +0000 (16:01 -0800)]
[cmake] Remove confusing condition argument from else() NFC

Summary:
This makes it look like an elseif and also the variable referenced
in the condition was removed from this function in r366622.

Reviewers: dsanders, beanz, smeenai, compnerd, phosek

Reviewed By: beanz

Subscribers: mgorny, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70159

4 years agoRevert "Forward declare Optional<T> in STLExtras.h"
Reid Kleckner [Thu, 14 Nov 2019 00:36:21 +0000 (16:36 -0800)]
Revert "Forward declare Optional<T> in STLExtras.h"

This reverts commit a36f316390d4bc1bcb0e9de0f55831385ab24099.

I did not intend to push this with the InitializePasses.h change.

4 years agoSink all InitializePasses.h includes
Reid Kleckner [Wed, 13 Nov 2019 21:15:01 +0000 (13:15 -0800)]
Sink all InitializePasses.h includes

This file lists every pass in LLVM, and is included by Pass.h, which is
very popular. Every time we add, remove, or rename a pass in LLVM, it
caused lots of recompilation.

I found this fact by looking at this table, which is sorted by the
number of times a file was changed over the last 100,000 git commits
multiplied by the number of object files that depend on it in the
current checkout:
  recompiles    touches affected_files  header
  342380        95      3604    llvm/include/llvm/ADT/STLExtras.h
  314730        234     1345    llvm/include/llvm/InitializePasses.h
  307036        118     2602    llvm/include/llvm/ADT/APInt.h
  213049        59      3611    llvm/include/llvm/Support/MathExtras.h
  170422        47      3626    llvm/include/llvm/Support/Compiler.h
  162225        45      3605    llvm/include/llvm/ADT/Optional.h
  158319        63      2513    llvm/include/llvm/ADT/Triple.h
  140322        39      3598    llvm/include/llvm/ADT/StringRef.h
  137647        59      2333    llvm/include/llvm/Support/Error.h
  131619        73      1803    llvm/include/llvm/Support/FileSystem.h

Before this change, touching InitializePasses.h would cause 1345 files
to recompile. After this change, touching it only causes 550 compiles in
an incremental rebuild.

Reviewers: bkramer, asbirlea, bollu, jdoerfert

Differential Revision: https://reviews.llvm.org/D70211

4 years agoForward declare Optional<T> in STLExtras.h
Reid Kleckner [Thu, 14 Nov 2019 00:28:12 +0000 (16:28 -0800)]
Forward declare Optional<T> in STLExtras.h

WIP stats

4 years ago[AMDGPU] Fixed mfma-loop test. NFC.
Stanislav Mekhanoshin [Wed, 13 Nov 2019 23:58:49 +0000 (15:58 -0800)]
[AMDGPU] Fixed mfma-loop test. NFC.

4 years ago[LLDB] Cleanup the DataEncoder utility. (NFC)
Jonas Devlieghere [Wed, 13 Nov 2019 23:38:43 +0000 (15:38 -0800)]
[LLDB] Cleanup the DataEncoder utility. (NFC)

This commit removes unused methods from the DataEncoder class and cleans
up the API by making all the internal methods private.

4 years agoRevert a hunk from 9634064cfa1b9bf7b7
Reid Kleckner [Wed, 13 Nov 2019 23:40:22 +0000 (15:40 -0800)]
Revert a hunk from 9634064cfa1b9bf7b7

This causes errors when building LLDB because the Windows implementation
doesn't implement this method:

C:\src\llvm-project\lldb\source\Plugins\ScriptInterpreter\Python\ScriptInterpreterPython.cpp(915,19): error: allocating an object of abstract class type 'lldb_private::ConnectionGenericFile'
              new ConnectionGenericFile(read_file, true));
                  ^
C:\src\llvm-project\lldb\include\lldb/Utility/Connection.h(174,28): note: unimplemented pure virtual method 'GetReadObject' in 'ConnectionGenericFile'
  virtual lldb::IOObjectSP GetReadObject() = 0;
                           ^

4 years ago[LLDB] Implement pure virtual method in MockConnection
Jonas Devlieghere [Wed, 13 Nov 2019 23:37:24 +0000 (15:37 -0800)]
[LLDB] Implement pure virtual method in MockConnection

I made GetReadObject pure virtual in the base class and forgot to add
the method to the mock class.

4 years agoSink MachineFunction private method out of line
Reid Kleckner [Wed, 13 Nov 2019 23:33:12 +0000 (15:33 -0800)]
Sink MachineFunction private method out of line

This method is private and only called from this file and doesn't need
to be inline. Saves a TargetMachine.h include in MachineFunction.h, a
popular header. The include was introduced in 98603a8153086 despite the
forward decl of LLVMTargetMachine.

4 years ago[X86] Don't treat mxcsr as a register name when parsing MS inline assembly.
Craig Topper [Wed, 13 Nov 2019 23:20:34 +0000 (15:20 -0800)]
[X86] Don't treat mxcsr as a register name when parsing MS inline assembly.

No instruction takes mxcsr as a an operand so we should always
treat it as an identifier name.

4 years ago[LLDB] Fix another set of -Wdocumentation warnings
Jonas Devlieghere [Wed, 13 Nov 2019 23:11:45 +0000 (15:11 -0800)]
[LLDB] Fix another set of -Wdocumentation warnings

At this point I'm just fixing issues as I see them pop up locally in
incremental builds.

4 years ago[LLDB] Remove dead code from StreamFile
Jonas Devlieghere [Wed, 13 Nov 2019 23:08:51 +0000 (15:08 -0800)]
[LLDB] Remove dead code from StreamFile

4 years ago[RegisterContext] Remove now unneded vestiges.
Davide Italiano [Wed, 13 Nov 2019 22:52:52 +0000 (14:52 -0800)]
[RegisterContext] Remove now unneded vestiges.

4 years ago[LLDB] Fix a bunch of -Wdocumentation warnings in ExpressionParser
Jonas Devlieghere [Wed, 13 Nov 2019 22:40:28 +0000 (14:40 -0800)]
[LLDB] Fix a bunch of -Wdocumentation warnings in ExpressionParser

4 years agoRemove redundant check. (NFC)
Adrian Prantl [Wed, 13 Nov 2019 22:19:01 +0000 (14:19 -0800)]
Remove redundant check. (NFC)

4 years agoUse cheaper, equivalent predicate. (NFC)
Adrian Prantl [Wed, 13 Nov 2019 22:16:11 +0000 (14:16 -0800)]
Use cheaper, equivalent predicate. (NFC)

4 years ago[X86] Don't set the operation action for i16 SINT_TO_FP to Promote just because SSE1...
Craig Topper [Wed, 13 Nov 2019 21:20:00 +0000 (13:20 -0800)]
[X86] Don't set the operation action for i16 SINT_TO_FP to Promote just because SSE1 is enabled.

Instead do custom promotion in the handler so that we can still
allow i16 to be used with fp80. And f64 without sse2.

4 years ago[X86] Fix typo in comment. NFC
Craig Topper [Wed, 13 Nov 2019 20:43:33 +0000 (12:43 -0800)]
[X86] Fix typo in comment. NFC

4 years ago[X86] Move all the FP_TO_XINT/XINT_TO_FP setOperationActions into the same !useSoftFl...
Craig Topper [Wed, 13 Nov 2019 20:37:02 +0000 (12:37 -0800)]
[X86] Move all the FP_TO_XINT/XINT_TO_FP setOperationActions into the same !useSoftFloat block. Qualify all of the Promote actions for these with !useSoftFloat too. NFCI

The Promote action doesn't apply until LegalizeDAG. By the time
we get there, we would have already softened all the FP operations
if useSoftFloat was true. So there wouldn't be any operation left
to Promote.

4 years agoRename clang-module-related *DWO* functions to *ClangModule* (NFC)
Adrian Prantl [Wed, 13 Nov 2019 22:05:10 +0000 (14:05 -0800)]
Rename clang-module-related *DWO* functions to *ClangModule* (NFC)

This avoids confusing them with fission-related functionality.

I also moved two accessor functions from DWARFDIE into static
functions in DWARFASTParserClang were their only use is located.

4 years ago[PGO][PGSO] Temporarily disable the large working set size behavior.
Hiroshi Yamauchi [Wed, 13 Nov 2019 21:49:33 +0000 (13:49 -0800)]
[PGO][PGSO] Temporarily disable the large working set size behavior.

Summary:
This temporarily disables the large working set size behavior in profile guided
size optimization due to internal benchmark regressions.

Reviewers: davidxl

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70207

4 years agolibc++ status page: Fix HTML.
Marek Kurdej [Wed, 13 Nov 2019 21:51:29 +0000 (22:51 +0100)]
libc++ status page: Fix HTML.

4 years ago[SimplifyCFG] add test for select with FMF; NFC
Sanjay Patel [Wed, 13 Nov 2019 21:45:17 +0000 (16:45 -0500)]
[SimplifyCFG] add test for select with FMF; NFC

4 years agoRename ParseTypeFromDWO to ParseTypeFromClangModule (NFC)
Adrian Prantl [Wed, 13 Nov 2019 18:52:30 +0000 (10:52 -0800)]
Rename ParseTypeFromDWO to ParseTypeFromClangModule (NFC)

Because that is what this function really does. The old name is
misleading.

4 years agoRevert "[RISCV] Use compiler-rt if no GCC installation detected"
Richard Smith [Wed, 13 Nov 2019 21:14:07 +0000 (13:14 -0800)]
Revert "[RISCV] Use compiler-rt if no GCC installation detected"

This change causes test failures for builds configured with
-DCLANG_DEFAULT_RTLIB=compiler-rt.

This reverts commit 3289352e6bb9d2949c678c625478024bf2a5fbfb.

4 years ago[SLP] fix miscompile on min/max reductions with extra uses (PR43948)
Sanjay Patel [Wed, 13 Nov 2019 20:41:26 +0000 (15:41 -0500)]
[SLP] fix miscompile on min/max reductions with extra uses (PR43948)

The bug manifests as replacing a reduction operand with an undef
value.

The problem appears to be limited to cases where a min/max reduction
has extra uses of the compare operand to the select.

In the general case, we are tracking "ExternallyUsedValues" and
an "IgnoreList" of the reduction operations, but those may not apply
to the final compare+select in a min/max reduction.

For that, we use replaceAllUsesWith (RAUW) to ensure that the new
vectorized reduction values are transferred to all subsequent users.

Differential Revision: https://reviews.llvm.org/D70148

4 years ago[clang-format] refactor the use of the SMDiagnostics in replacement warnings
mydeveloperday [Wed, 13 Nov 2019 20:54:42 +0000 (20:54 +0000)]
[clang-format] refactor the use of the SMDiagnostics in replacement warnings

Summary:
Review comments in {D69854} recommended a simpler approach of creating the SMDiagnostics to remove much of the complexity. (thanks @thakis)

@vlad.tsyrklevich I've rebuilt on both Windows and Linux (running Linux with Address and Undefined sanitizers) over the clang code base

Reviewers: thakis, klimek, mitchell-stellar, vlad.tsyrklevich

Reviewed By: thakis

Subscribers: cfe-commits, thakis, vlad.tsyrklevich

Tags: #clang-format, #clang

Differential Revision: https://reviews.llvm.org/D69921

4 years ago[LLD] [COFF] Fix automatically importing data symbols from DLLs with LTO
Martin Storsjö [Wed, 13 Nov 2019 07:40:30 +0000 (09:40 +0200)]
[LLD] [COFF] Fix automatically importing data symbols from DLLs with LTO

This broke in 51dcb292cc002, "[lld-link] diagnose undefined symbols
before LTO when possible" (very soon after the 9.0 branch, so
luckily the 9.0 release is unaffected).

The code for loading objects we believe might be needed for autoimport
(loadMinGWAutomaticImports()) does run before the new
reportUnresolvable() function, but it had a condition to only operate
on symbols from regular object files. This condition came from
resolveRemainingUndefines(), but as loadMinGWAutomaticImports() now
has to operate before the LTO, it has to operate on undefineds from
LTO objects as well.

Differential Revision: https://reviews.llvm.org/D70166

4 years agoAdd -disable-builtin option to opt
Dimitry Andric [Wed, 13 Nov 2019 19:15:22 +0000 (20:15 +0100)]
Add -disable-builtin option to opt

Summary:
The option allows to disable specific target library builtin functions,
instead of -disable-simplify-libcalls, which disables all of them.

This is a prerequisite for D70143, which fixes PR43081.

Reviewers: xbolva00, spatel, jdoerfert, efriedma

Reviewed By: efriedma

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70193

4 years ago[LLDB] Fix a bunch of -Wdocumentation warnings
Jonas Devlieghere [Wed, 13 Nov 2019 19:35:53 +0000 (11:35 -0800)]
[LLDB] Fix a bunch of -Wdocumentation warnings

4 years ago[dsymutil] Add -dump to llvm-bcanalyzer invocations
Francis Visoiu Mistrih [Wed, 13 Nov 2019 20:27:03 +0000 (12:27 -0800)]
[dsymutil] Add -dump to llvm-bcanalyzer invocations

4 years ago[TargetLowering] Increase the storage size of NumRegistersForVT to allow the type...
Craig Topper [Wed, 13 Nov 2019 20:09:34 +0000 (12:09 -0800)]
[TargetLowering] Increase the storage size of NumRegistersForVT to allow the type break down for v256i1 and other types to be stored correctly

v256i1 on X86 without avx512 breaks down to 256 i8 values when passed between basic blocks. But the NumRegistersForVT was sized at a byte for each VT. This results in 256 being stored as 0.

This patch enlarges the type to 16 bits and adds an assert to ensure that no information is lost when the entry is stored.

Differential Revision: https://reviews.llvm.org/D70138

4 years ago[mips] Reduce number of nested `if` statements. NFC
Simon Atanasyan [Tue, 12 Nov 2019 12:35:16 +0000 (15:35 +0300)]
[mips] Reduce number of nested `if` statements. NFC

4 years ago[mips] Add test to check ELF output for JAL XGOT expansion. NFC
Simon Atanasyan [Sun, 10 Nov 2019 11:53:36 +0000 (14:53 +0300)]
[mips] Add test to check ELF output for JAL XGOT expansion. NFC

4 years ago[mips] Add tests to check `jal sym+offset`. NFC
Simon Atanasyan [Sat, 9 Nov 2019 07:35:21 +0000 (10:35 +0300)]
[mips] Add tests to check `jal sym+offset`. NFC

4 years ago[LiveInterval] Allow updating subranges with slightly out-dated IR
Quentin Colombet [Wed, 13 Nov 2019 00:32:12 +0000 (16:32 -0800)]
[LiveInterval] Allow updating subranges with slightly out-dated IR

During register coalescing, we update the live-intervals on-the-fly.
To do that we are in this strange mode where the live-intervals can
be slightly out-of-sync (more precisely they are forward looking)
compared to what the IR actually represents.
This happens because the register coalescer only updates the IR when
it is done with updating the live-intervals and it has to do it this
way because updating the IR on-the-fly would actually clobber some
information on how the live-ranges that are being updated look like.

This is problematic for updates that rely on the IR to accurately
represents the state of the live-ranges. Right now, we have only
one of those: stripValuesNotDefiningMask.
To reconcile this need of out-of-sync IR, this patch introduces a
new argument to LiveInterval::refineSubRanges that allows the code
doing the live range updates to reason about how the code should
look like after the coalescer will have rewritten the registers.
Essentially this captures how a subregister index with be offseted
to match its position in a new register class.

E.g., let say we want to merge:
    V1.sub1:<2 x s32> = COPY V2.sub3:<4 x s32>

We do that by choosing a class where sub1:<2 x s32> and sub3:<4 x s32>
overlap, i.e., by choosing a class where we can find "offset + 1 == 3".
Put differently we align V2's sub3 with V1's sub1:
    V2: sub0 sub1 sub2 sub3
    V1: <offset>  sub0 sub1

This offset will look like a composed subregidx in the the class:
     V1.(composed sub2 with sub1):<4 x s32> = COPY V2.sub3:<4 x s32>
 =>  V1.(composed sub2 with sub1):<4 x s32> = COPY V2.sub3:<4 x s32>

Now if we didn't rewrite the uses and def of V1, all the checks for V1
need to account for this offset to match what the live intervals intend
to capture.

Prior to this patch, we would fail to recognize the uses and def of V1
and would end up with machine verifier errors: No live segment at def.
This could lead to miscompile as we would drop some live-ranges and
thus, miss some interferences.

For this problem to trigger, we need to reach stripValuesNotDefiningMask
while having a mismatch between the IR and the live-ranges (i.e.,
we have to apply a subreg offset to the IR.)

This requires the following three conditions:
1. An update of overlapping subreg lanes: e.g., dsub0 == <ssub0, ssub1>
2. An update with Tuple registers with a possibility to coalesce the
   subreg index: e.g., v1.dsub_1 == v2.dsub_3
3. Subreg liveness enabled.

looking at the IR to decide what is alive and what is not, i.e., calling
stripValuesNotDefiningMask.
coalescer maintains for the live-ranges information.

None of the targets that currently use subreg liveness (i.e., the targets
that fulfill #3, Hexagon, AMDGPU, PowerPC, and SystemZ IIRC) expose #1 and
and #2, so this patch also artificial enables subreg liveness for ARM,
so that a nice test case can be attached.

4 years ago[TTI] Fix cast cost on vector types.
Michael Liao [Wed, 13 Nov 2019 17:19:29 +0000 (12:19 -0500)]
[TTI] Fix cast cost on vector types.

- Only split vector types when both src and dst types are splittable.

4 years ago[llvm-bcanalyzer] Don't dump the contents if -dump is not passed
Francis Visoiu Mistrih [Wed, 13 Nov 2019 18:37:14 +0000 (10:37 -0800)]
[llvm-bcanalyzer] Don't dump the contents if -dump is not passed

With all the previous refactorings this slipped through and now we
always dump the contents of the bitcode files, even if -dump is not
passed.

4 years ago[AArch64][v8.3a] Add missing imp-defs on RETA*.
Ahmed Bougacha [Wed, 26 Jun 2019 01:44:39 +0000 (18:44 -0700)]
[AArch64][v8.3a] Add missing imp-defs on RETA*.

RETA always implicitly uses LR, unlike RET which merely has an
alias that defaults it to LR.
Additionally, RETA implicitly uses SP as well, which it uses as
a discriminator to authenticate LR.

This isn't usually noticeable, because RET_ReallyLR is used in most
of the backend.  However, the post-RA scheduler, if enabled, will
cause miscompiles if the imp-uses are missing.

While there, fix a typo in the lone affected testcase.

4 years ago[AArch64][v8.3a] Add LDRA '[xN]!' alias.
Ahmed Bougacha [Wed, 26 Jun 2019 01:38:20 +0000 (18:38 -0700)]
[AArch64][v8.3a] Add LDRA '[xN]!' alias.

The instruction definition has been retroactively expanded to
allow for an alias for '[xN, 0]!' as '[xN]!'.
That wouldn't make sense on LDR, but does for LDRA.

4 years ago[SLP] improve test readability; NFC
Sanjay Patel [Wed, 13 Nov 2019 17:58:43 +0000 (12:58 -0500)]
[SLP] improve test readability; NFC

4 years ago[BPF] fix clang test failure for bpf-attr-preserve-access-index-4.c
Yonghong Song [Wed, 13 Nov 2019 17:40:57 +0000 (09:40 -0800)]
[BPF] fix clang test failure for bpf-attr-preserve-access-index-4.c

Depending on different cmake configures, clang may generate different
IR name for slot variables. Let us use the regex instead of hard
coding the name. I did the same for other bpf-attr-preserve-access-index
tests with such an approach, but somehow did not do for this one.

4 years ago[RISCV] Use compiler-rt if no GCC installation detected
Edward Jones [Wed, 13 Nov 2019 17:00:39 +0000 (17:00 +0000)]
[RISCV] Use compiler-rt if no GCC installation detected

If a GCC installation is not detected, then this attempts to
use compiler-rt and the compiler-rt crtbegin/crtend
implementations as a fallback.

Differential Revision: https://reviews.llvm.org/D68407

4 years agoFix typo in DwarfDebug [NFC]
David Stenberg [Wed, 13 Nov 2019 17:06:16 +0000 (18:06 +0100)]
Fix typo in DwarfDebug [NFC]

4 years agoDon't set LLVM_NO_DEAD_STRIP on AIX
David Tenty [Wed, 13 Nov 2019 16:59:10 +0000 (11:59 -0500)]
Don't set LLVM_NO_DEAD_STRIP on AIX

Summary:
when building plugins, as AIX has symbols in it's standard library that
must be garbage collected or we will see link errors. Export lists will
handle this instead on AIX.

Reviewers: stevewan, sfertile, jasonliu, xingxue, DiggerLin

Reviewed By: DiggerLin

Subscribers: mgorny, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70130

4 years ago[BPF] add missing attribute in pragma-attribute-supported-attributes-list.test
Yonghong Song [Wed, 13 Nov 2019 16:50:42 +0000 (08:50 -0800)]
[BPF] add missing attribute in pragma-attribute-supported-attributes-list.test

Add the newly supported BPF specific __attribute__((preserve_access_index)
in the pragma-attribute-supported-attributes-list.test.

4 years ago[SLP] reduce code duplication for min/max vs. other reductions; NFCI
Sanjay Patel [Wed, 13 Nov 2019 16:25:22 +0000 (11:25 -0500)]
[SLP] reduce code duplication for min/max vs. other reductions; NFCI

4 years ago[BPF] Add preserve_access_index attribute for record definition
Yonghong Song [Sat, 2 Nov 2019 05:16:59 +0000 (22:16 -0700)]
[BPF] Add preserve_access_index attribute for record definition

This is a resubmission for the previous reverted commit
943436040121 with the same subject. This commit fixed the
segfault issue and addressed additional review comments.

This patch introduced a new bpf specific attribute which can
be added to struct or union definition. For example,
  struct s { ... } __attribute__((preserve_access_index));
  union u { ... } __attribute__((preserve_access_index));
The goal is to simplify user codes for cases
where preserve access index happens for certain struct/union,
so user does not need to use clang __builtin_preserve_access_index
for every members.

The attribute has no effect if -g is not specified.

When the attribute is specified and -g is specified, any member
access defined by that structure or union, including array subscript
access and inner records, will be preserved through
  __builtin_preserve_{array,struct,union}_access_index()
IR intrinsics, which will enable relocation generation
in bpf backend.

The following is an example to illustrate the usage:
  -bash-4.4$ cat t.c
  #define __reloc__ __attribute__((preserve_access_index))
  struct s1 {
    int c;
  } __reloc__;

  struct s2 {
    union {
      struct s1 b[3];
    };
  } __reloc__;

  struct s3 {
    struct s2 a;
  } __reloc__;

  int test(struct s3 *arg) {
    return arg->a.b[2].c;
  }
  -bash-4.4$ clang -target bpf -g -S -O2 t.c

A relocation with access string "0:0:0:0:2:0" will be generated
representing access offset of arg->a.b[2].c.

forward declaration with attribute is also handled properly such
that the attribute is copied and populated in real record definition.

Differential Revision: https://reviews.llvm.org/D69759

4 years agoFix comment spelling {addresing -> addressing} (NFC)
Matthew Malcomson [Wed, 13 Nov 2019 15:50:44 +0000 (15:50 +0000)]
Fix comment spelling {addresing -> addressing} (NFC)

4 years ago[profile] Factor out logic for mmap'ing merged profile, NFC
Vedant Kumar [Tue, 12 Nov 2019 18:23:12 +0000 (10:23 -0800)]
[profile] Factor out logic for mmap'ing merged profile, NFC

Split out the logic to get the size of a merged profile and to do a
compatibility check. This can be shared with both the continuous+merging
mode implementation, as well as the runtime-allocated counters
implementation planned for Fuchsia.

Lifted out of D69586.

Differential Revision: https://reviews.llvm.org/D70135

4 years ago[InstCombine] propagate fast-math-flags (FMF) to select when inverting fcmp+select
Sanjay Patel [Wed, 13 Nov 2019 15:35:34 +0000 (10:35 -0500)]
[InstCombine] propagate fast-math-flags (FMF) to select when inverting fcmp+select

As noted by the FIXME comment, this is not correct based on our current FMF semantics.
We should be propagating FMF from the final value in a sequence (in this case the
'select'). So the behavior even without this patch is wrong, but we did not allow FMF
on 'select' until recently.

But if we do the correct thing right now in this patch, we'll inevitably introduce
regressions because we have not wired up FMF propagation for 'phi' and 'select' in
other passes (like SimplifyCFG) or other places in InstCombine. I'm not seeing a
better incremental way to make progress.

That said, the potential extra damage over the existing wrong behavior from this
patch is very limited. AFAIK, the only way to have different FMF on IR in the same
function is if we have LTO inlined IR from 2 modules that were compiled using
different fast-math settings.

As seen in the tests, we may actually see some improvements with this patch because
adding the FMF to the 'select' allows matching to min/max intrinsics that were
previously missed (in the common case, the 'fcmp' and 'select' should have identical
FMF to begin with).

Next steps in the transition:

    Make similar changes in instcombine as needed.
    Enable phi-to-select FMF propagation in SimplifyCFG.
    Remove dependencies on fcmp with FMF.
    Deprecate FMF on fcmp.

Differential Revision: https://reviews.llvm.org/D69720

4 years agoDWARFDebugLoclists: Add an api to get the location lists of a DWARF unit
Pavel Labath [Fri, 8 Nov 2019 11:18:07 +0000 (12:18 +0100)]
DWARFDebugLoclists: Add an api to get the location lists of a DWARF unit

Summary:
This avoid the need to duplicate the location lists searching logic in
various users. The "inline location list dumping" code (which is the
only user actually updated to handle DWARF v5 location lists)  is
switched to this method. After adding v4 location list support, I'll
switch other users too.

Reviewers: dblaikie, probinson, JDevlieghere, aprantl, SouraVX

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D70084

4 years agoRemove commented out CHECK-NEXT to try and appease llvm-clang-x86_64-expensive-checks...
Simon Pilgrim [Wed, 13 Nov 2019 14:59:12 +0000 (14:59 +0000)]
Remove commented out CHECK-NEXT to try and appease llvm-clang-x86_64-expensive-checks-win buildbot

4 years agoPowerPC - fix uninitialized variable warnings. NFCI.
Simon Pilgrim [Wed, 13 Nov 2019 14:39:54 +0000 (14:39 +0000)]
PowerPC - fix uninitialized variable warnings. NFCI.

4 years agoFix uninitialized variable warning. NFCI.
Simon Pilgrim [Wed, 13 Nov 2019 14:39:19 +0000 (14:39 +0000)]
Fix uninitialized variable warning. NFCI.

4 years agoFix uninitialized variable warning. NFCI.
Simon Pilgrim [Wed, 13 Nov 2019 14:38:58 +0000 (14:38 +0000)]
Fix uninitialized variable warning. NFCI.

4 years agoFix uninitialized variable warning. NFCI.
Simon Pilgrim [Wed, 13 Nov 2019 14:38:36 +0000 (14:38 +0000)]
Fix uninitialized variable warning. NFCI.

4 years agoSparc - fix uninitialized variable warnings. NFCI.
Simon Pilgrim [Wed, 13 Nov 2019 14:15:17 +0000 (14:15 +0000)]
Sparc - fix uninitialized variable warnings. NFCI.

4 years agoPPCReduceCRLogicals - fix static analyzer warnings. NFC
Simon Pilgrim [Wed, 13 Nov 2019 13:46:43 +0000 (13:46 +0000)]
PPCReduceCRLogicals - fix static analyzer warnings. NFC
- Fix uninitialized variable warnings.
- Fix null dereference warnings.

4 years agoSLPVectorizer - make comparison operators + isInSchedulingRegion const
Simon Pilgrim [Wed, 13 Nov 2019 13:45:15 +0000 (13:45 +0000)]
SLPVectorizer - make comparison operators + isInSchedulingRegion const

Fixes cppcheck warnings.

4 years ago[clang][Tooling] Filter flags that generate output in SyntaxOnlyAdjuster
Kadir Cetinkaya [Wed, 13 Nov 2019 12:44:40 +0000 (13:44 +0100)]
[clang][Tooling] Filter flags that generate output in SyntaxOnlyAdjuster

Summary:
Flags that generate output could result in failures when creating
syntax only actions. This patch introduces initial logic for filtering out
those. The first such flag is "save-temps", which saves intermediate
files(bitcode, assembly, etc.) into a specified directory.

Fixes https://github.com/clangd/clangd/issues/191

Reviewers: hokein

Subscribers: ilya-biryukov, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D70173

4 years ago[clangd] Add bool return type to Index::refs API.
Haojian Wu [Wed, 13 Nov 2019 13:42:26 +0000 (14:42 +0100)]
[clangd] Add bool return type to Index::refs API.

Summary:
Similar to fuzzyFind, the bool indicates whether there are more xref
results.

Reviewers: ilya-biryukov

Reviewed By: ilya-biryukov

Subscribers: merge_guards_bot, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D70139

4 years ago[InstCombine] Avoid moving ops that do restrict undef across shuffles.
Florian Hahn [Wed, 13 Nov 2019 13:26:13 +0000 (13:26 +0000)]
[InstCombine] Avoid moving ops that do restrict undef across shuffles.

I think we have to be a bit more careful when it comes to moving
ops across shuffles, if the op does restrict undef. For example, without
this patch, we would move 'and %v, <0, 0, -1, -1>' over a
'shufflevector %a, undef, <undef, undef, 1, 2>'. As a result, the first
2 lanes of the result are undef after the combine, but they really
should be 0, unless I am missing something.

For ops that do fold to undef on undef operands, the current behavior
should be fine. I've add conservative check OpDoesRestrictUndef, maybe
there's a better existing utility?

Reviewers: spatel, RKSimon, lebedev.ri

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D70093

4 years agoRevert "[RISCV] Fix wrong CFI directives"
Luís Marques [Wed, 13 Nov 2019 13:28:33 +0000 (13:28 +0000)]
Revert "[RISCV] Fix wrong CFI directives"

test/DebugInfo/RISCV/relax-debug-frame.ll wasn't properly updated.

4 years ago[InstCombine] Precommit shuffle tests for D70093.
Florian Hahn [Mon, 11 Nov 2019 17:42:13 +0000 (17:42 +0000)]
[InstCombine] Precommit shuffle tests for D70093.

4 years ago[ARM][MVE] canTailPredicateLoop
Sjoerd Meijer [Wed, 13 Nov 2019 13:02:16 +0000 (13:02 +0000)]
[ARM][MVE] canTailPredicateLoop

This implements TTI hook 'preferPredicateOverEpilogue' for MVE.  This is a
first version and it operates on single block loops only. With this change, the
vectoriser will now determine if tail-folding scalar remainder loops is
possible/desired, which is the first step to generate MVE tail-predicated
vector loops.

This is disabled by default for now. I.e,, this is depends on option
-disable-mve-tail-predication, which is off by default.

I will follow up on this soon with a patch for the vectoriser to respect loop
hint 'vectorize.predicate.enable'. I.e., with this loop hint set to Disabled,
we don't want to tail-fold and we shouldn't query this TTI hook, which is
done in D70125.

Differential Revision: https://reviews.llvm.org/D69845

4 years ago[RISCV] Fix wrong CFI directives
Luís Marques [Wed, 13 Nov 2019 12:58:00 +0000 (12:58 +0000)]
[RISCV] Fix wrong CFI directives

Summary: Removes CFI CFA directives that could incorrectly propagate
beyond the basic block they were inteded for. Specifically it removes
the epilogue CFI directives. See the branch_and_tail_call test for an
example of the issue. Should fix the stack unwinding issues caused by
the incorrect directives.

Reviewers: asb, lenary, shiva0217
Reviewed By: lenary
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D69723

4 years ago[ARM,MVE] Add intrinsics for contiguous load/stores.
Simon Tatham [Mon, 11 Nov 2019 17:02:39 +0000 (17:02 +0000)]
[ARM,MVE] Add intrinsics for contiguous load/stores.

This patch adds the ACLE intrinsics for all the MVE load and store
instructions not already handled by D69791. These ones don't need new
IR intrinsics, because they can be implemented in terms of standard
LLVM IR constructions.

Some of the load and store instructions access less than 128 bits of
memory, sign/zero extending each value to a wider vector lane on load
or truncating it on store. These are represented in IR by a load of a
shorter vector followed by a zext/sext, and conversely, a trunc
followed by a short store. Existing ISel patterns already recognize
those combinations and turn them into the right MVE instructions.

The predicated forms of all these instructions are represented in the
same way, except that the ordinary load/store operation is replaced
with the existing intrinsics @llvm.masked.{load,store}. These are
currently only code-generated as predicated MVE load/store
instructions if you give LLVM the `-enable-arm-maskedldst` option; so
I've done that in the LLVM codegen test. When we make that the
default, that option can be removed.

In the Tablegen backend, I've had to add a handful of extra support
features:

* We need to be able to make clang::Address objects out of a
  pointer and an alignment (previously we only needed these when the
  user passed us an existing one).

* We can now specify vector types that aren't 128 bits wide (for use
  in those intermediate values in IR), the parametrized type system
  can make one starting from two existing vector types (using the lane
  count of one and the element type of the other).

* I've added support for code generation of pointer casts, and for
  specifying LLVM types as operands to IRBuilder operations (for zext
  and sext, though I think they'll come in useful again).

* Now not all IR construction operations need to be specified as
  Builder.CreateFoo; some don't involve a Builder at all, and one
  passes it as a parameter to a tiny static helper function in
  CGBuiltin.cpp.

Reviewers: ostannard, MarkMurrayARM, dmgreen

Subscribers: kristof.beyls, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D70088

4 years ago[X86][AVX] Add plausible schedule classes to MASKPAIR/VP2INTERSECT/VDPBF16PS instructions
Simon Pilgrim [Wed, 13 Nov 2019 11:58:19 +0000 (11:58 +0000)]
[X86][AVX] Add plausible schedule classes to MASKPAIR/VP2INTERSECT/VDPBF16PS instructions

These are really just placeholders that use approximately the right resources - once we have CPUs scheduler models that support these instructions they will need revisiting.

In the meantime this means that all instructions have a class of some kind., meaning models can be more easily flagged as complete.

4 years ago[libomptarget] Move supporti.h to support.cu
JonChesterfield [Wed, 13 Nov 2019 11:36:30 +0000 (11:36 +0000)]
[libomptarget] Move supporti.h to support.cu

Summary:
[libomptarget] Move supporti.h to support.cu
Reimplementation of D69652, without the unity build and refactors.
Will need a clean build of libomptarget as the cmakelists changed.

Reviewers: ABataev, jdoerfert

Reviewed By: jdoerfert

Subscribers: mgorny, jfb, openmp-commits

Tags: #openmp

Differential Revision: https://reviews.llvm.org/D70131

4 years agoRevert 57dd4b0 "[ValueTracking] Allow context-sensitive nullness check for non-pointers"
Hans Wennborg [Wed, 13 Nov 2019 10:59:22 +0000 (11:59 +0100)]
Revert 57dd4b0 "[ValueTracking] Allow context-sensitive nullness check for non-pointers"

This caused miscompiles of Chromium (https://crbug.com/1023818). The reduced
repro is small enough to fit here:

  $ cat /tmp/a.c
  unsigned char f(unsigned char *p) {
    unsigned char result = 0;
    for (int shift = 0; shift < 1; ++shift)
      result |= p[0] << (shift * 8);
    return result;
  }
  $ bin/clang -O2 -S -o - /tmp/a.c | grep -A4 f:
  f:                                      # @f
          .cfi_startproc
  # %bb.0:                                # %entry
          xorl    %eax, %eax
          retq

That's nicely optimized, but I don't think it's the right result :-)

> Same as D60846 but with a fix for the problem encountered there which
> was a missing context adjustment in the handling of PHI nodes.
>
> The test that caused D60846 to be reverted was added in e15ab8f277c7.
>
> Reviewers: nikic, nlopes, mkazantsev,spatel, dlrobertson, uabelho, hakzsam
>
> Subscribers: hiraditya, bollu, llvm-commits
>
> Tags: #llvm
>
> Differential Revision: https://reviews.llvm.org/D69571

This reverts commit 57dd4b03e4806bbb4760ab6150940150d884df20.

4 years ago[Mips] Add rematerialization support for ldi.fmt
Mirko Brkusanin [Wed, 13 Nov 2019 10:22:35 +0000 (11:22 +0100)]
[Mips] Add rematerialization support for ldi.fmt

Instruction ldi.fmt can be considered cheap enough to avoid spill and restore
of value that it produces since it's loaded from immediate.

Differential Revision: https://reviews.llvm.org/D69898

4 years ago[mips] Show an error if 64-bit target triple provided with 32-bit CPU
Simon Atanasyan [Wed, 13 Nov 2019 09:08:38 +0000 (12:08 +0300)]
[mips] Show an error if 64-bit target triple provided with 32-bit CPU

When a 64-bit triple is used emit an error if the CPU only supports
32-bit code.

Patch by Miloš Stojanović.

Differential Revision: https://reviews.llvm.org/D70018

4 years ago[mips][test] Add Mips CPU tests. NFC
Simon Atanasyan [Wed, 13 Nov 2019 09:06:01 +0000 (12:06 +0300)]
[mips][test] Add Mips CPU tests. NFC

Adding tests check all available CPUs on Mips.

Patch by Miloš Stojanović.

Differential Revision: https://reviews.llvm.org/D70017

4 years ago[OpenCL] Add remaining vector data builtin functions
Sven van Haastregt [Wed, 13 Nov 2019 10:16:33 +0000 (10:16 +0000)]
[OpenCL] Add remaining vector data builtin functions

Add the remaining half (fp16) vector data load and store builtin
functions from the OpenCL C specification.

Patch by Pierre Gondois and Sven van Haastregt.

4 years agoTemporarily revert "[InstCombine] Fold PHIs with equal incoming pointers"
Daniil Suchkov [Wed, 13 Nov 2019 10:14:11 +0000 (17:14 +0700)]
Temporarily revert "[InstCombine] Fold PHIs with equal incoming pointers"

Revert due to sanitizer-windows buildbot failure.

This reverts commit bbb29738b58aaf6f6518269abdcf8f64131665a9.