Fix wrong debug type.

Reviewed By: hanchung

Differential Revision: https://reviews.llvm.org/D148729

This makes it easier to reuse the legality part for other import
policies that wouldn't use thresholds.

Importing un-inlinable functions is also legal, because they could be
further specialized in a context-specific way, without inlining.

Differential Revision: https://reviews.llvm.org/D148838

Alan spotted another test failure that was a result of
https://reviews.llvm.org/D148546 when running expensive checks tests
locally on windows.

Reviewed By: ayzhao

Differential Revision: https://reviews.llvm.org/D148861

Without this dependency, it is possible that llvm-lib.exe will not be
built, in which case CMake will try to use lib.exe to build libraries,
but this tool cannot handle bitcode files.

Differential Revision: https://reviews.llvm.org/D148751

My reland of https://reviews.llvm.org/D148546 has caused a few windows
demangler tests to fail when run with -DLLVM_ENABLE_EXPENSIVE_CHECKS=ON
on windows.

I have a sneaking suspicion that MSVC's
std::string_view::iterator::operator* may be missing a nullptr check.

Link: https://lab.llvm.org/buildbot/#/builders/42/builds/9723/steps/7/logs/stdio

Reviewed By: ayzhao

Differential Revision: https://reviews.llvm.org/D148852

This patch adds assertions to prevent the compilation when we try to
bit cast a type that is not trivially copyable when using
__builtin_bit_cast, or when we try to bit cast a type that is not
trivially copyable and trivially constructable when using memcpy.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D148739

Summary:
The `+ptx` features correspond to the related CUDA version. We require a
certain set of features from the `ptxas` assembler, which is tied to the
CUDA version. Some of the ones set here were insufficient, so I am
simply setting a cutoff to the CUDA 9.0 release as the minimum. This
roughly corresponds to what should be required for sm_60 to be compiled
with the source.

Previously, `vecreduce_{and,or} vNi1` could lead to miscompilations because the legalizer first decides to `any_ext` the operand (which is correct for `vecreduce_{and,or}`) and then decides to use `vecreduce_u{min,max}` instead (for which `any_ext` is incorrect). This patch changes it so the `vecreduce_u{min,max}` operations use `sign_ext` instead of `any_ext`.

Issue: https://github.com/llvm/llvm-project/issues/62211

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D148672

Values in bounds are expected to have integer or index types. Enforce
this expectation by restricting the type to be IntOrIndex.

Reviewed By: razvanlupusoru

Differential Revision: https://reviews.llvm.org/D148839

This bug was caught by the aarch64 full build builder.

Differential Revision: https://reviews.llvm.org/D148844

In particular, make it `foo.a(foo.o)$ARCHIVE_OFFSET`. The goal is to
make it more similar to both ld64 implementation, which uses the
`foo.a(foo.o)$MODULE_ID` format. We dump some of these names in LTO
code, so matching ld64's format is helpful. This format is also more
similar to LLD-ELF's, which is `foo.a(foo.o at $ARCHIVE_OFFSET)`.

Reviewed By: #lld-macho, oontvoo

Differential Revision: https://reviews.llvm.org/D148828

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D148842

If the AVL is a virtual register defined by a vsetvli with the same
vlmax we need and the previous vsetvli we saw in the data flow also
has that vlmax, we can use the x0, x0 form when we insert a vsetvli.

Not only does this avoid an update of the VL physical register, but
it may allow doLocalPostpass to completely remove the inserted vsetvli
by rewriting the vtype of the previous vsetvli.

Differential Revision: https://reviews.llvm.org/D148735

Allow two currently erroneous cases of !DIR$ IGNORE_TKR errors: allocatable
and pointers, and IGNORE_TKR(R) on (other) arguments passed via descriptors.
Downgrade these cases to warnings when they appear in external interfaces,
since their implementations may well be in C.  But retain the error status
on these cases for module procedures, since the Fortran implementation
probably can't work.

Differential Revision: https://reviews.llvm.org/D148833

Currently conversions to interfaces may happen implicitly (e.g.
`Attribute -> TypedAttr`), failing a runtime assert if the interface
isn't actually implemented. This change marks the `Interface(ValueT)`
constructor as explicit so that a cast is required.

Where it was straightforward to I adjusted code to not require casts,
otherwise I just made them explicit.

Depends on D148491, D148492

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D148493

This allows implicit conversion from `ElementsAttr` to `TypedAttr`, but
required renaming the `ElementsAttr::getType()` interface method to
`getShapedType`.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D148492

This adds `arith::ConstantOp::materialize`, which builds a constant from
an attribute and type only if it would result in a valid op. This is
useful for dialect `materializeConstant` hooks, and allows for removing
the previous `Attribute, Type` builder which was only used during
materialization.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D148491

instruction, NFC.

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D148818

This is the first instance where we've really needed to add
documentation for a diagnostic group, but -Wwrite-strings really
deserves it.

This warning option changes the semantic behavior of code, so enabling
it can cause code to break (and disabling it can too). That's worth
calling out loudly in our documentation.

After the switch to `add_custom_target` to run integration tests, most
of them were not actually being run because of the difference in the way
the COMMAND value is treated between `add_custom_target` and
`add_custom_command`. This patch gets the integration tests to run
again by passing the correct set of arguments to `add_custom_target`.

Reviewed By: michaelrj

Differential Revision: https://reviews.llvm.org/D148786

Uint addition and subtraction normally use builtins which aren't
constexpr. This patch adds an rvalue overload version of the addition
and subtraction operation that is always constexpr.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D148759

Review: https://reviews.llvm.org/D148832

Includes macros:
    linux/SCHED_OTHER // posix req
    linux/SCHED_FIFO // posix req
    linux/SCHED_RR // posix req
    linux/SCHED_BATCH
    linux/SCHED_ISO
    linux/SCHED_IDLE
    linux/SCHED_DEADLINE

Includes types:
    struct sched_param { int sched_priority; }

Includes functions:
    sched_setparam
    sched_getparam
    sched_setscheduler
    sched_getscheduler
    sched_get_priority_max
    sched_get_priority_min
    sched_rr_get_interval

Reviewed By: michaelrj

Differential Revision: https://reviews.llvm.org/D148069

Previously unconditionally stored to the return value. This is
incorrect, we should only return if user value is non-null.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D148293

This commit adds a pattern that expands `math.roundeven` into
`math.round` + some ops from `arith`. This is needed to be able to run
`math.roundeven` in a vectorized manner.

Reviewed By: jpienaar

Differential Revision: https://reviews.llvm.org/D148285

This test is heavy on test resources and involves moving a large 2GB+ file
across an ssh connection when testing on remote devices. Exclude all tests
except for macOS testing (on host). Remote device testing for macOS may
eventually cause a problem on macOS as well w/o better handling of tests
that require more resource/time than we currently handle.
    Introduced by:
        https://reviews.llvm.org/D146189
    Subsequent exclusion of all but darwin by:
        https://reviews.llvm.org/D147094
    Subsequent exclusion of tvOS and watchOS by:
        https://reviews.llvm.org/D147502
    Subsequent exclusion of all but macOS...

rdar://107570309

Reviewed By: yln, thetruestblue

Differential Revision: https://reviews.llvm.org/D148727

As obvious from the paper's title this is an LWG issue and thus retroactively
applied to C++20. This change may the output for certain code points:
1 Considers 8477 extra codepoints as having a width 2 (as of Unicode 15)
  (mostly Tangut Ideographs)
2 Change the width of 85 unassigned code points from 2 to 1
3 Change the width of 8 codepoints (in the range U+3248 CIRCLED NUMBER
  TEN ON BLACK SQUARE ... U+324F CIRCLED NUMBER EIGHTY ON BLACK
  SQUARE) from 2 to 1, because it seems questionable to make an exception
  for those without input from Unicode

Note that libc++ already uses Unicode 15, while the Standard requires Unicode 12.
(The last time I checked MSVC STL used Unicode 14.)

So in practice the only notable change is item 3.

Implements
  P2675 LWG3780: The Paper
  format's width estimation is too approximate and not forward compatible

Benchmark before these changes
--------------------------------------------------------------------
Benchmark                          Time             CPU   Iterations
--------------------------------------------------------------------
BM_ascii_text<char>             3928 ns         3928 ns       178131
BM_unicode_text<char>          75231 ns        75230 ns         9158
BM_cyrillic_text<char>         59837 ns        59834 ns        11529
BM_japanese_text<char>         39842 ns        39832 ns        17501
BM_emoji_text<char>             3931 ns         3930 ns       177750
BM_ascii_text<wchar_t>          4024 ns         4024 ns       174190
BM_unicode_text<wchar_t>       63756 ns        63751 ns        11136
BM_cyrillic_text<wchar_t>      44639 ns        44638 ns        15597
BM_japanese_text<wchar_t>      34425 ns        34424 ns        20283
BM_emoji_text<wchar_t>          3937 ns         3937 ns       177684

Benchmark after these changes
--------------------------------------------------------------------
Benchmark                          Time             CPU   Iterations
--------------------------------------------------------------------
BM_ascii_text<char>             3914 ns         3913 ns       178814
BM_unicode_text<char>          70380 ns        70378 ns         9694
BM_cyrillic_text<char>         51889 ns        51877 ns        13488
BM_japanese_text<char>         41707 ns        41705 ns        16723
BM_emoji_text<char>             3908 ns         3907 ns       177912
BM_ascii_text<wchar_t>          3949 ns         3948 ns       177525
BM_unicode_text<wchar_t>       64591 ns        64587 ns        10649
BM_cyrillic_text<wchar_t>      44089 ns        44078 ns        15721
BM_japanese_text<wchar_t>      39369 ns        39367 ns        17779
BM_emoji_text<wchar_t>          3936 ns         3934 ns       177821

Benchmarks without "if(__code_point < (__entries[0] >> 14))"
--------------------------------------------------------------------
Benchmark                          Time             CPU   Iterations
--------------------------------------------------------------------
BM_ascii_text<char>             3922 ns         3922 ns       178587
BM_unicode_text<char>          94474 ns        94474 ns         7351
BM_cyrillic_text<char>         69202 ns        69200 ns        10157
BM_japanese_text<char>         42735 ns        42692 ns        16382
BM_emoji_text<char>             3920 ns         3919 ns       178704
BM_ascii_text<wchar_t>          3951 ns         3950 ns       177224
BM_unicode_text<wchar_t>       81003 ns        80988 ns         8668
BM_cyrillic_text<wchar_t>      57020 ns        57018 ns        12048
BM_japanese_text<wchar_t>      39695 ns        39687 ns        17582
BM_emoji_text<wchar_t>          3977 ns         3976 ns       176479

This optimization does carry its weight for the Unicode and Cyrillic
test. For the Japanese tests the gains are minor and for emoji it seems
to have no effect.

Reviewed By: ldionne, tahonermann, #libc

Differential Revision: https://reviews.llvm.org/D144499

When --threads= is unspecified, we set it to
`parallel::strategy.compute_thread_count()`, which uses
sched_getaffinity (Linux)/cpuset_getaffinity (FreeBSD)/std::thread::hardware_concurrency (others).
With extensive testing on many machines (many configurations from
{aarch64,x86-64} x {Linux,FreeBSD,Windows} x allocators(native,mimalloc,rpmalloc) combinations)
with varying workloads, we discovered that when the concurrency is larger than
16, the linking process is slower than using --threads=16 due to parallelism
overhead outweighs optimizations. This is particularly harmful for machines with
many cores or when the link job competes with other jobs.

Cap parallel::strategy when --threads= is unspecified.
For some workloads changing the concurrency from 8 to 16 has nearly no improvement.

--thinlto-jobs= is unchanged since ThinLTO backend compiles are embarrassingly
parallel.

Link: https://discourse.llvm.org/t/avoidable-overhead-from-threading-by-default/69160

Reviewed By: peter.smith, andrewng

Differential Revision: https://reviews.llvm.org/D147493

Reviewed By: ldionne, #libc

Differential Revision: https://reviews.llvm.org/D148459

As pointed out in
https://discourse.llvm.org/t/undeterministic-thin-index-file/69985, the
block count added to distributed ThinLTO index files breaks incremental
builds on ThinLTO - if any linked file has a different number of BBs,
then the accumulated sum placed in the index files will change, causing
all ThinLTO backend compiles to be redone.

The block count is only used for scaling of partial sample profiles, and
was added in D80403 for D79831.

This patch simply removes this field from the index files of non partial
sample profile compiles, which is NFC on the output of the compiler.

We subsequently need to see if this can be removed for partial sample
profiles without signficant performance loss, or redesigned in a way
that does not destroy caching.

Differential Revision: https://reviews.llvm.org/D148746

Fixed issue where {tu,cu}-index fixup code for DWARF5 that would report an error when
fixup map is empty. Which is the case when seciton(s) are not over 4GB or
--manaully-generate-unit-index is not specified

Differential Revision: https://reviews.llvm.org/D148578

Defaults to ON for x86_64 && (Linux | Darwin).

If enabled, checks that /proc/self/map_files is readable. Some systems are configured so that getdents fails with EPERM.

Reviewed By: rafauler

Differential Revision: https://reviews.llvm.org/D148742

Review: https://reviews.llvm.org/D147572

We were treating vXi8 multiply as the sum of a trunc(mul(extend(),extend())) which diverged from the costs from llvm-mcaonce we extended beyond legal types

Use a modified version of the D103695 script to determine more accurate throughput/latency/codesize/size-latency cost estimates

Helps address some of the regressions identified in D148806

In https://reviews.llvm.org/D141285 the optional definitions of `HWCAP`
macros were moved to before their usage. However, they were also moved
to before the inclusion of system headers which can optionally define
them. If any of those system headers then actually defined any of the
`HWCAP` macros, it would result in a redefinition error.

Move the system header includes to just before the optional definitions,
to avoid this problem.

Reviewed By: ilinpv

Differential Revision: https://reviews.llvm.org/D148819

We only want to make PCH imports visible once for the the TU, not
repeatedly after every subsequent import. This causes some incorrect
behaviour with submodule visibility, and causes us to get extra module
dependencies in the scanner. So far I have only seen obviously incorrect
behaviour when building with -fmodule-name to cause a submodule to be
textually included when using the PCH, though the old behaviour seems
wrong regardless.

rdar://107449644

Differential Revision: https://reviews.llvm.org/D148176