[clang] Exclude non-template classes when checking if constraint refers to containing template arguments (#74265)

When checking if the constraint uses any enclosing template parameters
for [temp.friend]p9, if a containing record is used as argument, we
assume that the constraint depends on enclosing template parameters
without checking if the record is templated. The reproducer from the bug
is included as a test case.

Fixes #71595

This instruction produces a 32-bit sign extended value if the SEW is less than or
equal to 32.

Adds a warning, issued by the clang semantic analysis. The patch warns
on switch which don't have the default branch.

This is a counterpart of gcc's Wswitch-default.

Fixed #41854.

Reverts llvm/llvm-project#74360

As I wrote in the analysis of #74360:

Since
https://github.com/llvm/llvm-project/commit/bc4e0c048aa3cd940b0cea787014c7e8680e5040
we will not add PluginAdaptors into the container of all plugin adaptors
before the plugin is not ready. The error is thereby gone. When and old
HSA loads other libraries they can call register_image but that will
simply not register the image with the plugin we are currently
initializing. That seems like reasonable behavior, thought it is good to
keep in mind if we ever want a kernel library (@jhuber6 @mjklemm). We
can still have a standalone kernel library though or load it late after
all plugins are setup (which seems reasonable).

I did not expect one our tests actually doing exactly what this will not
allow anymore, at least when you use rocm <5.5.0. Need to figure out if
we want this behavior (for rocm <5.5.0).

The `ForallRewriter` pattern used to generate invalid IR:
```
mlir/test/Dialect/SparseTensor/GPU/gpu_combi.mlir:0:0: error: 'scf.for' op expects region #0 to have 0 or 1 blocks
mlir/test/Dialect/SparseTensor/GPU/gpu_combi.mlir:0:0: note: see current operation:
"scf.for"(%8, %2, %9) ({
^bb0(%arg5: index):
  // ...
  "scf.yield"() : () -> ()
^bb1(%10: index):  // no predecessors
  "scf.yield"() : () -> ()
}) : (index, index, index) -> ()
```

This commit fixes tests such as
`mlir/test/Dialect/SparseTensor/GPU/gpu_combi.mlir` when verifying the
IR after each pattern application (#74270).

`createQuickSort` used to generate invalid IR:
```
"func.func"() <{function_type = (index, index, memref<?xindex>, memref<?xf32>, memref<?xi32>) -> (), sym_name = "_sparse_qsort_0_1_index_coo_1_f32_i32", sym_visibility = "private"}> ({
^bb0(%arg0: index, %arg1: index, %arg2: memref<?xindex>, %arg3: memref<?xf32>, %arg4: memref<?xi32>):
  %0:2 = "scf.while"(%arg0, %arg1) ({
  ^bb0(%arg5: index, %arg6: index):
    // ...
    "scf.condition"(%3, %arg5, %arg6) : (i1, index, index) -> ()
  }, {
  ^bb0(%arg5: index, %arg6: index):
    // ...
    %7:2 = "scf.if"(%6) ({
      %8 = "arith.cmpi"(%2, %3) <{predicate = 7 : i64}> : (index, index) -> i1
      // ...
      "scf.yield"(%9#0, %9#1) : (index, index) -> ()
      %10 = "arith.constant"() <{value = 0 : index}> : () -> index
    }, {
      "scf.yield"(%arg5, %arg5) : (index, index) -> ()
    }) : (i1) -> (index, index)
    "scf.yield"(%7#0, %7#1) : (index, index) -> ()
  }) : (index, index) -> (index, index)
  "func.return"() : () -> ()
}) : () -> ()
within split at mlir/test/Dialect/SparseTensor/buffer_rewriting.mlir:76 offset :11:1: error: 'scf.yield' op must be the last operation in the parent block
```

This commit fixes tests such as
`mlir/test/Dialect/SparseTensor/buffer_rewriting.mlir` when verifying
the IR after each pattern application (#74270).

The test fails intermittently due to non-unique file name %T.

Use %t based file names instead.

Fixes: https://github.com/llvm/llvm-project/issues/74472

Op verifiers should verify only local properties of an op. The dynamic
sizes of a `tensor.generate` op should not be verified. Dynamic sizes
that have a negative constant value should not prevent the
`tensor.generate` op from verifying.

Also share some code between the `tensor.empty` and `tensor.generate`
"dynamic dim -> static dim" canonicalization patterns.

Remove the `invalid-canonicalize.mlir` file and move the test case to
`canonicalize.mlir`. Canonicalization no longer produces IR that does
not verify (and leaves the op as is).

Pointer jump stubs content was incorrectly using 1-byte alignment. This bug
flew under the radar because the ordinary instruction streams (and aligned
section starts) usually left the block 4-byte aligned anyway. The bug was
observed when an out-of-tree pass introduced new content with arbitrary sizes
into the text section.

No testcase yet: We don't have a good way to replicate the perturbation that
was caused by the out-of-tree pass.

RWBuffer doesn't have a default type in dxc's implementation, so it
shouldn't have one in clang either.

Reviewers: llvm-beanz, python3kgae

Reviewed By: python3kgae, llvm-beanz

Pull Request: https://github.com/llvm/llvm-project/pull/71265

The partial bufferization framework has been replaced with One-Shot
Bufferize. SCF-specific canonicalization patterns for
`to_memref`/`to_tensor` are no longer needed.

If a template is defined via an external AST source, it won't have a
location. When we emit warnings about misusing such templates we
shouldn't emit a "template is declared here" warning with no location,
as that's just confusing.

Reviewers: llvm-beanz, erichkeane, AaronBallman

Reviewed By: erichkeane, AaronBallman

Pull Request: https://github.com/llvm/llvm-project/pull/71264

Add missing constant propogation folder for LeftShiftLogical,
RightShift[Logical|Arithmetic].

Implement additional folding when Shift value is 0.

This helps for readability of lowered code into SPIR-V.

Part of work for #70704

Before we expected all symbols in the device image to be backed up with
data that we could read. However, uninitialized values are not. We now
check for this case and avoid reading random memory.

This also replaces the correct readGlobalFromImage call with a
isSymbolInImage check after
https://github.com/llvm/llvm-project/pull/74550 picked the wrong one.

Fixes: https://github.com/llvm/llvm-project/issues/74582

`*` is not friendly to the MLIR attribute parser and will fail to be
parsed. Switch the `*` enum representation to `star`.

We probably should use a temporary name, but having stable names helps
debugging.

Remove `DelayedBinDesc` as it is not necessary since
https://github.com/llvm/llvm-project/commit/bc4e0c048aa3cd940b0cea787014c7e8680e5040.
See
https://github.com/llvm/llvm-project/pull/74360#issuecomment-1843603736
for details.

These tests started passing after this PR landed:
https://github.com/llvm/llvm-project/pull/74580

The reorganisation assists with identifying information that's relevant
to the reader by using sections, note/warning blocks, and highlighted
lists.

Some rewording was necessary to fit the new structure and some to
improve flow. Changes to the intention of the documentation have not
been made.

---------

Co-authored-by: Will Hawkins <whh8b@obs.cr>

getOptionVisibilityMask already returns options::FlangOption in FlangMode,
so this assignment is entirely pointless.

Original message:
We still have to keep the noCommonBitsSet call to handle multiple reassociations in one pass. We'll lose the flag on the first reassociation.

By default we now only look for the plugins we build, but the user can
overwrite that with `LIBOMPTARGET_PLUGINS_TO_LOAD="cuda,amdgpu,x86_64"`

This commit reverts the changes in
https://github.com/llvm/llvm-project/pull/71780 and all of its follow-up
patches.

We got reports of the `.debug_names/.debug_gnu_pubnames/gdb_index/etc.`
sections growing by a non-trivial amount for some large projects. While
GCC emits definitions for static data member constants into the Names
index, they do so *only* for explicitly `constexpr` members. We were
indexing *all* constant-initialized const-static members, which is
likely where the significant size difference comes from. However, only
emitting explicitly `constexpr` variables into the index doesn't seem
like a good way forward, since from clang's perspective `const`-static
integrals are `constexpr` too, and that shouldn't be any different in
the debug-info component. Also, as new code moves to `constexpr` instead
of `const` static for constants, such solution would just delay the
growth of the Names index.

To prevent the size regression we revert to not emitting definitions for
static data-members that have no location.

To support access to such constants from LLDB we'll most likely have to
have to make LLDB find the constants by looking at the containing class
first.

Removes at in favor of curr; also makes method delegates consistent

Replaces an llvm::cast that assumed that all Binary instances were either
Archive or MachOUniversalBinary instances with a dyn_cast. The cast was
triggering an assert in StaticLibraryDefinitionGenerator::Load if that method
was given a path or MemoryBuffer containing a relocatable object file.
Switching to dyn_cast causes the operation to error out with a bad-format
error as expected.

Fixes rdar://119262300

This reverts commit 78964457.

Looks like I didn't rebase this correctly before commit

We still have to keep the noCommonBitsSet call to handle multiple reassociations in one pass. We'll lose the flag on the first reassociation.

The current vectorization of 1D depthwise convolutions in Linalg is
_sub-optimal_ for tensor with a low number of channel dimensions, e.g.:

```mlir
linalg.depthwise_conv_1d_nwc_wc
    {dilations = dense<1> : vector<1xi64>,
    strides = dense<1> : vector<1xi64>}
    ins(%input, %filter : tensor<1x8x3xi8>, tensor<1x3xi8>)
    outs(%output : tensor<1x8x3xi8>) -> tensor<1x8x3xi8>
```

That's due to the fact that ultimately (i.e. at LLVM level),
vectorization happens along the trailing dimension (i.e. the channel
dimension). In this case it leads to vectors with 3 elements (or worse,
if there's e.g. only 1 channel dimension). For comparison, a 128 bit
wide vector registers can hold 16 x i8.

Instead, this patch adds an option to flatten/collapse the channel
dimension into the width dimension of the input/filter/output using
`vector.shape_cast` operation:

```mlir
    %sc_input = vector.shape_cast %input : vector<1x8x3xi8> to vector<1x24xi8>
    %sc_output = vector.shape_cast %output : vector<1x8x3xi8> to vector<1x24xi8>
    %b_filter = vector.broadcast %filter : vector<3xi8> to vector<1x8x3xi8>
    %sc_filter = vector.shape_cast %b_filter : vector<1x8x3xi8> to vector<1x24xi8>
```

This new vectorization mode is implemented in `depthwiseConv` by
inserting `vector.shape_cast` Ops before and after 
`depthwiseConv1dSliceAsMulAcc` is invoked. It can be selected through
e.g. a transform dialect attribute:

```mlir
  transform.structured.vectorize_children_and_apply_patterns %conv {flatten_1d_depthwise_conv}
```

A forthcoming patch will implement a strategy to automatically switch
between the two implementations, depending on the shape of the input
tensors.

Co-authored by: Bradley Smith <bradley.smith@arm.com>

This adds a consistent usage with `at` for everything that refers to the
current loop nesting. This cleans up some redundant legacy code from
when we were still using topSort inside sparsifier code.

I discovered two issues for when a CFI dispatch table entry is used as a
direct call.
# Inlining
There is the possibility that the dispatch table entry contains only a
single function pointer:
```
; Function Attrs: naked nocf_check
define private void @.cfi.jumptable() #6 align 8 { entry:
  call void asm sideeffect "jmp ${0:c}@plt\0Aint3\0Aint3\0Aint3\0A", "s"(ptr @_Z7throw_ei)
  unreachable
}
```
If this function is inlined, the unreachable follows and ruins the
containing function.
# Exception Handling
The dispatch table is always marked NoUnwind. This is fine if the
entries are never used directly, but if a direct call is used which the
containing function expects to throw, it will no longer throw and the
exception handling code will be lost.

To avoid compiler errors on some platforms introduced
by #66583, now using std::vector to pass list of
matchers into main matcher, and removed static variable
as it could introduce some other issues.

Finds constants and function calls to math functions that can be
replaced with c++20's mathematical constants from the 'numbers' 
header and offers fix-it hints.
Does not match the use of variables with that value, and instead,
offers a replacement at the definition of those variables.