Completely refactor the cache file to simplify it and remove unnecessary
variables. The main functional change here is that the non-PGO builds
now use two stages, so `ninja -C build stage2-package` can be used with
both PGO and non-PGO builds.

@jimingham I am wondering if you are ok removing this test? It caused
failures in some of the build bots because the user time was less than a
microsecond. Alternatively we can increase the number of loops or maybe
I need some other approach? I had commented it out just to not impact
others

Add the callsite intrinsic. 

Structurally, it is very similar to the counter intrinsics, hence the inheritance relationship. We can probably rename `InstrProfCntrInstBase` to `InstrProfIndexedBase` later - because the "counting" aspect is really left to derived types of `InstrProfCntrInstBase`, and it only concerns itself with the index aspect (which is what we care about for `callsite`, too)

(Tracking Issue: #89287, RFC referenced there)

Related to clean-up opportunities discussed at #90013.

After these cleanups, the `RelaxAll` parameter from
`createMCObjectStreamer` can be removed as well. As
`createMCObjectStreamer` is a more user-facing API and used by two files
in mlir/, we postpone the cleanup to the future.

Use [[fallthrough]] directly (similarly to
https://reviews.llvm.org/D131346).

This recommits eb05a2e8.

By expressing the conditions covered by MAP_ALLOWNOMEM in a more
direct way.

To correctly lower multi_reduction of 1-dim scalable vector, e.g., <[4]xf32>

Related to clean-up opportunities discussed at #90013.

Added GISelShouldIgnore property to class Pattern in TargetSelectionDAG.td; it's similar to FastISelShouldIgnore. This bit can be put on a record to avoid its pattern import within GlobalISelEmitter. This allows one to avoid the record's GISel .td implementation, .inc generation, and any skipped pattern warnings from -warn-on-skipped-patterns.

Extend LoopAccessAnalysis to support different strides and as a
consequence non-constant distances between dependences using SCEV to
reason about the direction of the dependence.

In multiple places, logic to rule out dependences using the stride has
been updated to only be used if StrideA == StrideB, i.e. there's a
common stride.

We now also may bail out at multiple places where we may have to set
FoundNonConstantDistanceDependence. This is done when we need to bail
out and the distance is not constant to preserve original behavior.

Fixes https://github.com/llvm/llvm-project/issues/87336

PR: https://github.com/llvm/llvm-project/pull/88039

The preprocessor define `__HLSL_ENABLE_16_BIT` should be set to 1 if
native 16-bit types are enabled and not set if they are not.

Previously we were setting the value to match the HLSL active language
version, and we had no test coverage verifing the value was set and not
set as expected.

Fixes #89787

Adds a test for `inferConvolutionDimsImpl` to exercise the logic for
depthwise convs.

Remove unneeded LLVM_FALLTHROUGH added after https://reviews.llvm.org/D131346

Rename `OutgoingNotification` to `OutgoingMessage`, since the same
callback function type will be used in a future commit to represent
outgoing requests, in addition to outgoing notifications.

Add a unit test exercising `lsp::MessageHanlder::outgoingNotification`.
Split off from pull request #90076, as requested.

This reverts commit eb05a2e8.
(The comming broke Fortran runtime builders:
https://lab.llvm.org/buildbot/#/builders/267/builds/1277)

These functions do not operate on PortableMemInfoBlock.  This patch
moves them outside the class.

This is a change I really don't like posting, but I think we're out of
other options.  As can be seen in the test differences, we have cases
where adding the freeze inhibits real optimizations.
    
Given no other target handles the undef semantics correctly here, I
think the practical answer is that we shouldn't either.  Yuck.
    
As examples, consider:
* combineMulSpecial in X86.
* performMulCombine in AArch64
    
The only other real option I see here is to move all of the strength
reduction code out of ISEL.  We could do this either via tablegen rules,
or as an MI pass, but other than shifting the point where we ignore
undef
semantics, I don't this is meaningfully different.
    
Note that the particular tests included here would be fixed if we added
SHA/SHL to canCreateUndefOrPoison. However, a) that's already been tried
twice and exposes its own set of regressions, and b) these are simply
examples.  You can create many alternate examples.

Linked to https://github.com/gnustep/libobjc2/pull/289.

More information can be found in issue: #88273.

My solution involves creating a new message-send function for this
calling convention when targeting MSVC. Additional information is
available in the libobjc2 pull request.

I am unsure whether we should check for a runtime version where
objc_msgSend_stret2_np is guaranteed to be present or leave it as is,
considering it remains a critical bug. What are your thoughts about this
@davidchisnall?

[Clang][Sema] Diagnose class member access expressions naming non-existent members of the current instantiation prior to instantiation in the absence of dependent base classes (#84050)

Consider the following:
```cpp
template<typename T>
struct A
{
    auto f()
    {
        return this->x;
    }
};
```
Although `A` has no dependent base classes and the lookup context for
`x` is the current instantiation, we currently do not diagnose the
absence of a member `x` until `A<T>::f` is instantiated. This patch
moves the point of diagnosis for such expressions to occur at the point
of definition (i.e. prior to instantiation).

This patch provides the initial implementation for the "Step Into
Specific/Step In Targets" feature in VSCode DAP.

The implementation disassembles all the call instructions in step range
and try to resolve operand name (assuming one operand) using debug info.
Later, the call target function name is chosen by end user and specified
in the StepInto() API call.

It is v1 because of using the existing step in target function name API.
This implementation has several limitations:
* Won't for indirect/virtual function call -- in most cases, our
disassembler won't be able to solve the indirect call target
address/name.
* Won't work for target function without debug info -- if the target
function has symbol but not debug info, the existing
ThreadPlanStepInRange won't stop.
* Relying on function names can be fragile -- if there is some middle
glue/thunk code, our disassembler can only resolve the glue/thunk code's
name not the real target function name. It can be fragile to depend
compiler/linker emits the same names for both.
* Does not support step into raw address call sites -- it is a valid
scenario that in Visual Studio debugger, user can explicitly choose a
raw address to step into which land in the function without debug
info/symbol, then choose UI to load the debug info on-demand for that
module/frame to continue exploring.

A more reliable design could be extending the ThreadPlanStepInRange to
support step in based on call-site instruction offset/PC which I will
propose in next iteration.

---------

Co-authored-by: jeffreytan81 <jeffreytan@fb.com>

These tests pass on AIX.

llvm/cmake/modules/HandleLLVMOptions.cmake adds -Wextra.
-Wcast-function-type-mismatch was recently added to -Wextra, leading to
a warning for the `get_wrap_symbols` code (https://reviews.llvm.org/D44235).

Suppress the diagnostic.

Pull Request: https://github.com/llvm/llvm-project/pull/89994

This fixes https://github.com/llvm/llvm-project/issues/57589, and aligns
Clang with the behavior of current versions of gcc. There is a new
option, -mdaz-ftz, to control the linking of the file that sets FTZ/DAZ
on startup, and this flag is on by default if -ffast-math is present and
-shared isn't.

This also partially reverts fa7cd549 in that it disables the attempt
to set the IR denormal-fp-math attribute based on whether or not
-ffast-math is applied as it is insufficiently reliable.

Add fallthrough annotations to avoid warnings if -Wimplicit-fallthrough
is enabled.

Test plan: ninja check-all

[InstCombine] Canonicalize `gep T, (gep i8, base, C1), (Index + C2)` into `gep T, (gep i8, base, C1 + C2 * sizeof(T)), Index` (#76177)

This patch tries to canonicalize `gep T, (gep i8, base, C1), (Index +
C2)` into `gep T, (gep i8, base, C1 + C2 * sizeof(T)), Index`.

Alive2: https://alive2.llvm.org/ce/z/dxShKF
Fixes regressions found in
https://github.com/llvm/llvm-project/pull/68882.

We changed the name of one of the types, which is consumed by LLDB.  My
patch local build + CI didn't catch it, but a build bot did!  This
commit fixes it by updating the name in LLDB.

This patch removes getFullSchema in MemProfTest.cpp in favor of
llvm::memprof::PortableMemInfoBlock::getFullSchema as they do exactly
the same thing.

OpenACC is going to need an array sections implementation that is a
simpler version/more restrictive version of the OpenMP version. 

This patch moves `OMPArraySectionExpr` to `Expr.h` and renames it `ArraySectionExpr`,
 then adds an enum to choose between the two.

This also fixes a couple of 'drive-by' issues that I discovered on the way,
but leaves the OpenACC Sema parts reasonably unimplemented (no semantic
analysis implementation), as that will be a followup patch.

This commit closes #80850 where author suggests adding a
readability check to detect missing parentheses around mathematical
expressions when operators of different priorities are used.

Signed-off-by: 11happy <soni5happy@gmail.com>

This commit implements runtime verification for LinalgStructuredOps
using the existing `RuntimeVerifiableOpInterface`. The verification
checks that the runtime sizes of the operands match the runtime sizes
inferred by composing the loop ranges with the op's indexing maps.

Currently, GDBRemoteCommunicationServerLLGS::Handle_qfThreadInfo asserts
if the number of processes under debug isn’t 1 and the multiprocess
feature isn’t supported. This is so that we don't string IDs of threads
belonging to different processes together without including the IDs of
the processes themselves in the response when there are multiple
processes under debug. However, it’s conceivable that we have no process
under debug and the multiprocess feature isn’t supported. So, have
GDBRemoteCommunicationServerLLGS::Handle_qfThreadInfo assert if the
number of processes under debug is greater than 1 and the multiprocess
feature isn’t supported.

Some assembly mnemonics may assemble to instructions of different
lengths. The longer form is to support instructions like a long branch.
On X86, -mrelax-all enables `MCRelaxAll`, which expands instructions to
the long form regardless of whether a short form suffices, while
-mno-relax-all only expands instructions when needed.

```
// x86 example
void foo(int a) {
  // -mno-relax-all or gas: short jump (2 bytes)
  // -mrelax-all: near jump (6 bytes)
  if (a) bar();
}
```

The -mrelax-all default for non-RISCV -O0 appears to only affect x86 and
increases code size without any compile time difference for a stage-2
x86-64 build of lld.
```
-mrelax-all:    file size: 60.9MiB   VM size: 52.4MiB
-mno-relax-all: file size: 58.2MiB   VM size: 49.7MiB
```

There is no compile time difference (other than noise) GNU assembler
doesn't expand instructions by default. Let's remove the -mrelax-all default.

Pull Request: https://github.com/llvm/llvm-project/pull/90013