aboutsummaryrefslogtreecommitdiff
AgeCommit message (Collapse)AuthorFilesLines
2026-02-09address reviewerusers/DataCorrupted/ExposeDirectMethod-optPeter Rong4-78/+67
2026-01-12change test flag namePeter Rong1-1/+1
2026-01-12Add a cache to remember previously realized classesPeter Rong5-13/+86
2026-01-12Add a cache to remember all classes that should've been realized by loadPeter Rong2-14/+47
2026-01-12simplify testsPeter Rong1-51/+33
2026-01-12formatPeter Rong1-2/+4
2026-01-12fix some lint warningsPeter Rong1-3/+3
2026-01-12evict weak classPeter Rong1-1/+1
2026-01-12fix mac testsPeter Rong1-5/+8
2026-01-12update test and fix incorrect heuristicPeter Rong3-37/+160
2026-01-12[ExposeObjCDirect] OptimizationsPeter Rong2-5/+81
In many cases we can infer that class object has been realized
2026-01-09formatPeter Rong1-1/+1
2026-01-09Add helper function backPeter Rong2-0/+17
2026-01-09rebase to helper renamingPeter Rong8-19/+19
2026-01-09update commentsPeter Rong1-3/+5
2026-01-09fix mac testPeter Rong1-11/+0
2026-01-09formatPeter Rong1-2/+1
2026-01-09amend mac testsPeter Rong1-1/+3
2026-01-09add testsPeter Rong7-2/+846
2026-01-09[ExposeDirectMethod] Nil chech thunk generationPeter Rong4-7/+258
- Generation - Dispatch
2026-01-09Rename to precondition thunkPeter Rong1-2/+2
2026-01-09update driver behavior and a testPeter Rong1-0/+1
2026-01-09[ExposeObjCDirect] Adding a flag to allow new objc direct ABIPeter Rong1-0/+6
1. Add a flag 2. Clean up and set up helper functions to implement later Signed-off-by: Peter Rong <PeterRong@meta.com>
2026-01-05rebase to name changesPeter Rong1-5/+1
2026-01-05formatPeter Rong2-4/+5
2026-01-05fix errorPeter Rong1-1/+2
2026-01-05[ExposeObjCDirect] Setup helper functionsPeter Rong3-22/+88
1. GenerateDirectMethodsPreconditionCheck: Move some functionalities to a separate functions. Those functions will be reused if we move precondition checks into a thunk 2. Create `DirectMethodInfo`, which will be used to manage true implementation and its thunk
2026-01-05[ObjCDirectPreconditionThunk] Adding a flag to with objc_direct symbols' ↵Peter Rong8-12/+81
prefix (#170616) ## TL;DR This is a stack of PRs implementing features to expose direct methods ABI. You can see the RFC, design, and discussion [here](https://discourse.llvm.org/t/rfc-optimizing-code-size-of-objc-direct-by-exposing-function-symbols-and-moving-nil-checks-to-thunks/88866). The stack of the following four PRs completes the whole feature. https://github.com/llvm/llvm-project/pull/170616 **Flag `-fobjc-direct-precondition-thunk` set up** https://github.com/llvm/llvm-project/pull/170617 Code refactoring to ease later reviews https://github.com/llvm/llvm-project/pull/170618 Thunk generation https://github.com/llvm/llvm-project/pull/170619 Optimizations, some class objects can be known to be realized ## Implementation details 1. Add a flag. I used `-fobjc-direct-precondition-thunk` instead of `-fobjc-direct-caller-thunks` as discussed in this PR. 2. Clean up and set up helper functions to implement later a. `canMessageReceiverBeNull` / `canClassObjectBeUnrealized` these two functions will be helpful later to determine which function (true implementation or nil check thunk) we should dispatch a call to. Formatting. b. `getSymbolNameForMethod` has a new argument `includePrefixByte`, which allows us to erase the prefixing `\01` when the flag is enabled c. `usePreconditionThunk` is the single source of truth of what we should do. It not only checks for the flag, but also whether the method is qualified and we are in the right runtime. A method that `usePreconditionThunk` is either `shouldHavePreconditionThunk` or `shouldHavePreconditionInline`. ## Tests Driver tests --------- Signed-off-by: Peter Rong <PeterRong@meta.com> Co-authored-by: Kyungwoo Lee <kyulee@meta.com>
2026-01-05[clang][Modules] Fix unexpected warnings triggered by a PCH and a module ↵Qiongsi Wu6-1/+100
with config macros (#174034) When a PCH is compiled with macro definitions on the command line, such as `-DCONFIG1`, an unexpected warning can occur if the macro definitions happen to belong to an imported module's config macros. The warning may look like the following: ``` definition of configuration macro 'CONFIG1' has no effect on the import of 'Mod1'; pass '-DCONFIG1=...' on the command line to configure the module ``` while `-DCONFIG1` is clearly on the command line when `clang` compiles the source that uses the PCH and the module. The reason this can happen is a combination of two things: 1. The logic that checks for config macros is not aware of any command line macros passed through the PCH ([here](https://github.com/llvm/llvm-project/blob/7976ac990000a58a7474269a3ca95e16aed8c35b/clang/lib/Frontend/CompilerInstance.cpp#L1562)). 2. `clang` _replaces_ the predefined macros on the command line with the predefined macros from the PCH, which does not include any builtins ([here](https://github.com/llvm/llvm-project/blob/7976ac990000a58a7474269a3ca95e16aed8c35b/clang/lib/Frontend/CompilerInstance.cpp#L679)). This PR teaches the preprocessor to recognize the command line macro definitions passed transitively through the PCH, so that the error check does not miss these definitions by mistake. The config macro itself works fine, and it is only the error check that needs fixing. rdar://95261458
2026-01-05[mlir][Python] use maybeDowncast for PyType/PyAttribute returns after ↵Maksim Levental8-36/+52
#174156 (#174489) #174156 made all gettors return `Py*` but skipped downcasting where possible. So restore it by calling `.maybeDowncast`.
2026-01-05[VPlan] Remove VPWidenSelectRecipe, use VPWidenRecipe instead (NFCI). (#174234)Florian Hahn16-241/+110
All extra state has been removed from VPWidenSelectRecipe at this point. There's no benefit of having a separate recipe and Select can easily be handled by the existing VPWidenRecipe. PR: https://github.com/llvm/llvm-project/pull/174234
2026-01-05[SLP] Report the correct operand to getArithmeticInstrCost() when duplicated ↵Ryan Buchner2-3/+52
scalars (#174442) Before, we were selecting the wrong operand in cases when Scalars contained duplicate values. Stems from #135797. Using: `opt -passes=slp-vectorizer -mtriple=riscv64 -mattr=+v t.ll` ``` target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128" target triple = "riscv64" define void @foo(ptr noalias %A, ptr noalias %B) { entry: %0 = load i32, ptr %B %add = add nsw i32 %0, 1 store i32 %add, ptr %A %arrayidx.1 = getelementptr inbounds nuw i8, ptr %B, i64 4 %1 = load i32, ptr %arrayidx.1 %add.1 = add nsw i32 %1, 1 %arrayidx2.1 = getelementptr inbounds nuw i8, ptr %A, i64 4 store i32 %add.1, ptr %arrayidx2.1 %arrayidx.2 = getelementptr inbounds nuw i8, ptr %B, i64 8 %2 = load i32, ptr %arrayidx.2 %add.2 = add nsw i32 %2, 1 %arrayidx2.2 = getelementptr inbounds nuw i8, ptr %A, i64 8 store i32 %add.2, ptr %arrayidx2.2 %arrayidx.3 = getelementptr inbounds nuw i8, ptr %B, i64 12 %arrayidx2.3 = getelementptr inbounds nuw i8, ptr %A, i64 12 store i32 %add, ptr %arrayidx2.3 %arrayidx.4 = getelementptr inbounds nuw i8, ptr %B, i64 16 %4 = load i32, ptr %arrayidx.4 %add.4 = add nsw i32 %4, 1 %arrayidx2.4 = getelementptr inbounds nuw i8, ptr %A, i64 16 store i32 %add.4, ptr %arrayidx2.4 %arrayidx.5 = getelementptr inbounds nuw i8, ptr %B, i64 20 %5 = load i32, ptr %arrayidx.5 %add.5 = add nsw i32 %5, 1 %arrayidx2.5 = getelementptr inbounds nuw i8, ptr %A, i64 20 store i32 %add.5, ptr %arrayidx2.5 %arrayidx.6 = getelementptr inbounds nuw i8, ptr %B, i64 24 %6 = load i32, ptr %arrayidx.6 %add.6 = add nsw i32 %6, 1 %arrayidx2.6 = getelementptr inbounds nuw i8, ptr %A, i64 24 store i32 %add.6, ptr %arrayidx2.6 %arrayidx.7 = getelementptr inbounds nuw i8, ptr %B, i64 28 %7 = load i32, ptr %arrayidx.7 %add.7 = add nsw i32 %7, 1 %arrayidx2.7 = getelementptr inbounds nuw i8, ptr %A, i64 28 store i32 %add.7, ptr %arrayidx2.7 ret void } ``` The following trace is produced, note the wrong operand is used for `Idx > 2` Before: ``` GetScalarCost(), Idx=0 UniqueValues[Idx]: %add = add nsw i32 %0, 1 Op1: %0 = load i32, ptr %B, align 4 GetScalarCost(), Idx=1 UniqueValues[Idx]: %add.1 = add nsw i32 %1, 1 Op1: %1 = load i32, ptr %arrayidx.1, align 4 GetScalarCost(), Idx=2 UniqueValues[Idx]: %add.2 = add nsw i32 %2, 1 Op1: %2 = load i32, ptr %arrayidx.2, align 4 GetScalarCost(), Idx=3 UniqueValues[Idx]: %add.4 = add nsw i32 %3, 1 Op1: %0 = load i32, ptr %B, align 4 GetScalarCost(), Idx=4 UniqueValues[Idx]: %add.5 = add nsw i32 %4, 1 Op1: %3 = load i32, ptr %arrayidx.4, align 4 GetScalarCost(), Idx=5 UniqueValues[Idx]: %add.6 = add nsw i32 %5, 1 Op1: %4 = load i32, ptr %arrayidx.5, align 4 GetScalarCost(), Idx=6 UniqueValues[Idx]: %add.7 = add nsw i32 %6, 1 Op1: %5 = load i32, ptr %arrayidx.6, align 4 ``` After: ``` GetScalarCost(), Idx=0 UniqueValues[Idx]: %add = add nsw i32 %0, 1 Op1: %0 = load i32, ptr %B, align 4 GetScalarCost(), Idx=1 UniqueValues[Idx]: %add.1 = add nsw i32 %1, 1 Op1: %1 = load i32, ptr %arrayidx.1, align 4 GetScalarCost(), Idx=2 UniqueValues[Idx]: %add.2 = add nsw i32 %2, 1 Op1: %2 = load i32, ptr %arrayidx.2, align 4 GetScalarCost(), Idx=3 UniqueValues[Idx]: %add.4 = add nsw i32 %3, 1 Op1: %3 = load i32, ptr %arrayidx.4, align 4 GetScalarCost(), Idx=4 UniqueValues[Idx]: %add.5 = add nsw i32 %4, 1 Op1: %4 = load i32, ptr %arrayidx.5, align 4 GetScalarCost(), Idx=5 UniqueValues[Idx]: %add.6 = add nsw i32 %5, 1 Op1: %5 = load i32, ptr %arrayidx.6, align 4 GetScalarCost(), Idx=6 UniqueValues[Idx]: %add.7 = add nsw i32 %6, 1 Op1: %6 = load i32, ptr %arrayidx.7, align 4 ```
2026-01-05[Fuchsia] Set libcxx baremetal options (#173825)Prabhu Rajasekaran1-0/+4
Setting LIBCXX_HAS_RT_LIB and LIBCXX_HAS_PTHREAD_LIB to OFF to prevent POSIX dependencies creeping in.
2026-01-05[MemProf] Include matching calls in the dot graph node label (#174247)Teresa Johnson2-0/+11
After initially matching stack nodes to summary we may have multiple calls per node, e.g. in the case of indirect calls with multiple profiled callee targets. It is useful to see all of these calls, which will show up in the poststackupdate dot graph.
2026-01-05[AMDGPU] Add new llvm.amdgcn.wave.shuffle intrinsic (#167372)saxlungs7-1/+459
This intrinsic will be useful for implementing the OpGroupNonUniformShuffle operation in the SPIR-V reference --------- Signed-off-by: Domenic Nutile <domenic.nutile@gmail.com> Co-authored-by: Jay Foad <jay.foad@gmail.com>
2026-01-05Reapply "clang/AMDGPU: Stop looking for oclc_daz_opt_* control libraries ↵Matt Arsenault6-51/+28
(#134805)" (#174483) This reverts commit ccfb97b42174eab118a4e4222c25e986db876563. 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.
2026-01-05[libunwind][WebAssembly] Fix typos (NFC) (#173745)Heejin Ahn1-7/+7
2026-01-05[mlir][Python] fix NV examples after #172892 (#174481)Maksim Levental3-17/+17
2026-01-05[lit] Make not still fail if the called process returns a signalAiden Grossman6-2/+47
This is the behavior of the main not binary that was not preserved in the internal shell. Make it so that the builtin not command does actually fail if we end up with a signal rather than just a non-zero exit code. Reviewers: petrhosek, ilovepi, jdenny-ornl, arichardson Pull Request: https://github.com/llvm/llvm-project/pull/174298
2026-01-05[bazel] Add comment about DEFAULT_TARGETS (#174479)Keith Smiley1-0/+1
This should not include targets that aren't enabled in cmake. 6ccf97674b2deaa03e271725306b18a712a56113
2026-01-05[mlir][Python] use canonical Python `isinstance` instead of ↵Maksim Levental21-216/+224
`Type.isinstance` (#172892) We've been able to do `isinstance(x, Type)` for a quite a while now (since https://github.com/llvm/llvm-project/commit/bfb1ba752655bf09b35c486f6cc9817dbedfb1bb) so remove `Type.isinstance` and the the special-casing (`_is_integer_type`, `_is_floating_point_type`, `_is_index_type`) in some places (and therefore support various `fp8`, `fp6`, `fp4` types).
2026-01-05[flang][cuda] Add CUFLaunchAttachAttr pass (#174465)Valentin Clement (バレンタイン クレメン)5-0/+101
CUF kernel are generated via gpu.launch and then outlined. The resulting launch operation needs to hava a CUDA attribute attached so it will be distinguishable from other launch.
2026-01-05[Bazel] Port 735b1c284d6a3e838c08699944707ae8c303fa8f (#174476)Aiden Grossman1-0/+1
2026-01-05[clang] Fix IO sandbox violations in diagnostic filenames (#173107)Ben Langmuir2-2/+2
Update TextDiagnostic and SARIFDiagnostic emitFilename to use the FileManager's makeAbsolutePath instead of directly calling make_absolute. This fixes IO sandbox violation errors.
2026-01-05[clang] Reuse configured VFS for chained includes (#173288)Jan Svoboda1-1/+3
This PR propagates the already-configured VFS when handling chained includes, preventing unexpected use of the real FS and sandbox violations.
2026-01-05[ISel] Introduce llvm.clmul intrinsic (#168731)Ramkumar Ramachandra20-2/+13877
In line with a std proposal to introduce the llvm.clmul family of intrinsics corresponding to carry-less multiply operations. This work builds upon 727ee7e ([APInt] Introduce carry-less multiply primitives), and follow-up patches will introduce custom-lowering on supported targets, replacing target-specific clmul intrinsics. Testing is done on the RISC-V target, which should be sufficient to prove that the intrinsics work, since no RISC-V specific lowering has been added. Ref: https://isocpp.org/files/papers/P3642R3.html Co-authored-by: Craig Topper <craig.topper@sifive.com>
2026-01-05[InstCombine] Canonicalize `switch(X^C)` expressions to `switch(X)`Antonio Frighetto3-24/+183
`switch(X^C)` expressions can be folded to `switch(X)`. Minor opportunity to generalize simplifications in `visitSwitchInst` via an inverse function helper as well. Proof: https://alive2.llvm.org/ce/z/TMRy_3. Fixes: https://github.com/llvm/llvm-project/issues/174255. Fixes: https://github.com/llvm/llvm-project/issues/143368.
2026-01-05[Clang][NFC] remove getUnqualifiedType() when it's already unqualified (#172504)Rose Hudson4-17/+9
Since 8c4950951269ec58296afbeba14e99aef467f84d, getCanonicalTypeUnqualified() calls getUnqualifiedType(), so there's no point in calling that again on its return value.
2026-01-05[flang][acc] Introduce interface for rematerializable ops (#174467)Razvan Lupusoru4-1/+45
During the outlining process when offloading acc regions, the body of the compute kernel is separated from its original location and live-in values are handled in various ways including becoming function arguments. However, some operations are purely synthetic and only make sense when included with another operation (usually such operations exist to simplify IR design). Bounds and shapes are examples where during outlining they should be recreated inside to capture the full information. Therefore, introduce a new operation interface named OutlineRematerializationOpInterface meant to be attached to such operations. It is currently expected that all such operations are memory effect free to ensure there are no considerations needed when moving or cloning them into outlined regions. The interface is attached to the following operations: - acc.bounds (directly in TableGen) - fir.shape (via external model) - fir.shape_shift (via external model) - fir.shift (via external model) - fir.field_index (via external model) The pass that will use this interface and associated testing will follow in another pull request.
2026-01-05[clang][modules] Diagnose config mismatches more generally from precompiled ↵Cyndy Ishida10-37/+104
files (#174260) PCHs (but also modules generated from several implicit invocations like swiftc) previously reported a confusing diagnostic about module caches being mismatched by subdir. This is an implementation detail of the module machinery, and not very useful to the end user. Instead, report this case as a configuration mismatch when the compiler can confirm the module cache was passed the same between the current TU & previously compiled products. Ideally, each argument that could result in this error would be uniquely reported (e.g., O3), but as a starting point, providing something more general is strictly better than pointing the user to the module cache. This patch also includes NFCs for renaming variable names from Module to AST and formatting cleanup in related areas. resolves: rdar://167453135
generated by cgit v1.2.3 (git 2.39.1) at 2026-05-16 06:26:41 +0000