)]}' { "log": [ { "commit": "e2224153b0246e1199f0aa188603f13ad2e7ff49", "tree": "8b3fcd8047e98e063da9666ef1d794042fdc771a", "parents": [ "43cd878f9b3cb905c31d0114282f563ceb22f932" ], "author": { "name": "Nick Desaulniers", "email": "ndesaulniers@google.com", "time": "Thu Nov 12 15:43:15 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 15:49:24 2026 -0800" }, "message": "Revert \"[CIR] Implement \u0027ctor/dtor returns this\u0027\" (#242107)\n\nReverts llvm/llvm-project#240954\n\nclang/test/CIR/CodeGen/constructor-destructor-return-this.cpp is\nfailing, blocking presubmits.\n\nGitOrigin-RevId: 3b9974f902dbce577702743074af2095ca658dae\n" }, { "commit": "43cd878f9b3cb905c31d0114282f563ceb22f932", "tree": "c62ed25282609812311c27e988a59146f19d533f", "parents": [ "81e7d79b633ee25005211a5193ee8271f0b200e0" ], "author": { "name": "Vitaly Buka", "email": "vitalybuka@google.com", "time": "Thu Nov 12 15:40:31 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 15:48:42 2026 -0800" }, "message": "[test][Clang][CFI] Precommit tests for cfi-vcall checked loads (#241638)\n\nRecord current `cfi-vcall` codegen before emitting\n`llvm.type.checked.load` for non-trapping modes. Cover cross-DSO CFI,\nVFE, relative vtables, speculative devirtualization, MS ABI secondary\nvfptrs, virtual object deletion, ignorelists, debug info, sanitizer\nstats, and `EmitVTablePtrCheck` in trap and minimal-runtime modes.\n\nAdd compiler-rt tests for two false traps, marked XFAIL:\n- deleting a base object through a same-layout derived pointer with\n `-fwhole-program-vtables`;\n- `-fdevirtualize-speculatively` with a public class whose vtable comes\n from a non-LTO object.\n\nNeeded by #241639, #241640, #241643, #241644, #241646, and #241647.\n\nAssisted-by: Gemini\nGitOrigin-RevId: 0e97dc248f1cc1c339cd2211b713367f357c9abd\n" }, { "commit": "81e7d79b633ee25005211a5193ee8271f0b200e0", "tree": "890e55037c44f15e644ae2d6c149d8473083fe86", "parents": [ "70a01895f781b70f8bf13a11e2e488164389830d" ], "author": { "name": "Jan Svoboda", "email": "jan_svoboda@apple.com", "time": "Thu Nov 12 15:20:47 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 15:29:39 2026 -0800" }, "message": "[clang] Flush the PCM stream before querying the file status (#242050)\n\nThe `writeImpl()` function queries the status of the written file\nthrough its file descriptor while the `raw_fd_ostream` may still hold\nbuffered contents. This can report a size smaller than the actual file.\nThis PR flushes the stream first.\n\nNothing observed this bug. The file did get flushed right at the end of\nthe scope before being moved into the final destination, so no other\nprocesses ever saw a partially-written file. The current Clang instance\nnever reads back the written buffer, it keeps using its in-memory\nversion. The only observable thing is the out-parameter `Size` being\nwrong, but implicit modules never use it and instead hash contents of\nthe file for consistency checks.\n\nGitOrigin-RevId: 4d619e562cce75aaf331a667530dd6537eab7aa2\n" }, { "commit": "70a01895f781b70f8bf13a11e2e488164389830d", "tree": "42ece24d8a18bf08d417c3093bcdad70f0b77016", "parents": [ "b14e2736cc6d3281172967bab79ae3914e812139" ], "author": { "name": "Helena Kotas", "email": "hekotas@microsoft.com", "time": "Thu Nov 12 15:15:14 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 15:29:07 2026 -0800" }, "message": "[HLSL][NFC] Move common defines to the main header (hlsl.h) (#241626)\n\nMove common `#define`s from included headers to the main header\n(`hlsl.h`). Includes `_HLSL_BUILTIN_ALIAS`, `HLSL_AVAILABILITY` and\n`_HLSL_AVAILABILITY_STAGE`.\n\nGitOrigin-RevId: 3abdc343c5cc1dbcbf455d91eeb95a680ec77e68\n" }, { "commit": "b14e2736cc6d3281172967bab79ae3914e812139", "tree": "597f2d5ba8e8f5c4483fa529e4e108938395de72", "parents": [ "399c8db860cc19d6eab962ddbba0560058f5f49c" ], "author": { "name": "Helena Kotas", "email": "hekotas@microsoft.com", "time": "Thu Nov 12 15:08:28 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 15:26:42 2026 -0800" }, "message": "[HLSL] Add Shader Model 6.10 and DXIL version 1.10 to the target triple (#241066)\n\nAdds Shader Model 6.10 and DXIL version 1.10 as supported versions of\nthe target triple.\n\nGitOrigin-RevId: 25026c90e87390089eacd7f812b7c8bd9cc43abb\n" }, { "commit": "399c8db860cc19d6eab962ddbba0560058f5f49c", "tree": "4bda0b042edfd0871ecce8a2745e6c7d3d1c2cfb", "parents": [ "26aaebd2204d7d901f0788dc1a31b44a8d1789be" ], "author": { "name": "Adam Smith", "email": "adams@nvidia.com", "time": "Thu Nov 12 16:51:30 2026 -0600" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 14:57:47 2026 -0800" }, "message": "[CIR] Add inbounds and nuw flags to cir.get_element (#234774)\n\nCIR lowered array subscripts to a getelementptr with no no-wrap flags,\nwhere classic codegen marks them inbounds, and also nuw when the index\nis unsigned.\n\ncir.get_element now has optional inbounds and nuw flags, and CIRGen sets\nthem where classic does: inbounds unless -fwrapv-pointer is given, nuw\nfor an unsigned index expression, and no inbounds for an OpenMP device\nsubscript of an array of arrays whose base is not an alloca. The flags\nlive on the op because both depend on language options and on the C\nindex type, which the lowering cannot see. LowerToLLVM maps them to the\nGEP\u0027s no-wrap flags.\n\nAssisted-by: Cursor / Claude Opus 5.5\nGitOrigin-RevId: e550f70bdd531509b9e413946b8206fba1d49e4d\n" }, { "commit": "26aaebd2204d7d901f0788dc1a31b44a8d1789be", "tree": "ef23ad6d246e5ce1b806877c4068c69d590894fc", "parents": [ "3951f23c11e475414366c21dcc5e221348c7b8d9" ], "author": { "name": "Richard Howell", "email": "rmaz@users.noreply.github.com", "time": "Thu Nov 12 14:30:35 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 14:39:07 2026 -0800" }, "message": "[clang] Add -fpch-base-dir-is-cwd (#239791)\n\nSerialize PCH input paths relative to the current working directory, so\na build tree can be moved while preserving its relative layout. Reuse\nthe existing relocatable AST prefix stripping without storing a reader\nbase directory or enabling sysroot-based re-rooting.\n\nThis is the PCH equivalent of -fmodule-file-home-is-cwd.\n\nAssisted-by: Muse\nGitOrigin-RevId: efda9c9e4fd4b0f4b5c3558b152d13725c8e710c\n" }, { "commit": "3951f23c11e475414366c21dcc5e221348c7b8d9", "tree": "54dd4b77ae722a54b94cc99165e51667f6d7dd07", "parents": [ "9e9fea4e531e5eccfae35862c08c915e0f170ec8" ], "author": { "name": "Cyndy Ishida", "email": "cyndy_ishida@apple.com", "time": "Thu Nov 12 14:19:21 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 14:23:05 2026 -0800" }, "message": "[clang][deps][test] Migrate more module graph tests to scan-deps-filter, NFC (#242019)\n\nPipe these tests\u0027 output through `scan-deps-filter` so they only assert\non the fields they check. Unrelated fields that are added or reordered\nno longer break them.\n\nGitOrigin-RevId: f25418e7b1a9d7504dde66689942beae6e7715d3\n" }, { "commit": "9e9fea4e531e5eccfae35862c08c915e0f170ec8", "tree": "4d8b00315bb429a41214f8b3d0ff3bcec3dcca01", "parents": [ "96d22224f3fbf4c1150c9551de02009085ba9362" ], "author": { "name": "Erich Keane", "email": "ekeane@nvidia.com", "time": "Thu Nov 12 14:14:16 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 14:19:09 2026 -0800" }, "message": "[CIR] Implement \u0027ctor/dtor returns this\u0027 (#240954)\n\nThis is a blocking feature for a few of the ABIs, plus is a command line\noption for existing itanium ABIs. This patch fills in the handful of\n\"NYI\" spots.\n\nThe real \u0027interesting\u0027 part of this is the tests. This was well-tested\nin the CodeGenCXX classic codegen test, so this patch enables those of\nwhich can be enabled trivially (or with minor SSA name removals).\n\nSOME of those run-lines are not yet converted because they have\nunsupported platforms in their RUN lines, but we intend to convert those\neventually.\n\nGitOrigin-RevId: 8c3d123eb9541dca1acfd0e48c9315db19543a20\n" }, { "commit": "96d22224f3fbf4c1150c9551de02009085ba9362", "tree": "2b4c64f11696362a6e994241325d103917490a95", "parents": [ "334866c2a5d28899f6193802a71599584f334f90" ], "author": { "name": "Vitaly Buka", "email": "vitalybuka@google.com", "time": "Thu Nov 12 12:26:16 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 12:33:01 2026 -0800" }, "message": "[NFC][Clang][CFI] Pass SourceLocation to EmitVTableTypeCheckedLoad (#241637)\n\nPass the call-site `SourceLocation` to `EmitVTableTypeCheckedLoad`. It\nis\nunused for now; #241646 uses it to report failed checked loads with\ntheir\nsource location.\n\nAssisted-by: Gemini\nGitOrigin-RevId: 9b03826eee6790ae2d4130f1aaeb03b1f29f1c16\n" }, { "commit": "334866c2a5d28899f6193802a71599584f334f90", "tree": "c534571e22f6a67e1fc231aad5061687ee45a105", "parents": [ "03944014e4965f78fb68badf137e0f543714d2f4" ], "author": { "name": "Aman Maurya", "email": "amanmaurya9209@gmail.com", "time": "Fri Nov 13 01:44:54 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 12:32:30 2026 -0800" }, "message": "[CIR] Support ConstantExpr in CIRGenFunction::emitLValue (#237421)\n\n### Description\n\nThis patch adds `ConstantExpr` handling to `CIRGenFunction::emitLValue`,\nestablishing parity with classic Clang CodeGen\n(`clang/lib/CodeGen/CGExpr.cpp:1825-1830`) and completing AST visitor\ncoverage across CIR.\n\nIn ClangIR, `ConstantExpr` was already supported in `ScalarExprEmitter`\n(`CIRGenExprScalar.cpp:147`) and `AggExprEmitter`\n(`CIRGenExprAggregate.cpp:344`), but was stubbed with `errorNYI` in\n`CIRGenFunction::emitLValue`:\n```cpp\ncase Expr::ConstantExprClass:\n getCIRGenModule().errorNYI(e-\u003egetSourceRange(), \"emitLValue: ConstantExpr\");\n return LValue();\n```\n\nClassic Clang unwraps and evaluates the subexpression when evaluating an\nlvalue. This patch brings `CIRGenFunction::emitLValue` in line with\nclassic Clang by delegating to\n`emitLValue(cast\u003cConstantExpr\u003e(e)-\u003egetSubExpr())`.\n\nFixes #237403.\n\nGitOrigin-RevId: 9099ddf1e0aae2397fb8a0aea85dc74e4ed4e698\n" }, { "commit": "03944014e4965f78fb68badf137e0f543714d2f4", "tree": "16c14ba4ea5d52269d3fb5c4bd0085131cb75efb", "parents": [ "16a9aa86e60f2df8fd093a84d9d15622f0e929b7" ], "author": { "name": "Yashas Andaluri", "email": "yandalur@qti.qualcomm.com", "time": "Fri Nov 13 01:43:07 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 12:18:20 2026 -0800" }, "message": "[Hexagon] Return HLX vectors in registers (#242014)\n\nWith HLX enabled, functions returning an HLX vector or vector pair\nreturned it indirectly through memory. Mark these returns inreg, as is\ndone for HVX vectors, so that they are returned in HLX registers.\n\nAssisted by AI\n\nGitOrigin-RevId: 267deae940034d617a2778d41438661b04c4e297\n" }, { "commit": "16a9aa86e60f2df8fd093a84d9d15622f0e929b7", "tree": "539f7fad5f8a759fbb919bb11a6f55c3974d9c92", "parents": [ "1476c1e916fe0954ae5a9fa9fea6b14a05d7710f" ], "author": { "name": "Prabhu Rajasekaran", "email": "prabhukr@google.com", "time": "Thu Nov 12 15:03:34 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 12:08:13 2026 -0800" }, "message": "[clang] Fix target-dependent default for -mincremental-linker-compatible (#222099)\n\nPatch #188800 modified the target specific default for this flag which\nresulted in non MSVC, non UEFI targets getting this flag set. This\ncaused issue https://github.com/llvm/llvm-project/issues/219457 which\nthis patch should fix.\n\nAssisted-by: Gemini\nGitOrigin-RevId: 08d064a71dbdfa1b50f352d34c103ad245070611\n" }, { "commit": "1476c1e916fe0954ae5a9fa9fea6b14a05d7710f", "tree": "9b60b0ef9d8ab0d0c418fd75e24c655a1bd5cc12", "parents": [ "6590d77396573f851c3411a3f2f5f73cb23d044a" ], "author": { "name": "Sean Perry", "email": "perry@ca.ibm.com", "time": "Thu Nov 12 14:20:47 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 11:36:44 2026 -0800" }, "message": "[SystemZ] Update clang lit tests for default trigraphs on z/OS (#238549)\n\nUpdate the clang lit tests so they work (or marked as xfail) on z/OS\nwith trigraphs being on by default.\n\nGitOrigin-RevId: 6aa8d348164a30737238618708ac86833d376688\n" }, { "commit": "6590d77396573f851c3411a3f2f5f73cb23d044a", "tree": "722eb3bcc34d43b0913bf425949c378dceb8521a", "parents": [ "8c4c0df21298e05d8531b18d6b5d3ba4a55dca6b" ], "author": { "name": "Sean Perry", "email": "perry@ca.ibm.com", "time": "Thu Nov 12 14:19:04 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 11:34:07 2026 -0800" }, "message": "[SystemZ][z/OS] Mark scan-dep tests as unsupported/xfail (#238557)\n\nMost of these are unsupported b/c z/OS doesn\u0027t have the __clangast\nsection just like in AIX. The others are like not handling multiple\nthreads.\n\nGitOrigin-RevId: 1464a03abec0ff474028ef94c14c4269658f51c3\n" }, { "commit": "8c4c0df21298e05d8531b18d6b5d3ba4a55dca6b", "tree": "137eb64c227c340748aad4497d6b9c7822232880", "parents": [ "1971a7dcc6775636a8258286f4672a30650db97a" ], "author": { "name": "agozillon", "email": "Andrew.Gozillon@amd.com", "time": "Thu Nov 12 20:09:22 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 11:14:21 2026 -0800" }, "message": "[OpenMP][OpenMPIRBuilder] Enable non-pointee storage present propagation in mappers (#241537)\n\nAdd a boolean to emitUserDefinedMapper to toggle on present propagation\nfor non-pointee storage members, effectively storage shared with/part of\nthe top-level parent. This will remain off for the main clang path for\nthe moment but is switched on for Flang/CIR. This is due to the fact\nthat Clang still predominantly uses OBJ_AND_PTR style maps, whereas the\nMLIR path only uses it for declare target at the moment, which bypasses\nthe mapper route making it a non-issue for this change.\n\nGitOrigin-RevId: f453643ee0cccfd2def20c7e487eab57861b2637\n" }, { "commit": "1971a7dcc6775636a8258286f4672a30650db97a", "tree": "b2b0c628aaecdaf43aaa13cc58eb8e057d2f4115", "parents": [ "4d6efd85c9eec8d7c4781d7b73c78a17766363fa" ], "author": { "name": "Andy Kaylor", "email": "akaylor@nvidia.com", "time": "Thu Nov 12 11:05:00 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 11:13:17 2026 -0800" }, "message": "[CIR] Use target-specific maximum for _BitInt alignment (#241466)\n\nCIR had been using a hard-coded 64-bit maximum alignment for _BitInt\ntypes, which matches x86-64 targets but not AArch64. This change adds an\nMLIR data layout entry to record the target\u0027s maximum _BitInt alignment\nso that this value can be retrieved by post-codegen queries.\n\nAssisted-by: Cursor / various models\nGitOrigin-RevId: 573aba3e54eb6dabd6b157aa69f5005c0504b431\n" }, { "commit": "4d6efd85c9eec8d7c4781d7b73c78a17766363fa", "tree": "0a36bf963a581dd6578915896fdd9380e10b91e8", "parents": [ "50a2ae035aec77b0f6dd4f9e900a64594127c7ac" ], "author": { "name": "Akash Manna", "email": "akash.manna.mymail@gmail.com", "time": "Fri Nov 13 00:14:28 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 10:57:52 2026 -0800" }, "message": "[Clang][OpenMP] Emit device kernels for target regions in lambdas without an enclosing function (#240840)\n\nFixes #136652\n\nA `target` region in a lambda at namespace scope no longer crashes the\nhost after #226691, but with offloading the device still emitted no\nkernel for it. The device finds target regions by scanning function\nbodies, and a lambda in a global initializer, a default argument or a\ndefault member initializer is reached from none of them. The host\nregisters the entry, the device image has nothing behind it.\n\nThe device scan now walks those initializers and names a region the way\nthe host does: after the enclosing non-closure function, or after the\ncall operator itself when there is none, through one helper shared by\nboth sides. Such closures also get a mangling number when offloading, as\nunder CUDA, so the two compilations agree on the name.\n\nAI Assisted!\n\nGitOrigin-RevId: 48371514fb1ea71ae2e4adfcbfcc21b0662310b8\n" }, { "commit": "50a2ae035aec77b0f6dd4f9e900a64594127c7ac", "tree": "d620b9126e9149cd8710a4e40e6bda08e5907234", "parents": [ "11a53bd8bc2f0b68543c279a2a8b9cadf5dbcf39" ], "author": { "name": "Akash Manna", "email": "akash.manna.mymail@gmail.com", "time": "Fri Nov 13 00:08:43 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 10:42:40 2026 -0800" }, "message": "[Clang][OpenMP] Fix crash on a tuple-like structured binding in a task (#241357)\n\nFixes #116258\n\nA tuple-like structured binding (`std::pair`, `std::tuple`) referenced\nin a `task` or named in `firstprivate` crashed clang right after the\n\"not yet supported\" error. `isOpenMPCapturedDecl` maps struct and array\nbindings to their `DecompositionDecl` but let tuple-like ones fall into\nthe data-member path, which dereferences a private copy that a binding\u0027s\nDSA never has.\n\n`isOpenMPCapturedDecl` now reports the holding variable for a privatized\ntuple-like binding, so its uses still reach the capture checks instead\nof a null private copy; unprivatized ones keep returning null. Outside\nany construct, `getDSA` now classifies a tuple-like binding by the\nstorage of the variable it decomposes, so a namespace-scope one is\nshared like that variable.\n\nLLM tools were used for this contribution.\n\nGitOrigin-RevId: 2fdabe9a555a511708b5191dfac7c6394bbc54b8\n" }, { "commit": "11a53bd8bc2f0b68543c279a2a8b9cadf5dbcf39", "tree": "8ff4e6118990bdbc268ab75d0692c75aee685e49", "parents": [ "753a69340b9fad3f273b729e8cc148bf2d5dfe88" ], "author": { "name": "Hongyu Chen", "email": "xxs_chy@outlook.com", "time": "Fri Nov 13 01:59:58 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 10:42:10 2026 -0800" }, "message": "[RISCV][P-ext] Add Scalar Zip/UnZip Intrinsics (#241430)\n\nThis patch adds scalar zip/unzip intrinsics for P-ext. Also, it supports\nmissing codegen for zip*hp.\n\nGitOrigin-RevId: 5c3918c95b0b3285406c376bb22e2e91bfd95c79\n" }, { "commit": "753a69340b9fad3f273b729e8cc148bf2d5dfe88", "tree": "b12eadd611e1252faa80de172ded0f7407e68681", "parents": [ "1bfd537815d60f782f7ee1165062fbb21a8a5718" ], "author": { "name": "Helena Kotas", "email": "hekotas@microsoft.com", "time": "Thu Nov 12 09:59:21 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 10:03:27 2026 -0800" }, "message": "[HLSL][NFC] Fix fragile tests (#241625)\n\n- Remove unstable attribute-group number\n- Add anchor for function declaration to scope the checks to the\nintended function.\n\nGitOrigin-RevId: 6b90c7effac6a7b5526ccdef179c87e0abad79c6\n" }, { "commit": "1bfd537815d60f782f7ee1165062fbb21a8a5718", "tree": "86f517c45d26db82c60209011d6068d40aef8136", "parents": [ "02365829074742daabda39ba2f1f95aae5d40e0a" ], "author": { "name": "Andy Kaylor", "email": "akaylor@nvidia.com", "time": "Thu Nov 12 09:50:56 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 09:59:03 2026 -0800" }, "message": "[CIR] Add CIR unit tests to check-clang-cir (#238102)\n\nThis change updates the CMake handling of the check-clang-cir and adds\nlit config files to include CIR unit tests with the check-clang-cir\nbuild target. This will ensure that these tests are run as part of the\nmlir CI testing, which uses on check-clang-cir and not the full\ncheck-clang target.\n\nAssisted-by: Cursor / Grok 4.7\nGitOrigin-RevId: 2c7c89d94b4598b1daddecbf0c4c451ce72c3bf4\n" }, { "commit": "02365829074742daabda39ba2f1f95aae5d40e0a", "tree": "8079b871feb5d76dad6a9545a2f11a3d3822e1a3", "parents": [ "48734f031a06db0d10f96ec2f01b2e3eb663fed2" ], "author": { "name": "Alexey Karyakin", "email": "akaryaki@qti.qualcomm.com", "time": "Thu Nov 12 11:48:00 2026 -0600" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 09:56:48 2026 -0800" }, "message": "Hexagon hmx (#240595)\n\nAdd driver options and assembler-level support for Hexagon Matrix\neXtensions (HMX), implemented in Hexagon v68 and v73. HMX is an\naccelerator for matmul and convolution machine learning workloads.\n\nThere are several classes of instructions:\n\n- matrix multiply: activation and weight; these two instructions must be\n paired in one packet,\n- bias load and store,\n- convert transfer,\n- write convert state,\n- accumulator clear.\n\nOnly IEEE fp16 (hf) ISA variant is supported at this time.\n\nThe new target feature \"hmx\" is added to enable the HMX extensions.\nIn addition, new command line options are added: \"-mhmx\" in llvm-mc and\n\"-mhmx\" / \"-mno-hmx\" in the clang driver.\n\nIntrinsics and builtins will follow.\n\nReference:\n\nhttps://docs.qualcomm.com/doc/80-N2040-62/80-N2040-62_REV_AA_Qualcomm_Hexagon_V81_HMX_Programmers_Reference_Manual.pdf\n\nCo-Authored-By: Brendon Cahoon \u003cbcahoon@quicinc.com\u003e\nGitOrigin-RevId: 34f2fb6b56d03b62885900c86f5c83e4c9d63641\n" }, { "commit": "48734f031a06db0d10f96ec2f01b2e3eb663fed2", "tree": "3ca02d1771e7cbba6268e11d1b899a8f1c211be0", "parents": [ "19f8fe4dbdfbbac200377312a5a99a7f3bbdca34" ], "author": { "name": "Helena Kotas", "email": "hekotas@microsoft.com", "time": "Thu Nov 12 09:47:13 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 09:56:14 2026 -0800" }, "message": "[HLSL] Handle `this` in dynamic resource indexing validation (#239876)\n\nFixes an `llvm_unreachable` assert in dynamic resource index validation\nwhen a resource member is accessed in a member function using `this`.\n\nFixes #239858\n\nGitOrigin-RevId: acd1c6c90f3f36b8f320d4eb39ce8de4882fe461\n" }, { "commit": "19f8fe4dbdfbbac200377312a5a99a7f3bbdca34", "tree": "145f3516af5f4076fa429e0d78af90a936a0bd8b", "parents": [ "f48a3eca60fc2cb90aad55ed5320924b07c87e81" ], "author": { "name": "Kaitlin Peng", "email": "kaitlinpeng@microsoft.com", "time": "Thu Nov 12 09:34:31 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 09:37:56 2026 -0800" }, "message": "[Matrix] Implement matrix support for the `mad` intrinsic (#239357)\n\nCloses #184510.\n\nThis PR implements the matrix api for `mad` and adds matrix codegen\ntests and matrix sema tests. Matrix backend tests were not added because\nno DirectX or SPIRV backend changes were made.\n\nAssisted-by: Claude Opus 5.5\nGitOrigin-RevId: 70882fe35c716c07514f5cb3054e3041f99c902d\n" }, { "commit": "f48a3eca60fc2cb90aad55ed5320924b07c87e81", "tree": "500921e24a8a4ec5fdb1f7af2fee9f1859164fbd", "parents": [ "aad2f349caa80cd4cc7a32280d10c4b98edbae90" ], "author": { "name": "Fangrui Song", "email": "i@maskray.me", "time": "Thu Nov 12 09:24:00 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 09:28:35 2026 -0800" }, "message": "[Support] Pass tool options to parseCommandLineOptions (#240009)\n\nAdd a `parseCommandLineOptions` overload that takes the libraries a tool\nopts in to. Replace the global state `RegisterMCToolOptions` with\n`mc::getToolOptions` and `codegen::getToolOptions`, which tools pass\nwhere they parse, including lld\u0027s `-mllvm`, gold\u0027s `-plugin-opt`, and\nlibLTO\u0027s debug options.\n\n`RegisterCodeGenFlags` no longer registers the MC options.\n\nclang-linker-wrapper does not pass the MC options, so it now rejects\nthem in `-mllvm` and `--offload-opt`, which it forwards to the device\nand host link steps. Its wrapper module uses the module\u0027s own context\ninstead of the user\u0027s `-mllvm`, as discussed on #233513.\n\n`LibraryOptionsContext` records the library that parsed each argument,\nso `assign`, which runs on each ThinLTO backend thread, replays without\nlooking up the registry.\n\nLLM-aided\n\nGitOrigin-RevId: bda37de90d5db6ecdbbe23acad5c0b7cc3cd101f\n" }, { "commit": "aad2f349caa80cd4cc7a32280d10c4b98edbae90", "tree": "350fdd5c0de6f8cdae6c56fa5a45001f8bf158f9", "parents": [ "7275dc6b8385bb74cb31d5778c79fd3462e791f9" ], "author": { "name": "Ivan R. Ivanov", "email": "iivanov@nvidia.com", "time": "Thu Nov 12 18:04:12 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 09:12:04 2026 -0800" }, "message": "[offload] Add OpenACC Offload Kind (#239432)\n\nAdd the new OFK_OpenACC to the OffloadKind enum to represent device code\ngenerated by OpenACC compilation. The __tgt_acc_register_lib function\nwill be implemented by the offload/libacctarget runtime.\n\nAssisted-by: Claude\n\n---------\n\nCo-authored-by: Yury Plyakhin \u003cyury.plyakhin@intel.com\u003e\nGitOrigin-RevId: 276396965805972d40f1afb18c48d7be653fad63\n" }, { "commit": "7275dc6b8385bb74cb31d5778c79fd3462e791f9", "tree": "8ab55141e694b1e717f3eb8ad5154d4d01015da6", "parents": [ "479a58f19458929fb75408c88ffe973d655e5f64" ], "author": { "name": "Mintsuki", "email": "mintsuki@protonmail.com", "time": "Thu Nov 12 17:56:09 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 09:04:19 2026 -0800" }, "message": "[MSP430] Define the registers of the exception pointer and selector\n\nGCC passes the exception pointer and the selector in R12 and R13: its\n`__builtin_eh_return_data_regno` gives 12, 13 and 14 for 0, 1 and 2. Use\nR12 and R13, whose DWARF numbers are 12 and 13 in LLVM too, so that an\nunwinder that sets them with `_Unwind_SetGR` reaches them, and make\nClang\u0027s `__builtin_eh_return_data_regno` return those numbers for\nMSP430, where it gave -1.\n\nAssisted-by: Claude Code (Claude Opus 5.5)\nSigned-off-by: Mintsuki \u003cmintsuki@protonmail.com\u003e\nGitOrigin-RevId: 6cfa183986ae2b79a1e0e35aab346ae62e321a94\n" }, { "commit": "479a58f19458929fb75408c88ffe973d655e5f64", "tree": "fbd65694e2d72f6ed00232d3c2c1f1138f50c675", "parents": [ "4b9e22e13b354190cb36b5f6e79441b2eb883184" ], "author": { "name": "Chi-Chun, Chen", "email": "chichun.chen@hpe.com", "time": "Thu Nov 12 10:27:19 2026 -0600" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 08:32:47 2026 -0800" }, "message": "[OpenMP][clang][flang] Compare tied context selectors with each other (#233905)\n\nWhen several context selectors tie for the highest score, a selector\nthat is a strict subset of another tied selector loses, and otherwise\nthe first one wins. getBestVariantMatchForContext only compared each\ncandidate with the current best, which isn\u0027t enough because that\ncomparison isn\u0027t transitive:\n\n```\n #pragma omp metadirective \\\n when(user\u003d{condition(1)}, \\\n implementation\u003d{vendor(score(4): llvm)}: taskyield) \\\n when(device\u003d{kind(cpu)}, \\\n implementation\u003d{vendor(score(3): llvm)}: taskwait) \\\n when(user\u003d{condition(1)}, device\u003d{kind(host)}, \\\n implementation\u003d{vendor(score(3): llvm)}: barrier)\n```\n\nAll three score 5 on the host. The first is a strict subset of the\nthird, so TASKWAIT should be selected. The scan kept the first over the\nsecond, replaced it with the third, and never compared the third with\nthe second, so it selected BARRIER.\n\nThis patch collects all variants with the highest score, drops any that\nis a strict subset of another of them, and selects the first one left.\nAs before, a strict subset only loses to a variant with the same score.\nThis affects metadirective and declare variant selection in Clang and\nFlang.\n\nFlang reruns the selection for each runtime alternative, so comparing\nevery tied variant with all the others can slow compilation when there\nare many. The tie-break first checks the supersets it has already found,\nso tied variants that share a few supersets need only a few comparisons\neach. The header now documents the tie-break.\n\nFixes #233893.\n\nAssisted-by: GitHub Copilot (Claude Opus 5.5)\nGitOrigin-RevId: 56adb6e7a7e3f1cec31bcd75cb19f00bc7903f21\n" }, { "commit": "4b9e22e13b354190cb36b5f6e79441b2eb883184", "tree": "408baa8fb27c1d63fbfda467082f92cf2d6b9450", "parents": [ "47bebae94f3eebcd50eb41fb21066546ee68fdb0" ], "author": { "name": "Matt Arsenault", "email": "Matthew.Arsenault@amd.com", "time": "Thu Nov 12 08:24:00 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 08:28:26 2026 -0800" }, "message": "OpenMP: Stop using TTI to pick the partial unroll factor in OMPIRBuilder (#240030)\n\nWhen the unrolled loop\u0027s CanonicalLoopInfo is needed by an enclosing\nloop directive and no factor was given, unrollLoopPartial computed one\nby constructing its own TargetMachine with default TargetOptions and a\nstandalone FunctionAnalysisManager, then running LoopUnrollPass\u0027s cost\nmodel on the function while it was still being built.\n\nThis is a layering violation. Frontends should not be using codegen\noptimization information to construct the original program. This will\nnot have a properly configured TargetTransformInfo from a pass\nmanager. The cost model is also not meaningful on incomplete,\nunoptimized IR.\n\nUse a factor of 2, matching what Sema\u0027s OpenMP loop transformation\nuses for a partial clause without a factor. This removes the hidden\n-openmp-ir-builder-unroll-threshold-factor option.\n\nCo-authored-by: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\nGitOrigin-RevId: bc5acaac64c49bcbb77fd2d92eec917e7e6ccbbe\n" }, { "commit": "47bebae94f3eebcd50eb41fb21066546ee68fdb0", "tree": "ffeb5c74f2866b6411549cab56747976ca92cff6", "parents": [ "5d7e66b75be65cd004d568fb376f622be0ce1c14" ], "author": { "name": "Vitaly Buka", "email": "vitalybuka@google.com", "time": "Thu Nov 12 08:18:44 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 08:23:03 2026 -0800" }, "message": "[NFCI][Clang][CFI] Extract EmitVTableCheck helper (#241636)\n\nExtract `CodeGenFunction::EmitVTableCheck` from `EmitVTablePtrCheck`, so\n`EmitVTableTypeCheckedLoad` can report failed checks the same way.\n\nWith `-fsanitize-stats`, the existing `@__sanitizer_stat_report` call\nnow\ncomes after `@llvm.type.test` instead of before it. The two instructions\ndo not depend on each other, so the order has no effect.\n\nAssisted-by: Gemini\nGitOrigin-RevId: 1db8ce340a37c766dc90632bacd3c653989af218\n" }, { "commit": "5d7e66b75be65cd004d568fb376f622be0ce1c14", "tree": "5a676b1229b39ec580caa250e5592a806872c360", "parents": [ "5f827c49487720d9db168c753bba4a896b65b06a" ], "author": { "name": "Benedek Kaibas", "email": "82393336+benedekaibas@users.noreply.github.com", "time": "Thu Nov 12 17:18:17 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 08:22:37 2026 -0800" }, "message": "[NFC][analyzer] Add regression tests from bisection commits to the CSA\u0027s test suite (#241058)\n\nSome issues reported in the Clang Static Analyzer got fixed by bisected\ncommits, but those commits did not include the motivating examples from\nthe reported issues. This PR adds some of those test cases to existing\ntest files.\n\nThe issues that were not covered by test cases and now are by this PR\nare the following: https://github.com/llvm/llvm-project/issues/56264,\nhttps://github.com/llvm/llvm-project/issues/47932,\nhttps://github.com/llvm/llvm-project/issues/39448,\nhttps://github.com/llvm/llvm-project/issues/60694,\nhttps://github.com/llvm/llvm-project/issues/26595,\nhttps://github.com/llvm/llvm-project/issues/59870,\nhttps://github.com/llvm/llvm-project/issues/64647,\nhttps://github.com/llvm/llvm-project/issues/43015,\nhttps://github.com/llvm/llvm-project/issues/3206,\nhttps://github.com/llvm/llvm-project/issues/40797,\nhttps://github.com/llvm/llvm-project/issues/49330,\nhttps://github.com/llvm/llvm-project/issues/20099,\nhttps://github.com/llvm/llvm-project/issues/19340\n\nWhile adding https://github.com/llvm/llvm-project/issues/60694 I used\n`creduce` to reduce the size of the code example mentioned in the issue.\n\n*AI Policy*: These bisection commits were found by AI. I have revisited\neach issue and verified that the mentioned commits actually fix the\nissue. I have not used AI for editing the existing test cases. To make\nsure the test cases this PR adds are not present in the test suite I\nhave asked an LLM to scan the test suite of the Clang Static Analyzer\nwhich I verified manually as well.\n\nGitOrigin-RevId: d4d7ec0475e2f6aac6a7ace58cb3a4148d9e1620\n" }, { "commit": "5f827c49487720d9db168c753bba4a896b65b06a", "tree": "b23c9bde08d224758d300451c2ba160617a42a26", "parents": [ "81979bf0dc9035dd685fcddbaba06d737edd1214" ], "author": { "name": "Chaitanya", "email": "Krishna.Sankisa@amd.com", "time": "Thu Nov 12 21:36:37 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 08:12:31 2026 -0800" }, "message": "[CIR] Flatten one-field structs in CallConvLowering (#241190)\n\nSummary:\n\n- On AMDGPU, struct whose only field is an aggregate such as `struct {\nint V[4]; }` is passed as that field.\n- OGCG gives `[4 x i32]`. Example:\n[func_single_array_element_struct_arg](https://github.com/llvm/llvm-project/blob/7d25e4f9cd72686727beaf0b661ce3e910ee1349/clang/test/CodeGenOpenCL/amdgpu-abi-struct-coerce.cl#L309)\ntest in `amdgpu-abi-struct-coerce.cl`.\n- CIR passed the whole struct, because `getFlattenedCoercedType` in\n`CIRABIRewriteContext` skipped struct with one field.\n- This patch lets the rewriter flatten struct with one field. Returns\nare not flattened as in OGCG.\n\nRelated to issue: https://github.com/llvm/llvm-project/issues/220471\n\nAssisted by: claude opus 5.5\n\nGitOrigin-RevId: b226c304483110e109fa7d36bd23b79e68b95c1a\n" }, { "commit": "81979bf0dc9035dd685fcddbaba06d737edd1214", "tree": "620177fc8b6f8f104d3809ecb5f1672c54c86d85", "parents": [ "1235ebfb1b1cddba1cd45d660f73f27f7408bcde" ], "author": { "name": "Mintsuki", "email": "mintsuki@protonmail.com", "time": "Thu Nov 12 16:55:43 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 08:02:54 2026 -0800" }, "message": "[clang][Sema] Drop the doubled space of the unsupported type diagnostic (#241816)\n\n`err_target_unsupported_type` puts the optional bit size in a select\nthat is followed by a space, so when the select is empty, the message\nhas two spaces after \"requires\":\n\n```\nerror: \u0027x\u0027 requires \u0027long double\u0027 type support, but target \u0027x86_64-unknown-linux-gnu\u0027 does not support it\n```\n\nThe size is left out for long double on targets without it, such as x86\nwith -mno-80387, for float and double return types on targets without\nfloating-point returns, and for `_BitInt` in device compilation.\n\nMove the space into the select, so the message reads\n`requires \u0027long double\u0027 type support` without the size, and\n`requires 128 bit size \u0027__float128\u0027 type support` with it, as before.\n\nAssisted-by: Claude Code (Claude Opus 5.5)\nGitOrigin-RevId: e6d8bed099a2cc1d7146f9146c50683b15637b0b\n" }, { "commit": "1235ebfb1b1cddba1cd45d660f73f27f7408bcde", "tree": "52db081bf2909fb34d2df4dba376f175d429bcc5", "parents": [ "1ef266db5269ce5f4cb0bbdc86255dca4ad87a47" ], "author": { "name": "Henrich Lauko", "email": "xlauko@mail.muni.cz", "time": "Thu Nov 12 16:19:43 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 07:22:22 2026 -0800" }, "message": "[CIR] Match classic codegen for saturating fixed-point negation and padded conversions (#236702)\n\nThree ClangIR fixed-point paths diverged from classic codegen:\n\n- Unary minus on a saturating type emitted a plain subtraction. It is\nnow a saturating `0 - x`.\n- With `-fpadding-on-unsigned-fixed-point`, float to unsigned\nfixed-point conversion produced the `fptosi.sat` result in an unsigned\nCIR type, so its clamp to 0 compared as unsigned and never fired. It now\nconverts and clamps in a signed type, like `llvm::FixedPointBuilder`,\nand casts to the unsigned type afterwards.\n- A saturating conversion that kept the width but changed signedness\nreturned a value of the source CIR type, which broke the `cir.cmp`\nverifier and needed a pointer bitcast to store. It now casts to the\ndestination type.\n\nAssisted-by: Claude Code (Claude Opus 5.5).\nGitOrigin-RevId: c3daf3f3b7cedc8e6f33591a3d87377d36618449\n" }, { "commit": "1ef266db5269ce5f4cb0bbdc86255dca4ad87a47", "tree": "00ff4fe5c8ecf74ccf64d1f5246def8d5c249fac", "parents": [ "b7db050d910d645f3d6cd450bd0ed8f6337c8d80" ], "author": { "name": "Erich Keane", "email": "ekeane@nvidia.com", "time": "Thu Nov 12 06:59:50 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 07:03:11 2026 -0800" }, "message": "[CIR] Add relative_cxx_abi_vtables to lowering-options (#241040)\n\nThis is a first in quite a few patches that I intend to commit towards\ngetting us \u0027relative-vtable\u0027 support. I believe that CIR-Gen should\nabstract this away, so I am going to move get_virtual_fn_addr to return\nan abstract type instead. This patch will let LowerToLLVM handle all the\n\u0027heavy lifting\u0027 of figuring out the differences between pointer and\noffset addresses.\n\nGitOrigin-RevId: 6b8b5622d97111ae0994e180e90433037cc18edc\n" }, { "commit": "b7db050d910d645f3d6cd450bd0ed8f6337c8d80", "tree": "1a1b5f0da0067de804293fc83e4a70b1d1eadde6", "parents": [ "1859fa8fb7d49b8be38570094f8f46716317d82e" ], "author": { "name": "Erich Keane", "email": "ekeane@nvidia.com", "time": "Thu Nov 12 06:46:08 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 06:53:36 2026 -0800" }, "message": "[CIR] Disable ClangIR modules test on ARM too (#241901)\n\nThe commit #239686 introduced some run lines here, but the lack of the\narm abi caused some build bots to fail on this. AIX had the same problem\nand was disabled the same here: #241451\n\nGitOrigin-RevId: deb776437ad9e900da5f78ef27bd3a30f1d26ce1\n" }, { "commit": "1859fa8fb7d49b8be38570094f8f46716317d82e", "tree": "ba9a052af5be37d6b9942968e18c22736ba88bd0", "parents": [ "3ad9da591c0734edca8bf77886e09e7e2251b733" ], "author": { "name": "Daniel Petrovic", "email": "daniel-dev@hotmail.de", "time": "Thu Nov 12 15:27:33 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 06:34:01 2026 -0800" }, "message": "[Clang] Accept mixed GNU and standard attributes in namespace scope (#240734)\n\nParse leading GNU and standard attributes in namespace scope using the\nsame loop as global declarations. Previously, a standard attribute\nfollowing a GNU attribute was rejected with \"an attribute list cannot\nappear here\".\n\nFixes #240537.\n\nGitOrigin-RevId: 4e9d349ab86dc0a913db3cba11aa2763d6985aff\n" }, { "commit": "3ad9da591c0734edca8bf77886e09e7e2251b733", "tree": "f190d6e79bc3dd26a6a7ac5b12c202fdbf10584f", "parents": [ "96e505279c867ed1252ceabeb6ae55759a3a921d" ], "author": { "name": "Soham Karandikar", "email": "43989259+skadewdl3@users.noreply.github.com", "time": "Thu Nov 12 19:43:12 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 06:18:01 2026 -0800" }, "message": "[Clang] [OpenMP] Use raw AST statement as a fallback when transformed statement isn\u0027t available for jump check (#234881)\n\nFixes #230923.\n\nTaking a look at the segfault in a debugger and playing around with the\nreproducer, the error happened only when:\n- There\u0027s a non-empty statement before the nested for-loop (I still have\nno clue why this is the case)\n- There\u0027s an interchange directive present\n\nThe segfault was happening in `JumpScopeChecker::BuildScopeInformation`\nwhen processing the statement as an `OMPExecutableDirective`. In case of\nan interchange directive being present, the transformation gets\ndeferred, `getStructuredBlock` returns `nullptr` causing `getBeginLoc`\nto segfault.\n\nThis PR adds a fallback so if the transformed statement returned by\n`getStructuredBlock()` isn\u0027t available, the raw AST statement is used\ninstead.\n\nAssisted-by: GPT Sol 6.1\nI used Sol to decide upon where the fix should be placed.\nGitOrigin-RevId: 01a45a3c8dcd2e4643d60caf80e7ae693a68bccd\n" }, { "commit": "96e505279c867ed1252ceabeb6ae55759a3a921d", "tree": "5cda9edfb6ee156430608577592c5d47a0c16c17", "parents": [ "a3b6e7932e1be2372e2d7bad5cd46161978a3a50" ], "author": { "name": "Erich Keane", "email": "ekeane@nvidia.com", "time": "Thu Nov 12 05:57:32 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 06:12:03 2026 -0800" }, "message": "[CIR] Abstract the \u0027load\u0027 of a vtable pointer into the get_virtual_fn op (#241048)\n\nIn anticipation of \u0027relative\u0027 vtables, this abstracts the \u0027relativeness\u0027\nas well as the type of the vtable pointer back into cir.vptr. With this\npatch, we\u0027ve removed all of this down to the LowerToLLVM, which can use\nthe relative_cxx_abi_vtables lowering flag (see #241040) to eventually\nemit this correctly.\n\nAt the moment, this is effectively a \u0027NFC\u0027, as everything that actually\nuses relative layouts is either already a CIRGen NYI (left in place\n until those are implemented), or will be added as an unreachable-NYI\ndiagnostic in #241040.\n\nGitOrigin-RevId: 95b733c2fb85d76d6ec74ebb8b381ef004e9f18f\n" }, { "commit": "a3b6e7932e1be2372e2d7bad5cd46161978a3a50", "tree": "ade04a1c8c7e17d62693d7e6bbce51e04eb966c5", "parents": [ "320598cb9192965c564119d65773210560ab675f" ], "author": { "name": "Edd Dawson", "email": "edd.dawson@sony.com", "time": "Thu Nov 12 13:06:52 2026 +0000" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 05:12:28 2026 -0800" }, "message": "[Driver][TySan][PS5] Enable -fsanitize\u003dtype for PS5. (#241338)\n\nBased on work originally authored by @gbMattN.\n\nGitOrigin-RevId: 2a73e5d8b90b0c9584c80cbd73254544289d9563\n" }, { "commit": "320598cb9192965c564119d65773210560ab675f", "tree": "8f8091f864ab9deac7d29bb4606246a2e8dd0ac9", "parents": [ "e40a8d4fc269f07f3ee57160444512bd1525a414" ], "author": { "name": "Akimasa Watanuki", "email": "mencotton0410@gmail.com", "time": "Thu Nov 12 06:56:51 2026 -0600" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 05:03:32 2026 -0800" }, "message": "[CIR][NFC] Separate shared metadata lowering from OpenCL conversions (#240990)\n\nExtract function metadata dispatch and aggregation into\n`LowerToLLVMMetadata`, including the maximum work-group size conversion\nintroduced in #227715. Keep OpenCL-specific conversions in\n`LowerToLLVMOpenCLMetadata`.\n\nAssisted-by: Codex / GPT-6\nGitOrigin-RevId: 2f9eef9d33ac91c3d2ac48a8a2cdc008d03986cd\n" }, { "commit": "e40a8d4fc269f07f3ee57160444512bd1525a414", "tree": "8aff10eee2c81a23a040ae97c529011ff35ce1b8", "parents": [ "862e473630df4ff03ddb277649699541b49a43f5" ], "author": { "name": "Macro Terra", "email": "13541229370@qq.com", "time": "Thu Nov 12 20:14:28 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 04:19:05 2026 -0800" }, "message": "[Clang][NFC] Re-enable consumed analysis test for chained logical OR (#240849)\n\nRe-enable the three-way logical OR test in `testComplexConditionals1` by\nremoving its `#if 0` guard and stale FIXME. The operands start in the\nunknown consumed state. Keep all six existing expected warnings in the\ntrue and false branches unchanged.\n\nPrepared with assistance from Codex.\n\nGitOrigin-RevId: 041eb01e544d2452f02b74006e3a713a86529540\n" }, { "commit": "862e473630df4ff03ddb277649699541b49a43f5", "tree": "2f484adbce8d9750f7ed99b995dabd1fb4a4439d", "parents": [ "1c0f5a80592972cf97f73c6d63132147a7e3b5c1" ], "author": { "name": "Macro Terra", "email": "13541229370@qq.com", "time": "Thu Nov 12 20:14:06 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 04:18:29 2026 -0800" }, "message": "[Clang][NFC] Re-enable local-class static member diagnostic test (#240834)\n\nRe-enable the local-class static data member example in\n`clang/test/SemaTemplate/instantiate-static-local.cpp`. Keep its\nexisting `expected-error {{static}}`, remove the `#if 0` guard and stale\ncrash FIXME, and remove the conflicting `expected-no-diagnostics`\nmarker.\n\nThe crash was fixed by e2def106757534b07a2d3ff15ddd48e14b69a66d\n(#182707). A targeted AArch64 snapshot search followed by native X86\nassertions-on builds confirms the boundary: the parent\nc1c4a5a870eb668fec480771c4d002e17362ed18 crashes in\n`LocalInstantiationScope::findInstantiationOf` on both attempts, while\nthe fix emits only the expected illegal-static-member diagnostic and\nexits normally on both attempts.\n\nPrepared with assistance from Codex.\n\nGitOrigin-RevId: d9ff2ae303d278efd8fde8b6436580fdfdaebce1\n" }, { "commit": "1c0f5a80592972cf97f73c6d63132147a7e3b5c1", "tree": "ac290c1fded94198430ba91d195c099daa39aca8", "parents": [ "3b7804cc8552b5a933503f4916ab5ac07bacdc64" ], "author": { "name": "Mariya Podchishchaeva", "email": "Mariya.Podchishchaeva@amd.com", "time": "Thu Nov 12 12:59:31 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 04:03:43 2026 -0800" }, "message": "[CIR] Use `cir.vec.fcmp` op for AArch64 builtins (#241404)\n\nI\u0027ve been splitting integer and floating comparisons and stumbled upon\nthese.\n\nclaude updated the test.\n\nGitOrigin-RevId: b02e25dc58dc147d625fa0cb162576eb137b144f\n" }, { "commit": "3b7804cc8552b5a933503f4916ab5ac07bacdc64", "tree": "4505f775f65b7b73aa25fae59ef084b791f5856f", "parents": [ "7fafe48504f21e257ef6e15e1e42387f61fb790c" ], "author": { "name": "Jonathan Zakharov", "email": "jzakharov@nvidia.com", "time": "Thu Nov 12 03:23:31 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 03:30:11 2026 -0800" }, "message": "[HLSL][SPIRV] Implement -fvk-support-nonzero-base-instance (#239362)\n\nFixes #136952, per discourse\n[here](https://discourse.llvm.org/t/rfc-hlsl-spir-v-what-should-sv-instanceid-sv-vertexid-default-to/92447/4)\n\n@dnovillo @inbelic\n\nAssisted by AI\n\nGitOrigin-RevId: 76c4166728856f7d6ffb19b0d49798aaec2854e4\n" }, { "commit": "7fafe48504f21e257ef6e15e1e42387f61fb790c", "tree": "203bcdfdc5f1393c005e723b477df5d7ffca1f8a", "parents": [ "c6326b0d29a20595d4a82e044f162b488aaf7cf8" ], "author": { "name": "Jonathan Zakharov", "email": "jzakharov@nvidia.com", "time": "Thu Nov 12 03:23:00 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 03:29:23 2026 -0800" }, "message": "[VK][HLSL][SPIRV] Implement vk::builtin attribute (#231268)\n\nFixes https://github.com/llvm/llvm-project/issues/136898\n\nPR 1/2 in a series; Prerequisite to\nhttps://github.com/llvm/llvm-project/pull/231275 and also related to\nhttps://github.com/llvm/offload-test-suite/pull/1567\n\n@dnovillo @inbelic\n\nAssisted by LLM\n\nGitOrigin-RevId: f4871b2283856093d47f1bdf964a3126db8d49ce\n" }, { "commit": "c6326b0d29a20595d4a82e044f162b488aaf7cf8", "tree": "c2e05606e78049bcca0fa814e6a990d8d054bc84", "parents": [ "30782a069d4964809ff92739a2c608ff7cff1961" ], "author": { "name": "Steffen Larsen", "email": "sholstla@amd.com", "time": "Thu Nov 12 11:50:43 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 03:00:16 2026 -0800" }, "message": "[clang][PCH] Add x86 target requirement to PCH test (#241763)\n\nPCH/pch-clangast-raw-section.c was failing on non-x86 systems in\npost-commit testing of #237198. This is fixed by adding the appropriate\ntarget requirement to the test.\n\nSigned-off-by: Steffen Holst Larsen \u003csholstla@amd.com\u003e\nGitOrigin-RevId: 342b8447062701d5efea8a57f1950e8cbf98ccb7\n" }, { "commit": "30782a069d4964809ff92739a2c608ff7cff1961", "tree": "7801fd0489d48643813cd7e0103c8f2cd7d945d4", "parents": [ "d1fae7b3886e4c0f4373042c2beb99b83b6fb26d" ], "author": { "name": "Donát Nagy", "email": "donat.nagy@ericsson.com", "time": "Thu Nov 12 10:33:49 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 01:38:05 2026 -0800" }, "message": "[analyzer] Correct logic in 3-param addTransition (#241415)\n\nThe method `CheckerContext::addTransition` should create a new exploded\nnode when (1) there was a state change or (2) an explicit non-null `Tag`\nis specified.\n\nHowever, in the 3-parameter overload (which accepts an exploded node\n`Pred` to allow chaining transitions) this logic compared the `State` to\n`getState()` (the state at the construction of `CheckerContext`) and\nreturned early with `getPredecessor()` (the node that was passed to the\nconstructor of `CheckerContext`) instead of `Pred`.\n\nThis commit corrects this oversight by using `Pred` and its state -- but\npreserves the corner case that a null `Pred` is accepted and handled as\nif it was `getPredecessor()`. (Note that passing a null node to\n`addTransitionImpl` is equivalent to passing `getPredecessor()`, some\ncheckers -- at least `FuchsiaHandleChecker` leak reporting -- rely on\nthat `addTransition` can gracefully handle a null node argument.)\n\nThis may perturb node creation in rare cases (within the LIT tests there\nwere two occurrences of the same difference: the new early return\nactivated in `MPIChecker` in cases where the old one didn\u0027t fire) but I\ndon\u0027t expect a significant impact. I verified that `MPIChecker` will\nfunction correctly with the new behavior of`addTransition`.\n\nGitOrigin-RevId: 16da446d09a4eff8b6e7946e8f3429c9156f717f\n" }, { "commit": "d1fae7b3886e4c0f4373042c2beb99b83b6fb26d", "tree": "309daa679f9f6d608ced7cfeebdf0599ddb3160d", "parents": [ "330f12cdf791084fea1f02899f39ec2a354c1abc" ], "author": { "name": "Hamed", "email": "hamedghorbani17@live.com", "time": "Thu Nov 12 01:28:47 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 01:34:04 2026 -0800" }, "message": "[clang][ARM] Add the MSVC compiler barriers for 32-bit Arm (#240572)\n\n_ReadWriteBarrier, _ReadBarrier and _WriteBarrier were not builtins on\n32-bit Arm. intrin.h declares them for every target, so a call became a\ncall to an external function that doesn\u0027t exist, and #pragma intrinsic\nwarned about them. Add them as builtins. They lower to a single-thread\nfence through MSVCIntrin, which x86 and AArch64 now use too. sqlite\nuses them.\n\nAssisted-by: Claude Opus 5.5 (Anthropic)\nGitOrigin-RevId: 439f34c95f16945ce0e8c8e7904283a503e1c662\n" }, { "commit": "330f12cdf791084fea1f02899f39ec2a354c1abc", "tree": "2010bfacd910874e666d25a85e767f1f5618763c", "parents": [ "12ea68c5ad6be6967e8d87c77841554fd2f6c3a6" ], "author": { "name": "Hamed", "email": "hamedghorbani17@live.com", "time": "Thu Nov 12 01:26:47 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 01:33:29 2026 -0800" }, "message": "[clang][ARM] Accept pcs(\"aapcs-vfp\") on Windows (#240569)\n\nWindows on ARM always uses AAPCS-VFP, but pcs(\"aapcs-vfp\") gave a\nwarning there that the calling convention is not supported, which is an\nerror with -Werror. Skia uses the attribute. Ignore it on Windows, like\nthe x86 conventions, so it means the default convention, which it is.\n\nAssisted-by: Claude Opus 5.5 (Anthropic)\nGitOrigin-RevId: 2034dbe97220c044d6a4f4488e0379d43ea41bd8\n" }, { "commit": "12ea68c5ad6be6967e8d87c77841554fd2f6c3a6", "tree": "76ecebe33f9295ead13564e59cd9e036b579ee75", "parents": [ "4560ef2ee90f217f81e73c81d2366a019836c2f4" ], "author": { "name": "Donát Nagy", "email": "donat.nagy@ericsson.com", "time": "Thu Nov 12 10:24:52 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 01:32:48 2026 -0800" }, "message": "[NFC][analyzer] Simplify CheckerContext::generateNonFatalErrorNode (#241346)\n\nThe three-parameter form of `CheckerContext::generateNonFatalErrorNode`\nhad an early return statement which wasn\u0027t activated in any of the 6\ncall sites of this method, because each caller always passed a non-null\nstate (the state of an existing exploded node).\n\nBy the way, this early return was also logically incorrect, so a\ndownstream checker that relied on it would be already very buggy.\n\nI also refactored four of the six call sites of this overload to use the\nsimpler two-parameter overload (without passing any argument, using the\ndefaults).\n\nGitOrigin-RevId: c6b5a98c107ff7d69728cf706338c74c5f91bcba\n" }, { "commit": "4560ef2ee90f217f81e73c81d2366a019836c2f4", "tree": "d38f24deb8ced00bbf4da7f55d2140219b2f128f", "parents": [ "02da45584f40700f47947e2207fdc08e2fb34d14" ], "author": { "name": "Donát Nagy", "email": "donat.nagy@ericsson.com", "time": "Thu Nov 12 10:23:58 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 01:28:19 2026 -0800" }, "message": "[NFC][clang][Sema] Move three tests to SemaCXX (#240864)\n\nMove three C++ lit test files from `clang/test/Analysis` to\n`clang/test/SemaCXX` because that is where the bulk of the tests for\nthese compiler warnings (`-Wthread-safety`, `-Wuninitialized`,\n`-Wconsumed` and `-Wunreachable-code-aggressive`) are located.\n\nFor background information, see\nhttps://discourse.llvm.org/t/untangle-static-analyzer-and-ssaf-tests/91835/13\n\nGitOrigin-RevId: 7de9fe9afe1abe002f7076068ff6deb6a82262cb\n" }, { "commit": "02da45584f40700f47947e2207fdc08e2fb34d14", "tree": "a8ceea346ad0568731d4d1120a11059af6a7f61f", "parents": [ "5675e2497d73bf9ff6afd15b85ffa962e49235f9" ], "author": { "name": "AMullen", "email": "andreas.mullen@arm.com", "time": "Thu Nov 12 09:19:58 2026 +0000" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 01:23:49 2026 -0800" }, "message": "[Clang][Docs] Improve -fcs-profile-generate documentation (#239616)\n\nThe current documentation for `-fcs-profile-generate` does not clearly\nexplain the additional profiling stage or the need to combine the base\nand context-sensitive profiles to produce a CSPGO-optimized executable.\nThis PR makes these requirements clearer, and also expands on the\ncurrent example to highlight these points.\n\nGitOrigin-RevId: 9b066693658e5e6c86602a51c0ca6a828ae62b82\n" }, { "commit": "5675e2497d73bf9ff6afd15b85ffa962e49235f9", "tree": "e183633a1cd578b83af75b7349f897a843a7d8fd", "parents": [ "afa83565de1a094683d28947c33182e6b3324454" ], "author": { "name": "Steffen Larsen", "email": "sholstla@amd.com", "time": "Thu Nov 12 10:15:59 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 01:23:17 2026 -0800" }, "message": "[Clang][PCH] Use llvm.raw.sections for embedding the AST (#237198)\n\nTo avoid the size restriction that may be imposed by the address spaces\nof the target, embed the serialized Clang AST through the\n`llvm.raw.sections` named metadata onto a section that is not loaded,\ninstead of placing it in the `__clang_ast` global variable. The section\nnames and the 8-byte alignment are unchanged.\n\nSecond and last part of #214135.\n\nAssisted-by: Claude Opus 5.5\n\nSigned-off-by: Steffen Holst Larsen \u003csholstla@amd.com\u003e\nGitOrigin-RevId: 0536d9f5224ed4b2b9aa2200b92975d4d86a88b8\n" }, { "commit": "afa83565de1a094683d28947c33182e6b3324454", "tree": "109e749506b698e45480d5599223e9931c37d17b", "parents": [ "ee3ffc478b2309f4eb56d60c42bb4d70dd28b7ea" ], "author": { "name": "Markus Böck", "email": "mboeck@nvidia.com", "time": "Thu Nov 12 10:15:14 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 01:22:40 2026 -0800" }, "message": "[mlir][ODS] Include headers required by generated bytecode readers and writers (#241226)\n\nThe generated bytecode readers and writers use `DialectBytecodeReader`,\n`DialectBytecodeWriter` and `TypeSwitch` but did not include the headers\ndeclaring them. They therefore relied on the surrounding source file to\ninclude these headers first.\n\nThe generated readers and writers now include `BytecodeImplementation.h`\nand `TypeSwitch.h`.\n\nAs these includes must not be expanded within a namespace, all users now\ninclude the generated file at global scope rather than within their\nanonymous namespace.\n\nSimilar to #236155 and #238735 but for the bytecode generator.\n\nCo-authored-by: Claude \u003cnoreply@anthropic.com\u003e\nGitOrigin-RevId: b4984ca753763a36a92610441393e9c557b9206a\n" }, { "commit": "ee3ffc478b2309f4eb56d60c42bb4d70dd28b7ea", "tree": "c0dfe640127d46474f6a52f241e57b95d8ddf5c6", "parents": [ "96429d116ab5dbe5914383b1fb0599910e79ad36" ], "author": { "name": "Markus Böck", "email": "mboeck@nvidia.com", "time": "Thu Nov 12 09:25:48 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 00:31:25 2026 -0800" }, "message": "[mlir][ODS] Fully qualify names in generated bytecode readers and writers (#241224)\n\nThe code generated by `-gen-bytecode` referred to MLIR and LLVM\nconstructs such as `MLIRContext`, `DialectBytecodeReader`,\n`LogicalResult`, `TypeSwitch` and `getChecked` without qualification, as\ndid the snippets in `BytecodeBase.td`. It therefore only compiled in a\nsource file with `using namespace mlir;`.\n\nThis PR fully qualifies all names emitted by the generator and by\n`BytecodeBase.td`. Performed a minor refactor while doing so that also\nfixes an error message and makes the code less ad-hoc\n\nCo-authored-by: Claude \u003cnoreply@anthropic.com\u003e\nGitOrigin-RevId: 20bd5249b9518eb148c122ff07af1858ea9290d0\n" }, { "commit": "96429d116ab5dbe5914383b1fb0599910e79ad36", "tree": "54d2be6a96715ded9746a0b521858e887f8885a3", "parents": [ "77547c5c8b6961608f5ad51b78fc1e4f7dcda559" ], "author": { "name": "SiHuaN", "email": "liyongtai@iscas.ac.cn", "time": "Thu Nov 12 15:57:49 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Nov 12 00:02:25 2026 -0800" }, "message": "[RISCV][P-ext] Add scalar ssh1sadd intrinsic (#239234)\n\nAdd the scalar Shift-Add intrinsic from the P-extension intrinsic spec:\n\nhttps://github.com/riscv/riscv-p-spec/blob/master/P-ext-intrinsics.adoc#shift-add\n\n- `__riscv_ssh1sadd_i32`\n\nGitOrigin-RevId: f1ca157013423590536b9585a5a64f8802d4d5f3\n" }, { "commit": "77547c5c8b6961608f5ad51b78fc1e4f7dcda559", "tree": "78815648e944e28df8f48c0f8e16300ebc673ed3", "parents": [ "cd7ed6d610b984f93e23dc5bc7d0eb30a4385d7d" ], "author": { "name": "Timm Baeder", "email": "tbaeder@redhat.com", "time": "Thu Nov 12 07:12:59 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 22:17:39 2026 -0800" }, "message": "[clang][bytecode] Optimize `CheckStore()` (#241345)\n\nBlock pointers are the only writable pointer type, which is why we\u0027re\nchecking for them very early in `CheckStore`. Make the distinction more\nexplicit and use the block and the `PtrView` of the pointer explicitly.\n\nGitOrigin-RevId: 2b19c3f3cdb3ea11052abcc09db948f8644c289d\n" }, { "commit": "cd7ed6d610b984f93e23dc5bc7d0eb30a4385d7d", "tree": "e34a828da098e994ad7a6e45199fb22e4c56fb84", "parents": [ "fa6a910b9be7e19aacd4aa7af88c1c537ffc0323" ], "author": { "name": "Timm Baeder", "email": "tbaeder@redhat.com", "time": "Thu Nov 12 06:49:34 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 22:10:19 2026 -0800" }, "message": "[clang][bytecode] Remove unreachable code in __builtin_object_size (#240742)\n\nIn the current interpreter, removing the equivalent if statement breaks\nthis test:\n```c\ngi \u003d __builtin_object_size(\u0026t[1].v[14], 3);\n```\n\nbut in the bytecode interpeter, removing this if does not break anything\nas it\u0027s unreachable, see\nhttps://github.com/llvm/llvm-project/pull/213017/changes#r4475679359\n\nGitOrigin-RevId: a60677dca3a8536a0cf5bec1b846610014a983fd\n" }, { "commit": "fa6a910b9be7e19aacd4aa7af88c1c537ffc0323", "tree": "8740aba2529f46955314cd3a1c866acfb949f70d", "parents": [ "0379dcb94f52269c099c5fff6204e9f2512762fe" ], "author": { "name": "Arseniy Obolenskiy", "email": "arseniy.obolenskiy@amd.com", "time": "Thu Nov 12 05:58:00 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 21:14:54 2026 -0800" }, "message": "[CIR] Fix va_list lowering for non-zero default address space (#238455)\n\nDrop the bitcast to a generic-AS `ptr`, which is invalid for `ptr\naddrspace(4)` on SPIR-V\n\nGitOrigin-RevId: 0a3a3dba31cd1cdfcd04751c34569507c5442764\n" }, { "commit": "0379dcb94f52269c099c5fff6204e9f2512762fe", "tree": "4d5e22fdc70a37f3ab26a422960e1cfac662464f", "parents": [ "314e4548fefc2e7d75a6ecb2d3e1210da7959c8a" ], "author": { "name": "Jan Svoboda", "email": "jan_svoboda@apple.com", "time": "Wed Nov 11 20:50:03 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 21:14:25 2026 -0800" }, "message": "[clang][deps] Give `LocalFS` precedence in the scanning VFS overlay (#241464)\n\nThe overlay of `LocalFS` and the global FS that\n`DependencyScanningWorkerFilesystem`\nforwards unhandled queries to (e.g. directory iteration) put the global\nFS on\ntop. That\u0027s inconsistent with `status()`, `openFileForRead()` and\n`getRealPath()`,\nwhich look at `LocalFS` first. Put `LocalFS` on top of the overlay\ninstead.\n\nDepends on #241463, only review the second commit.\n\nGitOrigin-RevId: 62ac48c44f922bf60e91d7e43f522223aa7d1eb9\n" }, { "commit": "314e4548fefc2e7d75a6ecb2d3e1210da7959c8a", "tree": "8b19bf7d3d11c444e95245804e84515326a4e342", "parents": [ "b14c31ee44934b76d16348b19666799784a58f8c" ], "author": { "name": "Arseniy Obolenskiy", "email": "arseniy.obolenskiy@amd.com", "time": "Thu Nov 12 05:49:51 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 21:12:44 2026 -0800" }, "message": "[CIR][OpenMP] Add registerTargetGlobalVariable for global variables (#234077)\n\nLimit the NYI to declare target variables so other globals emit under\n-fopenmp\n\nGitOrigin-RevId: 01c460448069b3e596f869158f6ab64fba11bf3a\n" }, { "commit": "b14c31ee44934b76d16348b19666799784a58f8c", "tree": "cbc09fd648cb1c98851576ca0e2c32fb924d0374", "parents": [ "8dc1de22aa57a969d7ec4ec1a8aa322b7d407795" ], "author": { "name": "Arseniy Obolenskiy", "email": "arseniy.obolenskiy@amd.com", "time": "Thu Nov 12 05:49:02 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 21:11:22 2026 -0800" }, "message": "[CIR] Keep source address space in vtable.get_vptr result (#238495)\n\nThe result was always in the default address space, so objects in other\naddress spaces left an unrealized_conversion_cast and failed LLVM\ntranslation\n\nGitOrigin-RevId: 98f1b00ed78751166d08a8563c56d65d5f3b5872\n" }, { "commit": "8dc1de22aa57a969d7ec4ec1a8aa322b7d407795", "tree": "c076be4c3fe254856f5b7b5a9ba39eba2437483b", "parents": [ "90d934d0070c8ec2dfc88b7e5819d553636bca87" ], "author": { "name": "Chaitanya", "email": "Krishna.Sankisa@amd.com", "time": "Thu Nov 12 10:08:44 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 20:43:27 2026 -0800" }, "message": "[CIR] Track flexible array members in record layout and pass to ABI (#241158)\n\nSummary:\n\n- In CIR, a record which has flexible array member was wrongly\nclassified and passed in register.\n- Classic CodeGen passes them byval and returns them through sret.\n- Added `flexible_array_member` to `#cir.record_layout`, set it from\n`RecordDecl::hasFlexibleArrayMember()` in CIRGen. Mapped it to\n`RecordFlags::HasFlexibleArrayMember` in the bridge.\n- Classic test reference:\n[amdgpu-abi-struct-coerce.cl](https://github.com/llvm/llvm-project/blob/main/clang/test/CodeGenOpenCL/amdgpu-abi-struct-coerce.cl)\n([func_flexible_array_arg](https://github.com/llvm/llvm-project/blob/f1185ef40be8c0d6647e8f70a43f1c963d2d3d46/clang/test/CodeGenOpenCL/amdgpu-abi-struct-coerce.cl#L318),\n[func_flexible_array_ret](https://github.com/llvm/llvm-project/blob/f1185ef40be8c0d6647e8f70a43f1c963d2d3d46/clang/test/CodeGenOpenCL/amdgpu-abi-struct-coerce.cl#L450)).\n\nRelates to issue : https://github.com/llvm/llvm-project/issues/220471\n\nAssisted by: claude opus 5.5\n\nGitOrigin-RevId: 0b86baa6b5eb027237e995461e525b0e1808f334\n" }, { "commit": "90d934d0070c8ec2dfc88b7e5819d553636bca87", "tree": "fb117dc3ec20d967196d5618e4f80ac38ab38f01", "parents": [ "e95bb2bbb9adc622a9280df853725f7e1364e250" ], "author": { "name": "Jan Svoboda", "email": "jan_svoboda@apple.com", "time": "Wed Nov 11 19:59:27 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 20:02:29 2026 -0800" }, "message": "[clang][deps] Don\u0027t share real paths from `LocalFS` via the shared cache (#241463)\n\n`DependencyScanningWorkerFilesystem::getRealPath()` computed uncached\nreal paths via the overlay of `LocalFS` and the global FS, but stored\nthe result in the shared cache. Since `LocalFS` is specific to the\nworker, this leaked its real paths to other workers, and could hide\n`LocalFS` real paths behind results cached by other workers.\n\nResolve real paths of files that exist in `LocalFS` through `LocalFS`\nand only cache them locally, the same way `getOrCreateFileSystemEntry()`\ndoes for status and contents. Compute all other real paths through the\nglobal FS.\n\nGitOrigin-RevId: 2934809c8aeea143924dd7e052367ab97df8912e\n" }, { "commit": "e95bb2bbb9adc622a9280df853725f7e1364e250", "tree": "779663c0992de4e8c05c9c0e1c3eb6d08ffa2f55", "parents": [ "e6aee489e3307861394fe3b84cc934865acc54fa" ], "author": { "name": "Akash Manna", "email": "akash.manna.mymail@gmail.com", "time": "Thu Nov 12 09:00:00 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 19:32:03 2026 -0800" }, "message": "[Clang][CodeGen][NFC] Add regression test for repeated `target_clones` redeclarations (#237830)\n\nFixes #149372\n\nA `target_clones` definition, a bare redeclaration and then one\nrepeating the attribute asserted with `Default or Generic case must be\nlast` in the `x86` resolver. Sema did not link the bare redeclaration to\nthe definition, so CodeGen saw two functions named `f` and collected\nevery version twice.\n\n#169259 fixed the `Sema` side, so the attribute is inherited and each\nversion is collected once. This adds the shape from the report.\n\nAssisted-by : LLM tools.\nGitOrigin-RevId: 7d415490eb8b793e900aae869912552176b6b13b\n" }, { "commit": "e6aee489e3307861394fe3b84cc934865acc54fa", "tree": "f7095975b701cfe12ccfaa10aef992024d75fc00", "parents": [ "375660253878a50f394a97c90b14043282f1ee93" ], "author": { "name": "Alexey Karyakin", "email": "akaryaki@qti.qualcomm.com", "time": "Wed Nov 11 18:46:42 2026 -0600" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 17:09:12 2026 -0800" }, "message": "[Hexagon] NFC: Refactor repetitive code in getTargetDefines (#237318)\n\nGitOrigin-RevId: d2de796c607c7a765acd01ae60190267dd23a427\n" }, { "commit": "375660253878a50f394a97c90b14043282f1ee93", "tree": "3e60f23e16f53cb526f40c8dd8d78e93ce5e717d", "parents": [ "f2b73f5ad2417af2f4925426a17820aad449ec00" ], "author": { "name": "Kaitlin Peng", "email": "kaitlinpeng@microsoft.com", "time": "Wed Nov 11 15:46:30 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 15:53:10 2026 -0800" }, "message": "[Matrix] Implement matrix support for the `clamp` intrinsic (#240969)\n\nCloses #184498.\n\nThis PR implements the matrix api for `clamp` and adds matrix codegen\ntests and matrix sema tests. Matrix backend tests were not added because\nno DirectX or SPIRV backend changes were made.\n\nAssisted-by: Claude Opus 5.5\nGitOrigin-RevId: b796466db4cdcc972ce82d38e3df341b782dc886\n" }, { "commit": "f2b73f5ad2417af2f4925426a17820aad449ec00", "tree": "5634000f0c6bd539009cd6bf67adef367fd3aa5b", "parents": [ "3dcddc7276001059b90d9c134a84cdf869fffb75" ], "author": { "name": "Hamed", "email": "hamedghorbani17@live.com", "time": "Wed Nov 11 14:59:00 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 15:03:07 2026 -0800" }, "message": "[clang][MS ABI] Use image-relative throw info on 32-bit Arm (#240571)\n\nMSVC\u0027s runtime reads the throw information (ThrowInfo, CatchableTypeArray and\nCatchableType) as offsets from the image base on every target but 32-bit x86,\nand _CxxThrowException looks up the image base from the ThrowInfo. clang wrote\nthese tables with absolute pointers on 32-bit Arm, so a throw crashed in the\nruntime there. Make them image-relative on Arm, as on 64-bit targets.\nThe RTTI stays absolute, as on MSVC.\n\nAssisted-by: Claude Opus 5.5 (Anthropic)\nGitOrigin-RevId: a7cbbc2a9c973ce951c4c5ecf62ae0cdea97e08d\n" }, { "commit": "3dcddc7276001059b90d9c134a84cdf869fffb75", "tree": "eb1431544153c5319e86a144de37eefc8ea4381f", "parents": [ "014cb26c51413555ada0fb8b0f99ccd93c87cd95" ], "author": { "name": "Yashas Andaluri", "email": "yandalur@qti.qualcomm.com", "time": "Thu Nov 12 03:28:13 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 14:03:36 2026 -0800" }, "message": "[Hexagon] Add HLX driver checks, intrinsic header and Sema diagnostics (#241435)\n\n- Driver: report an error for -mhlx and -mno-hlx on CPUs older than v85,\nwhich do not have HLX (they enabled the HLX feature and defined the\n__HLX__ macros), and pass -mhlx to the assembler when HLX is enabled and\nthe integrated assembler is not used.\n- Headers: add hlx_hexagon_protos.h with the Q6_* macros for the V85 HLX\nbuiltins, include it from hexagon_protos.h when HLX is enabled, and\ndefine the HLX_Vector and HLX_VectorPair types in hexagon_types.h.\n- Sema: HLX vectors can only be local variables, and there are no\ninstructions for element access or vector pair loads and stores. When\nHLX is enabled, reject global, static and variable sized objects of HLX\ntypes, struct, union and class members of HLX types, subscripts of HLX\nvectors, and dereferences of HLX_VectorPair pointers.\n\nCo-authored-by: Adarsha Regmi \u003caregmi@qti.qualcomm.com\u003e\nCo-authored-by: Alexey Karyakin \u003cakaryaki@qti.qualcomm.com\u003e\nCo-authored-by: Ankit Aggarwal \u003caankit@qti.qualcomm.com\u003e\nCo-authored-by: Ikhlas Ajbar \u003ciajbar@qti.qualcomm.com\u003e\nCo-authored-by: Rishabh Bali \u003crbali@qti.qualcomm.com\u003e\n\nAssisted by AI\n\n---------\n\nCo-authored-by: Adarsha Regmi \u003caregmi@qti.qualcomm.com\u003e\nCo-authored-by: Alexey Karyakin \u003cakaryaki@qti.qualcomm.com\u003e\nCo-authored-by: Ankit Aggarwal \u003caankit@qti.qualcomm.com\u003e\nCo-authored-by: Ikhlas Ajbar \u003ciajbar@qti.qualcomm.com\u003e\nCo-authored-by: Rishabh Bali \u003crbali@qti.qualcomm.com\u003e\nGitOrigin-RevId: de84021f326f2ef14c734eead26cb1d208c2c1c8\n" }, { "commit": "014cb26c51413555ada0fb8b0f99ccd93c87cd95", "tree": "31be0ef44778acdae22e7fba5b4981adce94b4c3", "parents": [ "a9bfcc58b940ea8d670043048220fc0988a7bbba" ], "author": { "name": "Adrian Prantl", "email": "aprantl@apple.com", "time": "Wed Nov 11 13:09:42 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 13:12:30 2026 -0800" }, "message": "[clang] Relocate DWARF across sections on Mach-O firmware targets (#239797)\n\nChange the Darwin Toolchain to pass -grelocatable-dwarf for embedded\nMach-O targets when emitting DWARF 5 or later.\n\nAssisted-by: Claude\n\nDepends on https://github.com/llvm/llvm-project/pull/239793\nThis enables workflows that involve doing incremental links that\nconcatenate DWARF contributions.\n\nGitOrigin-RevId: a1ac3cde91e657b5642529ebcf1e40479a6074eb\n" }, { "commit": "a9bfcc58b940ea8d670043048220fc0988a7bbba", "tree": "1902e1137dff040fac195a8adf41266e15415e73", "parents": [ "7d748d7274d050a04f86de6a476fd31f1a2612f3" ], "author": { "name": "Hubert Tong", "email": "hubert.reinterpretcast@gmail.com", "time": "Wed Nov 11 15:01:25 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 12:07:43 2026 -0800" }, "message": "[NFC][test] AIX: Skip Clang IR steps in module-funcs-from-imports.cppm (#241451)\n\nThe Clang IR steps encounter:\n```\nerror: C++ ABI \u0027xl\u0027 is not yet supported by CIR\n```\n\nTracking issue: https://github.com/llvm/llvm-project/issues/241433\n\nResolves CI failure from\nhttps://github.com/llvm/llvm-project/pull/239686.\n\nCo-authored-by: Hubert Tong \u003chstong@ca.ibm.com\u003e\nGitOrigin-RevId: cc6fe608a116ceef9b665cdc4af5bcf188e00c0c\n" }, { "commit": "7d748d7274d050a04f86de6a476fd31f1a2612f3", "tree": "6e1ee50a101a8b971591309b9d9f742a7b1b323d", "parents": [ "d05a644ba3c1779c84fbce3e244eecbe751c0c84" ], "author": { "name": "Bruno Cardoso Lopes", "email": "bruno.cardoso@gmail.com", "time": "Wed Nov 11 11:34:30 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 12:07:17 2026 -0800" }, "message": "[CIR] Prune CIR type subtrees when collecting symbol uses (#234811)\n\napplyReplacements builds a symbol user map over the whole module to find\nthe users of replaced functions, and the generic walk descends into\nevery nested record of every function signature even though no symbol\nreference can hide in a CIR type. Collect only the users of the\nfunctions being replaced, in one pre-order walk that skips CIR type\nsubtrees, and keep chained replacements working by merging the old users\ninto the new symbol\u0027s set.\n\nOn SemaExpr.cpp, applyReplacements drops from 1.7% to 0.4% of compile\nsamples (perf, cycles).\n\nGitOrigin-RevId: 9dce18f0e44d68e45623d9c1b60779c9b70734d7\n" }, { "commit": "d05a644ba3c1779c84fbce3e244eecbe751c0c84", "tree": "543df241866fb3a2b006396d9fd069011338a59f", "parents": [ "b9fb3848bc7516cda87ba636696fc4343bbd4acd" ], "author": { "name": "Bruno Cardoso Lopes", "email": "bruno.cardoso@gmail.com", "time": "Wed Nov 11 11:33:53 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 11:37:25 2026 -0800" }, "message": "[mlir][llvm] Add tbaa_struct field attrs to the LLVM dialect (#238991)\n\nThe LLVM dialect models `!tbaa` access tags but has no representation\nfor `!tbaa.struct`, so aggregate copies with struct TBAA lose that\nmetadata in both directions: export drops it, import warns unhandled\nmetadata.\n\nThis PR adds the missing pieces.\n\nThe importer declines nodes it cannot model (null tags, empty nodes)\nrather than representing them (warning on unhandled metadata), the\nMLIR-side verification of LLVM\u0027s non-decreasing-offset rule is left out.\n\nGitOrigin-RevId: 8a166cd52fff14512f50a92eb85ce289e689b302\n" }, { "commit": "b9fb3848bc7516cda87ba636696fc4343bbd4acd", "tree": "8937c8ae40b7ecc5bb4186897188b8e7a80e1a5e", "parents": [ "1780c221f5031af1c5a4253faa40307f78df970d" ], "author": { "name": "Teresa Johnson", "email": "tejohnson@google.com", "time": "Wed Nov 11 11:32:12 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 11:36:59 2026 -0800" }, "message": "[MemProf][NFC] Standardize on MemPGO terminology and expand docs (#241353)\n\nDocumentation, option help strings, header comments, and tests have\nreferred to LLVM\u0027s memory profile-guided optimization framework\ninconsistently as \"profile guided heap optimization\" (PGHO),\n\"profile-guided memory optimization\", and \"MemProf\". Because the\ncompiler pipeline optimizes both dynamic heap allocations and static\ndata layout using profiles from the MemProf instrumentation and runtime,\nstandardize on \"profile-guided memory optimization (MemPGO)\" for the\noverall optimization framework while reserving \"MemProf\" for the\nprofiler, runtime, and profile format.\n\nUpdates and expands documentation that was in MemProf.md (now moved to\nMemPGO.md with a redirect stub in the old file) to describe more details\nabout how allocation contexts are classified, the hint values used, and\nhow matching and LTO whole-program context disambiguation work.\n\nAssisted-by: Gemini\nGitOrigin-RevId: 8cd8d4c951ced5a8c1c4125cb232f8bf21470c5e\n" }, { "commit": "1780c221f5031af1c5a4253faa40307f78df970d", "tree": "49b15a588ea86762fd5ff006b4c05ccc3c851e17", "parents": [ "2024fd3b21357977730a83fecb1077df6f53b7a6" ], "author": { "name": "Kazu Hirata", "email": "kazu@google.com", "time": "Wed Nov 11 11:15:12 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 11:23:01 2026 -0800" }, "message": "[CIR] Remove unused CIRGenModule::getMLIRVisibility overload (NFC) (#241103)\n\nThe last callers of the CIRGenModule::getMLIRVisibility overload taking\nclang::Visibility were removed on November 10, 2026 in commit\n07a5b49195119547bc6420bde79a5b714153faca.\n\nAssisted-by: Antigravity\nGitOrigin-RevId: fcdef03f49466a34fb8d2c067922985a7e10a601\n" }, { "commit": "2024fd3b21357977730a83fecb1077df6f53b7a6", "tree": "3cf3355c8fa7fdcaf579198a01be14796c0f7d7c", "parents": [ "d071c8f590b978174e275844311d2ce094b8467b" ], "author": { "name": "stephenduong1004", "email": "stephenduong@google.com", "time": "Wed Nov 11 13:09:16 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 10:14:18 2026 -0800" }, "message": "[clang][AST] Fix crash attaching doc comments to decls from header macros (#234708)\n\n[clang][AST] Fix crash attaching doc comments to decls from header macros\n\nAfter #65481, `getLocsForCommentSearch` returns both the macro expansion\nlocation and the macro definition\u0027s spelling location, which may be in a\ndifferent file (e.g., an `#include`d header). Unlike\n`getRawCommentNoCache`, `attachCommentsToJustParsedDecls` passed\n`*CommentsInThisFile` (the first declaration\u0027s file) for every\n`DeclLoc`. #221605 extends this to `-Wdocumentation-html`.\n\nWhen a declaration group (e.g., an Objective-C `@implementation`)\ncontains a main-file declaration followed by a declaration expanded from\na header macro, `getRawCommentNoCacheImpl` searches the main file\u0027s\ncomment map using the header\u0027s byte offset (`LocDecomp.second`). If an\nunattached comment in the main file straddles `LocDecomp.second`,\n`LocDecomp.second - CommentEndOffset` underflows `unsigned` and crashes\nin `StringRef::find_last_of`.\n\n- Look up the `FileID` and comment map for each `DeclLoc` in\n`attachCommentsToJustParsedDecls` (reusing `CommentsInThisFile` when\n`DeclFile \u003d\u003d File`).\n- Guard against `CommentEndOffset \u003e LocDecomp.second` in\n`getRawCommentNoCacheImpl`.\n- Add regression test `clang/test/AST/ast-crash-doc-macro-in-header.m`.\n\nAI-assisted\n\nGitOrigin-RevId: 2247ad95b94b4457a9d8526bc7062d520d78dc4d\n" }, { "commit": "d071c8f590b978174e275844311d2ce094b8467b", "tree": "4349c5be64cdc56c30a2d5f648985091c6e78f1a", "parents": [ "9c2acf6bf1e176aa8b0550d25fcc9427b58ec954" ], "author": { "name": "newgre", "email": "jannewger@gmail.com", "time": "Wed Nov 11 17:32:43 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 08:38:24 2026 -0800" }, "message": "[CopyProf] Intercept new/delete and track per-object cost. (#235455)\n\nThis intercepts the C++ `new` and `delete` operators, so that all\nheap memory owned by a copy is tracked.\n\nUsing per-object state tracking, reports now tell the user how much\naccumulated memory a copy is worth and whether it allocated.\nThis is useful for filtering of reports.\n\nRFC:\nhttps://discourse.llvm.org/t/rfc-copysanitizer-csan-detecting-unneccessary-object-copies-at-runtime/91038\n\nStack: #237763 → #232851 → #235453 → #235454 → #235455 (this PR)\nGitOrigin-RevId: 5e721edefebe889c9c91a29e676e97f580d221bb\n" }, { "commit": "9c2acf6bf1e176aa8b0550d25fcc9427b58ec954", "tree": "2242fb78350099d9ee8f3b521682cbcb6e7f8995", "parents": [ "e4595278202d9f9843fd60de23b04857310ebfe2" ], "author": { "name": "Jan Voung", "email": "jvoung@google.com", "time": "Wed Nov 11 11:06:39 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 08:29:54 2026 -0800" }, "message": "[clang-tidy] For bugprone-unchecked-optional-access add simple BoolSolverModel option (#240848)\n\nAdds a checker option to select the Simple BoolModel (introduced in the\nframework in https://github.com/llvm/llvm-project/pull/231061) for\nhttps://github.com/llvm/llvm-project/issues/224638. This gives users the\nchoice of faster analysis (e.g., one way to address\nhttps://github.com/llvm/llvm-project/issues/69369), but requires \"has\nvalue\" checks to be more straightforward (no correlated branch\nconditions for example).\n\n---------\n\nCo-authored-by: EugeneZelenko \u003ceugene.zelenko@gmail.com\u003e\nCo-authored-by: Zeyi Xu \u003czeyi2@nekoarch.cc\u003e\nGitOrigin-RevId: 85a7e596c0f0e01c2598897133c1232352742173\n" }, { "commit": "e4595278202d9f9843fd60de23b04857310ebfe2", "tree": "c47e0920f1aff573b07641d9535733ca35c3b247", "parents": [ "c8bb0c798d7c64792e5f467da516d8467471fb69" ], "author": { "name": "Joel E. Denny", "email": "jdenny.ornl@gmail.com", "time": "Wed Nov 11 10:13:20 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 07:21:00 2026 -0800" }, "message": "[NvlinkWrapper] opt-remarks-filter -\u003e opt-remarks-passes (#241045)\n\n`renderRemarksOptions` in Clang\u0027s `CommonArgs.cpp` translates\n`-foptimization-record-passes` to `--plugin-opt\u003dopt-remarks-passes`.\n`lld` expects that, but `clang-nvlink-wrapper` currently expects\n`--plugin-opt\u003dopt-remarks-filter` instead, producing errors like:\n\n```\nclang-nvlink-wrapper: Unknown command line argument opt-remarks-passes\u003dinline\n```\n\nIn `NVLinkOpts.td` for `clang-nvlink-wrapper`, this patch adds\n`--plugin-opt\u003dopt-remarks-passes` and makes\n`--plugin-opt\u003dopt-remarks-filter` an alias. If there are no users of the\nlatter, we can consider removing it.\n\nAssisted-by: LLM\nGitOrigin-RevId: 67fe07fc6ca1e1c7e9188e677fa4e368207f6812\n" }, { "commit": "c8bb0c798d7c64792e5f467da516d8467471fb69", "tree": "7ada3c8b5c21269b3aca6d6bb2d9bfe2e0ca0aa4", "parents": [ "2016733da7d5730377ce995c6069352ccb0d2c06" ], "author": { "name": "Henrich Lauko", "email": "xlauko@mail.muni.cz", "time": "Wed Nov 11 16:09:23 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 07:16:07 2026 -0800" }, "message": "[CIR] Make cir.condition exit to the loop, not to itself (#240971)\n\nThe loop-exit successor of `cir.condition` was\n`RegionSuccessor(getOperation())`, which names the condition op instead\nof the loop around it. An operation successor names the op that control\nresumes after, and the loop\u0027s own `getSuccessorRegions` names the loop\nfor the same edge. MLIR\u0027s dataflow analyses use that op.\nDeadCodeAnalysis recorded the exit as a predecessor of the point after\n`cir.condition`, so dense analyses never saw control leave a\n`cir.while`, `cir.for` or `cir.do` without a cleanup region. The forward\nstate after such a loop kept its initial value, and the backward state\nat the condition ignored the code after the loop. The fix names the\nloop. The control-flow unit test hid the mismatch because it compared op\nand terminator successors by region only. It now compares whole\nsuccessors, so it fails without the fix.\n\nFor example, in this function:\n\n```mlir\ncir.func @f(%c : !cir.bool, %p : !cir.ptr\u003c!s32i\u003e, %v : !s32i) -\u003e !s32i {\n cir.store %v, %p : !s32i, !cir.ptr\u003c!s32i\u003e\n cir.while {\n cir.condition(%c)\n } do {\n cir.yield\n }\n %x \u003d cir.load %p : !cir.ptr\u003c!s32i\u003e, !s32i\n cir.return %x : !s32i\n}\n```\n\nthe only exit from `cir.while` is the false edge of `cir.condition`.\nDeadCodeAnalysis recorded that edge at the end of the condition region,\nso the point after `cir.while` had no predecessors. A forward dense\nanalysis such as last-modified never joined any state into that point.\nAt the `cir.load` it reported nothing, not the `cir.store` before the\nloop. With the fix the edge reaches the point after `cir.while`, and the\nstore reaches the load.\n\nAssisted-by: Claude Code (claude-opus-5-5).\nGitOrigin-RevId: 80742e8e1fa32d4e9e5d6b6c43824f068652b701\n" }, { "commit": "2016733da7d5730377ce995c6069352ccb0d2c06", "tree": "18b1019b5bdab89c9c660a7b1abd2a1124d34d31", "parents": [ "b556dcd264e2f3deceb4e09932044ad986f0f719" ], "author": { "name": "Chen Miao", "email": "chenmiao.ku@gmail.com", "time": "Wed Nov 11 21:48:51 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 05:56:28 2026 -0800" }, "message": "[clang][C23] Delay thread storage recovery for constexpr diagnostics (#236758)\n\nWhen a thread storage specifier conflicts with another storage class,\nDeclSpec clears it before diagnosing a C23 constexpr conflict. The\nsecond diagnostic then reports \"unspecified\" and loses the specifier\u0027s\nsource range.\n\nFor a block-scope declaration:\n\nBefore:\n```\na:2:22: error: cannot combine with previous \u0027register\u0027 declaration specifier\n 2 | constexpr register _Thread_local int x \u003d 1;\n | ~~~~~~~~ ^\na:2:3: error: cannot combine with previous \u0027unspecified\u0027 declaration specifier\n 2 | constexpr register _Thread_local int x \u003d 1;\n | ^\n2 errors generated.\n```\n\nAfter:\n```\na:2:22: error: cannot combine with previous \u0027register\u0027 declaration specifier\n 2 | constexpr register _Thread_local int x \u003d 1;\n | ~~~~~~~~ ^\na:2:3: error: cannot combine with previous \u0027_Thread_local\u0027 declaration specifier\n 2 | constexpr register _Thread_local int x \u003d 1;\n | ^ ~~~~~~~~~~~~~\n2 errors generated.\n```\n\nDelay clearing the thread storage specifier until after the constexpr\ndiagnostic.\n\nGitOrigin-RevId: 97e0e67e311cbb4ddc68f14d20c585fed3769533\n" }, { "commit": "b556dcd264e2f3deceb4e09932044ad986f0f719", "tree": "e970bd4d5fd64b19d3bf7805c31a65feec912a3a", "parents": [ "3bbe9eae1251a748e37110f839d40fd3a6b02539" ], "author": { "name": "Chen Miao", "email": "chenmiao.ku@gmail.com", "time": "Wed Nov 11 21:48:19 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 05:55:55 2026 -0800" }, "message": "[clang][Sema][NFC] Consolidate checking for vector reduction builtins (#238663)\n\nShare argument conversion, element type extraction, and result type\nassignment across the single-argument reduction builtins. Preserve each\nbuiltin\u0027s element type restrictions and diagnostics, and reuse\ngetVectorElementType for fixed-size and sizeless vectors.\n\nGitOrigin-RevId: f72bef205768471f7114a76e2575c8ed27a98ad7\n" }, { "commit": "3bbe9eae1251a748e37110f839d40fd3a6b02539", "tree": "857bd7307455ad6ca37e48b62ca06acc6fec767e", "parents": [ "544a499ee99810965146d5c8ab791677ba85d793" ], "author": { "name": "CatherineMoore", "email": "catmoore@amd.com", "time": "Wed Nov 11 08:47:50 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 05:55:27 2026 -0800" }, "message": "[NFC][OpenMP] Track Fortran OpenMP support status only in the Flang docs (#240917)\n\nThe Clang OpenMP support page had a \"Fortran Status\" column in its 5.2,\n6.0 and 6.1 tables that was not kept up to date with the Flang page. For\nOpenMP 5.2 alone it disagreed with flang/docs/OpenMPSupport.md on 13\nrows (e.g. \"if clause on teams\", \"otherwise clause\" and \"pure procedure\nsupport\" were listed as unclaimed for Fortran but are done in Flang).\n\nMake the Flang page the only place Fortran status is recorded. The Clang\npage now covers Clang (C/C++) and the LLVM OpenMP runtimes, whose status\napplies to both languages.\n\nGitOrigin-RevId: 7ff67394ff7adfa9d7a8618afb86d28fffab44df\n" }, { "commit": "544a499ee99810965146d5c8ab791677ba85d793", "tree": "6c7dd8e2b6ed202d0a856d5a1945d7abb84a4bb6", "parents": [ "973c43acaf7a91ec6f26cd4fbeebba6b64fab4b8" ], "author": { "name": "Brian Cain", "email": "brian.cain@oss.qualcomm.com", "time": "Wed Nov 11 07:41:11 2026 -0600" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 05:49:18 2026 -0800" }, "message": "[csan] Add support for Hexagon (#241091)\n\nGitOrigin-RevId: e19e7790275513040b9b0e12b50e624b0e6ce780\n" }, { "commit": "973c43acaf7a91ec6f26cd4fbeebba6b64fab4b8", "tree": "fd46f1d001be4d3438b8b5b3459fd242f577fd17", "parents": [ "0610f01d1fbf83353144b772c942d8cbd6df1147" ], "author": { "name": "Gábor Horváth", "email": "xazax.hun@gmail.com", "time": "Wed Nov 11 12:45:00 2026 +0000" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 04:49:55 2026 -0800" }, "message": "[LifetimeSafety] Analyze blocks that cannot reach the exit in liveness (#240890)\n\nThe backward liveness analysis was seeded only at the exit block, so\nblocks that cannot reach it, e.g., the body of an infinite loop, were\nnever analyzed and uses in them were missed.\n\nFixes #222273\n\nAssisted-by: Claude Opus 5.5\n\nCo-authored-by: Gabor Horvath \u003cgaborh@apple.com\u003e\nGitOrigin-RevId: bff401d6a3fb2a430c03e1aa3de1f18a62602f6b\n" }, { "commit": "0610f01d1fbf83353144b772c942d8cbd6df1147", "tree": "99167c3be5ffea6f766ae9191c463e03f3d67109", "parents": [ "7352da324e8743268cd6ab0a52499254d1b83341" ], "author": { "name": "Nikita Popov", "email": "npopov@redhat.com", "time": "Wed Nov 11 11:47:44 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 02:54:44 2026 -0800" }, "message": "[IRBuilder] Enable GEP canonicalization by default (#240885)\n\nThis enables canonicalization of GEPs to ptradd form by default, with a\ndisableGEPCanonicalization() opt-out. The opt-out is added to all the\nplaces where this would currently affect test output.\n\ndisableGEPCanonicalization() calls will get dropped in followup changes,\ntogether with the necessary test adjustments.\n\nAI was used to insert the disableGEPCanonicalization() calls in all the\nnecessary places.\n\nGitOrigin-RevId: 847ab254208538d1aed8cb124599b8d18e96a4ff\n" }, { "commit": "7352da324e8743268cd6ab0a52499254d1b83341", "tree": "0136596a6bc97f391a355bff7e0b71766205c495", "parents": [ "c7bf66182d3baea47557133e69c49876cd05993a" ], "author": { "name": "Mariya Podchishchaeva", "email": "Mariya.Podchishchaeva@amd.com", "time": "Wed Nov 11 11:15:33 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 02:21:47 2026 -0800" }, "message": "[CIR][X86] Add support for cmpps/cmppd* builtins (#228491)\n\nThis basically ports corresponding OGCG code using `cir.vec.fcmp`\ninstruction and/or intrinsic calls.\nI used claude to help with testing and mechanical code changes.\n\nGitOrigin-RevId: 4f75128a46440587704cf8b5c950a5d217c3358b\n" }, { "commit": "c7bf66182d3baea47557133e69c49876cd05993a", "tree": "77162c6c3bf270587ea5f9ea2e1bcbde7ae3eb00", "parents": [ "f10dc23bf297aecbead2139b8c71a4f549cc9ff6" ], "author": { "name": "Timm Baeder", "email": "tbaeder@redhat.com", "time": "Wed Nov 11 10:00:10 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 01:08:02 2026 -0800" }, "message": "[clang][bytecode] Optimize `OffsetHelper` usage (#240908)\n\nDo (most of) the checks outside of the helper. Most of the checks are\nonly possible on a subset of call sites.\n\nGitOrigin-RevId: 56077b8e684d54efd7387e8e4a822b3857ee7b82\n" }, { "commit": "f10dc23bf297aecbead2139b8c71a4f549cc9ff6", "tree": "f2dd7e7b01a9ce65767cbcfc29deb0144a9d9214", "parents": [ "c9a7cbc8e4bb581d9dbd904849f725365d455ebe" ], "author": { "name": "Pengcheng Wang", "email": "wangpengcheng.pp@bytedance.com", "time": "Wed Nov 11 16:37:28 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 00:44:49 2026 -0800" }, "message": "[clang] Cache default target feature attributes\n\nReuse the canonical default target-features attribute for functions\nwithout per-function target settings. Keep target-specific overrides\nseparate from the default cache.\n\nThree interleaved CTMark pairs reduce median compile time by 1.505%\nand geomean by 1.950%.\n\nAssisted-by: TRAE CLI (GPT-6)\n\nReviewers:\nkito-cheng, nikic, AaronBallman, mshockwave, lukel97, asl, hubert-reinterpretcast, efriedma-quic\n\nPull Request: https://github.com/llvm/llvm-project/pull/240338\n\nGitOrigin-RevId: 9218ff711edac6107e736ada42e0f7a7eb4f5b18\n" }, { "commit": "c9a7cbc8e4bb581d9dbd904849f725365d455ebe", "tree": "b13bb24b4eb90919e19b35fbf9e7733f4099a360", "parents": [ "66a38c50711190be7ff38c2045283eb4790a6742" ], "author": { "name": "Younan Zhang", "email": "zyn7109@gmail.com", "time": "Wed Nov 11 16:37:25 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 00:44:06 2026 -0800" }, "message": "[clang][NFC] Don\u0027t track OpaqueValueExpr in bad conversion sequence (#238243)\n\nThose OVE might be allocated on stack and we have already tracked their\ntypes for diagnostics. So let\u0027s leave them null.\n\nGitOrigin-RevId: bc128fb01b02a4cedf6d320dec82b71f1a1158e5\n" }, { "commit": "66a38c50711190be7ff38c2045283eb4790a6742", "tree": "255ed4820143ab901dc4ff76598f46ef559b2262", "parents": [ "ea72dbd2343d4c67646dd7300a4ae9c4020f62d1" ], "author": { "name": "Arseniy Obolenskiy", "email": "arseniy.obolenskiy@amd.com", "time": "Wed Nov 11 09:19:41 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Nov 11 00:29:15 2026 -0800" }, "message": "[CIR] Mark inline asm convergent in GPU code (#233772)\n\nInline asm may hide a barrier, so it must not be moved into divergent\ncontrol flow\n\ndepends on https://github.com/llvm/llvm-project/pull/233770\n\nGitOrigin-RevId: cbafc6590bb3372dd8d588ab5d3ebed43ebdaaf5\n" }, { "commit": "ea72dbd2343d4c67646dd7300a4ae9c4020f62d1", "tree": "dfff984aca1814087b4980e4c434af529ebf87b5", "parents": [ "e3efbae5062d12585c3c6081dcc728ee1b46148d" ], "author": { "name": "Brandon Wu", "email": "brandon.wu@sifive.com", "time": "Wed Nov 11 11:48:45 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Nov 10 19:52:03 2026 -0800" }, "message": "Revert \"[RISCV][clang] Support XSfmm ABI attributes (#206260)\" (#241096)\n\nThis reverts commit 09dcbfb936c629a1285fb04fff9974456f041804. Revert for\nnow and investigate why compile time increase.\n\nGitOrigin-RevId: d5cbc178a32886fc5f8646c5df965de84e589b4e\n" }, { "commit": "e3efbae5062d12585c3c6081dcc728ee1b46148d", "tree": "c39303165e6f1b5de1258978b6b2a18ecfa259e0", "parents": [ "eb21c34631ed38bffc19efb61c8cf0965fd243ca" ], "author": { "name": "Pepper Gray", "email": "hello@peppergray.xyz", "time": "Wed Nov 11 04:46:55 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Nov 10 19:51:32 2026 -0800" }, "message": "[Clang][Driver][MIPS] Fix musl dynamic linker path (#240826)\n\nClang\u0027s Linux driver selects an incorrect dynamic linker for several\nMIPS musl configurations.\n\n- select the correct musl dynamic linker for MIPS targets\n- add driver tests\n\nCloses: #240798\n\nSigned-off-by: Pepper Gray \u003chello@peppergray.xyz\u003e\nGitOrigin-RevId: 3731f4119f461c3b8f9e910831516f509c09a500\n" }, { "commit": "eb21c34631ed38bffc19efb61c8cf0965fd243ca", "tree": "a6843ff095c3f5e4b51c76ba4681ba6926ae6abc", "parents": [ "93da4255ddff4a82c69fdbd65966e47fb9e96b70" ], "author": { "name": "Akash Manna", "email": "akash.manna.mymail@gmail.com", "time": "Wed Nov 11 08:47:43 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Nov 10 19:25:06 2026 -0800" }, "message": "[Clang][CodeGen] Fix crash in implicit assignment operators after a `noreturn` member copy (#240447)\n\nFixes #111612\n\nWhen an implicit assignment operator copies a member or base through a\n`noreturn` `operator\u003d`, `EmitStmt` emits `unreachable` and clears the\ninsertion point. `AssignmentMemcpyizer` then batches the trivially\ncopyable members that follow into one memcpy and emits it directly, so\nthe field GEPs were built on a null block.\n\n`FieldMemcpyizer::emitMemcpy` now drops the batch when there is no\ninsertion point, as `EmitStmt` does for any other unreachable statement.\nThe operator then ends with the call and `unreachable`, like a\nhand-written one with the same statements.\n\nAI Assisted!\n\nGitOrigin-RevId: c8b08b01f9fd0f4cd600ac85be17f221580d0f5a\n" }, { "commit": "93da4255ddff4a82c69fdbd65966e47fb9e96b70", "tree": "56781f5d36d32ef76b980a4c61d69b3bf30a1fe2", "parents": [ "a628ae47516e3568542c10de491398cd86537d94" ], "author": { "name": "Andres-Salamanca", "email": "andrealebarbaritos@gmail.com", "time": "Tue Nov 10 21:46:00 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Nov 10 18:52:52 2026 -0800" }, "message": "[CIR] Add presplitcoroutine attribute and enable coroutine lowering to LLVM IR (#239447)\n\nEmit the `presplitcoroutine` function attribute when lowering a\n`cir.func` that has a coroutine kind. `LLVM\u0027s` coroutine passes use this\nattribute to identify the functions they need to split, so `CIR`\ncoroutines can now be lowered to LLVM IR\n\nGitOrigin-RevId: c6f436ff6f6671681a263eb3aa404b4543915cad\n" }, { "commit": "a628ae47516e3568542c10de491398cd86537d94", "tree": "15de4bdd22091a08b2a643ada04a23f1d1cf7270", "parents": [ "091d1f5a2dfba42174271d3c2777c108ef6a582c" ], "author": { "name": "Sam Elliott", "email": "aelliott@qti.qualcomm.com", "time": "Tue Nov 10 18:14:31 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Nov 10 18:17:19 2026 -0800" }, "message": "[clang][nfc] Add baremetal runtime path checks (#238184)\n\nShowing how the multilib configuration affects runtime include and lib\npaths.\n\nThis change also adds comments to describe why the include paths and lib\npaths have slightly different handling.\n\nAssisted-by: AI\nGitOrigin-RevId: eb690e603226b08320f5c58c39dc416c59ebb11f\n" }, { "commit": "091d1f5a2dfba42174271d3c2777c108ef6a582c", "tree": "741007e413e6b7cf069c151aaef280deef8c53f1", "parents": [ "afaf1e184649c301ceba1040f6de865793fd587e" ], "author": { "name": "Balaji V. Iyer.", "email": "balaji.iyer@intel.com", "time": "Tue Nov 10 17:04:29 2026 -0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Nov 10 17:15:22 2026 -0800" }, "message": "[CIR][SYCL] Emit sycl_in_device field to the for cir.lowering_options (#238117)\n\nAdd sycl_is_device field to the `cir.lowering_options`. This is\npopulated during CIRGen time from `langOpts.SYCLIsDevice`.\n\nGitOrigin-RevId: 69ba8a304aaf0078fcf78c53424397d6b94024ea\n" }, { "commit": "afaf1e184649c301ceba1040f6de865793fd587e", "tree": "ba7e749c359c8ce006af07bb2f0dd27485e0c046", "parents": [ "347fa066a06076a2ed99ebdd3bfb615e2d548452" ], "author": { "name": "Qi Ye", "email": "qi.ye@intel.com", "time": "Tue Nov 10 20:02:21 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Nov 10 17:14:28 2026 -0800" }, "message": "[CIR][SYCL][NFC] Port some CodeGenSYCL tests from clang to CIR CodeGen (#236512)\n\nPorts some SYCL CodeGen tests from clang to CIR CodeGen.\n\nGitOrigin-RevId: ad36cf397999167651105452ef48c1636db93285\n" } ], "next": "347fa066a06076a2ed99ebdd3bfb615e2d548452" }