LLVM and SPIRV-LLVM-Translator pulldown (WW12 2026)#21560
Draft
LLVM and SPIRV-LLVM-Translator pulldown (WW12 2026)#21560
Conversation
Implement the OpenMPRequiredSimdAlign kind of UnaryExprOrTypeTraitExpr
The by-module-name scanning APIs are fairly spread out. There's the main `CompilerInstanceWithContext` class that provides a constructor, `initialize()` and `computeDependencies()`. Then there's the `DependencyScanningWorker` that optionally owns `CompilerInstanceWithContext` and re-exposes two `initialize()` and one `computeDependencies()` functions. Lastly, there's `DependencyScanningTool` that again re-exposes two variants of `initialize()` and one `computeDependencies()`. The current setup makes it unnecessarily difficult to make changes to these APIs (as observed in swiftlang/llvm-project#12453). This PR makes `CompilerInstanceWithContext` standalone, and hides the construct + initialize pattern behind a static factory function. This makes it harder to misuse the API (forgetting to call `initialize()`, calling it twice, etc.) and means changes now need to only touch single class instead of three classes spread over multiple files.
Summary: This was previously a Trieber stack, which is a perfectly fine generic and lock-free data structure. However, this used some expensive CAS operations and had issues with ABA. Because the only user of this was the slab cache mechanism, we can pretty safely specialize it. Instead, we simply search a fixed size buffer for some sentinal values and CAS into it. For allocations that only ever hit the cache, this improves performance from ~9000 cycles to ~6000 cycles and similar improvements for workloads that feel the pain of small thread counts hitting the cache.
Summary: This implementation only worked if the lane mask passed in was uniform, but this is against the expected usage where the user may be wishing to check if a value is uniform *within* a mask subset. Also remove redundant sync_lanes, the ballots and shuffles already have synchronizing behavior.
… HLSL casts in the new constant interpreter for basic matrix constexpr evaluation in HLSL (#184840) Forgot to change the target branch before merging. This PR is a cherry-pick of the squashed-and-merged PR commit b16aa4b7ec665911c74300cd7442659b70973d13 from 183424 This PR fixes #182963 This PR is an extension of #178762 which has already been merged. This PR adds support for `ConstantMatrixType` and the HLSL casts `CK_HLSLArrayRValue`, `CK_HLSLMatrixTruncation`, `CK_HLSLAggregateSplatCast`, and `CK_HLSLElementwiseCast` to the bytecode constexpr evaluator. The implementations of CK_HLSLAggregateSplatCast and CK_HLSLElementwiseCast are incomplete, as they still need to support struct and array types to enable use of the experimental new constant interpreter on other existing HLSL constexpr tests. The completion of the implementations of these casts will be tracked in a separate issue (#183426) and implemented in a separate PR. Assisted-by: claude-opus-4.6
* This will allow us to setup clang-stage1-RA jobs as multi branch pipelines for release branches without setting up all the downstream jobs. I will need to cherry-pick the jenkinsfile to the release/22.x branch which I will do after this lands
Moving getResidentPages out of linux.cpp and adding it to MemMapBase
…_TYPE` (#184793) The `ASSIGN_TYPE` instruction should not be referenced anymore at this point. So we can free its memory. Follow up of llvm/llvm-project#182330
…#184843) Reverts llvm/llvm-project#183794 It broke a couple of tests from Fujitsu testsuite.
returning anything from the callback. Fixing this on the off chance that is what is causing the linux-only failure in this test after PR: llvm/llvm-project#184272
…classes (#182706) Inherited constructors in `dllexport` classes are now exported for ABI-compatible cases, matching MSVC behavior. Constructors with variadic arguments or callee-cleanup parameters are not yet supported and produce a warning. This aims to partially resolve llvm/llvm-project#162640. Assisted by : Cursor // Claude Opus 4.6
Add additional tests with multiple argmin/argmax reductions for llvm/llvm-project#184729.
…ment. (#184839) We don't have instructions for an i8 sshlat.
This just adds this as an NYI, but this will suppress the Wswitch warning.
…#184714) Replace "compile" with "assemble" in formatter_bytecode. This is in preparation for the addition of a Python to formatter bytecode compiler. It will be more clear to have one meaning for "compile".
…yFixIrreducibleControlFlow" (#181755) (#184441) Re-application of #181755. Includes fixes to issues found after the original's merge.
…84853) It is a convention to use uppercase names of directives and clauses in diagnostic messages, but getting such names is somewhat cumbersome: ``` parser::ToUpperCaseLetters(llvm::omp::getOpenMPDirectiveName(dirId)); parser::ToUpperCaseLetters(llvm::omp::getOpenMPClauseName(clauseId)); ``` Implement `GetUpperName` (overloaded for clauses and directives) to shorten it to ``` GetUpperName(dirId, version); GetUpperName(clauseId, version); ``` This patch replaces existing instances of this pattern, adding the use of OpenMP version where it was previously missing.
This is a speculative fix for the openmp-offload-sles-build-only build
bot that fails after #184376 with:
```
clang/lib/Tooling/DependencyScanningTool.cpp:336:12: error: could not convert ‘CIWithContext’ from ‘clang::tooling::CompilerInstanceWithContext’ to ‘std::optional<clang::tooling::CompilerInstanceWithContext>’
return CIWithContext;
^~~~~~~~~~~~~
```
…#179568) Fixes #168944 Depends on #178762 being merged first. The commit adds implicit conversion warnings for HLSL matrix types in Clang's semantic checker by extending SemaChecking.cpp to detect precision loss when converting between matrix types (specifically, ConstantMatrixType). A new test validates float precision warnings, and an existing test was updated to expect the new diagnostics. Assisted-by: claude-opus-4.5
to see the private state to do so until we are done hijacking.
Avoid error like: ``` error: runtime derived type info descriptor was not generated ```
Some compilers (e.g. on AIX) do not pack by default. Use LLVM_PACKED to ensure the VPIRFlags struct is packed as expected on all platforms. This matches what we already do in other places for AIX, e.g. in llvm/include/llvm/CodeGen/SelectionDAGNodes.h (added in 844a02e), although it uses the more general LLVM_PACKED unconditionally PR: llvm/llvm-project#184687
Implicitly discovered module maps that reference files outside their module directory cause order dependent behavior when using implicitly discovered module maps. This adds an off by default diagnostic about these cases with the long term goal of removing import order dependent behavior. Module maps found via `-fmodule-map-file=` are not a problem because they are all loaded at the start of translation. Assisted-by: claude-opus-4.6
CONFLICT (content): Merge conflict in clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
…tate" mode. (#184272)" This reverts commit 97572c1. This patch seems to cause TestWatchpointCommandPython.py to time out on the ubuntu buildbots (but nowhere else that I can find so far.) The timeout is weird too, the TEST FILE is timing out but the individual tests aren't being shown and there's no other output. Grrr... Anyway I'll revert this and then see if I can do some guessing about how this change might cause the test to fail.
Update tests for llvm-project commit llvm/llvm-project@63074da25d16: `("[DebugInfo][DwarfDebug] Move emission of globals from beginModule() to endModule() (5/7) (#184219)", 2026-03-04)`. Original commit: KhronosGroup/SPIRV-LLVM-Translator@9981b92e6cc099b
To avoid name clash between builtins with `half` and `bfloat` return types, introduce return type postfix for `bfloat nan` builtin in reverse translation to SPIR-V Friendly IR. Original commit: KhronosGroup/SPIRV-LLVM-Translator@3ce38f4ecb3da94
…3656) The llc command already works with the `SPV_INTEL_variable_length_array` extension enabled Original commit: KhronosGroup/SPIRV-LLVM-Translator@f66eceee367a2a2
Newer SPIRV-Tools spirv-dis versions print the textual instruction name instead of the instruction number. Make the FileCheck check lines accept both for compatibility. Original commit: KhronosGroup/SPIRV-LLVM-Translator@12645e343582ce6
A recent version of spirv-val is now reporting "OpConstantComposite must not have spec constant operands" for some tests. Fix by letting `addCompositeConstant` defer to `addSpecConstantComposite` when attempting to add a constant that has one or more specialization constant operands. Original commit: KhronosGroup/SPIRV-LLVM-Translator@13f1c79e6272d49
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
2330267 enabled UseNewOffloadingDriver for HIP. We need to update the customization to switch HIP to NewOffloadingDriver as well. We can NOT enable all targets using `C.getActiveOffloadKinds() != Action::OFK_None` yet. Fixes: Clang :: Driver/hip-code-object-version.hip Clang :: Driver/hip-options.hip Clang :: Driver/hip-sanitize-options.hip Clang :: Driver/hip-save-temps.hip Clang :: Driver/hip-thinlto.hip Clang :: Driver/hip-toolchain-opt.hip
Community has removed the option (-bugpoint-enable-legacy-pm) to run legacy pm with opt recently. llvm/llvm-project#182320 This is causing 1000+ intel specific test failures because we still have RUN lines that test legacy pm with opt. In most cases, we have two RUN lines, one for legacy PM the other for new PM. Since we already switched to new PM for quite a while, I think it is time for us to remove the legacy PM RUN lines now
This change aligns with the upstream removal of -bugpoint-enable-legacy-pm (llvm/llvm-project#182320) .
Commit e2f7f83 added logic to detect address space mismatches and insert additional addrspacecast instructions to avoid invalid pointer casts. This updates the SYCL test expectations to match the new IR patterns with the extra addrspacecast operations. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com> Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Fix CodeGenCUDA/flush-denormals.cu and CodeGenSYCL/flush-denormals.cpp tests that broke after commit 5a88dff which removed the wchar_size module flag for standard targets. The commit changed module.flags to no longer include wchar_size for standard targets, reducing the number of module flags. Updated the test CHECK lines to match the new module.flags pattern with fewer elements. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com> CMPLRLLVM-73948 Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
…ies (#134805)" (#174483) This reverts commit ccfb97b. This was reverted due to the unfortunate reliance on external device library installations, which ship the last rocm released bitcode. The last attempt was 8 months ago, so hopefully the buildbots are now caught up to a more recent build that no longer needs the old control library.
9bec7a5 revert the changes and removed the tests. We did not remove the tests so it is failing.
Fix annotations-field.c tests that broke after commit b3d99ac which changed Clang's CGBuilder to use DataLayout-aware TargetFolder instead of ConstantFolder. The commit causes GEP constant expressions to be emitted in canonical getelementptr i8 form instead of typed form, and affects when the nuw qualifier is emitted. Updated CHECK lines to reflect the new IR: - Changed from typed GEPs like `getelementptr (%struct.foo, ...)` to canonical i8 form `getelementptr (i8, ptr, i64 offset)` Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Fix extra warnings/errors like:
```
nblog2.log: 13595: In file included from /netbatch/build_comp63266_00/ics_top/llvm/llvm/include/llvm/IR/Metadata.h:35:
nblog2.log: 13596: /netbatch/build_comp63266_00/ics_top/llvm/llvm/include/llvm/ADT/ArrayRef.h:79:35: error: implicit conversion changes signedness: 'long' to 'size_type' (aka 'unsigned long') [-Werror,-Wsign-conversion]
nblog2.log: 13597: 79 | : Data(begin), Length(end - begin) {
nblog2.log: 13598: | ~~~~~^~~~~~~
nblog2.log: ...
nblog2.log: 13843: make[2]: *** [tools/sycl-jit/passes/CMakeFiles/SYCLJITPasses.dir/build.make:78: tools/sycl-jit/passes/CMakeFiles/SYCLJITPasses.dir/SYCLJITPasses.cpp.o] Error 1
```
jsji
force-pushed
the
llvmspirv_pulldown
branch
from
043272e to
20925eb
Compare
…171b8d) After the NVPTX refactoring in a171b8d and 21d93c2, createNVVMInternalAddrspaceWrap unconditionally adds ReadOnly attribute to byval parameters. This causes a verifier error for kernel parameters with local_accessor (writable shared memory), which are already marked writeonly by the frontend. The fix checks for existing WriteOnly attribute before adding ReadOnly, preventing the attribute conflict while preserving the readonly optimization for truly read-only parameters. Fixes compilation of structs_with_special_types_as_kernel_paramters.cpp test which failed with: Attributes 'readonly and writeonly' are incompatible! Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
After the NVPTX refactoring in a171b8d, NVPTXLowerArgs unconditionally adds grid_constant attribute to all byval parameters via createNVVMInternalAddrspaceWrap. This causes issues with writeonly parameters (marked by SPIRV backend with initializes attribute) that need to be written to for initialization. The grid_constant attribute requires parameters to be read-only inputs in PTX .param space, but writeonly parameters are outputs that must be written to. This causes ptxas errors: "Illegal to write to function input parameter". The fix ensures writeonly byval parameters: 1. Are forced to use copyByValParam (local memory copy) early in lowerKernelByValParam 2. Skip getting grid_constant/ReadOnly attributes in createNVVMInternalAddrspaceWrap This allows writeonly parameters to be properly handled as outputs in local memory where they can be written to, fixing compilation with -fsycl-use-spirv-backend-for-spirv-gen. Fixes: structs_with_special_types_as_kernel_paramters.cpp test Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
LLVM: llvm/llvm-project@337fed3
SPIRV-LLVM-Translator: KhronosGroup/SPIRV-LLVM-Translator@13f1c79e6272d49