Skip to content

Sycl web - #21545

Closed
zahiraam wants to merge 4198 commits into
syclfrom
sycl-web
Closed

Sycl web#21545
zahiraam wants to merge 4198 commits into
syclfrom
sycl-web

Conversation

@zahiraam

Copy link
Copy Markdown
Contributor

No description provided.

JDevlieghere and others added 30 commits March 12, 2026 14:36
…5541)

In #162162, I added support for OSC 9;4 graphical progress. I put it
behind the `show-progress` setting because I didn't have a reliable way
to detect whether the escape code was supported by the terminal.

Since then, more tools have added support for it, most notably Claude
Code and Homebrew. While I still don't have a good way to detect this,
there are a handful of known terminals that are easy enough to identify.

This PR toggles the default of `show-progress` to on again and puts
showing the progress behind a check for those known terminals (Windows
Terminal, ConEmu & Ghostty).

This means that if you're running in one of those, you'll get the visual
progress by default unless you set `show-progress` to off. The downside
is that if you're on an unrecognized terminal, you can't force it on any
longer by setting `show-progress` to on. I think that's a fair trade-off
as the setting wasn't really advertised and I doubt many people are
using that. As a workaround, they can set `OSC_PROGRESS` to spoof an
OSC-supporting terminal.
This PR adds WaveActiveBitXor function to HLSL, with spirv and DXIL code
generation.
Fixes llvm/llvm-project#99168
Use the same format string in DWARFCompileUnit.cpp as we do everywhere
else to report the error from GetAttributeAddressRanges. I noticed the
inconsistency when looking at our system log:

```
2026-03-12 05:36:48.600914 -0700	0x2e66: DIE has no address range information
2026-03-12 05:36:48.600950 -0700	0x2e8c: DIE has no address range information
2026-03-12 05:36:48.601068 -0700	DIE(0x1dc2): DIE has no address range information
2026-03-12 05:36:48.631000 -0700	DIE(0x80f4): DIE has no address range information
```
Instrumentation emitted for overflow by division was not checking with the sanitizer case list's type entries.

The original type-based ignorelist support (#107332) added `isTypeIgnoredBySanitizer` calls to `CanElideOverflowCheck`, which covers `+`, `-`, `*`, `++`, `--`. However, division and remainder have a separate code path in `EmitUndefinedBehaviorIntegerDivAndRemCheck` that never calls `CanElideOverflowCheck` or checks the ignorelist directly.

Add a check so that the SCL is honored for the div/rem case.
Some VPlan transforms, like vectorizing fmin without fast-math,
introduce AnyOfs, which have costs assigned in the VPlan-based cost
model, but not the legacy cost model. Account for their cost like done
for other similar VPInstrctions, like EVL.

Fixes llvm/llvm-project#185867.
…#186243)

This pass handles `acc routine` directive by creating specialized
functions with appropriate parallelism information that can be used for
eventual creation of device function.

For each acc.routine that is not bound by name, the pass creates a new
function (the "device" copy) whose body is a single acc.compute_region
containing a clone of the original (host) function body. Parallelism is
expressed by one acc.par_width derived from the routine's clauses (seq,
vector, worker, gang). The device copy created is simply a staging place
for eventual move to device module level function.

---------

Co-authored-by: Delaram Talaashrafi <dtalaashrafi@nvidia.com>
…loading (#185514)

Weak symbols may be overridden during linking, and this may cause
profile mismatch when compiling the weak symbols, while the profile was
created based on the overriding function. Skip counting the weak symbol
while checking the mismatched function profiles to avoid false alarm on
rejecting legit profiles.
…186023)

`iswdigit` was implemented
(llvm/llvm-project#181635) but not declared in
the generated wctype.h (wctype.yaml) and was missing from the Windows
config entrypoints.

This pr declares iswdigit in wctype.h and enables it for the Windows
config.

also is part of llvm/llvm-project#185136.
…inlining (#186266)

When the MLIR inliner inlines a callable region that has more than one
block, it calls `handleTerminator(op, Block *newDest)` for the
terminator of every inlined block. `TestInlinerInterface` only
implemented the single-block variant (`handleTerminator(op,
ValueRange)`), so the default `llvm_unreachable` was hit when inlining a
`test.functional_region_op` whose body contained multiple blocks (e.g.
an explicit `cf.br` jump to a successor block whose terminator was
`test.return`).

Fix: add the missing `handleTerminator(op, Block *)` override to
`TestInlinerInterface`. Mirror the pattern used by
`FuncDialectInlinerExtension`: if the terminator is a `TestReturnOp`,
replace it with a `cf.br` to `newDest` carrying the return operands. Any
other terminator (e.g. `cf.br` for intra-region branches) is left
untouched — the existing `ControlFlowInlinerInterface` no-op already
handles those correctly.

Add a regression test in `test/Transforms/inlining.mlir` that inlines a
`call_indirect` into a `test.functional_region_op` with two blocks.

Fixes #185350

Assisted-by: Claude Code
…5547)

Currently these functions take a target pointer. In Cross-language
projects this means it's down to chance what typesystem the resulting
value will be in, since the implementation returns the first scratch
type system in the target, which depends on the order of images and
their implementation language.

By passing in an execution context the selected frame is used to
determine the typesystem, which is usually the expected outcome when
using DIL.

This should be entirely NFC for Clang-only LLDBs, but is a necessity for
LLDBs with additional type system plugins such as the Swift plugin.

Assisted by Claude to patch the call sites.
Summary:
We use this `dyn_ptr` argument in Clang/OpenMP to handle the
`KernelLaunchEnvironment`. This is a per-kernel argument used to share
some information. Currenetly, it's prepended to the argument list and we
generate storage for it in the runtime.

This is bad for a few reasons:
1. It changes the ABI by shifting user arguments
2. It cannot be trivially be left uninitialized if unused
3. The runtime must allocate its own memory for it

This PR changes it to be appended instead. Additionally, space for this
is always emitted. This means the OMPIRBuilder itself will provide the
storage, we simply need to populate it in the runtime if it is used.
This means that if it's unused we don't always pay the cost and it's
easier for non-OpenMP users to ignore it.

Backward compatibility is maintained by auto-upgrading the kernel
arguments. In `libomptarget` we completely allocate a new buffer to
store this in the new format. The plugins still need to respect the old
ABI of the called device object, so we simply rotate it if it's the old
version.
…to be zero (#186267)

The zext_inreg test changes are interesting - we mostly get closer to
the VPBLENDW(XMM,0) pattern, but the fast-shuffle targets still fail to
recognize the upper half of the YMM is zero and don't simplify the
AND(YMM,C) as much as they could.

Helps with regression identified on #182200
Change the default of `LLDB_ENABLE_ARM64E_DEBUGSERVER` so we build
debugserver fat for arm64 and arm64e when running on macOS Tahoe (26) or
later on ptrauth-capable hardware.
WaitcntGenerator state depends on MF attributes, so create a new WCG object per MF
until we have a better solution. This patch also adds a test that exercises this.

Even though we stopped creating a new WCG instance in #177689, the behavior
didn't change because SIInsertWaitcnts gets recreated on every MF
(so this patch is practically an NFC).
Remove `NoNaNsFPMath` uses in X86 part. This should be the last part.
Fixes a redefinition warning on Ubuntu 25.10 coming from `ptrace.h`
including `hwcap.h`.

```
/home/jonas/llvm/llvm-project/lldb/source/Plugins/Process/Linux/NativeRegisterContextLinux_arm64.cpp:84:9: warning: ‘HWCAP2_POE’ redefined
   84 | #define HWCAP2_POE (1ULL << 63)
      |         ^~~~~~~~~~
In file included from /usr/include/aarch64-linux-gnu/asm/ptrace.h:25,
                 from /home/jonas/llvm/llvm-project/lldb/source/Plugins/Process/Linux/NativeRegisterContextLinux_arm64.h:19,
                 from /home/jonas/llvm/llvm-project/lldb/source/Plugins/Process/Linux/NativeRegisterContextLinux_arm64.cpp:11:
/usr/include/aarch64-linux-gnu/asm/hwcap.h:141:9: note: this is the location of the previous definition
  141 | #define kkk              (1UL << 63)
      |         ^~~~~~~~~~
```
This extends the cir.delete_array lowering code to introduce a loop that
calls destructors when the array being deleted represents a destructed
type. The lowering introduces the destructors by way of a cir.array.dtor
operation, which is further expanded during LoweringPrepare. This also
required updating the cir.array.dtor operation to accept a raw pointer
to the element type and a value representing the number of elements to
be destructed.

This does not yet handle the possibility of destructors throwing
exceptions.
…186055)

Some trailing kill/dead register operands may added by
MachineInstr::addRegisterKilled or MachineInstr::addRegisterDead, which
uses the overlap registers same with the operand 1-4.

Let's remove them here as only 5 operands are assert existing.
…939)

Our use case requires making two (or more) hashes for the same input but
with different keys.
The library already supports this - we just need to expose the API
Add the architecture-specific pieces needed for the ASan and UBSan
sanitizer runtimes to build and run on hexagon-unknown-linux-musl.

Without this patch, building sanitizer runtimes for Hexagon Linux fails
with:

  sanitizer_linux.cpp: error: member access into incomplete type
      'struct stat64'

because musl libc does not provide struct stat64. This patch routes
Hexagon through the statx() syscall path (like LoongArch) to avoid the
stat64 dependency entirely.

Changes:

* asan_mapping.h: Add ASAN_SHADOW_OFFSET_CONST (0x20000000) for Hexagon
with shadow layout documentation.
* sanitizer_linux.cpp: Implement internal_clone() for Hexagon using
inline assembly (trap0 syscall, generic clone argument order: flags,
stack, ptid, ctid, tls). Route Hexagon through the statx() path for stat
operations since musl lacks struct stat64.
* sanitizer_linux.h: Add Hexagon to the internal_clone() declaration
guard.
* sanitizer_stoptheworld_linux_libcdep.cpp: Add Hexagon to the
StopTheWorld architecture guard with register definitions.
* sanitizer_asm.h: Define ASM_TAIL_CALL as 'jump' for Hexagon.
* CMakeLists.txt: Add -fno-emulated-tls for Hexagon targets. Hexagon
Linux uses native TLS via the UGP register; emulated TLS produces broken
sanitizer runtimes with unresolvable __emutls references.
TranslateOpenMPTargetArgs drops OPT_m_Group options, including
`-mlinker-version`, when the device triple differs from the host triple.
This can cause an assertion failure in MachO::getLinkerVersion when
running `offload-Xarch.c` and `openmp-offload-gpu.c` on Darwin as it
expects the linker version obtained via `-mlinker-version` and the
version cached during a previous invocation to match.

Fix this by preserving `-mlinker-version` when filtering OPT_m_Group
options.

The tests currently pass because 6758bec ("[ObjC] Support emission
of selector stubs calls instead of objc_msgSend."), which triggered the
assertion, was reverted. This fix is a prerequisite for reapplying that
commit.
Adds CIRGen support for a subset of the __sync_fetch_and_* builtins,
including arithmetic (add, sub) and bitwise (and, or, xor, nand)
variants.
…ng (#186172)

Now we need to provide a location when call `load()`, e.g.
```python
with Context(), Location.unknown():
  MyDialect.load()
```

But it's actually weird: IRDL is just an implementation details, so for
users they don't know why they need to provide a location for loading a
dialect, which is unrelated to constructing an IR module.

This PR made location optional for dialect loading.
…ch as possible (#179178)"

This reverts commit 8eb0dfe.

Breaks some clang header module targets. See the original PR for
discussion/reproducers.
JDevlieghere and others added 28 commits March 13, 2026 16:14
…r (#186326)

Collect allocation traces for tagged memory when using the
`darwin-mte-launcher` to help debug MTE crashes.
The callee_type metadata is expected to be a list of generalized type
metadata by the IR verifier. But for indirect calls with internal
linkage the type metadata is just an integer. Avoid including them in
callee_type metadata.

This will reduce the precision of the generated call graph as the edges to internal linkage functions whose address were taken will not be present anymore. We need to handle this in the future.
…823)

Restrict PHI nodes that getRangeRef is allowed to recursively examine so
we don't need a "visited" set. And fix createSCEVIter so it creates all
the relevant SCEV nodes before getRangeRef tries to examine them.

The tests that are affected have induction variables that aren't
AddRecs. (Other cases are theoretically affected, but don't seem to show
up in our tests.)
Summary:
These linked to the extern "C" versions which did not exist in test
builds.
Add a bit to TargetInfo to specify that vectors are element-aligned
rather than naturally aligned. This is needed to match DirectX's Data
Layout in LLVM.

Note that this removes the `Opts.HLSL` early exit from
`checkDataLayoutConsistency` so that we actually get these checks when
compiling HLSL. This check looks like it was put there because of
similarity between OpenCL and HLSL, but it isn't actually necessary.

Resolves #123968
The test expects crash after commit essentially null-dereferencing.
Just check that it's nullptr directly.

Fixes asan/ubsan buildbot.
Follow-up of 9b96ebc. There are binary_def.inc and unary_def.inc in
header directory.
- clc_ep.inc -> clc_ep_decl.inc
- relational/binary_def.inc -> relational/relational_binary_def.inc
- relational/unary_def.inc -> relational/relational_unary_def.inc
…mmediates (#185643)

This is a follow-up to #182456. This PR adds support for short forward
branches where branches are from Qualcomm uC `Xqcibi` extension.
…6538)

On some systems, looking up an result in a timeout.

Error here is not a sign of compiler-rt issue.

Fixes flakiness on
https://lab.llvm.org/buildbot/#/builders/sanitizer-ppc64le-linux
…erated (#186516)

Update lowering to not process operands if the operation will not be
generated because of a `if(.false.)` clause. This was breaking when
enabling CUDA Fortran on the same test.
I've been FreeBSD src contributor since 2022 and started working for the
FreeBSD Foundation starting from January. I created and have been
actively working on #180061 to achieve [LLDB improvement on
FreeBSD](https://wiki.freebsd.org/SummerOfCodeIdeas#Improve_LLDB_on_FreeBSD)
which was originally Google Summer of Code idea.

The initial LLDB support for userspace and kernel debugging on FreeBSD
was initiated by Moritz system back in 2020, but there are some missing
bits like architecture support due to lack of time. Due to this, FreeBSD
developers still depend on GDB-derived kernel debugger and it still
remains de-facto tier 1 debugger (e.g. no scripts in source tree for
lldb).

To track what/how I'm working on this task, please take a look at my
foundation [status report on LLDB
improvement](https://github.com/FreeBSDFoundation/status-updates/blob/main/Minsoo_Choo/kdb.md).
As you see, I will also improve Lua scripting on LLDB as Lua remains
most FreeBSD developer's choice for scripting language over Python.

Until school starts again in May, I will work full time in the
foundation which means I can actively create new features, fix bugs, and
review other's code changes. I will have less time to dedicate myself to
contribute to open source projects after May, but I will still commit
myself to code reviews and bug fixes that do not take too much time.

My FreeBSD contributions:
https://cgit.freebsd.org/src/log/?qt=author&q=Minsoo+Choo
My FreeBSD Phabricator profile:
https://reviews.freebsd.org/p/minsoochoo0122_proton.me/

Signed-off-by: Minsoo Choo <minsoochoo0122@proton.me>
Signed-off-by: Minsoo Choo <minsoochoo0122@proton.me>
…mment (#186544)

Align with cmake function name.
All other fields in these tracts are already initialized.
Fixes Msan report in DebugInfoCodeViewTests after #183172.

https://lab.llvm.org/buildbot/#/builders/sanitizer-x86_64-linux-fast
…lers (#185926)

[CUDA/HIP][SYCL] Deduplicate deferred diagnostics across multiple
callers

Deferred diagnostics for a function were emitted once per caller that
forced the function into device context. When multiple device functions
called the same host-device function containing errors, the diagnostics
were repeated for each caller, producing noisy duplicate output.

Change the deferred diagnostic emission to a two-pass approach:
1. During the call graph walk, collect callers in DeviceKnownEmittedFns
   (now storing multiple callers per function) and mark functions that
   need diagnostics, but don't emit yet.
2. After the walk completes, emit diagnostics once per function with
   all callers listed as notes.

Call chain notes now use "called by" for the first caller in each chain
and "which is called by" for subsequent callers in the chain, making it
easy to distinguish separate call chains.

Also add documentation for deferred diagnostics and the concept of
HD-promoted functions to the HIP and CUDA docs.

Fixes: llvm/llvm-project#180638
  CONFLICT (content): Merge conflict in clang/lib/Sema/SemaDecl.cpp
  CONFLICT (content): Merge conflict in libclc/CMakeLists.txt
  CONFLICT (content): Merge conflict in libclc/clc/lib/amdgpu/math/clc_exp.cl
  CONFLICT (content): Merge conflict in clang/lib/CodeGen/Targets/NVPTX.cpp
  CONFLICT (content): Merge conflict in clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
  CONFLICT (content): Merge conflict in libclc/CMakeLists.txt
  CONFLICT (add/add): Merge conflict in libclc/test/CMakeLists.txt
  CONFLICT (add/add): Merge conflict in libclc/test/lit.cfg.py
  CONFLICT (add/add): Merge conflict in libclc/test/lit.site.cfg.py.in
  CONFLICT (content): Merge conflict in clang/lib/Driver/ToolChain.cpp
  CONFLICT (content): Merge conflict in clang/lib/Sema/Sema.cpp
@zahiraam zahiraam closed this Mar 17, 2026
run: |
./llvm/utils/release/github-upload-release.py \
--token ${{ github.token }} \
--release ${{ inputs.release-version }} \

Check failure

Code scanning / zizmor

code injection via template expansion Error

code injection via template expansion
name: Test SYCL
runs-on: ubuntu-24.04
container:
image: ghcr.io/llvm/ci-ubuntu-24.04:latest

Check failure

Code scanning / zizmor

unpinned image references Error

unpinned image references
name: Test SPIR-V
runs-on: ubuntu-24.04
container:
image: ghcr.io/llvm/ci-ubuntu-24.04:latest

Check failure

Code scanning / zizmor

unpinned image references Error

unpinned image references
container:
image: ghcr.io/llvm/ci-ubuntu-24.04:latest
steps:
- uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1

Check warning

Code scanning / zizmor

credential persistence through GitHub Actions artifacts Warning

credential persistence through GitHub Actions artifacts
container:
image: ghcr.io/llvm/ci-ubuntu-24.04:latest
steps:
- uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1

Check warning

Code scanning / zizmor

credential persistence through GitHub Actions artifacts Warning

credential persistence through GitHub Actions artifacts
repo = github.Github(auth=github.Auth.Token("${{ github.token }}")).get_repo("${{ github.repository }}")
pr = repo.get_pull(${{ github.event.issue.number }})
comment = pr.get_issue_comment(${{ github.event.comment.id }})
add_text = "> _Running benchmarks in ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}/job/${{ job.check_run_id }}_"

Check notice

Code scanning / zizmor

code injection via template expansion Note

code injection via template expansion
run: |
digest_file="sha256"
echo "$INPUTS_DIGEST -" > $digest_file
cat ${{ steps.download-artifact.outputs.download-path }}/* | sha256sum -c $digest_file

Check notice

Code scanning / zizmor

code injection via template expansion Note

code injection via template expansion
- name: Rename attestation file
shell: bash
run: |
for f in ${{ steps.download-artifact.outputs.download-path }}/*; do

Check notice

Code scanning / zizmor

code injection via template expansion Note

code injection via template expansion
shell: bash
run: |
for f in ${{ steps.download-artifact.outputs.download-path }}/*; do
cp ${{ steps.provenance.outputs.bundle-path }} $(basename $f).jsonl

Check notice

Code scanning / zizmor

code injection via template expansion Note

code injection via template expansion
--token ${{ github.token }} \
--release ${{ inputs.release-version }} \
upload \
--files ${{ steps.download-artifact.outputs.download-path }}/* *.jsonl

Check notice

Code scanning / zizmor

code injection via template expansion Note

code injection via template expansion
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.