LockedAccess provides pointer-like access to a value while holding a
lock. All accessors are rvalue-ref-qualified, restricting usage to
temporaries to prevent accidental lock lifetime extension. A with_ref
method is provided for multi-statement critical sections.
When a class is marked `__declspec(dllexport)`, Clang eagerly creates
inherited constructors via `findInheritingConstructor` and propagates
the dllexport attribute to all members. This bypasses overload
resolution, which would normally filter out constructors whose requires
clause is not satisfied. As a result, Clang attempted to instantiate
constructor bodies that should never be available, causing spurious
compilation errors.
Add constraint satisfaction checks in `checkClassLevelDLLAttribute` to
match MSVC behavior:
1. Before eagerly creating inherited constructors, verify that the base
constructor's `requires` clause is satisfied. Skip creation otherwise.
2. Before applying dllexport to non-inherited methods of class template
specializations, verify constraint satisfaction. This handles the case
where `dllexport` propagates to a base template specialization whose own
members have unsatisfied constraints.
Inherited constructors skip the second check since their constraints
were already verified at creation time.
Fixes#185924
Followup to https://github.com/llvm/llvm-project/pull/182706
Assisted by: Cursor // Claude Opus 4.6
In LoongArch and RISC-V, the relaxation pass iterates over input sections
within executable output sections. When a linker script places a synthetic
section (e.g., .got) into such an output section, the linker would crash
because synthetic sections do not have the relaxAux field initialized.
The relaxAux data structure is only allocated for non-synthetic sections
in initSymbolAnchors. This patch adds the necessary null checks in the
relaxation loops (relaxOnce and finalizeRelax) to skip sections that
do not require relaxation.
A null check is also added to elf::initSymbolAnchors to ensure the
subsequent sorting of anchors is safe.
Fixes: #184757
Reviewers: MaskRay
Pull Request: https://github.com/llvm/llvm-project/pull/184758
Value is guaranteed to be zero after the loop:
for (I = 0; Value; ++I, Value >>= 7)
Therefore the subsequent `if (Value)` condition is always false.
Remove the unreachable code. Reported by PVS-Studio.
Fixed: #170122
`MCPseudoProbeDecoder` cannot be copeied/moved due to its address
dependence on the DummyInlineRoot member address. Explicitly delete the move constructor.
There's no point constructing a dominator tree or similar on
known-broken IR. Generally, functions should be able to assume that IR
is valid (i.e., passes the verifier). Users of this "feature" were:
- Verifier, fixed by verifying existence of terminators first.
- FuzzMutate, worked around by temporarily inserting terminators.
- OpenMP to run analyses while building the IR, worked around by
temporarily inserting terminators.
- Polly to work with an empty dominator tree, fixed by temporarily
adding an unreachable inst.
- MergeBlockIntoPredecessor, inadvertently, fixed by adding terminator
before updating MemorySSA.
- Some sloppily written unit tests.
The first code example in the "confusing standard behavior" section
had a comment claiming `[[unlikely]]` makes the branch unlikely,
contradicting a later example showing the same placement being ignored.
Rewords the comment to clarify this is the C++ Standard's
recommendation that Clang does not follow, since the attribute is not on
the substatement.
Continues the work from #126372.
Fixes#126362.
Use single quotes for string arguments inside f-strings or otherwise the
version of black that we use fails to parse. Also reformat the file
given that hasn't been working for a while (wholesale or incrementally)
to the above issue.
Previously, PrivateLabelPrefix was default-initialized to "L", so basic
block labels were added to the symbol table. This seems like an
oversight, so use ".L" for all private labels.
This reverts commit af22b50fac2311ff3f859e4e8bdec552c7aa8d5a.
This seems to have had no noticeable effect on the frequency of failures
so likely was not the issue.
BranchProbabilityInfo will compute it's own dominator tree and
post-dominator tree if none is specified; avoid this by using the
analysis manager/pass manager to get the analysis, which will reuse the
previously computed DomTree.
This patch supersedes PR #151970 by adding the option
``AllowShortRecordOnASingleLine`` that allows the following formatting:
```c++
struct foo {};
struct bar { int i; };
struct baz
{
int i;
int j;
int k;
};
```
---------
Co-authored-by: owenca <owenpiano@gmail.com>
Remove the DenseMap handling lambda paramter mappings from
`EvalEmitter`. This was always unused. Remove it and use `if constexpr`
to keep things compiling.
The current clang-format configuration
option AllowShortFunctionsOnASingleLine uses a single enum
(ShortFunctionStyle) to control when short function definitions can be
merged onto a single line. This enum provides predefined combinations of
conditions
(e.g., None, Empty only, Inline only, Inline including Empty, All).
This approach has limitations:
1. **Lack of Granularity:** Users cannot specify arbitrary combinations
of conditions. For example, a user might want to allow merging
for both empty functions and short top-level functions, but not for
short functions defined within classes. This is not possible with the
current enum options except by choosing All, which might merge more than
desired.
2. **Inflexibility:** Adding new conditions for merging (e.g.,
distinguishing between member functions and constructors, handling
lambdas specifically) would require adding many new combined enum
values, leading to a combinatorial explosion and making the
configuration complex.
3. **Implicit Behavior:** Some options imply others
(e.g., Inline implies Empty), which might not always be intuitive or
desired.
The goal is to replace this single-choice enum with a more flexible
mechanism allowing users to specify a set of conditions that must be met
for a short function to be merged onto a single line.
---------
Co-authored-by: owenca <owenpiano@gmail.com>
The extensible dialect system defined `compareProperties` to false
because it doesn't use properties. However, this should have been
`true`, as the empty properties are trivially always equal to
themselves. Doing otherwise means that no operations in extensible
dialects that aren't the exact same operation will ever compare equal
for the purposes of operations like CSE.
Session::addService now returns a reference to the added Service. This
allows clients to hold a reference for further direct interaction with
the Service object.
This commit also introduces a new Session::createService convenience
method that creates the service and returns a reference to it.
The name "Service" better reflects the general purpose of this class: It
provides *something* (often resource management) to the Session, is
owned by the Session, and receives notifications from the Session when
the controller detaches / is detached, and when the Session is shut
down.
An example of a non-resource-managing Service (to be added in an
upcoming patch) is a detach / shutdown notification service: Clients can
add this service to register arbitrary callbacks to be run on detach /
shutdown. The advantage of this over the current Session detach /
shutdown callback system is that clients can control both the order of
the callbacks, and their order relative to notification of other
services.
On AIX, this test sometimes fails with error `Assertion failed: y ==
true`. The test assumes `compare_exchange_weak` should succeed on a
single call, however according to the standard:
> A weak compare-and-exchange operation may fail spuriously. That is,
even when the contents of memory referred to by expected and ptr are
equal, it may return false and store back to expected the same memory
contents that were originally there.
This spurious failure enables implementation of compare-and-exchange on
a broader class of machines, e.g., load-locked store-conditional
machines. A consequence of spurious failure is that nearly all uses of
weak compare-and-exchange will be in a loop.
[atomics.ref.ops]/27
This includes several changes:
- `Dialect.load(reload=False)` will fail if the dialect was already
loaded in a different context. To prevent the further program abortion.
- `Dialect.load(reload=True)` implies `replace=True` in
dialect/operation registering.
- `PyGlobals::registerDialectImpl` now has a parameter `replace`.
- `register_dialect` and `register_operation` is no longer exposed in
`mlir.dialects.ext`.
This should solve the registering problem found in writing transform
test cases by @rolfmorel.
Recognize table based log2 implementations like
```
unsigned log2(unsigned v) {
static const unsigned char table[] = {
0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30,
8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31
};
v |= v >> 1;
v |= v >> 2;
v |= v >> 4;
v |= v >> 8;
v |= v >> 16;
return table[(unsigned)(v * 0x07C4ACDDU) >> 27];
}
```
and replaces with 31 - llvm.ctlz(v).
Similar for i64 log2. Other sizes can be supported with correct multiply
constant and table values, but I have not found examples yet.
This code is based on the existing tryToRecognizeTableBasedCttz. Like
that function, we support
any combination of multiply constant and table values that produce the
correct result.
It handles the same pattern as #177110, but does not match the outer
subtract from that patch. It is assumed that InstCombine or other
optimizations can combine (sub 31 (sub 31, cttz V)) later.
I have limited this to targets that have a fast ctlz. The backend does
not yet have a table based lowering for ctlz so this reduces the chance
of regressions.
Split the monolithic cir.unary operation (which dispatched on a
UnaryOpKind enum) into four separate operations: cir.inc, cir.dec,
cir.minus, and cir.not.
Changes:
- Add CIR_UnaryOpInterface with getInput()/getResult() methods
- Add CIR_UnaryOp and CIR_UnaryOpWithOverflowFlag base classes
- Define IncOp, DecOp, MinusOp, NotOp with per-op folds
- Add Involution trait to NotOp for not(not(x)) -> x folding
- Replace createUnaryOp() with createInc/Dec/Minus/Not builders
- Split LLVM lowering into four separate patterns
- Split LoweringPrepare complex-type handling per unary op
- Update CIRCanonicalize and CIRSimplify for new op types
- Update all codegen files to use bool params instead of UnaryOpKind
- Remove CIR_UnaryOpKind enum and old CIR_UnaryOp definition
Assembly format change:
cir.unary(inc, %x) nsw : !s32i, !s32i -> cir.inc nsw %x : !s32i
cir.unary(not, %x) : !u32i, !u32i -> cir.not %x : !u32i
Introduce `readability-redundant-qualified-alias` to flag identity type
aliases that repeat a qualified name and suggest using-declarations when
safe. The check is conservative: it skips macros, elaborated keywords,
dependent types, and templates. `OnlyNamespaceScope` controls whether
local/class scopes are included (default `false`).
Depends on: #183940#183941
This is a scaled down version of https://reviews.llvm.org/D136315.
The intent is largely the same as before[^1], but I've scaled down the
scope to try to avoid the issues that the previous patch caused:
- the changes are now opt-in based on enabling `CLANG_USE_XCSELECT`
- this only works when targeting macOS on a macOS host (this is the only
case supported by `libxcselect`[^2])
- calling `libxcselect` is done only when the target is `*-apple-macos*`
to avoid breaking many tests
Another reason to leave this as opt-in for now is that there are some
bugs in libxcselect that need fixing before it is safe to use by default
for all users. This has been reported to Apple as FB16081077.
[^1]: See also https://reviews.llvm.org/D109460 and #45225.
[^2]: https://developer.apple.com/documentation/xcselect?language=objc
