platform/upstream/llvm.git
3 years ago[llvm-readobj] Refactor ELFDumper::printAttributes()
Jozef Lawrynowicz [Tue, 17 Aug 2021 20:28:31 +0000 (13:28 -0700)]
[llvm-readobj] Refactor ELFDumper::printAttributes()

The current implementation of printAttributes makes it fiddly to extend
attribute support for new targets.

By refactoring the code so all target specific variables are
initialized in a switch/case statement, it becomes simpler to extend
attribute support for new targets.

Reviewed By: jhenderson, MaskRay

Differential Revision: https://reviews.llvm.org/D107968

3 years ago[lldb] Fix#2 of DW_AT_ranges DW_FORM_sec_offset not using DW_AT_rnglists_base (used...
Jan Kratochvil [Tue, 17 Aug 2021 20:16:43 +0000 (22:16 +0200)]
[lldb] Fix#2 of DW_AT_ranges DW_FORM_sec_offset not using DW_AT_rnglists_base (used by GCC)

Fix D98289 so that it works even for 2nd..nth compilation unit
(.debug_rnglists).

Reviewed By: dblaikie, ikudrin

Differential Revision: https://reviews.llvm.org/D106466

3 years agoRemove unused imports.
Dan Albert [Tue, 17 Aug 2021 20:03:27 +0000 (13:03 -0700)]
Remove unused imports.

3 years agoSimplify setting up LLVM as bazel external repo
Christian Sigg [Sun, 8 Aug 2021 13:19:33 +0000 (15:19 +0200)]
Simplify setting up LLVM as bazel external repo

Only require one intermediate repository instead of two.
Fewer parameters in llvm_config.
Remove bazel_skylib dependency.

Reviewed By: goncharov

Differential Revision: https://reviews.llvm.org/D107714

3 years ago[mlir][linalg] Don't drop existing attributes when creating ops
Lei Zhang [Tue, 17 Aug 2021 19:44:31 +0000 (15:44 -0400)]
[mlir][linalg] Don't drop existing attributes when creating ops

Reviewed By: mravishankar

Differential Revision: https://reviews.llvm.org/D108219

3 years agoSimplify testcase from c411c1b
Weverything [Tue, 17 Aug 2021 19:07:18 +0000 (12:07 -0700)]
Simplify testcase from c411c1b

3 years ago[PassBuilder] Use loop-mssa for licm
Nikita Popov [Mon, 16 Aug 2021 18:58:31 +0000 (20:58 +0200)]
[PassBuilder] Use loop-mssa for licm

Currently specifying -licm or -passes=licm will implicitly create
-passes=loop(licm). This does not match the intended default (used
by the legacy PM and by the default pipeline) of using the
MemorySSA-based LICM implementation. As I plan to drop the non-MSSA
implementation, this will stop working entirely...

This special-cases licm to create a loop-mssa manager instead. At
this point it's still possible to use -passes='loop(licm)' to opt
into the AST-based implementation.

Differential Revision: https://reviews.llvm.org/D108155

3 years ago[llvm-profgen] Clean up code dealing with multiple binaries
wlei [Thu, 12 Aug 2021 01:01:37 +0000 (18:01 -0700)]
[llvm-profgen] Clean up code dealing with multiple binaries

As we decided to support only one binary each time, this patch cleans up the related code dealing with multiple binaries. We can use `llvm-profdata` to merge profile from multiple binaries.

Reviewed By: hoy, wenlei

Differential Revision: https://reviews.llvm.org/D108002

3 years ago[OpenMP] Fix the usage of sscanf on MinGW
Martin Storsjö [Tue, 17 Aug 2021 10:15:49 +0000 (13:15 +0300)]
[OpenMP] Fix the usage of sscanf on MinGW

KMP_SSCANF only evaluates to sscanf_s within
    #if KMP_OS_WINDOWS && KMP_MSVC_COMPAT
so we need to pass the sscanf_s specific parameters within a similar
condition.

Differential Revision: https://reviews.llvm.org/D108196

3 years ago[sanitizers] Fix building on 32 bit Windows after 7256c05ecb7
Martin Storsjö [Mon, 16 Aug 2021 08:50:10 +0000 (11:50 +0300)]
[sanitizers] Fix building on 32 bit Windows after 7256c05ecb7

On 32 bit, 'long' (which is 32 bit on Windows) is used as base
type for SIZE_T and similar.

Differential Revision: https://reviews.llvm.org/D108191

3 years ago[InstCombine] add TODO about another min/max fold; NFC
Sanjay Patel [Tue, 17 Aug 2021 17:46:08 +0000 (13:46 -0400)]
[InstCombine] add TODO about another min/max fold; NFC

Suggested in post-commit for d0975b7cb0e1

3 years ago[InstCombine] add tests for min/max with 'not' ops; NFC
Sanjay Patel [Tue, 17 Aug 2021 15:38:10 +0000 (11:38 -0400)]
[InstCombine] add tests for min/max with 'not' ops; NFC

3 years ago[RISCV] Use RISCV::RVVBitsPerBlock for RGK_ScalableVector in getRegisterBitWidth.
Craig Topper [Tue, 17 Aug 2021 17:52:34 +0000 (10:52 -0700)]
[RISCV] Use RISCV::RVVBitsPerBlock for RGK_ScalableVector in getRegisterBitWidth.

I might be wrong, but I think this is should be width of the known
min size we use for scalable vectors. It shouldn't scale with
minimum vlen.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D107945

3 years ago[flang] Add missing call to BeginReadingRecord()
peter klausler [Fri, 13 Aug 2021 18:57:58 +0000 (11:57 -0700)]
[flang] Add missing call to BeginReadingRecord()

NAMELIST input needs to start with a call to BeginReadingRecord().
Internal unit input doesn't care (so unit tests were passing), but
external unit input does need the call and will assert without it.

Differential Revision: https://reviews.llvm.org/D108051

3 years agoSelectionDAGBuilder::visitInlineAsm - don't dereference dyn_cast<> results.
Simon Pilgrim [Tue, 17 Aug 2021 17:20:04 +0000 (18:20 +0100)]
SelectionDAGBuilder::visitInlineAsm - don't dereference dyn_cast<> results.

dyn_cast<> can return nullptr if the cast is illegal, use cast<> instead which will assert that the cast is correct.

Fixes static analyser warning.

3 years ago[AArch64] AArch64DAGToDAGISel::tryReadRegister/tryWriteRegister - don't dereference...
Simon Pilgrim [Tue, 17 Aug 2021 17:15:33 +0000 (18:15 +0100)]
[AArch64] AArch64DAGToDAGISel::tryReadRegister/tryWriteRegister - don't dereference dyn_cast<> results.

dyn_cast<> can return nullptr if the cast is illegal, use cast<> instead which will assert that the cast is correct.

Fixes static analyser warnings.

3 years ago[ARM] ARMDAGToDAGISel::tryReadRegister/tryWriteRegister - don't dereference dyn_cast...
Simon Pilgrim [Tue, 17 Aug 2021 17:13:59 +0000 (18:13 +0100)]
[ARM] ARMDAGToDAGISel::tryReadRegister/tryWriteRegister - don't dereference dyn_cast<> results.

dyn_cast<> can return nullptr if the cast is illegal, use cast<> instead which will assert that the cast is correct.

Fixes static analyser warnings.

3 years ago[libc++] Update the version of CMake in the Docker image
Louis Dionne [Tue, 17 Aug 2021 17:35:50 +0000 (13:35 -0400)]
[libc++] Update the version of CMake in the Docker image

3 years agoRevert "[tsan] Fix GCC 8.3 build after D107911"
Marco Elver [Tue, 17 Aug 2021 17:26:19 +0000 (19:26 +0200)]
Revert "[tsan] Fix GCC 8.3 build after D107911"

This reverts commit 797fe59e6b9512652c0ae5a6b69a3c6f5a573fcd.

The use of "EventType type : 3" is replicated for all Event structs and
therefore was still present. As a result this still caused failures on
older GCCs (9.2 or 8.3 or earlier).

The particular bot that was failing due to buggy GCC was fixed by
fef39cc472a773fae4761deaab1c701024ad13ec.

Therefore, no reason to keep the workaround around; revert it.

Reviewed By: vitalybuka

Differential Revision: https://reviews.llvm.org/D108192

3 years agoscan-build-py: Force the opening in utf-8
Sylvestre Ledru [Tue, 17 Aug 2021 17:17:30 +0000 (19:17 +0200)]
scan-build-py: Force the opening in utf-8

It fails on ubuntu bionic otherwise with:
```
scan-build-py-14: Run 'scan-view /tmp/scan-build-2021-08-09-09-14-36-765350-nx9s888s' to examine bug reports.
scan-build-py-14: Internal error.
Traceback (most recent call last):
  File "/usr/lib/llvm-14/lib/libscanbuild/__init__.py", line 125, in wrapper
    return function(*args, **kwargs)
  File "/usr/lib/llvm-14/lib/libscanbuild/analyze.py", line 72, in scan_build
    number_of_bugs = document(args)
  File "/usr/lib/llvm-14/lib/libscanbuild/report.py", line 35, in document
    for bug in read_bugs(args.output, html_reports_available):
  File "/usr/lib/llvm-14/lib/libscanbuild/report.py", line 282, in read_bugs
    for bug in parser(bug_file):
  File "/usr/lib/llvm-14/lib/libscanbuild/report.py", line 421, in parse_bug_html
    for line in handler.readlines():
  File "/usr/lib/python3.6/encodings/ascii.py", line 26, in decode
    return codecs.ascii_decode(input, self.errors)[0]
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc3 in position 3360: ordinal not in range(128)
scan-build-py-14: Please run this command again and turn on verbose mode (add '-vvvv' as argument).
```

I guess it is caused by a problem in Python 3.6

Reviewed By: phosek, isthismyaccount

Differential Revision: https://reviews.llvm.org/D107887

3 years agohwasan: Move stack ring buffer initialization before InitStackAndTls.
Peter Collingbourne [Tue, 17 Aug 2021 03:13:42 +0000 (20:13 -0700)]
hwasan: Move stack ring buffer initialization before InitStackAndTls.

D104248 moved the call to GetThreadStackAndTls to before the
initialization of the ring buffer TLS slot. As a result, if libc
is instrumented we crash in pthread_getattr_np which is called from
__sanitizer::GetThreadStackTopAndBottom.

Fix the problem by moving the stack ring buffer initialization before
the call to InitStackAndTls.

Differential Revision: https://reviews.llvm.org/D108184

3 years ago[flang] Fix regression from recent runtime input fix
peter klausler [Mon, 16 Aug 2021 20:15:01 +0000 (13:15 -0700)]
[flang] Fix regression from recent runtime input fix

A recent runtime I/O change[1] was meant to improve the handling of
input from external files missing a terminal newline on their last
records; the change was "triggered" by the wrong circumstances and
causing reads that should have pulled more data into the buffer to be
treated as EOFs.  So fix that, and also don't retain input data
in the buffer once an input record has been finished unless it's
known that list-directed or NAMELIST input of a repeated input item
may need to backspace a non-positionable external unit to return
to the beginning of the repeated item.

[1] 6578893a0453384346f149479f8574dfff977ace

Differential Revision: https://reviews.llvm.org/D108164

3 years ago[AMDGPU] Fix lowering of AMDGPU::G_CTTZ_ZERO_UNDEF to AMDGPU::G_AMDGPU_FFBL_B32
Simon Pilgrim [Tue, 17 Aug 2021 16:57:01 +0000 (17:57 +0100)]
[AMDGPU] Fix lowering of AMDGPU::G_CTTZ_ZERO_UNDEF to AMDGPU::G_AMDGPU_FFBL_B32

As mentioned on D107474, there was a copy+paste typo repeating G_CTLZ_ZERO_UNDEF that coverity reported as dead code.

Differential Revision: https://reviews.llvm.org/D108210

3 years ago[VP] Add vector-predicated reduction intrinsics
Fraser Cormack [Mon, 14 Jun 2021 08:42:00 +0000 (09:42 +0100)]
[VP] Add vector-predicated reduction intrinsics

This patch adds vector-predicated ("VP") reduction intrinsics corresponding to
each of the existing unpredicated `llvm.vector.reduce.*` versions. Unlike the
unpredicated reductions, all VP reductions have a start value. This start value
is returned when the no vector element is active.

Support for expansion on targets without native vector-predication support is
included.

This patch is based on the ["reduction
slice"](https://reviews.llvm.org/D57504#1732277) of the LLVM-VP reference patch
(https://reviews.llvm.org/D57504).

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D104308

3 years ago[libc][NFC] Disable double precision cos, sin and tan on Windows.
Siva Chandra Reddy [Tue, 17 Aug 2021 16:42:34 +0000 (16:42 +0000)]
[libc][NFC] Disable double precision cos, sin and tan on Windows.

The current x86_64 implementations do not build on the windows bot
machine. We will enable them back after fixing the problem.

3 years ago[OpenMP][NFC] Add option to print module after OpenMPOpt for debugging
Joseph Huber [Mon, 16 Aug 2021 17:06:54 +0000 (13:06 -0400)]
[OpenMP][NFC] Add option to print module after OpenMPOpt for debugging

This patch adds an extra option to print the module after running one of
the OpenMPOpt passes if debugging is enabled. This makes it much easier
to inspect the effects of this pass when doing debugging.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D108146

3 years ago[SCEVExpander] Stop hoisting IR when reusing phis
Philip Reames [Tue, 17 Aug 2021 16:26:59 +0000 (09:26 -0700)]
[SCEVExpander] Stop hoisting IR when reusing phis

his is a fix for PR43678, and is an alternate patch to D105723.

The basic issue we're running into is that LSR + SCEVExpander are moving the very instruction whose operand we're in the process of expanding. This breaks the subtle and ill-documented invariant which let LSR work. (Full story can be found here: https://reviews.llvm.org/D105723#2878473)

Rather than attempting a fix, this change just removes the optimization entirely. The code is entirely untested, and removing it appears to have no impact I can find.  This code was added back in 2014 by 1e12f8563d4b7 with a single test which does not seem to actually test the hoisting logic.

From a philosophical standpoint, it also seems very strange to have the expander implementing optimizations which should live in a dedicated transform pass.

Differential Revision: https://reviews.llvm.org/D106178

3 years ago[analyzer] Adjust JS code of analyzer's HTML report for IE support.
Denys Petrov [Tue, 3 Aug 2021 16:09:00 +0000 (19:09 +0300)]
[analyzer] Adjust JS code of analyzer's HTML report for IE support.

Summary: Change and replace some functions which IE does not support. This patch is made as a continuation of D92928 revision. Also improve hot keys behavior.

Differential Revision: https://reviews.llvm.org/D107366

3 years agoFix 2: [MCParser] Correctly handle CRLF line ends when consuming line comments
Tozer [Tue, 17 Aug 2021 16:09:44 +0000 (17:09 +0100)]
Fix 2: [MCParser] Correctly handle CRLF line ends when consuming line comments

Fixes an issue with revision 5c6f748c and ad40cb88.

Adds an mcpu argument to the test command, preventing an invalid default
CPU from being used on some platforms.

3 years ago[llvm-objdump] -T: print symbol versions
Fangrui Song [Tue, 17 Aug 2021 16:10:50 +0000 (09:10 -0700)]
[llvm-objdump] -T: print symbol versions

Similar to D94907 (llvm-nm -D).

The output will match GNU objdump 2.37.
Older versions don't use ` (version)` for undefined symbols.

Reviewed By: jhenderson

Differential Revision: https://reviews.llvm.org/D108097

3 years agoAllow setting attributes in build method generated by YAML-gen.
MaheshRavishankar [Tue, 17 Aug 2021 15:59:13 +0000 (08:59 -0700)]
Allow setting attributes in build method generated by YAML-gen.

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D108182

3 years ago[Object] Move llvm-nm's symbol version utility to ELFObjectFile::readDynsymVersions
Fangrui Song [Tue, 17 Aug 2021 16:06:39 +0000 (09:06 -0700)]
[Object] Move llvm-nm's symbol version utility to ELFObjectFile::readDynsymVersions

The utility can be reused by llvm-objdump -T.

Reviewed By: jhenderson

Differential Revision: https://reviews.llvm.org/D108096

3 years ago[libc++][NFC] Format expression-equivalent wrappers consistently
Louis Dionne [Mon, 16 Aug 2021 16:41:58 +0000 (12:41 -0400)]
[libc++][NFC] Format expression-equivalent wrappers consistently

Differential Revision: https://reviews.llvm.org/D108144

3 years ago[X86] Lower insertions into upper half of an 256-bit vector as broadcast+blend (PR50971)
Roman Lebedev [Tue, 17 Aug 2021 15:42:22 +0000 (18:42 +0300)]
[X86] Lower insertions into upper half of an 256-bit vector as broadcast+blend (PR50971)

Broadcast is not worse than extract+insert of subvector.
https://godbolt.org/z/aPq98G6Yh

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D105390

3 years ago[Polly] Fix possibly infinite loop.
Michael Kruse [Tue, 17 Aug 2021 15:34:59 +0000 (10:34 -0500)]
[Polly] Fix possibly infinite loop.

The loop had no side-effect since first committed in 642594ae87aca.
While it is obvious what was intended, the code seems to never trigger.

3 years ago[scudo] Fix format string specifiers
Kostya Kortchinsky [Mon, 16 Aug 2021 22:23:48 +0000 (15:23 -0700)]
[scudo] Fix format string specifiers

Enable `-Wformat` again, and fix the offending instances.

Differential Revision: https://reviews.llvm.org/D108168

3 years ago[libc++] Do not require movability in __non_propagating_cache::__emplace_deref
Louis Dionne [Wed, 11 Aug 2021 18:26:33 +0000 (14:26 -0400)]
[libc++] Do not require movability in __non_propagating_cache::__emplace_deref

As explained in http://eel.is/c++draft/range.nonprop.cache#note-1, we
should allow copy and move elision to happen when calling emplace_deref
in non-propagating-cache. Before this change, the only way to emplace
into the non-propagating-cache was to call `__set(*it)`, which materialized
`*it` when binding it to the reference argument of `__set` and disabled
move elision.

As a fly-by change, this also renames `__set` to `__emplace` for consistency
and adds tests for it.

Differential Revision: https://reviews.llvm.org/D107932

3 years agoFix: [MCParser] Correctly handle CRLF line ends when consuming line comments
Tozer [Tue, 17 Aug 2021 15:14:12 +0000 (16:14 +0100)]
Fix: [MCParser] Correctly handle CRLF line ends when consuming line comments

Fixes an issue with revision 5c6f748c.

Move the test added in the above commit into the X86 folder, ensuring
that it is only run on targets where its triple is valid.

3 years agotsan: test: Initialize all fields of Params struct
Marco Elver [Tue, 17 Aug 2021 14:54:07 +0000 (16:54 +0200)]
tsan: test: Initialize all fields of Params struct

Some compilers started complaining about the test:

tsan_trace_test.cpp:128:21: error: missing field 'type' initializer

Fix it by initializing all 5 fields, even though the type field will be
reset in the for loop.

Differential Revision: https://reviews.llvm.org/D108207

3 years ago[MCParser] Correctly handle CRLF line ends when consuming line comments
Tozer [Tue, 17 Aug 2021 14:38:45 +0000 (15:38 +0100)]
[MCParser] Correctly handle CRLF line ends when consuming line comments

Fixes issue: https://bugs.llvm.org/show_bug.cgi?id=47983

The AsmLexer currently has an issue with lexing line comments in files
with CRLF line endings, in which it reads the carriage return as being
part of the line comment. This causes an error for certain valid comment
layouts; this patch fixes this by excluding the carriage return from the
line comment.

Differential Revision: https://reviews.llvm.org/D90234

3 years ago[libc++][NFC] Fix indentation of documentation
Louis Dionne [Tue, 17 Aug 2021 14:52:12 +0000 (10:52 -0400)]
[libc++][NFC] Fix indentation of documentation

3 years ago[Bitcode] Remove unused declaration writeBitcodeHeader (NFC)
Kazu Hirata [Tue, 17 Aug 2021 14:10:51 +0000 (07:10 -0700)]
[Bitcode] Remove unused declaration writeBitcodeHeader (NFC)

The corresponding definition was removed on Nov 29, 2016 in commit
5a0a2e648c267d99111b21482ca709f580e9ccc2.

3 years ago[LIT]Accept cat_64 command name on AIX in shtest
Jinsong Ji [Tue, 17 Aug 2021 14:05:28 +0000 (14:05 +0000)]
[LIT]Accept cat_64 command name on AIX in shtest

AIX may use cat_64 for 64 bit cat, this is just update the lit test to accept the name as well.

Reviewed By: #powerpc, shchenz

Differential Revision: https://reviews.llvm.org/D108149

3 years ago[SVE] Remove usage of getMaxVScale for AArch64, in favour of IR Attribute
Dylan Fleming [Tue, 17 Aug 2021 13:00:47 +0000 (14:00 +0100)]
[SVE] Remove usage of getMaxVScale for AArch64, in favour of IR Attribute

Removed AArch64 usage of the getMaxVScale interface, replacing it with
the vscale_range(min, max) IR Attribute.

Reviewed By: paulwalker-arm

Differential Revision: https://reviews.llvm.org/D106277

3 years ago[libc++][NFC] Refactor tests for transform_view
Louis Dionne [Tue, 17 Aug 2021 13:20:24 +0000 (09:20 -0400)]
[libc++][NFC] Refactor tests for transform_view

Adjust the names of helper function objects to represent better what
they do, as suggested in the review of D107098.

3 years ago[ARM] Enable subreg liveness
David Green [Tue, 17 Aug 2021 13:10:33 +0000 (14:10 +0100)]
[ARM] Enable subreg liveness

This enables subreg liveness in the arm backend when MVE is present,
which allows the register allocator to detect when subregister are
alive/dead, compared to only acting on full registers. This can helps
produce better code on MVE with the way MQPR registers are made up of
SPR registers, but is especially helpful for MQQPR and MQQQQPR
registers, where there are very few "registers" available and being able
to split them up into subregs can help produce much better code.

Differential Revision: https://reviews.llvm.org/D107642

3 years ago[ARM] Add MQQPR and MQQQQPR spill and reload pseudo instructions
David Green [Tue, 17 Aug 2021 12:51:34 +0000 (13:51 +0100)]
[ARM] Add MQQPR and MQQQQPR spill and reload pseudo instructions

As a part of D107642, this adds pseudo instructions for MQQPR and
MQQQQPR register classes, that can spill and reloads entire registers
whilst keeping them combined, not splitting them into multiple D subregs
that a VLDMIA/VSTMIA would use. This can help certain analyses, and
helps to prevent verifier issues with subreg liveness.

3 years ago[InstCombine] remove unused function argument; NFC
Sanjay Patel [Tue, 17 Aug 2021 12:10:16 +0000 (08:10 -0400)]
[InstCombine] remove unused function argument; NFC

3 years ago[InstCombine] fold signed min/max intrinsics with negated operands
Sanjay Patel [Tue, 17 Aug 2021 11:31:05 +0000 (07:31 -0400)]
[InstCombine] fold signed min/max intrinsics with negated operands

If both operands are negated, we can invert the min/max and do
the negation after:
smax (neg nsw X), (neg nsw Y) --> neg nsw (smin X, Y)
smin (neg nsw X), (neg nsw Y) --> neg nsw (smax X, Y)

This is visible as a remaining regression in D98152. I don't see
a way to generalize this for 'unsigned' or adapt Negator to
handle it. This only appears to be safe with 'nsw':
https://alive2.llvm.org/ce/z/GUy1zJ

Differential Revision: https://reviews.llvm.org/D108165

3 years ago[InstCombine] add tests for smin/smax intrinsics with negated ops; NFC
Sanjay Patel [Mon, 16 Aug 2021 19:27:55 +0000 (15:27 -0400)]
[InstCombine] add tests for smin/smax intrinsics with negated ops; NFC

3 years ago[GlobalISel] Add combine for PTR_ADD with regbanks
Sebastian Neubauer [Tue, 17 Aug 2021 11:58:16 +0000 (13:58 +0200)]
[GlobalISel] Add combine for PTR_ADD with regbanks

Combine two G_PTR_ADDs, but keep the register bank of the constant.
That way, the combine can be used in post-regbank-select combines.

Introduce two helper methods in CombinerHelper, getRegBank and
setRegBank that get and set an optional register bank to a register.
That way, they can be used before and after register bank selection.

Differential Revision: https://reviews.llvm.org/D103326

3 years ago[lldb] Make TestAArch64AdrpAdd depend on the AArch64 target
Raphael Isemann [Tue, 17 Aug 2021 10:46:44 +0000 (12:46 +0200)]
[lldb] Make TestAArch64AdrpAdd depend on the AArch64 target

LLDB is using LLVM's target-specific disassembler which is only available when
the respective LLVM target has been enabled in the build config.

This patch just skips the test if there is no arm64 target (and its
disassembler) available in the current build config.

Reviewed By: jasonmolenda

Differential Revision: https://reviews.llvm.org/D108145

3 years ago[CodeGenPrepare] The instruction to be sunk should be inserted before its user in...
Tiehu Zhang [Tue, 17 Aug 2021 10:50:54 +0000 (18:50 +0800)]
[CodeGenPrepare] The instruction to be sunk should be inserted before its user in a block

In current implementation, the instruction to be sunk will be inserted before the target instruction without considering the def-use tree,
which may case Instruction does not dominate all uses error. We need to choose a suitable location to insert according to the use chain

Reviewed By: dmgreen

Differential Revision: https://reviews.llvm.org/D107262

3 years ago[DebugInfo][InstrRef] Honour too-much-debug-info cutouts
Jeremy Morse [Tue, 17 Aug 2021 10:32:41 +0000 (11:32 +0100)]
[DebugInfo][InstrRef] Honour too-much-debug-info cutouts

This reapplies 54a61c94f93, its follow up in 547b712500e, which were
reverted 95fe61e63954. Original commit message:

VarLoc based LiveDebugValues will abandon variable location propagation if
there are too many blocks and variable assignments in the function. If it
didn't, and we had (say) 1000 blocks and 1000 variables in scope, we'd end
up with 1 million DBG_VALUEs just at the start of blocks.

Instruction-referencing LiveDebugValues should honour this limitation too
(because the same limitation applies to it). Hoist the relevant command
line options into LiveDebugValues.cpp and pass it down into the
implementation classes as an argument to ExtendRanges. I've duplicated all
the run-lines in live-debug-values-cutoffs.mir to have an
instruction-referencing flavour.

Differential Revision: https://reviews.llvm.org/D107823

3 years ago[AArch64] LowerCONCAT_VECTORS - merge getNumOperands() calls. NFCI.
Simon Pilgrim [Tue, 17 Aug 2021 10:22:49 +0000 (11:22 +0100)]
[AArch64] LowerCONCAT_VECTORS - merge getNumOperands() calls. NFCI.

Improves on the unused variable fix from rG4357562067003e25ab343a2d67a60bd89cd66dbf

3 years ago[AggressiveInstCombine] Add shift left instruction to `TruncInstCombine` DAG
Anton Afanasyev [Thu, 12 Aug 2021 11:51:57 +0000 (14:51 +0300)]
[AggressiveInstCombine] Add shift left instruction to `TruncInstCombine` DAG

Add `shl` instruction to the DAG post-dominated by `trunc`, allowing
TruncInstCombine to reduce bitwidth of expressions containing left shifts.

The only thing we need to check is that the target bitwidth must be wider
than the maximal shift amount: https://alive2.llvm.org/ce/z/AwArqu

Part of https://reviews.llvm.org/D107766

Differential Revision: https://reviews.llvm.org/D108091

3 years ago[Test][AggressiveInstCombine] Add test for shifts
Anton Afanasyev [Thu, 12 Aug 2021 11:51:57 +0000 (14:51 +0300)]
[Test][AggressiveInstCombine] Add test for shifts

Precommit test for D107766/D108091. Also move fixed test for PR50555
from SLPVectorizer/X86/ to PhaseOrdering/X86/ subdirectory.

3 years ago[hwasan] Prevent reordering of tag checks.
Florian Mayer [Wed, 11 Aug 2021 13:25:50 +0000 (14:25 +0100)]
[hwasan] Prevent reordering of tag checks.

They were previously unconstrained, which allowed them to be reordered
before the shadow memory write.

Reviewed By: eugenis

Differential Revision: https://reviews.llvm.org/D107901

3 years ago[NFC] Fix typos
PeixinQiao [Tue, 17 Aug 2021 09:16:37 +0000 (17:16 +0800)]
[NFC] Fix typos

Initial commit test.

3 years ago[mlir][linalg] Remove duplicate methods (NFC).
Tobias Gysi [Tue, 17 Aug 2021 07:04:21 +0000 (07:04 +0000)]
[mlir][linalg] Remove duplicate methods (NFC).

Remove duplicate methods used to check iterator types.

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D108102

3 years ago[X86] [AMX] Replace bitcast with specific AMX intrinsics with X86 specific cast.
Bing1 Yu [Thu, 5 Aug 2021 09:01:21 +0000 (17:01 +0800)]
[X86] [AMX] Replace bitcast with specific AMX intrinsics with X86 specific cast.

There is some discussion on the bitcast for vector and x86_amx at https://reviews.llvm.org/D99152. This patch is to introduce a x86 specific cast for vector and x86_amx, so that it can avoid some unnecessary optimization by middle-end. On the other way, we have to optimize the x86 specific cast by ourselves. This patch also optimize the cast operation to eliminate redundant code.

Reviewed By: LuoYuanke

Differential Revision: https://reviews.llvm.org/D107544

3 years agoAMDGPU: During img instruction ret value construction cater for non int values
David Stuttard [Wed, 5 Feb 2020 14:22:18 +0000 (14:22 +0000)]
AMDGPU: During img instruction ret value construction cater for non int values

Make sure return type is int type.

Differential Revision: https://reviews.llvm.org/D108131

Change-Id: Ic02f07d1234cd51b6ed78c3fecd2cb1d6acd5644

3 years ago[flang][nfc] Tweak the FrontendAction class
Andrzej Warzynski [Fri, 13 Aug 2021 16:27:46 +0000 (16:27 +0000)]
[flang][nfc] Tweak the FrontendAction class

This patch refactors the `FrontendAction` class. It merely moves code
around so that re-using it is easier. No new functionality is
introduced.

1. Three new member methods are introduced: `RunPrescan`, `RunParse`,
`RunSemanticChecks`.
2. The following free functions are re-implemented as member methods:
  * `reportFatalSemanticErrors`
  * `reportFatalScanningErrors`
  * `reportFatalParsingErrors`
  * `reportFatalErrors`
`reportFatalSemanticErrors` is updated to resemble the other error
reporting functions and to make the API more consistent.
3. The `BeginSourceFileAction` methods are simplified and the unused
input argument is deleted.

Differential Revision: https://reviews.llvm.org/D108130

3 years ago[Polly][Isl] Use isl::val::sub instead of isl::val::sub_ui. NFC
Riccardo Mori [Tue, 17 Aug 2021 07:34:28 +0000 (09:34 +0200)]
[Polly][Isl] Use isl::val::sub instead of isl::val::sub_ui. NFC

This is part of an effort to reduce the differences between the custom C++ bindings used right now by polly in `lib/External/isl/include/isl/isl-noxceptions.h` and the official isl C++ interface.

Changes made:
 -  Use `isl::val::sub` instead of `isl::val::sub_ui`
 - `isl-noexceptions.h` has been generated by https://github.com/patacca/isl/commit/355e84163ae78ff637c71fb532f36d15277a2b1b

Depends on D107225

Reviewed By: Meinersbur

Differential Revision: https://reviews.llvm.org/D107293

3 years ago[MLIR] [Python] Allow 'operation.parent' to return 'None'
John Demme [Tue, 17 Aug 2021 05:37:14 +0000 (22:37 -0700)]
[MLIR] [Python] Allow 'operation.parent' to return 'None'

This is more Pythonic and better matches the C++ and C APIs.

Reviewed By: stellaraccident

Differential Revision: https://reviews.llvm.org/D108183

3 years ago[analyzer] Add option to SATest.py for extra checkers
Deep Majumder [Tue, 17 Aug 2021 05:12:30 +0000 (10:42 +0530)]
[analyzer] Add option to SATest.py for extra checkers

This patch adds the flag `extra-checkers` to the sub-command `build` for
passing a comma separated list of additional checkers to include.

Differential Revision: https://reviews.llvm.org/D106739

3 years ago[tests] precommit tests for D107692
Yunde Zhong [Tue, 17 Aug 2021 05:02:23 +0000 (13:02 +0800)]
[tests] precommit tests for D107692

3 years ago[AsmParser] Remove MDConstant (NFC)
Kazu Hirata [Tue, 17 Aug 2021 04:21:11 +0000 (21:21 -0700)]
[AsmParser] Remove MDConstant (NFC)

The last use was removed on Sep 22, 2016 in commit
fcee2d80017f8e2db6a8ac3a70bdc0653afa7d01.

3 years ago[LNICM] Fix infinite loop
Whitney Tsang [Tue, 17 Aug 2021 03:50:13 +0000 (12:50 +0900)]
[LNICM] Fix infinite loop

There is a bug introduced by https://reviews.llvm.org/D107219 which causes an infinite loop, when there are more than 2 levels PHINode chain.

Reviewed By: uint256_t

Differential Revision: https://reviews.llvm.org/D108166

3 years ago[AVR][clang] Improve search for avr-libc installation path
Ben Shi [Sat, 7 Aug 2021 08:29:27 +0000 (16:29 +0800)]
[AVR][clang] Improve search for avr-libc installation path

Search avr-libc path according to avr-gcc installation at first,
then other possible installed pathes.

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D107682

3 years ago[AMDGPU] Skip pseudo MIs in hazard recognizer
Christudasan Devadasan [Fri, 13 Aug 2021 02:54:42 +0000 (22:54 -0400)]
[AMDGPU] Skip pseudo MIs in hazard recognizer

Instructions like WAVE_BARRIER and SI_MASKED_UNREACHABLE
are only placeholders to prevent certain unwanted
transformations and will get discarded during assembly
emission. They should not be counted during nop insertion.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D108022

3 years ago[lld-macho][test] Remove ld64.lld: prefix in a diagnostic
Fangrui Song [Tue, 17 Aug 2021 02:41:11 +0000 (19:41 -0700)]
[lld-macho][test] Remove ld64.lld: prefix in a diagnostic

The convention is not to check the prefix before `error: `.
This gives flexibility if we need to rename ld64.lld to something else,
(e.g. a while ago we used ld64.lld.darwinnew).

3 years ago[split-file] Default to --no-leading-lines
Fangrui Song [Tue, 17 Aug 2021 02:23:11 +0000 (19:23 -0700)]
[split-file] Default to --no-leading-lines

It turns out that the --leading-lines may be a bad default.
[[#@LINE+-num]] is rarely used.

3 years ago[MLIR] [Python] Fix out-of-tree Windows python bindings
John Demme [Tue, 17 Aug 2021 02:18:23 +0000 (19:18 -0700)]
[MLIR] [Python] Fix out-of-tree Windows python bindings

MSVC needs to know where to put the archive (.lib) as well as the runtime
(.dll). If left to the default location, multiple rules to generate the same
file will be produced, creating a Ninja error.

Differential Revision: https://reviews.llvm.org/D108181

3 years ago[lld-macho] Refactor parseSections to avoid creating isec on LLVM segments
Vincent Lee [Mon, 16 Aug 2021 22:19:43 +0000 (15:19 -0700)]
[lld-macho] Refactor parseSections to avoid creating isec on LLVM segments

Address post follow up comment in D108016. Avoid creating isec for
LLVM segments since we are skipping over it.

Reviewed By: #lld-macho, int3

Differential Revision: https://reviews.llvm.org/D108167

3 years ago[AMDGPU] Make BVH isel consistent with other MIMG opcodes
Carl Ritson [Tue, 17 Aug 2021 01:24:49 +0000 (10:24 +0900)]
[AMDGPU] Make BVH isel consistent with other MIMG opcodes

Suffix opcodes with _gfx10.
Remove direct references to architecture specific opcodes.
Add a BVH flag and apply this to diassembly.
Fix a number of disassembly errors on gfx90a target caused by
previous incorrect BVH detection code.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D108117

3 years ago[mlir][Analysis][NFC] Clean up FlatAffineValueConstraints
Matthias Springer [Tue, 17 Aug 2021 01:27:41 +0000 (10:27 +0900)]
[mlir][Analysis][NFC] Clean up FlatAffineValueConstraints

* Rename ids to values in FlatAffineValueConstraints.
* Overall cleanup of comments in FlatAffineConstraints and FlatAffineValueConstraints.

Differential Revision: https://reviews.llvm.org/D107947

3 years agoFix missing qualifier in template type diffing
Weverything [Mon, 16 Aug 2021 23:54:10 +0000 (16:54 -0700)]
Fix missing qualifier in template type diffing

Handle SubstTemplateTypeParmType so qualifiers do not get dropped from
the diagnostic message.

3 years ago[mlir][Analysis][NFC] Split FlatAffineConstraints class
Matthias Springer [Tue, 17 Aug 2021 01:08:08 +0000 (10:08 +0900)]
[mlir][Analysis][NFC] Split FlatAffineConstraints class

* Extract "value" functionality of `FlatAffineConstraints` into a new derived `FlatAffineValueConstraints` class. Current users of `FlatAffineConstraints` can use `FlatAffineValueConstraints` without additional code changes, thus NFC.
* `FlatAffineConstraints` no longer associates dimensions with SSA Values. All functionality that requires this, is moved to `FlatAffineValueConstraints`.
* `FlatAffineConstraints` no longer makes assumptions about where Values associated with dimensions are coming from.

Differential Revision: https://reviews.llvm.org/D107725

3 years ago[SamplePGO][NFC] Dump function profiles in order
Hongtao Yu [Mon, 16 Aug 2021 21:17:43 +0000 (14:17 -0700)]
[SamplePGO][NFC] Dump function profiles in order

Sample profiles are stored in a string map which is basically an unordered map. Printing out profiles by simply walking the string map doesn't enforce an order. I'm sorting the map in the decreasing order of total samples to enable a more stable dump, which is good for comparing two dumps.

Reviewed By: wenlei, wlei

Differential Revision: https://reviews.llvm.org/D108147

3 years ago[clang] Expose unreachable fallthrough annotation warning
Nathan Chancellor [Tue, 17 Aug 2021 00:11:07 +0000 (17:11 -0700)]
[clang] Expose unreachable fallthrough annotation warning

The Linux kernel has a macro called IS_ENABLED(), which evaluates to a
constant 1 or 0 based on Kconfig selections, allowing C code to be
unconditionally enabled or disabled at build time. For example:

int foo(struct *a, int b) {
    switch (b) {
    case 1:
        if (a->flag || !IS_ENABLED(CONFIG_64BIT))
            return 1;
        __attribute__((fallthrough));
    case 2:
        return 2;
    default:
        return 3;
    }
}

There is an unreachable warning about the fallthrough annotation in the
first case because !IS_ENABLED(CONFIG_64BIT) can be evaluated to 1,
which looks like

        return 1;
        __attribute__((fallthrough));

to clang.

This type of warning is pointless for the Linux kernel because it does
this trick all over the place due to the sheer number of configuration
options that it has.

Add -Wunreachable-code-fallthrough, enabled under -Wunreachable-code, so
that projects that want to warn on unreachable code get this warning but
projects that do not care about unreachable code can still use
-Wimplicit-fallthrough without having to make changes to their code
base.

Fixes PR51094.

Reviewed By: aaron.ballman, nickdesaulniers

Differential Revision: https://reviews.llvm.org/D107933

3 years ago[MLIR] Add a bitcast method to DenseElementsAttr
Geoffrey Martin-Noble [Thu, 5 Aug 2021 23:29:59 +0000 (16:29 -0700)]
[MLIR] Add a bitcast method to DenseElementsAttr

This method bitcasts a DenseElementsAttr elementwise to one of the same
shape with a different element type.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D107612

3 years ago[tsan] Another attempt to fix GCC 8.3 build after D107911
Douglas Yung [Tue, 17 Aug 2021 00:01:57 +0000 (17:01 -0700)]
[tsan] Another attempt to fix GCC 8.3 build after D107911

This removes the -Werror compilation flag for x64 linux to work around a gcc bug.

GCC 8.3 reports '__tsan::v3::Event::type’ is too small to hold all values of ‘enum class __tsan::v3::EventType’
incorrectly which gets promoted to an error and causes the build to fail.

3 years ago[lldb] Fix -Wunused-but-set-variable
Fangrui Song [Mon, 16 Aug 2021 23:41:17 +0000 (16:41 -0700)]
[lldb] Fix -Wunused-but-set-variable

3 years ago[SamplePGO] Fixing a memory issue when creating profiles on-demand
Hongtao Yu [Mon, 16 Aug 2021 16:21:13 +0000 (09:21 -0700)]
[SamplePGO] Fixing a memory issue when creating profiles on-demand

There is a on-dmeand creation of function profile during top-down processing in the sample loader when merging uninlined callees.  During the profile creation, a stack string object is used to store a newly-created MD5 name, which is then used by reference as hash key in the profile map. This makes the hash key a dangling reference when later on the stack string object is deallocated.

The issue only happens with md5 profile use and was exposed by context split work for CS profile. I'm making a fix by storing newly created names in the reader.

Reviewed By: wenlei, wmi, wlei

Differential Revision: https://reviews.llvm.org/D108142

3 years agoClean up test for -f{,no-}implicit-modules-uses-lock
Duncan P. N. Exon Smith [Mon, 16 Aug 2021 23:18:11 +0000 (16:18 -0700)]
Clean up test for -f{,no-}implicit-modules-uses-lock

@arichardson pointed out in post-commit review for
https://reviews.llvm.org/D95583 (b714f73defc8e075) that `-verify` has an
optional argument that works a lot like `FileCheck`'s `-check-prefix`.
Use it to simplify the test for `-fno-implicit-modules-use-lock`!

3 years ago[tsan] Fix GCC 8.3 build after D107911
Vitaly Buka [Mon, 16 Aug 2021 23:15:57 +0000 (16:15 -0700)]
[tsan] Fix GCC 8.3 build after D107911

gcc 8.3 reports:
__tsan::v3::Event::type’ is too small to hold all values of ‘enum class __tsan::v3::EventType’

3 years ago[NFC] Remove/replace some confusing attribute getters on Function
Arthur Eubanks [Mon, 16 Aug 2021 23:07:47 +0000 (16:07 -0700)]
[NFC] Remove/replace some confusing attribute getters on Function

3 years ago[sanitizer] Define 32bit uptr as uint
Vitaly Buka [Sat, 14 Aug 2021 23:51:10 +0000 (16:51 -0700)]
[sanitizer] Define 32bit uptr as uint

This makes it consistent with uintptr_t.

It's 45138f788c9b3c4ac5d9ae4479841c411c15190e with Darwin fix.

Reviewed By: kstoimenov

Differential Revision: https://reviews.llvm.org/D108163

3 years agoRevert "[sanitizer] Fix MAC build after D108163"
Vitaly Buka [Mon, 16 Aug 2021 22:42:17 +0000 (15:42 -0700)]
Revert "[sanitizer] Fix MAC build after D108163"

They still fail to fix Darwin builds
https://green.lab.llvm.org/green/job/clang-stage1-RA/23399/consoleFull#462858634a1ca8a51-895e-46c6-af87-ce24fa4cd561

This reverts commit ae0628f716cc05ad28adf963538a67e69d58d21d.
This reverts commit 2c6448cdc2f68f8c28fd0bd9404182b81306e6e6.

3 years ago[M68k] Do not pass llvm::Function& to M68kCCState
Min-Yih Hsu [Mon, 16 Aug 2021 00:19:33 +0000 (17:19 -0700)]
[M68k] Do not pass llvm::Function& to M68kCCState

Previously we're passing `llvm::Function&` into `M68kCCState` to lower
arguments in fastcc. However, that reference might not be available if
it's a library call and we only need its argument types. Therefore,
now we're simply passing a list of argument llvm::Type-s.

This fixes PR-50752.

Differential Revision: https://reviews.llvm.org/D108101

3 years ago[AsmPrinter] fix nullptr dereference for MBBs with hasAddressTaken property without BB
Afanasyev Ivan [Mon, 16 Aug 2021 22:26:57 +0000 (15:26 -0700)]
[AsmPrinter] fix nullptr dereference for MBBs with hasAddressTaken property without BB

Basic block pointer is dereferenced unconditionally for MBBs with
hasAddressTaken property.

MBBs might have hasAddressTaken property without reference to BB.
Backend developers must assign fake BB to MBB to workaround this issue
and it should be fixed.

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D108092

3 years ago[libc] dedup handling of size 4 for memset
Guillaume Chatelet [Mon, 16 Aug 2021 22:28:53 +0000 (22:28 +0000)]
[libc] dedup handling of size 4 for memset

3 years ago[sanitizer] Fix MAC build after D108163
Vitaly Buka [Mon, 16 Aug 2021 22:21:54 +0000 (15:21 -0700)]
[sanitizer] Fix MAC build after D108163

3 years ago[libc] Optimize Loop strategy
Guillaume Chatelet [Mon, 16 Aug 2021 22:12:00 +0000 (22:12 +0000)]
[libc] Optimize Loop strategy

Since the precondition for loop is `size >= T::kSize` we always expect
at least one run of the loop. This patch transforms the for-loop into a
do/while-loop which saves at least one test.

We also add a second template parameter to allow the Tail operation to
differ from the loop operation.

3 years ago[ARM] Create MQQPR and MQQQQPR register classes
David Green [Mon, 16 Aug 2021 21:58:12 +0000 (22:58 +0100)]
[ARM] Create MQQPR and MQQQQPR register classes

Similar to the MQPR register class as the MVE equivalent to QPR, this
adds MQQPR and MQQQQPR register classes for the MVE equivalents of QQPR
and QQQQPR registers. The MVE MQPR seemed have worked out quite well,
and adding MQQPR and MQQQQPR allows us to a little more accurately
specify the number of registers, calculating register pressure limits a
little better.

Differential Revision: https://reviews.llvm.org/D107463

3 years ago[scudo] Use stdint types for internal types (redo)
Kostya Kortchinsky [Mon, 16 Aug 2021 18:59:15 +0000 (11:59 -0700)]
[scudo] Use stdint types for internal types (redo)

This is a redo of D108089 that broke some 32-bit builds.

`scudo::uptr` was defined as an `unsigned long` on 32-b platform,
while a `uintptr_t` is usually defined as an `unsigned int`.
This worked, this was not consistent, particularly with regard to
format string specifiers.

As suggested by Vitaly, since we are including `stdint.h`, define
the internal scudo integer types to those.

Differential Revision: https://reviews.llvm.org/D108152

3 years ago[sanitizer] Define 32bit uptr as uint
Vitaly Buka [Sat, 14 Aug 2021 23:51:10 +0000 (16:51 -0700)]
[sanitizer] Define 32bit uptr as uint

This makes it consistent with uintptr_t.

It's 45138f788c9b3c4ac5d9ae4479841c411c15190e with Darwin fix.

Reviewed By: kstoimenov

Differential Revision: https://reviews.llvm.org/D108163

3 years ago[mlir][tosa] Fixed depthwise conv parallel/reduction indices order
Rob Suderman [Mon, 16 Aug 2021 20:47:00 +0000 (13:47 -0700)]
[mlir][tosa] Fixed depthwise conv parallel/reduction indices order

Reduction axis should come after all parallel axis to work with vectorization.

Reviewed By: NatashaKnk

Differential Revision: https://reviews.llvm.org/D108005

3 years ago[Remarks] Emit optimization remarks for atomics generating CAS loop
Anshil Gandhi [Mon, 16 Aug 2021 20:56:01 +0000 (14:56 -0600)]
[Remarks] Emit optimization remarks for atomics generating CAS loop

Implements ORE in AtomicExpand pass to report atomics generating a
compare and swap loop.

Differential Revision: https://reviews.llvm.org/D106891