[qt/clang/llvm-project]: Summary of bulk changes made

Git repository change summary for qt/clang/llvm-project
Pushed by mirror-service into branch 'upstream/main'.
Changed from 1fe1bf4e4e1eeacb50c8a00d44f5737ec9b1c10e to bd586dd26e3ba49616175c14cb2970c581fb94b7
Acknowledgement was received that this change introduces only existing code that has been pushed to another public open source repository.

This change contains the following new commits:

Git commit a4764e36c7d9ef00810fae196b7030b6c7354db7 by GitHub (on behalf of Bryth) on 06/08/2026 at 11:16..
[mlir][ArmSVE] Fix typo in BFMMLA op description (NFC) (#214201)
https://invent.kde.org/qt/clang/llvm-project/-/commit/a4764e36c7d9ef00810fae196b7030b6c7354db7

Git commit 0c91f5b25492e552f2f24f4d56750c3948dd6e90 by GitHub (on behalf of Bryth) on 06/08/2026 at 11:16..
[mlir][vector] Disallow vector broadcast along scalable dim (#212197)

Broadcasts from vectors to higher rank vectors get lowered recursively
by iterating over leading dimensions to create "size(dimension)" insert
ops. This cannot be done for scalable dimensions whose size is unknown.
This PR prevents such transforms from occurring by stopping if the
leading dimension is scalable.
https://invent.kde.org/qt/clang/llvm-project/-/commit/0c91f5b25492e552f2f24f4d56750c3948dd6e90

Git commit 50ff635538a52197ce708ce65c3fb1752be0f20f by GitHub (on behalf of Abid Qadeer) on 06/08/2026 at 11:21..
[flang][debug] Emit debug info for named constants (#213974)

Fixes https://github.com/llvm/llvm-project/issues/213966.

Only globals whose uniqued name deconstructs to `NameKind::VARIABLE`
were described. A Fortran named constant (PARAMETER) is mangled with EC
and deconstructs to `NameKind::CONSTANT`, so it got no debug info at all
and a debugger could not evaluate one, whether it was declared in a
module or inside a procedure.

A module constant is described the way a module variable already is,
with a `DIGlobalVariable` scoped to the `DIModule`. A constant local to
a procedure is described in the scope of that procedure.

Two related fixes for entities whose global has internal linkage, which
covers both a procedure local constant and a procedure local SAVE
variable that was already being described:

- `isLocalToUnit` was hardcoded to false, so these were marked
`DW_AT_external`. It now follows the linkage of the global.

- A linkage name was emitted for them. There is no external symbol for a
debugger to match against, so it is now omitted, which is also what
clang does for a function local static.
 
Assisted by: Cursor

---------

Co-authored-by: Cursor <[email protected]>
https://invent.kde.org/qt/clang/llvm-project/-/commit/50ff635538a52197ce708ce65c3fb1752be0f20f

Git commit db89f5c0c9b0ce60373924075c2bcd5d1c41415a by GitHub (on behalf of Simon Pilgrim) on 06/08/2026 at 11:45..
[X86] Tidyup atom fpu->gpr costs (#214454)

These had matching high latency and throughput numbers, which suggest
they hadn't been updated since the itinerary conversion.

Confirmed with Agner, uops.info and instlatx64 which report a consistent
throughput of 2cy, latency seems to be about 4cy (uops.info just says <=
6cy)

Noticed while trying to confirm #214191 vector reduction costs for atom
type cpus, and llvm-mca was reporting some very odd numbers for bonnell.
https://invent.kde.org/qt/clang/llvm-project/-/commit/db89f5c0c9b0ce60373924075c2bcd5d1c41415a

Git commit 1bfde5fd3b8f809fc7ab1ce32862276144783ef5 by GitHub (on behalf of Sergio Afonso) on 06/08/2026 at 12:12..
[MLIR][OpenMP] Refactor omp dialect utils headers, NFC (#213648)

This patch implements some non-functional refactoring changes:
- It consolidates the two existing "utils" headers into one.
- It renames fields and arguments to the `OffloadModuleOpts` structure
to follow the MLIR style guide and more closely match the corresponding
`OffloadModuleInterface` attributes.
- It removes comments in OpenMP-to-LLVMIR translation referring to Flang
frontend options associated to the `OffloadModuleInterface`.
https://invent.kde.org/qt/clang/llvm-project/-/commit/1bfde5fd3b8f809fc7ab1ce32862276144783ef5

Git commit 105e9f1be4962318a8ec8adedb7b37eec1674ed4 by GitHub (on behalf of Sergio Afonso) on 06/08/2026 at 12:12..
[MLIR][OpenMP] Ensure -fopenmp-force-usm doesn't override requirements (#213649)

The `mlir::omp::setOffloadModuleInterfaceAttributes` utility function
can currently override any pre-existing OpenMP `requires` clauses in the
module. This doesn't cause any problems at the moment because all calls
to this function happen before any other `requires` are processed.
However, it's safer to make sure it never deletes pre-existing flags in
case the same function is reused in a different context.
https://invent.kde.org/qt/clang/llvm-project/-/commit/105e9f1be4962318a8ec8adedb7b37eec1674ed4

Git commit a8f0bcbdc4d8c2e046ce107a0a169e310162c807 by GitHub (on behalf of Abid Qadeer) on 06/08/2026 at 12:26..
[flang][debug] Add accurate line number information to module. (#214011)

The line number in a module's debug info was a guess: the source line of
the first module member we encountered, minus one. That was only correct
when the first declaration happened to be on the line right after the
MODULE statement, so debuggers usually reported a module at the wrong
line.
    
The location of `fir.module_debug_imports` is that of the MODULE
statement, so use it instead. That operation is now generated for every
module and not only for those containing a USE statement, so the
location is available in all cases. It is still only generated when
debug info is requested.
    
Co-authored-by: Cursor <[email protected]>

---------

Co-authored-by: Cursor <[email protected]>
https://invent.kde.org/qt/clang/llvm-project/-/commit/a8f0bcbdc4d8c2e046ce107a0a169e310162c807

Git commit 1cc730fd7b31cdc240a6357b66e7a1b05bbfe135 by GitHub (on behalf of Tim Besard) on 06/08/2026 at 12:37..
[SPIRV] Legalize i1 min/max before selection (#204231)

LLVM can produce `llvm.{s,u}{min,max}.i1` from boolean value patterns.
In SPIR-V, however, LLVM `i1` is represented as `OpTypeBool`, not as an
integer type. The OpenCL/GLSL extended min/max instructions require
integer scalar or integer vector operands and results, so selecting
those instructions directly for `i1` can produce invalid SPIR-V.

This patch makes the SPIR-V legalizer reflect that type-system boundary.
Nonstandard integer widths remain legal when the relevant extensions are
enabled, but `s1` is not treated as an extended integer width. Scalar
`i1` min/max is widened to `i32` before instruction selection and
converted back to bool. Boolean vectors are scalarized first, then each
lane follows the same scalar legalization path.

The result is deliberately conservative: it preserves the existing
extended-integer behavior while preventing boolean values from reaching
integer-only SPIR-V extended instructions.

---

The Khronos SPIRV-LLVM-Translator follows the same type-model premise:
`SPIRVWriter::transType` maps LLVM `i1` to `OpTypeBool`, while wider
LLVM integer types map to `OpTypeInt`.

The translator also avoids using OpenCL/GLSL extended min/max
instructions for LLVM min/max intrinsics. Its `SPIRVWriter.cpp` lowering
for `llvm.umin`, `llvm.umax`, `llvm.smin`, and `llvm.smax` emits an
integer comparison followed by `OpSelect`. That compare/select strategy
does not provide evidence that boolean extended min/max is valid; it
sidesteps the integer-only extended-instruction constraint entirely.

There is also a useful precedent in the translator regularization pass
for shifts: because SPIR-V shift operands must be integer scalar/vector
types, LLVM `i1` operands are treated as boolean and extended to `i32`
before the integer operation, then converted back to bool. This patch
applies the same principle to the LLVM backend path that currently uses
extended min/max instructions.

---------

Co-authored-by: Codex <[email protected]>
https://invent.kde.org/qt/clang/llvm-project/-/commit/1cc730fd7b31cdc240a6357b66e7a1b05bbfe135

Git commit b73efb77b873a536814b4aec98a41f4d099bd58d by GitHub (on behalf of Brian Cain) on 06/08/2026 at 13:05..
[Hexagon] Avoid Handling Debug Instructions (#212918)

The change is made in HexagonFrameLowering::(optimizeSpillSlots)
function. This Pass is responsible for managing function's stack frame
and ensuring proper stack space allocation. Spill slots means temporary
memory locations used for storing register values that need to be
spilled from registers to memory during code execution. While Optimizing
these spill slots we should not handle debug instructions. Optimized the
pass by skipping debug instructions in optimizeSpillSlots function.

A DBG_VALUE describing a variable that lives in a spill slot has a frame
index operand, but it is neither a load from nor a store to that slot,
so the slot was marked as one that cannot be optimized and the
store/load pair using it was no longer replaced with register copies.
The code generated for a function therefore differed depending on
whether debug info was enabled.

Co-authored-by: Chandana Sinderikeri <[email protected]>
https://invent.kde.org/qt/clang/llvm-project/-/commit/b73efb77b873a536814b4aec98a41f4d099bd58d

Git commit 8610fba1de1acdbd91f93e808a780aa7da5869cb by GitHub (on behalf of Hassnaa Hamdi) on 06/08/2026 at 13:08..
[LV][NFC] Reorder creating vplans for UserVF and epilogueVF (#214379)

For the case when we force userVF and epilogueVF, we create only 2
vplans.vThis patch reorder the vplans creation to create the main loop
vplan firstly then the epilogue loop vplan.
https://invent.kde.org/qt/clang/llvm-project/-/commit/8610fba1de1acdbd91f93e808a780aa7da5869cb

Git commit 849c51e082b0958524246a0a880f46d468a55147 by GitHub (on behalf of Trevor Gross) on 06/08/2026 at 13:08..
[compiler-rt] Disable executable stack on `aeabi_u{read,write}*.S` (#214465)

These were missing `NO_EXEC_STACK_DIRECTIVE` to add `.note.GNU-stack`;
without it, a binary including any of these files will have the stack
marked executable. Add the directive here, matching other similar files.

Symtab diff before:

$ clang compiler-rt/lib/builtins/arm/aeabi_uread4.S
--target=arm-unknown-linux-gnueabi -c
    $ llvm-readelf aeabi_uread4.o -S
    There are 5 section headers, starting at offset 0xe4:

    Section Headers:
[Nr] Name Type Address Off Size ES Flg Lk Inf Al
[ 0] NULL 00000000 000000 000000 00 0 0 0
[ 1] .strtab STRTAB 00000000 0000a8 000039 00 0 0 1
[ 2] .text PROGBITS 00000000 000034 000020 00 AX 0 0 4
[ 3] .ARM.attributes ARM_ATTRIBUTES 00000000 000054 000022 00 0 0 1
[ 4] .symtab SYMTAB 00000000 000078 000030 10 1 2 4

After:

$ clang compiler-rt/lib/builtins/arm/aeabi_uread4.S
--target=arm-unknown-linux-gnueabi -c
    $ llvm-readelf aeabi_uread4.o -S
    There are 6 section headers, starting at offset 0xf4:

    Section Headers:
[Nr] Name Type Address Off Size ES Flg Lk Inf Al
[ 0] NULL 00000000 000000 000000 00 0 0 0
[ 1] .strtab STRTAB 00000000 0000a8 000049 00 0 0 1
[ 2] .text PROGBITS 00000000 000034 000020 00 AX 0 0 4
[ 3] .note.GNU-stack PROGBITS 00000000 000054 000000 00 0 0 1
[ 4] .ARM.attributes ARM_ATTRIBUTES 00000000 000054 000022 00 0 0 1
[ 5] .symtab SYMTAB 00000000 000078 000030 10 1 2 4

Fixes: 39413af931a7 ("[Compiler-rt] Implement AEABI Unaligned Read/Write
       Helpers in compiler-rt (#167913)")
https://invent.kde.org/qt/clang/llvm-project/-/commit/849c51e082b0958524246a0a880f46d468a55147

Git commit 5ae6dfd93e916f417d2015efdcc9d45c9c48f6a6 by GitHub (on behalf of Lucas Ramirez) on 06/08/2026 at 13:25..
[AMDGPU][CodeGen] Fix `EXPENSIVE_CHECKS` for multi-def rematerialization  (#214484)

After #197580 registers may have multiple definitions.
`EXPENSIVE_CHECKS` were not updated to reflect that.
https://invent.kde.org/qt/clang/llvm-project/-/commit/5ae6dfd93e916f417d2015efdcc9d45c9c48f6a6

Git commit 65e92e073539c3fe3523256c4fba35ac1dedcb92 by GitHub (on behalf of Jacek Caban) on 06/08/2026 at 13:27..
[LLD][COFF] Emit ARM64X relocations for TLS directory (#213529)

The current behavior of merging EC and native chunks to have a single
TLS directory for both views matches the MSVC linker, but it has its
shortcomings. In addition to merging, that solution requires all TLS
callbacks to use -arm64xsameaddress, leaving it to the CRT to handle. If
the appropriate EC object files are not pulled in by linked EC code and
thus never have a chance to mark the callback with -arm64xsameaddress,
this may lead to an invalid image that crashes at runtime.

This patch avoids the entire problem by using entirely separate TLS
directories for EC and native views along with the standard ARM64X
dynamic relocation mechanism. Since callback lists are now separate, a
missing -arm64xsameaddress is no longer a problem.

Also, mingw-w64-crt currently doesn't mark its TLS callbacks with
-arm64xsameaddress. That could be changed if needed, but with this
change, it is no longer necessary.
https://invent.kde.org/qt/clang/llvm-project/-/commit/65e92e073539c3fe3523256c4fba35ac1dedcb92

Git commit 8caa6c19ace019317008a981f4469260f22dbc84 by GitHub (on behalf of forking-google-bazel-bot[bot]) on 06/08/2026 at 13:39..
[Bazel] Fixes 1bfde5f (#214467)

This fixes 1bfde5fd3b8f809fc7ab1ce32862276144783ef5 (#213648).

Buildkite error link:
https://buildkite.com/llvm-project/upstream-bazel/builds?commit=1bfde5fd3b8f809fc7ab1ce32862276144783ef5

Co-authored-by: Google Bazel Bot <[email protected]>
https://invent.kde.org/qt/clang/llvm-project/-/commit/8caa6c19ace019317008a981f4469260f22dbc84

Git commit 1599b010ffe5e341ef643759f22d797ff1d91ff4 by GitHub (on behalf of Ian Li) on 06/08/2026 at 13:50..
[Clang] Add missing colons to Preprocessor lit test (#214342)

PR #211512 introduced new lit tests that are missing colons. This PR
adds them back so the lit tests introduced actually run.
https://invent.kde.org/qt/clang/llvm-project/-/commit/1599b010ffe5e341ef643759f22d797ff1d91ff4

Git commit c90a5cbb0d3366526a5d44032b06b3e624a5249b by GitHub (on behalf of Joseph Huber) on 06/08/2026 at 14:00..
[libclc] Improve libclc device tests with feature checks (#214217)

Summary:
Make sure the tests can conditionally run on the correct features.

I think that it would be nice if we could print from these tests, but
that would require linking against `libc` for GPUs and I'm unsure if
that's worth it, so for now I'm just sticking with traps as the failure
mechanism.
https://invent.kde.org/qt/clang/llvm-project/-/commit/c90a5cbb0d3366526a5d44032b06b3e624a5249b

Git commit ce880d0a4fe7679658de0377542ac039992447cb by GitHub (on behalf of Nick Sarnie) on 06/08/2026 at 14:03..
[offload][lit] Run unit tests in check-offload if libomptarget isn't built (#214334)

Right now we define no top-level `check-offload` target if
`libomptarget` isn't built.
Instead, just define the target to run the unit tests instead instead of
running both the unit tests and the `libomptarget` tests we would do if
`libomptarget` were built.
It's really useful to have a top-level target to run tests, especially
for CI.
This target will be used by our Windows buildbot, as `libomptarget`
isn't supported on Windows.

---------

Signed-off-by: Nick Sarnie <[email protected]>
https://invent.kde.org/qt/clang/llvm-project/-/commit/ce880d0a4fe7679658de0377542ac039992447cb

Git commit 4442d7022c94ff041b65236670243014fa40f4d1 by GitHub (on behalf of EuphoricThinking) on 06/08/2026 at 14:08..
[offload] add parameterized unit tests (#209115)

Previously, liboffload tests were parametrized only with different
devices. The following changes enable creating tests with additional
parameters, besides devices. The existing tests, which often used
C-style macros instead of incorporating Google Test features, are
rewritten so that macros are replaced fully with Google Test
parametrized fixtures.

This patch adds:
- a new fixture that allows for creating parameterized tests
- new macros for the instantiation of parameterized tests
- a helper header file that contains information related to
parameterized tests, including the definition of parameters
- new and refactored printers used in test instantiation

Moreover, selected existing tests are modified and parametrized in order
to make the code less repetitive and more concise. Tests with up to two
possible parameters are not parameterized.

Previously, the tests were already parameterized with
`TestEnvironment::Device`. However, this patch combines
`TestEnvironment::Device` with an additional parameter within
`OffloadParam = std::tuple<TestEnvironment::Device, T>`.
`OffloadParam<T>` is handled by a new fixture
`OffloadDeviceTestWithParam<T>`, which provides the `getTestParam()`
method for the parameter extraction. Other functionalities, such as
access to the `Host` and to the `Device`, are identical to the previous
version of `OffloadDeviceTest`.

In order to avoid code duplication, the unparameterized versions of
fixtures are aliases for parameterized fixtures, with `int` type chosen
arbitrarily as an ignored parameter type, for example: `using
OffloadDeviceTest = OffloadDeviceTestWithParam<int>;`. The single mock
parameter of value `0` is combined with the devices in the provided
macros, yielding tuples of type `std::tuple<TestEnvironment::Device,
int>`. The hidden `int` parameter is not used, but it enables users to
instantiate unparameterized tests without knowledge about the
implementation details. Moreover, it allows for modifying only one
version of the fixture (the one being aliased), without the need to also
change the other version: either parameterized or unparameterized.

The following macros are added in this patch:
- `OFFLOAD_TESTS_INSTANTIATE_DEVICE_FIXTURE_WITH_PARAM(FIXTURE, VALUES,
PRINTER)`: instantiates `FIXTURE` with parameters stored in the provided
container `VALUES` (a C-style array or an STL-style container) and with
the given printer. The used devices do not include the host.
- `OFFLOAD_TESTS_INSTANTIATE_HOST_DEVICE_FIXTURE_WITH_PARAM(FIXTURE,
VALUES, PRINTER)`: similar to
`OFFLOAD_TESTS_INSTANTIATE_DEVICE_FIXTURE_WITH_PARAM`, but includes the
`Host` in the tested devices.
- `OFFLOAD_TESTS_INSTANTIATE_HOST_DEVICE_FIXTURE(FIXTURE)`: instantiates
`FIXTURE` without additional parameters, but also includes the `Host` in
the used devices.
- `OFFLOAD_TESTS_INSTANTIATE_WITH_DEVICES(FIXTURE, DEVICES)`:
instantiates `FIXTURE` with the provided devices, internally using a
mock parameter. It is not intended for direct use and acts as a helper
macro for the instantiation of unparameterized tests.
- `OFFLOAD_TESTS_INSTANTIATE_WITH_DEVICES_WITH_PARAM(FIXTURE, VALUES,
DEVICES, PRINTER)`: instantiates `FIXTURE` with the provided devices,
parameters and the given printer. It is not intended for direct use and
acts as a helper macro for the instantiation of parameterized tests.

Since the new macros enable testing the `Host` as part of the used
devices, `offload/unittests/OffloadAPI/device/olGetHostInfo.cpp` is
deleted and its tests have been moved to
`offload/unittests/OffloadAPI/device/olGetDeviceInfo.cpp`.

The new helper header
`offload/unittests/OffloadAPI/common/Properties.hpp`:
- pairs properties with their sizes for tests similar to
`offload/unittests/OffloadAPI/device/olGetDeviceInfo.cpp`
- packs selected properties into containers or arrays for easier access
and instantiation of fixtures
https://invent.kde.org/qt/clang/llvm-project/-/commit/4442d7022c94ff041b65236670243014fa40f4d1

Git commit 6348771d4f563fabd22338e772fd0cad394c80c5 by GitHub (on behalf of Juan Manuel Martinez Caamaño) on 06/08/2026 at 14:21..
[AMDGPU][LibCallSimplify] Use target type's float-semantics in `ConstantFP::get` (#213721)

Compiler was crashing with:

```
Constants.cpp:1124: static llvm::ConstantFP* llvm::ConstantFP::get(llvm::Type*, const llvm::APFloat&):
    Assertion `Ty->getScalarType() == Type::getFloatingPointTy(Cont ext, V.getSemantics()) &&
    "ConstantFP type doesn't match the type implied by its value!"' failed.
```

Since the code was quite similar to `getConstantFloatVector`, I've ended
up modifying its implementation to also handle scalars and renamed it.
https://invent.kde.org/qt/clang/llvm-project/-/commit/6348771d4f563fabd22338e772fd0cad394c80c5

Git commit 848dfb98225281d68e9682e9eebc658ad7a4ff2c by GitHub (on behalf of Brian Cain) on 06/08/2026 at 14:26..
[Hexagon] Do not sign extend the unsigned comparison of short integers. (#212914)

HexagonTargetLowering::LowerSETCC widens i8/i16 SETCC operands to i32
with a sign extension when the extension is free, or when the compared
constant is negative in the narrow type. That is what the compare
instructions want for equality comparisons, since they can encode small
negative immediates, but for unsigned comparisons a constant with the
sign bit of the narrow type set becomes a large 32-bit value that has to
be materialized in a register or use a constant extender. For

  %v = load i16, ptr %p
  %c = icmp ult i16 %v, 65524

we generated

  r1 = #-12
  r0 = memh(r0+#0)
  p0 = cmp.gtu(r1,r0)

instead of

  r0 = memuh(r0+#0)
  p0 = cmp.gtu(r0,##65523)

Restrict the transformation to equality condition codes and let the
generic operand promotion pick the extension for everything else. Signed
comparisons are unaffected, as the generic promotion already
sign-extends them.

Note that only the quality of the generated code was affected: sign
extension preserves the unsigned ordering of the values of the narrower
type, so the code produced before this change was correct.

Co-authored-by: Sumanth Gundapaneni <[email protected]>
https://invent.kde.org/qt/clang/llvm-project/-/commit/848dfb98225281d68e9682e9eebc658ad7a4ff2c

Git commit a6b0af7536ef0ae8383ec729c5fbf23c73243776 by GitHub (on behalf of Kyungtak Woo) on 06/08/2026 at 14:29..
[NFC] Add missing <cmath> and <limits> includes across llvm, mlir, lldb, flang, and bolt (#214410)

This is basically a continuation of
https://github.com/llvm/llvm-project/pull/214349...

This change adds missing `#include <cmath>` and `#include <limits>`
headers to several files across `llvm`, `mlir`, `lldb`, `flang`, and
`bolt`.

In commit
https://github.com/llvm/llvm-project/commit/ada3786e91ca2058f3ac8255a024c12ae7d263ee,
`<random>` stopped transitively pulling in the top-level `<cmath>`
header in favor of internal granular `<__math/...>` headers.
Multiple files across the codebase were implicitly relying on transitive
`<cmath>` inclusions from headers like
`llvm/Support/RandomNumberGenerator.h` and other headers that pull in
`<random>`. I'm explicitly adding the missing dependency.

### Affected Files
* **bolt**: `bolt/lib/Passes/SplitFunctions.cpp` (`std::pow`)
* **flang**: `flang/lib/Optimizer/CodeGen/Target.cpp` (`std::ceil`)
* **lldb**: `lldb/source/Plugins/Language/ObjC/Cocoa.cpp` (`std::floor`)
* **llvm**:
  * `llvm/lib/Transforms/Utils/Debugify.cpp` (`std::log10`)
  * `llvm/lib/Transforms/Utils/LowerMemIntrinsics.cpp` (`std::round`)
* `llvm/tools/llvm-dwarfdump/Statistics.cpp` (`std::round`,
`std::numeric_limits`)
  * `llvm/tools/llvm-exegesis/lib/Analysis.cpp` (`std::sqrt`)
  * `llvm/tools/llvm-exegesis/lib/ResultAggregator.cpp` (`std::ceil`)
* `llvm/tools/llvm-exegesis/lib/SchedClassResolution.cpp` (`std::round`)
* **mlir**:
* `mlir/include/mlir/Dialect/Quant/Utils/UniformSupport.h`
(`std::round`)
  * `mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp` (`std::log2`)
* `mlir/lib/Target/LLVMIR/Dialect/NVVM/NVVMToLLVMIRTranslation.cpp`
(`std::log2`)
https://invent.kde.org/qt/clang/llvm-project/-/commit/a6b0af7536ef0ae8383ec729c5fbf23c73243776

Git commit 27f1aa4c9a42b80ffb7ae688d9e5dd4245a4318d by GitHub (on behalf of Wiktor Ludwichowski) on 06/08/2026 at 14:41..
[X86][AsmParser] Support parentheses around scale and index register in Intel syntax (#211003)

Old patterns and why they broke
 
- Decisions matching (CurrState, PrevState) tuples (e.g. CurrState ==
IES_REGISTER && PrevState != IES_MULTIPLY)
to gate committing a register to Base/Index aren't paren-transparent:
any (...) turns CurrState into IES_RPAREN, so the tuple stops matching
and
`[reg + (reg)]`, `[reg + (reg*2)]`, `[reg + 2*(reg)]` fell through the
guards.
- Duplicated commit blocks. Recognizing whether a register is Base or
Scale was placed in four
handlers (onPlus, onMinus, onRBrac, onRParen), forcing onRParen to
replay it just to catch (reg).
- Scale was committed immediately in onInteger; a parenthesized scale
like (2) * 4 or 2 * (4)
    had nowhere to accumulate.
 
New patterns
 
- Pending value. Two pending fields: TmpReg and TmpScale, that store the
register/scale which is currently
being processed. Commit happens only at (+, -, ]), therefore the
solution is parenthesis-independent and
    can handle more scale expressions.
- Identical in onPlus/onMinus/onRBrac: !BaseReg && !TmpScale → BaseReg
(unscaled);
otherwise → IndexReg with Scale. onRParen no longer commits, parens are
purely an infix-calculator concern.
- Multiplicative scale accumulator onInteger. onStar on IES_INTEGER
seeds it from the popped operand;
reset at +, -, [, ]. Handles 2 * 4, (2) * 4, 2 * (4), 2 * 2 * 2, (2 * (2
* rbx)) * 2, etc.
- Conflict detection. Pending state makes two invariants diagnosable:
TmpReg already set when a
new register arrives via * or ( means multiplication of two registers;
NegativeAdditiveTerm
at commit means negative scale. Both previously misassembled silently.
https://invent.kde.org/qt/clang/llvm-project/-/commit/27f1aa4c9a42b80ffb7ae688d9e5dd4245a4318d

Git commit 798984232ffecb0336645089a057cd05a8bdc20b by GitHub (on behalf of Chris Copeland) on 06/08/2026 at 14:44..
[libc] Add missing type/macro headers for generated public headers (#213395)
https://invent.kde.org/qt/clang/llvm-project/-/commit/798984232ffecb0336645089a057cd05a8bdc20b

Git commit 9fd1052125a393b8aa16fa3e8411168332b22610 by GitHub (on behalf of Durgesh Nandan Mohanty) on 06/08/2026 at 14:50..
[AArch64][CodeGen] Fold tbx(splat(0), table, idxs) to tbl(table, idxs) (#214268)

When the destination/background vector of a TBX instruction is a splat
of zero, the operation is equivalent to a TBL instruction. TBL
implicitly zeroes out any elements where the index is out of bounds,
matching the behavior of TBX with a zero background vector.

This patch adds a DAG combine to optimize this case, reducing
instruction latency and register pressure.

Fixes #214077
https://invent.kde.org/qt/clang/llvm-project/-/commit/9fd1052125a393b8aa16fa3e8411168332b22610

Git commit bc00b3aae17eb8d3f1dee2e8fbbdf5c3e4a36546 by GitHub (on behalf of Gaëtan Bossu) on 06/08/2026 at 15:04..
[AArch64][CostModel] Explicitly disable histograms for nxv1 types (#214470)

Currently, histograms are not disabled for vscale x 1 types, but LV will
implicitly not generate them because surrounding arithmetic instructions
are invalid for vscale x 1.

SDAG does not implement operand widening for histograms, so this patch
explicitly disables those cases.
https://invent.kde.org/qt/clang/llvm-project/-/commit/bc00b3aae17eb8d3f1dee2e8fbbdf5c3e4a36546

Git commit 708dafb270b5f6971c6666cc0afe96d227fb198d by GitHub (on behalf of Arseniy Obolenskiy) on 06/08/2026 at 15:05..
[SPIR-V] Build vendored SPIRV-Tools in Release mode (#214159)

Default build mode in SPIRV-Tools is Debug
https://invent.kde.org/qt/clang/llvm-project/-/commit/708dafb270b5f6971c6666cc0afe96d227fb198d

Git commit ac9e7f5e9d9170a56945c716b6c87126a3d0963d by GitHub (on behalf of Zeyi Xu) on 06/08/2026 at 15:13..
[clang-tidy][docs] Rename bugprone check docs to Markdown [1/4] (#214412)

Tracking issue: #201242

See the [migration guide] for more information.

[migration guide]:

https://llvm.org/docs/SphinxQuickstartTemplate.html#markdown-migration-guidelines

This is the mechanical rename for part 1/4 of the remaining `bugprone`
check documentation.
The rewrite is provided by the next PR in this stack.
https://invent.kde.org/qt/clang/llvm-project/-/commit/ac9e7f5e9d9170a56945c716b6c87126a3d0963d

Git commit 11ba1b712acbee90dcc07b36467a7b2bb5a9b07b by GitHub (on behalf of Zeyi Xu) on 06/08/2026 at 15:13..
[clang-tidy][docs] Rewrite bugprone check docs to Markdown [1/4] (#214413)

Tracking issue: #201242

See the [migration guide] for more information.

[migration guide]:

https://llvm.org/docs/SphinxQuickstartTemplate.html#markdown-migration-guidelines

This rewrites part 1/4 of the remaining bugprone check documentation
from reST to MyST Markdown.

AI Usage: This was prepared with rst2myst and GPT5.6-assisted cleanup.
I manually verified that the documentation renders as expected.

Preview site:
https://broken.life/llvm-staging/bugprone-markdown-port/
https://invent.kde.org/qt/clang/llvm-project/-/commit/11ba1b712acbee90dcc07b36467a7b2bb5a9b07b

Git commit c7317d79aacf76b5eb2dc65010e574a5269838ef by GitHub (on behalf of Erich Keane) on 06/08/2026 at 15:28..
[CIR][MLIR-LLVM] Add 'thread local mode/model' attribute support (#214337)

The main purpose of this patch is to implement the 'tls_model' attribute
for ClangIR. However, this required updating the LLVM-IR dialect to ALSO
support this, not just as a bool. This patch threads it into both sides.

CIR tries to refer to it as "TLS_Model" to match the C/C++ attribute
closely, the LLVM Dialect refers to it as ThreadLocalMode to reflect
better what LLVM does.

Left as still 'not done' (in CIR only!) is the lowering of the 'other'
kinds of thread-local settings, which are intended to do various levels
of locking/initialization. Those are left for a future implementation
effort. The C++ test itself is taken directly from classic codegen.
https://invent.kde.org/qt/clang/llvm-project/-/commit/c7317d79aacf76b5eb2dc65010e574a5269838ef

Git commit 0a9cfa09242aa1cc162e55ddf1bbcc50664d763d by GitHub (on behalf of Andy Kaylor) on 06/08/2026 at 16:14..
[mlir][llvm] Add operations for coroutine intrinsics (#214099)

This adds operations to the LLVM dialect to represent coroutine
intrinsics that were previously missing from the dialect. A few
coroutine operations were already in place. This change adds operations
for the remaining intrinsics that cab be generated by Clang. There are
some additional coroutine intrinsics defined in LLVM IR that aren't
covered, but I'm omitting those until they are needed.

We are in the process of implementing coroutine support in CIR, and
these are the operations we'll need for lowering to LLVM.

Assisted-by: Cursor / Grok 4.5
https://invent.kde.org/qt/clang/llvm-project/-/commit/0a9cfa09242aa1cc162e55ddf1bbcc50664d763d

Git commit 3bace2206a56e39766a0697d1d63b05f537793a6 by GitHub (on behalf of Andy Kaylor) on 06/08/2026 at 16:15..
[CIR][AArch64] Update builtin handlers to use emitNeonCallToOp (#214075)

This is another change to prepare AArch64 builtin handling for the
transition to constrained FP handling. It replaces a number of places
where we were creating CIR operations directly with calls to
emitNeonCallToOp so that we will be able to centralize the constrained
FP handling.

This also updates the vrndns_f32 to eliminate a redundant load of the
operand, which is the only part of this change with a visible difference
in the output.

Assisted-by: Cursor / Grok 4.5
https://invent.kde.org/qt/clang/llvm-project/-/commit/3bace2206a56e39766a0697d1d63b05f537793a6

Git commit 02b08387ae43a19b5f2817641d36b9398a0e19e5 by GitHub (on behalf of Andy Kaylor) on 06/08/2026 at 16:16..
[mlir][llvm] Add more constrained FP operations (#213745)

This change adds special constrained forms of transcendental operations
for the remaining cases that lower to contrained fp intrinsic calls. It
also adds fast-math flag support to the constrained operations, which is
needed to handle combinations of Clang command-line options such as
"-ffinite-math-only -ftrapping-math".

Assisted-by: Cursor / various models
https://invent.kde.org/qt/clang/llvm-project/-/commit/02b08387ae43a19b5f2817641d36b9398a0e19e5

Git commit b2ba87ae2551808b36fc74dfac8199d54c043b4a by GitHub (on behalf of Jonas Devlieghere) on 06/08/2026 at 16:20..
[lldb] Split DynamicLoader binary loading into locate and load (NFCI) (#214372)

LoadBinaryWithUUIDAndAddress both searched for a binary and registered
it with the Target. Split it into LocateBinaries, which only searches,
and LoadBinaryInTarget, which mutates the Target, with
LocateAndLoadBinary keeping the single binary case a one-liner.

The eight binary parameters and the results of the search are bundled in
a new BinarySpec struct, and both entry points return an llvm::Expected.

The motivation is a follow-up that runs LocateBinaries in parallel on
the thread pool. NFC, except for some small improvements to the error
handling because we don't write to the async output stream directly (and
fixed the newline).
https://invent.kde.org/qt/clang/llvm-project/-/commit/b2ba87ae2551808b36fc74dfac8199d54c043b4a

Git commit 3705e63f8099e46c5a29ffef91a1cde9a91de392 by GitHub (on behalf of Daniel Chen) on 06/08/2026 at 16:20..
[flang] Fix incorrect offset for zero-size COMMON block members. (#214174)

Fixes #214171 

`ComputeOffsetsHelper::DoSymbol()` in
`flang/lib/Semantics/compute-offsets.cpp` returned early without calling
`symbol.set_offset()` when a symbol had zero size (e.g. CHARACTER*0). As
a result, every zero-size symbol in a COMMON block retained its default
offset of 0 — the block base address — instead of its correct sequential
position.

This incorrect offset caused two observable bugs:

1. **Wrong storage address**: LOC() and lowering always returned the
block base address for zero-size members instead of their actual
sequential position.

2. **False "cannot backward-extend" error**: When a zero-size COMMON
block member appeared in an EQUIVALENCE association, the backward-extend
check (`dep.offset > symbol.offset()`) incorrectly fired because
`symbol.offset()` was always 0. For example, the following valid code
was falsely rejected:
   ```fortran
   program p09
     integer(8) :: i8
     character(0) :: zc0
     character(8) :: c8
     common /blk/ i8, zc0
     equivalence (c8(5:8), zc0)
   end program

Error before fix:

`error: 'zc0' cannot backward-extend COMMON block /blk/ via EQUIVALENCE
with 'c8'`

After the fix, `zc0` is correctly assigned offset 8 (after `i8`), so
`dep.offset` (4) < `symbol.offset()` (8) and no error is raised.
https://invent.kde.org/qt/clang/llvm-project/-/commit/3705e63f8099e46c5a29ffef91a1cde9a91de392

Git commit 4e95a2df34548e032b3a93eef431b73db24839d7 by GitHub (on behalf of Kamlesh Kumar) on 06/08/2026 at 16:27..
[CodeGen] Add scalarization for vector (de)interleave (#212547)

This is required when NEON is disabled with -mattr=-neon.
https://invent.kde.org/qt/clang/llvm-project/-/commit/4e95a2df34548e032b3a93eef431b73db24839d7

Git commit 2b32e13e65a35fa8f757eee89992e172f5cb0da5 by GitHub (on behalf of Krzysztof Drewniak) on 06/08/2026 at 16:37..
[SDAG] Handle vector expansion when expanding FABS (#214341)

Before this, vector FABS that were scheduled for expansion would hit the
getSignAsIntValue() case and either assert there or fail later for lock
of instruction selection for a bitcast that's only looking at the scalar
size.

Now, we unroll to scalars as needed.

Test pre-committed in #214288
https://invent.kde.org/qt/clang/llvm-project/-/commit/2b32e13e65a35fa8f757eee89992e172f5cb0da5

Git commit 05ecef4367dad077961380d643c83db83794f023 by GitHub (on behalf of Anshil Gandhi) on 06/08/2026 at 16:38..
[AMDGPU][GlobalISel] Pre-commit tests for readanylane merge regbank combine (NFC) (#214355)

Add regbank-combiner tests covering a copy to vgpr whose source is a
merge or
build_vector of `G_AMDGPU_READANYLANE` results mixed with uniform
values.

These currently keep the round trip through sgprs. The tests also cover
the two
cases where the transform must not fire:

- the sgpr merge has another user, so it has to be kept;
- all merge sources are uniform, so moving the copy to the sources would
not
  remove any readanylane.

Pre-commit only, no functional change. The combine that removes the
round trip
is in the stacked PR.

Made with [Cursor](https://cursor.com)

---------

Co-authored-by: Cursor <[email protected]>
Co-authored-by: Matt Arsenault <[email protected]>
https://invent.kde.org/qt/clang/llvm-project/-/commit/05ecef4367dad077961380d643c83db83794f023

Git commit 79bc07942adcb4819d30075890c3e8cd75b2d8c1 by GitHub (on behalf of Alok Kumar Sharma) on 06/08/2026 at 16:48..
Reland "[MergeFunctions] Preserve instruction-level profile metadata during merging" (#208009) (#210138)

This relands #208009, which was reverted in #209987 after an ASan
heap-use-after-free surfaced in MergeFunctionsTest.TrueOutputModuleTest.

The failure was caused by MergeFunctionsTest destroying
FunctionAnalysisManager before ModuleAnalysisManager, while MAM holds a
cached proxy result that calls FAM.clear() on destruction. This PR adds
a commit reordering those members, so they are destroyed in the correct
order, fixing the use-after-free.

Original PR: #208009
Revert PR: #209987
https://invent.kde.org/qt/clang/llvm-project/-/commit/79bc07942adcb4819d30075890c3e8cd75b2d8c1

Git commit bd586dd26e3ba49616175c14cb2970c581fb94b7 by GitHub (on behalf of Jonas Devlieghere) on 06/08/2026 at 16:51..
[lldb] Use the standard GDB remote thread for a non-Wasm process (#214380)

CanDebug returns true whenever the plugin is requested by name, and the
architecture is not known until the stub reports it after connecting.
This means that a non-Wasm process can end up with a ThreadWasm whose
register context and unwinder have nothing to operate on.

Create the plain ThreadGDBRemote once the architecture is known, and add
a helper so that the check covers wasm64 as well as wasm32.

rdar://182229301
https://invent.kde.org/qt/clang/llvm-project/-/commit/bd586dd26e3ba49616175c14cb2970c581fb94b7