This reverts commit 7141c5b1
after the base commit D156363 was reverted.

Close https://github.com/llvm/llvm-project/issues/63947
Close https://github.com/llvm/llvm-project/issues/63595

This is suggested by @rsmith in
https://reviews.llvm.org/D154324#inline-1508868

Reviewed By: rsmith

Differential Revision: https://reviews.llvm.org/D156210

This reverts commit 8c3550b1.

clang/test/Driver/fsanitize.c has a mysterious failure worth
investigation.

We could use x0 form in vsetvli when we already know the vlmax and avl is equal to it.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D156404

When output range is only one entry, and input is low_pc/high_pc do not convert
to ranges. This helps with size of .debug_ranges/.debug_rnglists. It also helps
when either low_pc/high_pc is 0. We not generating potentially invalid ranges
that result in LLDB error.

Also fixed handling of DW_AT_subprogram with ranges. This can be created with
-fbasic-block-sections=all.

Reviewed By: maksfb

Differential Revision: https://reviews.llvm.org/D156374

This was probably the only really useful radar in our source code. It's
a request to be able to tell if __LINKEDIT has been mapped or not. I've
left a comment in the radar that we should update the corresponding code
if and when such an ability becomes available.

The existing comment already explains what the problem is. The radar
tracks caching negative lookups in xcrun. Having a backlink is handy,
but it's not necessary as the radar references the LLDB workaround.
Furthermore, we have other places in LLDB that work around xcrun not
caching negative that should potentially be reconsidered at that time.

This replaces two radar links with improved comments explaining the
underlying issues.

 - The first issue is working around a compiler bug that was fixed in
   f454dfb6.
 - The second issue is an invariant that doesn't actually hold. The
   latter was marked as FIXME but there was nothing in the radar about a
   possible alternative solution.

Both radars were closed.

This replaces the radar link with a comment explaining the underlying
problem, which is that when the (dynamic) type of an object changes, so
does their synthetic filter of choice, which requires us to regenerate
the filter.

The comment after the radar link already explains the issue. There's no
additional information in the radar and has been marked as closed by the
corresponding code change. This commit removes the link and reflows the
comment.

The code was commented out because it would trigger an assert in the
source manager. I can no longer reproduce the assert and none of the
ASan tests are failing with the code re-enabled.

The comment and radar referenced PyThreadState_Get which is no longer
used there and instead has been replaced to a call to
PyThreadState_GetDict which has different semantics. Unlike
PyThreadState_Get, it can return NULL and it is okay to call this
function when no current thread state is available.

Add new check to replace enable_if with C++20 constraints

Reviewed By: PiotrZSL

Differential Revision: https://reviews.llvm.org/D141892

Re-enable clang modules for source/Host when targetting macOS 14 or
later, in which the underlying issue has been fixed.

Remove some redundancy.

HasStdExtZve32f implies HasStdExtF
HasStdExtZve64d implies HasStdExtD
HasStdExtZvfbfwma implies HasStdExtZvfbfmin

The current algorithm to compute the symbol size is quadratic if there
are lots of symbols sharing the same addresses. This happens in a debug
build when lots of debug symbols get emitted in the symtab.

This patch improves the performance like `llvm-symbolizer` that relies
on the symbol size computation. Symbolizing a release+assert clang with
DebugInfo sees significant improvements from 3:40min to less than 1s.

Reviewed By: pete, mehdi_amini, arsenm, MaskRay

Differential Revision: https://reviews.llvm.org/D156603

This is consistent with other vendor extensions.

InstComine currently processes blocks in an arbitrary
depth-first order. This can break the usual invariant that the
operands of an instruction should be simplified before the
instruction itself, if uses across basic blocks (particularly
inside phi nodes) are involved.

This patch switches the initial worklist population to use RPO
instead, which will ensure that predecessors are visited before
successors (back-edges notwithstanding).

This allows us to fold more cases within a single InstCombine
iteration, in preparation for D154579. This change by itself
is a minor compile-time regression of about 0.1%, which will
be more than recovered by switching to single-iteration InstCombine.

Differential Revision: https://reviews.llvm.org/D75362

See D156171.

Shrink operands before creating the new instruction to make sure the
same evaluation order is used on all platforms. This fixes buildbot
failures due to different argument evaluation order on different
systems.

The ExecutionSession::removeJITDylibs operation will remove all JITDylibs in
the given list (i.e. first clear them, then remove them from the session).

ExecutionSession::endSession is updated to remove JITDylibs rather than just
clearing them. This prevents new code from being added to any JITDylib once
endSession has been called.

Make sure the full IR is checked for loop-vectorization-factors.ll and
to make sure nothing gets missed and add missing checks for type-shrinkage-insertelt.ll.

Also removes some undef ops from tests.

Fix variable style warnings (case, indentation, etc.)

Detects when a variable is both incremented/decremented and referenced inside a
complex condition and suggests moving them outside to avoid ambiguity in the
variable's value.

Reviewed By: xgupta

Differential Revision: https://reviews.llvm.org/D149015

[X86] X86DAGToDAGISel::matchAddressRecursively - use SelectionDAG::MaxRecursionDepth instead of hard coded constant. NFCI.

SelectionDAG::MaxRecursionDepth has the same max depth (with adjusted comparison).

[clang-tidy] Do not warn on macros starting with underscore and lowercase letter in bugprone-reserved-identifier

Fixes #64130

Differential Revision: https://reviews.llvm.org/D156608

This is to help catch problems in D156552 that only showed up in an
expensive checks build.

Remove the following warning.
```
WARNING: Change IR value name 'tmp4' or use --prefix-filecheck-ir-name to prevent possible conflict with scripted FileCheck name.
```

Fix incorrect setting NSW flag to non-overflowing indvar base (D154954)

Reviewed By: danilaml

Differential Revision: https://reviews.llvm.org/D156577

Support recursive record types in CTF, for example a struct that
contains a pointer to itself:

  struct S {
    struct S *n;
  };

We are now more lazy when creating LLDB types. When encountering a
record type (struct or union) we create a forward declaration and only
complete it when requested.

Differential revision: https://reviews.llvm.org/D156498

similar to 2a1edebd

AIX prefers rejecting certain unimplemented features like
-fprofile-sample-use, making clang_f_opts.c (legacy test file testing
many miscellaneous features) not suitable. We could specify --target=,
but this style change is probably not the best for this legacy file.