[MIPS] Use custom bitcast lowering to avoid excessive instructions
On Mips32r2 bitcast can be expanded to two sw instructions and an ldc1
when using bitcast i64 to double or an sdc1 and two lw instructions when
using bitcast double to i64. By introducing custom lowering that uses
mtc1/mthc1 we can avoid excessive instructions.
Patch by Mirko Brkusanin.
Differential Revision: https://reviews.llvm.org/D61069
[X86] Remove part of an if condition that should always be true.
The IndexReg will always be non-null at this point. Earlier in the function, if
IndexReg was null we set it to CurDAG->getRegister(0, VT) which made it
[JITLink] Make the JITLink MachO/x86-64 eh-frame test work on Windows.
This should fix the MachO/x86-64 eh-frame regression test by ensuring that
the symbols __ZTIi and ___gxx_personality_v0 are defined on all platforms.
[llvm-rtdyld] Add support for passing command line arguments to rtdyld-run code.
The --args option can now be used to pass arguments to code linked with
$ llvm-rtdyld file1.o file2.o --args a b c
is equivalent to:
$ ld -o program file1.o file2.o
$ ./program a b c
This is the rtdyld counterpart to the jitlink change in r359115, and makes
benchmarking and comparison between the tools easier.
Enable LoopVectorization by default.
When refactoring vectorization flags, vectorization was disabled by default in the new pass manager.
This patch re-enables is for both managers, and changes the assumptions opt makes, based on the new defaults.
Comments in opt.cpp should clarify the intended use of all flags to enable/disable vectorization.
Reviewers: chandlerc, jgorbe
Subscribers: jlebar, llvm-commits
Differential Revision: https://reviews.llvm.org/D61091
[llvm-objdump] errorToErrorCode+message -> toString
For test/Object/elf-invalid-phdr.test, the intended error message got lost due to errorToErrorCode().
[PowerPC][NFC]Update licence to Apache 2
Consolidate existing utilities for interpreting vector predicate maskes [NFC]
Update test to better check for the non-constexpr-ness of a move constructor. Fixes PR#41577.
Fix unused variable warning in LoopFusion pass.
Do not wrap the contents of printFusionCandidates in the LLVM_DEBUG macro. This
fixes an unused variable warning generated when compiling without asserts but
Differential Revision: https://reviews.llvm.org/D61035
[InstCombine] Be consistent w/handling of masked intrinsics style wise [NFC]
Use modern type trait implementations when available.
Teach libcxx to stop using various deprecated __has_* type traits, in favor of
the ("modern", C++11 era) __is_* type traits.
This is mostly just a simplification, but fixes at least one bug: _Atomic T
should be considered trivially-destructible, but is not considered to be POD by
Clang, and __has_trivial_destructor is specified in the GCC documentation as
returning false for non-POD non-class types.
Differential Revision: https://reviews.llvm.org/D48292
Fix typo in comment in r312851.
Thanks to Nico Weber for pointing this out!
[utils] Add a lldb data formatter for llvm::SmallString.
(lldb) p val
(llvm::SmallString<32>) $31 = "patatino"
Hide stderr output from lldb-argdumper
Under very specific circumstances the default shell /bin/sh might
print stuff to stderr before launching lldb-argdumper, which then
confuses the JSON parser. This patch suppresses stderr output from
lldb-argdumper to avoid this situation.
Differential Revision: https://reviews.llvm.org/D61101
PR41427: This has apparently been fixed already, just add a regression
Fix spelling error. NFC
Summary: Test commit.
Reviewers: msearles, jkorous
Reviewed By: jkorous
Subscribers: dexonsmith, arsenm, jvesely, nhaehnle, hiraditya, llvm-commits
Differential Revision: https://reviews.llvm.org/D61093
llvm-cvtres: Make new dupe resource error a bit friendlier
For well-known type IDs, include the name of the type.
To not duplicate the ID->name map, make llvm-readobj call this new
function as well. It has slightly different output, so this also
requires updating a few tests.
Differential Revision: https://reviews.llvm.org/D61086
posix_spawn should retry upon EINTR
We've seen cases of bots failing with:
clang: error: unable to execute command: posix_spawn failed: Interrupted system call
Add a small retry loop to posix_spawn in case this happens. Don't retry too much in case there's some systemic problem going on, but retry a few times.
Reviewers: Bigcheese, arphaman
Subscribers: jkorous, dexonsmith, kristina, llvm-commits
Differential Revision: https://reviews.llvm.org/D61096
Mark new jitlink test XFAIL for windows
[sanitizer_common] Remove some old commented-out printf statements, and fixed up the order of includes.
Recommitting r358783 and r358786 "[MS] Emit S_HEAPALLOCSITE debug info" with fixes for buildbot error (undefined assembler label).
This emits labels around heapallocsite calls and S_HEAPALLOCSITE debug
info in codeview. Currently only changes FastISel, so emitting labels still
needs to be implemented in SelectionDAG.
Subscribers: aprantl, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D61083
[codeview] Fix symbol names for dynamic initializers and atexit stubs
Add a new variant to GlobalDecl for these so that we can detect them
more easily during debug info emission and handle them appropriately.
Reviewers: rsmith, rjmccall, jyu2
Subscribers: aprantl, cfe-commits
Differential Revision: https://reviews.llvm.org/D60930
[DAGCombiner] scale repeated FP divisor by splat factor
If we have a vector FP division with a splatted divisor, use the existing transform
that converts 'x/y' into 'x * (1.0/y)' to allow more conversions. This can then
potentially be converted into a scalar FP division by existing combines (rL358984)
as seen in the tests here.
That can be a potentially big perf difference if scalar fdiv has better timing
(including avoiding possible frequency throttling for vector ops).
Differential Revision: https://reviews.llvm.org/D61028
[PowerPC] Allow using initial-exec TLS with PIC
Using initial-exec TLS variables is a reasonable performance
optimisation for system libraries. Use the correct PIC mechanism to get
hold of the GOT to avoid text relocations.
Differential Revision: https://reviews.llvm.org/D61026
Skip test introduced in r359140 on windows
Not sure what is or is not supposed to work on Windows and I have
no way to investigate this.
Add period at end of comment.
[X86] Attempt to fix use-after-poison from r359121.
[clang-tidy] Add new checker: llvm-prefer-isa-or-dyn-cast-in-conditionals
Looks at conditionals and finds cases of ``cast<>``, which will
assert rather than return a null pointer, and ``dyn_cast<>`` where
the return value is not captured. Additionally, finds cases that
match the pattern ``var.foo() && isa<X>(var.foo())``, where the
method is called twice and could be expensive.
.. code-block:: c++
// Finds cases like these:
if (auto x = cast<X>(y)) <...>
if (cast<X>(y)) <...>
// But not cases like these:
if (auto f = cast<Z>(y)->foo()) <...>
if (cast<Z>(y)->foo()) <...>
Reviewers: alexfh, rjmccall, hokein, aaron.ballman, JonasToth
Reviewed By: aaron.ballman
Subscribers: xbolva00, Eugene.Zelenko, mgorny, xazax.hun, cfe-commits
Tags: #clang-tools-extra, #clang
Differential Revision: https://reviews.llvm.org/D59802
[SystemInitializerFull] Fix header sorting (NFC)
Made some changes downstream that touched the header sorting.
Fix infinite recursion when calling C++ template functions
When we encounter a templated function in the debug information, we
were creating an AST that looked like this:
FunctionTemplateDecl 0x12980ab90 <<invalid sloc>> <invalid sloc> foo<int>
|-TemplateTypeParmDecl 0x12980aad0 <<invalid sloc>> <invalid sloc> class depth 0 index 0 T
|-FunctionDecl 0x12980aa30 <<invalid sloc>> <invalid sloc> foo<int> 'int (int)' extern
| |-TemplateArgument type 'int'
| `-ParmVarDecl 0x12980a998 <<invalid sloc>> <invalid sloc> t1 'int'
`-FunctionDecl 0x12980aa30 <<invalid sloc>> <invalid sloc> foo<int> 'int (int)' extern
|-TemplateArgument type 'int'
`-ParmVarDecl 0x12980a998 <<invalid sloc>> <invalid sloc> t1 'int'
Note that the FunctionTemplateDecl has 2 children which are identical (as
in have the same address). This is not what Clang is doing:
FunctionTemplateDecl 0x7f89d206c6f8 </tmp/template.cpp:1:1, line:4:1> line:2:5 foo
|-TemplateTypeParmDecl 0x7f89d206c4a8 <line:1:10, col:19> col:19 referenced typename depth 0 index 0 T
|-FunctionDecl 0x7f89d206c660 <line:2:1, line:4:1> line:2:5 foo 'int (T)'
| `-ParmVarDecl 0x7f89d206c570 <col:9, col:11> col:11 t1 'T'
`-FunctionDecl 0x7f89d206cb60 <line:2:1, line:4:1> line:2:5 used foo 'int (int)'
|-TemplateArgument type 'int'
`-ParmVarDecl 0x7f89d206ca68 <col:9, col:11> col:11 t1 'int':'int'
The 2 chidlren are different and actually repesent different things: the first
one is the unspecialized version and the second one is specialized. (Just looking
at the names shows another major difference which is that we create the parent
with a name of "foo<int>" when it should be just "foo".)
The fact that we have those 2 identical children confuses the ClangImporter
and generates an infinite recursion (reported in https://llvm.org/pr41473).
We cannot create the unspecialized version as the debug information doesn't
contain a mapping from the template parameters to their use in the prototype.
This patch just creates 2 different FunctionDecls for those 2 children of the
FunctionTemplateDecl. This avoids the infinite recursion and allows us to
call functions. As the XFAILs in the added test show, we've still got issues
in our handling of templates. I believe they are mostly centered on the fact
that we create do not register "foo" as a template, but "foo<int>". This is
a bigger change that will need changes to the debug information generation.
I believe this change makes sense on its own.
Reviewers: shafik, clayborg, jingham
Subscribers: aprantl, javed.absar, kristof.beyls, lldb-commits
Differential Revision: https://reviews.llvm.org/D61044
[AMDGPU] gfx1010 SOP instructions
Differential Revision: https://reviews.llvm.org/D61080
[ScriptInterpreterPython] find_first_of -> find (NFC)
Follow up to r357198.
[COFF] Don't emit .gfids sections when CFG is off
Put them on the list of GuardFidChunks instead of the main Chunks list,
even with CFG is off. It will be ignored if CFG is disabled.
[SLP] Fix crash after r358519, by V. Porpodas.
Summary: The code did not check if operand was undef before casting it to Instruction.
Reviewers: RKSimon, ABataev, dtemirbulatov
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D61024
Try once more to ensure constant initializaton of ManagedStatics
First, use the old style of linker initialization for MSVC 2019 in
addition to 2017. MSVC 2019 emits a dynamic initializer for
ManagedStatic when compiled in debug mode, and according to zturner,
also sometimes in release mode. I wasn't able to reproduce that, but it
seems best to stick with the old code that works.
When clang is using the MSVC STL, we have to give ManagedStatic a
constexpr constructor that fully zero initializes all fields, otherwise
it emits a dynamic initializer. The MSVC STL implementation of
std::atomic has a non-trivial (but constexpr) default constructor that
zero initializes the atomic value. Because one of the fields has a
non-trivial constructor, ManagedStatic ends up with a non-trivial ctor.
The ctor is not constexpr, so clang ends up emitting a dynamic
initializer, even though it simply does zero initialization. To make it
constexpr, we must initialize all fields of the ManagedStatic.
However, while the constructor that takes a pointer is marked constexpr,
clang says it does not evaluate to a constant because it contains a cast
from a pointer to an integer. I filed this as:
Once we do that, we can add back the
LLVM_REQUIRE_CONSTANT_INITIALIZATION marker, and so far as I'm aware it
compiles successfully on all supported targets.
[pstl] Make the default backend be the serial backend and always provide parallel policies
Before this change, the default backend was TBB but one could disable
anything related to TBB by removing the parallel policies. This change
uses the serial backend by default and removes the ability to disable
parallel policies, which is not useful anymore.
Reviewers: rodgert, MikeDvorskiy
Subscribers: mgorny, jkorous, dexonsmith, jdoerfert, libcxx-commits
Differential Revision: https://reviews.llvm.org/D59792
[compiler-rt] NFC fixed [whitespace/braces] LINT warning
[OPENMP]Initial support for non-rectangular loop nest.
Added basic semantic analysis for the non-rectangular loop nests for
OpenMP 5.0 support.
Add optional arg to profile count getters to filter
synthetic profile count.
Differential Revision: http://reviews.llvm.org/D61025
[X86] Prevent folding a load into an AND if that AND is really a ZEXT_INREG that should use movzx.
This can save a 32-bit immediate move.
We would shrink the load and fold it if it was non-volatile, but that's trickier to check for.
llvm-cvtres: Remove a default argument. No behavior change.
Revert using fcopyfile(3) to implement sys::fs::copy_file(Twine, int) on macOS
It turns out that I mesread the man page and fcopyfile(3) does not
actually support COPYFILE_CLONE for files.
[fuzzer] Fix reload.test on Linux/aarch64
The compiler generates a 'brk' instruction for __builtin_trap on aarch64
and Linux kernel issues a SIGTRAP. It is different from x86, where
compiler emits an 'ud2' and kernel issues a SIGILL.
A straightforward is to use abort instead.
[compiler-rt] Fix warning about virtual destructor in sanitizer_flag_parser.h
[EditLineTests] Call setenv() before editline is initialized.
[llvm-symbolizer] Quick fix for broken sanitizer bot
(sanitizer-x86_64-linux) until I can triage the issue properly. The
build has been broken due to the symbolizer build checks failing.
As the symbolizer build script relies on the old svn repo layout, it may
take a little while longer to find the responsible patch for the
breakage. This may be a completely valid fix, but I will need to confirm
it. For now, it unbreaks the build.
Build where the break first occurred: http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux/builds/21211
Changelist authors: @grimar, @maskray, @whitequark, @spatel, @dpreobra
For the authors, no action needed (yet). Will follow up when I can
identify the cause.
DebugInfo: Emit only declarations (not whole definitions) of non-unit user defined types into type units
While this doesn't come up in reasonable cases currently (the only user
defined types not in type units are ones without linkage - which makes
for near-ODR violations, because it'd be a type with linkage referencing
a type without linkage - such a type can't be validly defined in more
than one TU, so arguably it shouldn't be in a type unit to begin with -
but it's a convenient way to demonstrate an issue that will become more
revalent with homed modular debug info type definitions - which also
don't need to be in type units but more legitimately so).
Precursor to the Clang change to de-type-unit (by omitting the
'identifier') types homed due to strong linkage vtables. (making that
change without this one would lead to major type duplication in type
[X86] Remove dead nodes left after ReplaceAllUsesWith calls during address matching
ReplaceAllUsesWith doesn't remove the node that was replaced. So its left around in the graph messing up use counts on other nodes.
One thing to note, is that this isn't valid if the node being deleted is the root node of an LEA match that gets rejected. In that case the node needs to stay alive because the isel table walking code would still have a reference to it that its going to try to match next. I don't think that's the case here though because the nodes being deleted here should be "and", "srl", and "zero_extend" none of which can be the root node of an LEA match.
Differential Revision: https://reviews.llvm.org/D61048