)]}' { "log": [ { "commit": "983a50ab23748043c5eec905e4e561e5d21216e1", "tree": "d9a34df7e0664f6c32d51e3a4fbfbf11f0632a29", "parents": [ "04a03d37353a2eaedd06b9498d1da310c07c1405" ], "author": { "name": "Valentin Clement (バレンタイン クレメン)", "email": "clementval@gmail.com", "time": "Sat Jun 13 18:35:52 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Sat Jun 13 18:40:56 2026 -0700" }, "message": "Revert \"[flang][cuda] Set kernel intent(in) as const __restrict__\" (#203734)\n\nReverts llvm/llvm-project#203652\n\nbreaking some downstream cudafor code\n\nGitOrigin-RevId: e84e480bb5a53c794f303683df6460ddb44bb7ac\n" }, { "commit": "04a03d37353a2eaedd06b9498d1da310c07c1405", "tree": "055186d94802201335b8412dbc2dd2df2d29260e", "parents": [ "31d382ca1e5f0d9cbe8e27e0ed6b464b352edd84" ], "author": { "name": "Valentin Clement (バレンタイン クレメン)", "email": "clementval@gmail.com", "time": "Fri Jun 12 16:49:07 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 16:50:34 2026 -0700" }, "message": "[flang][cuda] Set kernel intent(in) as const __restrict__ (#203652)\n\nSet attributes on `intent(in)` so `ld.global.nc` is generated by the\nbackend.\n\nGitOrigin-RevId: 7670d88e7ea753095f753e6616b66f5e74da42ef\n" }, { "commit": "31d382ca1e5f0d9cbe8e27e0ed6b464b352edd84", "tree": "d9a34df7e0664f6c32d51e3a4fbfbf11f0632a29", "parents": [ "0fd07f826e9d62a90d1534950f1ba9423304e741" ], "author": { "name": "Zhen Wang", "email": "zhenw@nvidia.com", "time": "Fri Jun 12 16:25:36 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 16:28:16 2026 -0700" }, "message": "[flang][CUDA] Keep host literals from using unified-memory generic distance (#201257)\n\nFix CUDA generic resolution under `-gpu\u003dmem:unified` so unattributed\nliterals and expression temporaries are not treated as unified-memory\nactuals.\n\nPreviously, a host scalar literal such as `1.0` could score as\ncompatible with a `DEVICE` dummy and incorrectly select the\ndevice-scalar overload. This could pass a host stack address to a device\nhelper and fail at runtime. The fix applies the unified/managed memory\ndistance columns only to symbol-backed actuals.\n\nGitOrigin-RevId: af60d56719aa770dda70d19f37b0451b222e89e4\n" }, { "commit": "0fd07f826e9d62a90d1534950f1ba9423304e741", "tree": "982cce2212b1a6e2a00e31681965ead74ac27e7e", "parents": [ "d679c5d94719f1fe13f90b6f6ca3a85d993cc4c1" ], "author": { "name": "Zhen Wang", "email": "zhenw@nvidia.com", "time": "Fri Jun 12 16:08:53 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 16:11:55 2026 -0700" }, "message": "[flang][cuda] Fix host loads from CUDA constant globals (#203064)\n\nThis fixes CUDA Fortran lowering for scalar module variables with the\nconstant attribute that are read from host code, such as launch\nconfiguration expressions or CUF kernel loop bounds.\n\nPreviously, host-side declarations for these globals could be rewritten\nto device constant-memory addresses, causing host loads to dereference\nthe result of _FortranACUFGetDeviceAddress. The fix preserves host reads\nfrom the host-visible global while still using the device address for\nhost-to-device assignment updates.\n\nA FIR regression test covers host reads and assignment updates for\nscalar CUDA constant globals.\n\nGitOrigin-RevId: 6b82a0485dbd6a6a5006329f2fa5f4e8bc977488\n" }, { "commit": "d679c5d94719f1fe13f90b6f6ca3a85d993cc4c1", "tree": "8344aed9e037108530d8a3049ee1485d06f334e5", "parents": [ "ec12c3909c0fae7b54ee043c229f7d82cda1351b" ], "author": { "name": "Valentin Clement (バレンタイン クレメン)", "email": "clementval@gmail.com", "time": "Fri Jun 12 14:52:16 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 14:55:35 2026 -0700" }, "message": "[flang][cuda] Error out if pause statement is used in device code (#203642)\n\nGitOrigin-RevId: d2163f7c0c21390190e1ccb91e61958ad6ecc9f9\n" }, { "commit": "ec12c3909c0fae7b54ee043c229f7d82cda1351b", "tree": "1da8e2938ea2704a6df02e4e7b8e6206ef8fd733", "parents": [ "8f38bd307080eac199ec387a493957bc79ec5db9" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Fri Jun 12 13:18:10 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 11:19:45 2026 -0700" }, "message": "[flang][OpenMP] Add missing check for empty argument list (#203592)\n\nGitOrigin-RevId: f2957d7a56372cf8bd2b48aec81ae7c0fb38813f\n" }, { "commit": "8f38bd307080eac199ec387a493957bc79ec5db9", "tree": "c50aae539b0b6037b24e702e3f0c166c882dc400", "parents": [ "5404559e65190b735770482521862b7b7c6b434b" ], "author": { "name": "Kareem Ergawy", "email": "kergawy@nvidia.com", "time": "Fri Jun 12 17:29:33 2026 +0200" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 11:19:38 2026 -0700" }, "message": "[OpenACC][flang] Emit NYI when unstructured loops are associated with OpenACC directives (#202948)\n\nGitOrigin-RevId: bbeb2d519171b57bbabb25ca07223c286cdecea2\n" }, { "commit": "5404559e65190b735770482521862b7b7c6b434b", "tree": "69992d214f44208c6d11a2d58817574403b20a16", "parents": [ "6e2d73b28c68fbb168c655c713efdee0d11436bf" ], "author": { "name": "Zhen Wang", "email": "zhenw@nvidia.com", "time": "Fri Jun 12 08:25:51 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 08:30:10 2026 -0700" }, "message": "[flang] Avoid invalid declare_value for promoted dummy-scope variables (#202498)\n\nThis fixes a verifier failure in mem2reg after inlining a CUDA device\nprocedure. When a promoted FIR alloca had an associated fir.declare with\na dummy_scope, mem2reg could create a fir.declare_value at a loop header\nwhere the original dummy scope did not dominate.\n\nSkip creating block-argument fir.declare_value ops for such\ndeclarations, matching the existing replaced-value handling. Add a FIR\nmem2reg regression test for the loop-header block argument case.\n\nGitOrigin-RevId: bdc06892cdd37c0a52717373ad7774136ca72b63\n" }, { "commit": "6e2d73b28c68fbb168c655c713efdee0d11436bf", "tree": "318c78e1384865b679371877d6ea6c98c7044471", "parents": [ "7f4cacb1b3acd1fae703fb69de7a26821b5c2756" ], "author": { "name": "Aditya Trivedi", "email": "120598696+adit4443ya@users.noreply.github.com", "time": "Fri Jun 12 19:10:41 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 06:45:57 2026 -0700" }, "message": "[Flang][OpenMP] Fix crash when common block name is used in LINEAR clause (#203250)\n\n[Flang][OpenMP] Fix crash when common block name is used in LINEAR\nclause\n\n Using a common block name in a LINEAR clause (e.g. linear(/c/))\n caused\n a symbol-must-have-a-type crash during lowering. The semantic checker\n was not emitting an error because GetSymbolsInObjectList expands /c/\n to its member variables before the check runs, so the\n symbol-\u003ehas\u003cCommonBlockDetails\u003e() guard was never reached.\n\n Fix by checking for common block names directly on the OmpObjectList\n before the expansion, where the Name variant of OmpObject still holds\n the common block symbol.\n\n Fixes #202329\n\nGitOrigin-RevId: 5096057eb8254c7d56e1b70c4ba2fbf3ac215b73\n" }, { "commit": "7f4cacb1b3acd1fae703fb69de7a26821b5c2756", "tree": "cfa0b9f35b3d6a5593b32141d860225d25171b8a", "parents": [ "48adcb2241ae9f288866c3723c0e27e43cb055b6" ], "author": { "name": "Dan Bonachea", "email": "dobonachea@lbl.gov", "time": "Fri Jun 12 05:06:01 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 12 02:09:04 2026 -0700" }, "message": "[flang] __fortran_builtins: Update __builtin_team_type to meet PRIF specification (#202450)\n\nThe representation of `TEAM_TYPE` in the `ISO_FORTRAN_ENV` module is\nopaque to the compiler, but the size (and alignment) needs to match the\nPRIF specification for `prif_team_type` to ensure ABI compatibility with\nthe multi-image runtime library.\n\nFlang\u0027s old definition as a derived type containing only an\n`integer(int64)` component was leading to an 8-byte `TEAM_TYPE`. However\nPRIF specifies the `prif_team_type` component as a `pointer` to an\nopaque scalar type, which flang compiles to a 40-byte representation.\n\nThis mismatch was leading to incorrect behavior at runtime for programs\ncompiled with `-fcoarray` and using `TEAM_TYPE`, where `TEAM_TYPE`\nvalues returned by PRIF were being silently truncated by the compiler.\n\nChange the declaration of `__builtin_team_type` to match the PRIF\nspecification, thus ensuring the representation of `TEAM_TYPE` has the\nrequired size.\n\nGitOrigin-RevId: 1a4f405c318131c79df259008f76b0d0730d6130\n" }, { "commit": "48adcb2241ae9f288866c3723c0e27e43cb055b6", "tree": "e7ef4107b2483cf3b376cc7b92f6c35516d018c5", "parents": [ "29aefd82b39bbceeb638510129f316a2211e2b99" ], "author": { "name": "ShashwathiNavada", "email": "shashwathinavada@gmail.com", "time": "Fri Jun 12 10:45:01 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 22:21:27 2026 -0700" }, "message": "[Flang][Openmp]Prevent TODO abort on nothing directive (#202679)\n\nSince nothing is a no-op directive (OpenMP 5.2, 8.4), handle it during\nlowering instead of falling through to the generic unimplemented\nutility-directive path and triggering a TODO abort.\n\nGitOrigin-RevId: efb038f38f23ee201ac872ae98668c2ef922f0fa\n" }, { "commit": "29aefd82b39bbceeb638510129f316a2211e2b99", "tree": "434ab598ce5db7c85691eeb4d710cb7da8a9fdd5", "parents": [ "2ca6580dd1ea62d580f4ba711e00bc73ef0accd1" ], "author": { "name": "mleair", "email": "leairmark@gmail.com", "time": "Thu Jun 11 15:10:30 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 15:17:42 2026 -0700" }, "message": "[flang] Improve error for data component definition after CONTAINS in derived type (#203379)\n\nWhen a data component declaration appears after CONTAINS in a derived\ntype definition, flang previously emitted confusing \"expected \u0027FINAL\u0027\",\n \"expected \u0027GENERIC\u0027\", and \"expected \u0027PROCEDURE\u0027\" errors for each\n misplaced component.\n\n This patch adds a misplaced-component detector following the same\n pattern as `misplacedSpecificationStmt` in program-parsers.cpp.\n DataComponentDefStmt is tried as a last alternative in\n TypeBoundProcBinding\u0027s first(). When it matches, fail\u003c\u003e() fires\n with the message:\n\n error: component definition must precede CONTAINS in a derived type\n\n CombineFailedParses then replaces the three keyword-mismatch messages\n with this single targeted one, since the component parse advances\n further than the PROCEDURE/GENERIC/FINAL failures.\n\n Assisted-By: AI\n\nGitOrigin-RevId: 112fb2f79d7983be203957cad6b148865182ed47\n" }, { "commit": "2ca6580dd1ea62d580f4ba711e00bc73ef0accd1", "tree": "1b0f20a7f222d97039f1bc59bac981b379061bce", "parents": [ "6ad484d44d620ca7f8db697481c563c0ca7c1089" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Thu Jun 11 15:01:42 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 13:06:04 2026 -0700" }, "message": "[flang][OpenMP] Simplify code by using common functions, NFC (#203373)\n\nGitOrigin-RevId: 3b267d20846c8dc861fc302c1ac9a6d45d67aa02\n" }, { "commit": "6ad484d44d620ca7f8db697481c563c0ca7c1089", "tree": "f22ec503772b9c9695936e776535afe3b7ab730a", "parents": [ "c6d24064a513feacba7aa2cd4ed55aa237633f0e" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Thu Jun 11 12:58:31 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 11:00:33 2026 -0700" }, "message": "[flang][OpenMP] Add type-param-inquiry checks for non-list clauses (#203059)\n\nClauses that take lists of objects have these objects checked in a\nsingle place. There are still several clauses that take variables, but\nnot via a list. Those clauses still need individual checks for\ntype-param inquiries.\n\nGitOrigin-RevId: da88ce239d718d0f985fdd2268e0d111138ca506\n" }, { "commit": "c6d24064a513feacba7aa2cd4ed55aa237633f0e", "tree": "75bbb33f904c216a62f472f005a6192492d4038c", "parents": [ "96b5003c1846e46cef7eb340108bed9e52bfb548" ], "author": { "name": "Jason Van Beusekom", "email": "jason.van-beusekom@hpe.com", "time": "Thu Jun 11 12:31:33 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 10:35:09 2026 -0700" }, "message": "Revert \"[Flang][OpenMP] remove enable-delayed-privatization-staging to suppor…\" (#203348)\n\nReverts llvm/llvm-project#200952\n\ntest added in commit llvm/llvm-project#200952\n`/offload/test/offloading/fortran/target-firstprivate.f90` on x86\noffload, will fix test and open new PR with fixed changes\n\nGitOrigin-RevId: 5c8e727562cec51033f4ea0f5aa354c07be66587\n" }, { "commit": "96b5003c1846e46cef7eb340108bed9e52bfb548", "tree": "e85153d2a14c1336f503aab39a5f2ad6495707de", "parents": [ "bb4e8225faad15e7d1d2923db1c95aedc97d70ff" ], "author": { "name": "dpalermo", "email": "dan.palermo@amd.com", "time": "Thu Jun 11 11:16:37 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 09:20:59 2026 -0700" }, "message": "Revert \"[Flang][OpenMP][Sema] Add OpenMP warning when mapping local descriptors to device on enter without a corresponding exit\" (#203324)\n\nReverts llvm/llvm-project#201060\n\nWarning is actually reporting an error causing compilation to fail:\nerror: Semantic errors in OpenMPTargetStream.F90\n\nAbove error from Babelstream. Can be seen with either of the following\nreproducers:\naomp/test/smoke-fort/milestone-3-babel-copy\naomp/test/smoke-fort-dev/milestone-3-babel\nGitOrigin-RevId: 57197a0c7b9bb9c362e83e755f653c29996458b7\n" }, { "commit": "bb4e8225faad15e7d1d2923db1c95aedc97d70ff", "tree": "e1f738f52374e6221142c5af56b8c7b54a0693c9", "parents": [ "daa2d0dbc4b13197a1ff1de6d6af89dfc1f0a6ef" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Thu Jun 11 10:55:44 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 08:59:53 2026 -0700" }, "message": "[flang][OpenMP] Temporarily disable flaky test (#203307)\n\nGitOrigin-RevId: ab4286095e8df5ec10739866625e2b94b88952c4\n" }, { "commit": "daa2d0dbc4b13197a1ff1de6d6af89dfc1f0a6ef", "tree": "5882b7c8272a26109ffb688f39fd253fefcc1343", "parents": [ "eddeeab30fc5bac5f755a7821c59b496da0756c0" ], "author": { "name": "Jason Van Beusekom", "email": "jason.van-beusekom@hpe.com", "time": "Thu Jun 11 10:25:39 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 08:28:03 2026 -0700" }, "message": "[Flang][OpenMP] remove enable-delayed-privatization-staging to suppor… (#200952)\n\n…t target first private by default\n\nThis commit follows the decision in\nhttps://github.com/llvm/llvm-project/issues/182356 to remove the not yet\nimplemented for delayed privatization for firstprivate and private in\n`omp target` regions in flang\n\nFixes https://github.com/llvm/llvm-project/issues/182356\n\nAssisted with Opus\n\nGitOrigin-RevId: 1ee93f9e1ab7578c94ebd5f7e6aecaba9a33537d\n" }, { "commit": "eddeeab30fc5bac5f755a7821c59b496da0756c0", "tree": "4cdd2e1d427cdc3e4e8ca3d780aeea9c2902e5d3", "parents": [ "e3a359c0cc24f2b797113770c40f637f0b9ebe90" ], "author": { "name": "Philipp Rados", "email": "philipp.rados@openchip.com", "time": "Thu Jun 11 17:22:16 2026 +0200" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 08:25:37 2026 -0700" }, "message": "[Flang] Add target-abi to CodegenSpecifics (#201539)\n\nThis sets the target-abi, specified in TargetOptions on the MLIR-Module.\nThis allows it to be used by CodegenSpecifics in the TargetRewritePass.\n\nCurrently this is only needed by RISCV64, but can also be used by other\ntargets. This allows to reuse the logic from clang when determining the\nABI, needed for patch\nhttps://github.com/llvm/llvm-project/pull/198335#discussion_r3272341234.\n\nGitOrigin-RevId: dc4aa8452cc57f41995ed9ef14679f8f58ca15ee\n" }, { "commit": "e3a359c0cc24f2b797113770c40f637f0b9ebe90", "tree": "91e06580e5e6f72a825faad51705542244a52376", "parents": [ "7dcf1a87c2b32ee90f8f5a7846fd15551fcd141f" ], "author": { "name": "Leandro Lupori", "email": "leandro.lupori@linaro.org", "time": "Thu Jun 11 11:25:52 2026 -0300" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 07:31:04 2026 -0700" }, "message": "[flang][OpenMP] Fix the privatization of common blocks (#200446)\n\nCommon block objects were being replaced on every privatization,\nwhich caused errors for distinct privatizations on nested constructs.\n\nModifying a common block clone, instead of the original one, fixes the\nissue. Cloning is limited to DSA flags because target mapping lowering\nexpects to find the common block address, through its defining\noperation.\n\nFixes #178790\n\nGitOrigin-RevId: a21a57c483376359d16de7f7f0dce0d546800d9c\n" }, { "commit": "7dcf1a87c2b32ee90f8f5a7846fd15551fcd141f", "tree": "78afa7829517e8d95f996030d5d7471f2f0caadd", "parents": [ "384e254509a01732a65cf6abbc81013731e39201" ], "author": { "name": "Aditya Trivedi", "email": "120598696+adit4443ya@users.noreply.github.com", "time": "Thu Jun 11 18:47:31 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 06:20:29 2026 -0700" }, "message": "[Flang][OpenMP] Fix implicit symbol resolution for USE-renamed arrays (#189215)\n\n[Flang][OpenMP] Fix USE-renamed array DSA in OpenMP regions\n\n Problem: for a USE-renamed symbol (e.g. USE mod, ONLY: s_ary \u003d\u003e ary),\n the HostAssoc in the OMP scope was created under the original name\n \"ary\" instead of the local alias \"s_ary\".\n\n Fix: add a DeclareNewAccessEntity overload that takes an explicit\n SourceName, and call it with symbol-\u003ename() (the alias) rather than\n the ultimate symbol\u0027s name, so the HostAssoc is created under the\n name the user wrote.\n\n Fixes #185344\n\n Assisted-by: Claude Sonnet 4.6\n\nGitOrigin-RevId: 653865dd8bb337b24a79a288afa30ee5206f8342\n" }, { "commit": "384e254509a01732a65cf6abbc81013731e39201", "tree": "b74a334ca67c3ceb2ec55dcd238d55b1562899d8", "parents": [ "4dda5b4ba05304da4713b8a81645d1c49f5ed062" ], "author": { "name": "Jameson Nash", "email": "vtjnash@gmail.com", "time": "Thu Jun 11 08:58:36 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 06:00:33 2026 -0700" }, "message": "[docs] try again to handle doxygen everywhere (#203081)\n\nThe previous attempt at this (b7da9565017e32c18b927a7637714d1b660b558d)\nstill broke standalone builds. Now I have locally tested standalone\nflang, runtimes (with openmp), lldb, combined builds, and the utils\nscript. Hopefully that covers everything this time, and gets everything\ninto a more consistent state (always using the HandleDoxygen script in\nthe same way, included exactly once as required by the cmake design).\n\nGitOrigin-RevId: 7daddf2b6f44960344157a0accb3963a7d87ca03\n" }, { "commit": "4dda5b4ba05304da4713b8a81645d1c49f5ed062", "tree": "84f5ccbb87d71cf0a1f3d6985b09730870d8bd43", "parents": [ "977371582ad23761761ec70748082a005eae57da" ], "author": { "name": "Abid Qadeer", "email": "haqadeer@amd.com", "time": "Thu Jun 11 11:36:56 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 03:40:38 2026 -0700" }, "message": "[flang][OpenMP] Move TargetOMPContext to shared FlangOMPContext (NFC) (#202677)\n\nMoving the class to shared code makes it available for reuse by\nforthcoming DECLARE VARIANT lowering without any functional change to\nexisting metadirective lowering.\n\nGitOrigin-RevId: 076a0a3aacca9c01ca8b6602589752d28e5dbf38\n" }, { "commit": "977371582ad23761761ec70748082a005eae57da", "tree": "7fea863894f4d8fab193f2d0484926c0bf8a2c50", "parents": [ "90ea739f24c39a1ed29a8e4e2dba900f30ad65ed" ], "author": { "name": "jay0x", "email": "90309873+blazie2004@users.noreply.github.com", "time": "Thu Jun 11 15:46:18 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 03:20:49 2026 -0700" }, "message": "[Flang] Reject keyword arguments in statement function calls (#198610)\n\n**Problem**\nFlang silently accepted keyword arguments in calls to statement\nfunctions, violating F2018 C1535.\n\n**Standard: F2018 §15.5.1 C1535**: In a reference to a procedure whose\ninterface is implicit at the point of the reference, the actual argument\nshall not be a keyword argument.\n\nFlang silently compiles the following code without giving error` Keyword\nargument \u0027x\u0027 at (1) is invalid in a statement function\n`\n```\nprogram test\n integer :: f1, x, c\n f1(x) \u003d x / 2\n c \u003d f1(x\u003d10) ! Should be an error\nend program\n\n```\n\n**Summary**\nFixed an issue where statement functions were incorrectly treated as\nhaving an explicit interface, causing argument checks to be skipped.\nNow, statement functions are marked correctly so existing checks run and\nproper errors are shown\n\n**Fixes** : [198523](https://github.com/llvm/llvm-project/issues/198523)\n\n---------\n\nCo-authored-by: Jay Satish Kumar Patel \u003ckumarpat@pe31.hpc.amslabs.hpecorp.net\u003e\nGitOrigin-RevId: b01d0342c129405c30947f459cf055bbe8846974\n" }, { "commit": "90ea739f24c39a1ed29a8e4e2dba900f30ad65ed", "tree": "182d919217d1bc6d1040db2f57ff3b3c92f047ef", "parents": [ "e1663b12423d8fabbbf686acfa3b3dc254abfff6" ], "author": { "name": "khaki3", "email": "47756807+khaki3@users.noreply.github.com", "time": "Thu Jun 11 19:05:16 2026 +0900" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 11 03:15:40 2026 -0700" }, "message": "[flang][OpenACC] Don\u0027t hoist declare directive out of interface bodies (#202806)\n\nExample:\n```fortran\nprogram main\n real :: a(10, 60)\n interface\n subroutine compute(a)\n real :: a(10, 60)\n!$acc declare present(a)\n end subroutine\n end interface\n call compute(a)\nend program\n```\n\nIn this code, the `!$acc declare` inside the interface body is hoisted\ninto the\nhost program unit and lowered there, where its operand (the interface\ndummy)\nhas no IR value, so lowering aborts with \"symbol is not mapped to any IR\nvalue\".\nThis happens because interface-body procedures are not function-like\nunits, so\nthe directive is appended to the enclosing unit\u0027s evaluation list.\n\nFix: track interface-body nesting in the PFT builder and do not add an\n`OpenACCDeclarativeConstruct` to the enclosing unit\u0027s evaluation list\nwhile\ninside one. The directive is no longer hoisted out and is generated only\nin its\nproper scope (the procedure definition or a module-level declare).\nGitOrigin-RevId: 9673aae1fc67abcab756ed3f6e36dff846e8228c\n" }, { "commit": "e1663b12423d8fabbbf686acfa3b3dc254abfff6", "tree": "c6fb68fab9e671b48b456dbe003c9e2502357dbf", "parents": [ "b192ebb2c1b1231c1140d27f66ca725bceff525c" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Wed Jun 10 18:13:38 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 10 16:16:19 2026 -0700" }, "message": "[flang][OpenMP] Implicit declarations of procedures in DECLARE_TARGET (#201935)\n\nThis replaces commit 8f5df8891840b, since it was rejecting the following\ncase:\n```\n function baz(a)\n !$omp declare target to(baz)\n real, intent(in) :: a\n baz \u003d a\n end\n\n program main\n !$omp declare target(baz)\n integer, save :: baz ! error: \u0027baz\u0027 is already declared\n end\n```\nInstead of flagging an error, the \u0027baz\u0027 in the directive should be\nresolved to the explicitly declared variable.\n\nThe original motivating example was to allow the case where the main\nprogram (from the above snippet) looked like the following:\n```\n program main\n real :: a\n !$omp declare target(baz) ! \u0027baz\u0027 should be resolved to the\n !$omp target ! external function\n a \u003d baz(a) ! \u003c- because of this call\n !$omp end target\n end\n```\n\nThe problem is that \"declare_target(baz)\" despite being the same in both\ncases, should lead to two different outcomes in symbol resolution.\n\nThis fix will treat declarations introduced by a DECLARE_TARGET as\neligible for overriding with a potentially conflicting declaration\nstemming from the use of that name in a language construct or\nexpression. Since a mere mention of a name alone declares an object,\nsuch conflict occurs when the name would have been otherwise resolved to\na procedure.\nThis changes the behavior introduced in 8f5df8891840b, where\nyet-undeclared names on a DECLARE_TARGET were preferentially resolved to\nexternal procedures.\n\nThe function HandleProcedureName was modified to \"undeclare\" names\nimplicitly declared due to their appearance in a DECLARE_TARGET.\n\nGitOrigin-RevId: 4b5f74b154228f5d129ccd7dcb1e09e7193643ae\n" }, { "commit": "b192ebb2c1b1231c1140d27f66ca725bceff525c", "tree": "a43f254ec75e45e2a96afca5c96b33b4ae120c00", "parents": [ "8d3762fccad7659902a6d18928d1cb14e1b1bc49" ], "author": { "name": "Razvan Lupusoru", "email": "razvan.lupusoru@gmail.com", "time": "Wed Jun 10 15:09:03 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 10 15:10:40 2026 -0700" }, "message": "[flang][acc] Add tests for implicit `acc declare` of type descriptors (#203100)\n\nAdds 4 tests to cover different cases which requires implicit `acc\ndeclare` for type descriptors.\n\nGitOrigin-RevId: 6b281d1821bdc14ca92380cd39abb794347cf742\n" }, { "commit": "8d3762fccad7659902a6d18928d1cb14e1b1bc49", "tree": "c2b340610d02f965121b9a2f3bfffda8f28c931a", "parents": [ "5646030804ca040f8beccc13476b01694ed139a6" ], "author": { "name": "Andre Kuhlenschmidt", "email": "andre.kuhlenschmidt@gmail.com", "time": "Wed Jun 10 12:48:04 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 10 12:50:20 2026 -0700" }, "message": "[flang][semantics] Clean up -frelaxed-c-loc-checks warning flag handling (#203071)\n\nWhen I originally implemented this I was confused about how\nLanguageFeature warning flags work. Drop the separate UsageWarning::CLoc\nin favour of using LanguageFeature::RelaxedCLocChecks (renamed from\nRelaxedCLoc to match the feature flag) as the warning flag directly. The\nwarning is off by default since the user has already opted into this\nextension using the feature flag; use -Wrelaxed-c-loc-checks alongside\n-frelaxed-c-loc-checks to opt in. Update the test and docs accordingly.\n\nGitOrigin-RevId: 6f57b648974fb90cc6c495c83596e3022a42c146\n" }, { "commit": "5646030804ca040f8beccc13476b01694ed139a6", "tree": "f80dcf2739631f7528f6e1959b5d8c4570c5eac1", "parents": [ "cd0b1869b53287f46e025c89242b9c5bcae786ce" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Wed Jun 10 11:43:39 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 10 09:45:03 2026 -0700" }, "message": "[flang][OpenMP] Remove CheckSymbolName{,s}, NFC (#202811)\n\nThese functions checked if each OmpObject had a symbol, and emitted a\ndiagnostic if not. Name not having a symbol is an internal compiler\nerror (which will be detected separately), and not something actionable\nfor the user.\n\nRemove these functions since they don\u0027t serve any purpose anymore.\n\nGitOrigin-RevId: 826a4840096630bb8a84c4d7dbe9e3c5970cb511\n" }, { "commit": "cd0b1869b53287f46e025c89242b9c5bcae786ce", "tree": "51515b3ef089cf4ef5185f1521743abcefcf3543", "parents": [ "43e6fb9dad8985b0c9428a4b69f1cd67b86a24d4" ], "author": { "name": "ivanrodriguez3753", "email": "48269053+ivanrodriguez3753@users.noreply.github.com", "time": "Wed Jun 10 08:34:34 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 10 08:36:06 2026 -0700" }, "message": "[flang][semantic] parser node types and rewrite for explicit-shape-bounds-spec (#188447)\n\nThis commit lays the groundwork for semantic analysis of rank-1 integer array expressions being used as bounds in a declaration with explicit bounds.\n\nGitOrigin-RevId: f5a429456f3375c8d2e1a2a0b768cd0ee6651f1d\n" }, { "commit": "43e6fb9dad8985b0c9428a4b69f1cd67b86a24d4", "tree": "3e3fecd3257ef1cb6d64ede890a750bd40e6f36c", "parents": [ "a87ff4c64e574d9230cc39cf3a4482597be8b28f" ], "author": { "name": "Jameson Nash", "email": "vtjnash@gmail.com", "time": "Wed Jun 10 10:25:33 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 10 07:30:37 2026 -0700" }, "message": "[docs] update CI to use latest release of doxygen (#202420)\n\nResubmitting https://github.com/llvm/llvm-project/pull/191501 for\nreview, with added https://github.com/llvm/llvm-project/pull/202404 for\nSHA checksum verification and fix for cmake target name conflicts.\nTested with `./llvm/utils/release/build-docs.sh -no-sphinx` locally.\n\nCo-authored-by: Aiden Grossman \u003caidengrossman@google.com\u003e\nGitOrigin-RevId: b7da9565017e32c18b927a7637714d1b660b558d\n" }, { "commit": "a87ff4c64e574d9230cc39cf3a4482597be8b28f", "tree": "5a931234f7fea48f074495650acc1850aee430cd", "parents": [ "22c38b6625cc9b92024dbd1758ed145a109fd5bc" ], "author": { "name": "Leandro Lupori", "email": "leandro.lupori@linaro.org", "time": "Wed Jun 10 10:22:06 2026 -0300" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 10 06:25:23 2026 -0700" }, "message": "[flang][OpenMP] Fix privatization of linear in TARGET (#202443)\n\nLinear symbols are privatized by OpenMP IRBuilder, except when they are\nenclosed within TARGET, in which case their privatization must occur in\nDataSharingProcessor.\n\nFixes #201628\n\nGitOrigin-RevId: edf5305fee7e67f4d6fd5e1228970dba757ebc71\n" }, { "commit": "22c38b6625cc9b92024dbd1758ed145a109fd5bc", "tree": "73bc1bbdd1eb265d8ebc4a6f7f78615c8718c9e9", "parents": [ "95ed834241d376925b8136c948a98f9106e5a782" ], "author": { "name": "agozillon", "email": "Andrew.Gozillon@amd.com", "time": "Wed Jun 10 10:14:36 2026 +0200" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 10 01:16:13 2026 -0700" }, "message": "[Flang][OpenMP][Sema] Add OpenMP warning when mapping local descriptors to device on enter without a corresponding exit (#201060)\n\nThis PR aims to add a new warning to Flang that will emit when a user\ntries to map a local/temporary descriptor to device on an enter\ndirective without also applying it to a corresponding exit directive.\nThis problem can cause some pretty unique and difficult to track down\nerrors in programs as it can result in a user unintentionally locking\ninto place a stack allocated descriptor that has fallen out of scope,\nwhich can result in a later clash with another stack allocated variable\nthat\u0027s being mapped and just happens to reside in the old descriptor\naddress range.\n\nSo this PR attempts to warn about this problem to prevent users doing\nso, it\u0027s of note that we handle some of these cases in our\nMapInfoFinalization pass, but I believe we should still include these\ncases for portability reasons and incase we ever backtrack on our\ndecision to silently support some of these cases.\n\nMade this warning as it was a suggestion from Michael Klemm and seemed\nlike a good PR to add to guide users to avoid this pattern (as it\nunfortunately seems to be a common one that pops up). I\u0027ll perhaps look\ninto an optimization pass that tries to resolve some of these cases\nsilently in the future, but this will have to do in the meantime.\n\nGitOrigin-RevId: e1110dabc49d10c9353d6af7555c56f55fb0cca8\n" }, { "commit": "95ed834241d376925b8136c948a98f9106e5a782", "tree": "31e8507e34ae48cb67abb8288b4564e599870ba1", "parents": [ "296ffc85aecb74511f5ec933c621c1c0195c9382" ], "author": { "name": "Slava Zakharin", "email": "szakharin@nvidia.com", "time": "Tue Jun 09 15:30:14 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 15:35:48 2026 -0700" }, "message": "[flang] Support declarations scoping in FIR AA. (#201216)\n\nFurther experimentation with MLIR inlining showed that\nFIR AA becomes more conservative once a subprogram is inlined.\n\nFor example:\n```\nsubroutine caller(p1,p2)\n real, pointer :: p1,p1\n call callee(p1,p2)\nend\nsubroutine callee(a1,a2)\n real :: a1,a2\n a1 \u003d a2\nend\n```\n\nAfter `callee` is inlined, FIR AA assumes that `a1` and `a2`\nalias at the point of the assignment, because it classifies them\nas pointer accesses.\n\nThis patch adds a machinery to collect declaration information\nfor multiple subprogram scopes (as currently defined by\n`fir.dummy_scope`),\nso that FIR AA can use this information for better disambiguation.\n\nAssisted by Cursor\n\nGitOrigin-RevId: 04e06adc44f3c4cd7b7c62fd9d6c828cfe79eee2\n" }, { "commit": "296ffc85aecb74511f5ec933c621c1c0195c9382", "tree": "2a275fbab71b28f97031c2f03eb30158f730356f", "parents": [ "2c967d5cbc0d4268c2ba5116f4b37137c4e35ef5" ], "author": { "name": "Susan Tan (ス-ザン　タン)", "email": "zujunt@nvidia.com", "time": "Tue Jun 09 16:25:36 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 13:27:35 2026 -0700" }, "message": "[flang] Add RegionBranchOpInterface to fir.do_loop (#202418)\n\n`fir.do_loop` lacked RegionBranchOpInterface, causing dataflow analyses\nand others to treat host loops as opaque and conservatively block\noptimizations.\n\nImplement getSuccessorRegions, getEntrySuccessorOperands, and\ngetSuccessorInputs on fir.do_loop, and override\ngetMutableSuccessorOperands on fir.result to customize for accommodating\nthe occasionally present finalValue in `fir.result`.\n\nAs a side effect, alias analysis now resolves pass-through iter-carried\nreferences to MustAlias instead of MayAlias.\n\nGitOrigin-RevId: 8c0f84e7a71091df9e20fa06a2c5e67fe0d183f5\n" }, { "commit": "2c967d5cbc0d4268c2ba5116f4b37137c4e35ef5", "tree": "7427bfa6005e3b1bd02dd9f05631ecdd5ab64569", "parents": [ "dfe0610448c00d4b88e7f847dc649da0b2bb30a0" ], "author": { "name": "yebinchon", "email": "86588366+yebinchon@users.noreply.github.com", "time": "Tue Jun 09 12:19:04 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 12:20:34 2026 -0700" }, "message": "[FIR] add a fir.shape_extents operation (#199361)\n\nAdd fir.shape_extents op. This takes a !fir.shape\u003cn\u003e and unpacks it into\nn integer SSA values (one per dimension, row-to-column order). This\nsupports lowering when extent values are needed but the defining\nfir.shape is not visible.\n\nFIRToMemRef now inserts fir.shape_extents when recovering extents from a\nshape operand that is not a direct fir.shape / fir.shape_shift /\nexisting fir.shape_extents result (e.g. block arguments from\nfir.select_case).\n\nAlso allow live fir.shape values at fir-to-llvm lowering: !fir.shape\u003cn\u003e\nmaps to an n-field i64 LLVM struct, and live fir.shape ops lower via\nllvm.undef + llvm.insertvalue. fir.shape_extents lowers via\nllvm.extractvalue (+ cast to index).\n\nIn the original path, cg-rewrite still fuses shape-bearing fir.embox /\nfir.array_coor into fircg.* forms and shapes remain dead before LLVM\nlowering. The new path is for forwarded shapes that survive past memref\nlowering.\n\nTests: adds shape-extents.mlir (op + canonicalize) \u0026\nforwarded-shape.mlir (FIRToMemRef fir.shape_extents insertion)\nupdates convert-to-llvm-invalid.fir (live fir.shape is no longer an\ninvalid case).\n\n---------\n\nCo-authored-by: Yebin Chon \u003cychon@nvidia.com\u003e\nGitOrigin-RevId: 784836de471a74a45a4f4431974b7c3c588ffa81\n" }, { "commit": "dfe0610448c00d4b88e7f847dc649da0b2bb30a0", "tree": "549cc84082a7ae5a9771236031ed69124f1185b7", "parents": [ "fc9cfae2d4726e00f4d0b0d011c1166dfd2a31f1" ], "author": { "name": "Joseph Huber", "email": "huberjn@outlook.com", "time": "Tue Jun 09 14:15:42 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 12:20:24 2026 -0700" }, "message": "Reapply \"[Clang] Set default LTO mode for AMDGCN/SPIR-V targets to full\" (#202714) (#202736)\n\nThis reverts commit 655462209f3444ecaf526de6e1df2a84d5e54e4b.\n\nGitOrigin-RevId: 89b969eadbaf49a2de6b7722f248f67a42944d10\n" }, { "commit": "fc9cfae2d4726e00f4d0b0d011c1166dfd2a31f1", "tree": "a54e149a4e062d4eac0e135aaac53ec22b4434b6", "parents": [ "85cb7a17e8974c580e340ec49800f2da0329f168" ], "author": { "name": "John Otken", "email": "john@otken.com", "time": "Tue Jun 09 13:55:40 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 12:00:36 2026 -0700" }, "message": "[flang-rt] Add support for logical binary input edit descriptor (#193490)\n\nAdd binary \u0027B\u0027 input edit descriptor support for logical values to flang\nruntime EditLogicalInput(). The logical binary \u0027B\u0027 edit descriptor is\nalready supported in EditLogicalOutput().\n\nCo-authored-by: John Otken john.otken@hpe.com\nGitOrigin-RevId: 8dac043eb036250cfc2010c44c75b54d08d2d5ef\n" }, { "commit": "85cb7a17e8974c580e340ec49800f2da0329f168", "tree": "1c7e1f80d94f62ba09274071eda324faf67a8911", "parents": [ "b16f1968220d74a17b9d1f060a507e4e14457ece" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Tue Jun 09 13:25:50 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 11:30:23 2026 -0700" }, "message": "[flang][OpenMP] Move assumed-size array check to list item verification (#201385)\n\nThe presence of whole assumed-size arrays will now be diagnosed in a\nsingle location, together with the verification of list item kinds.\n\nGitOrigin-RevId: 2aaaff72e65f8a8e67515ca51768ef65f222d524\n" }, { "commit": "b16f1968220d74a17b9d1f060a507e4e14457ece", "tree": "c219751081c16af2ebd5336a58c4187ed03e77f5", "parents": [ "4128860bdb4993e33804a4fa49edb2cefe7dc275" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Tue Jun 09 13:09:30 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 11:12:01 2026 -0700" }, "message": "[flang][OpenMP] Move check for substring to semantic checks (#201384)\n\nMove it to CheckVarIsNotPartOfAnotherVar, which is also refactored a\nbit.\n\nGitOrigin-RevId: 9c3de270df396a85b01d9aa5dd6729a27f841f2b\n" }, { "commit": "4128860bdb4993e33804a4fa49edb2cefe7dc275", "tree": "89ec5019bced88242ad582934f3af9432e61b1b7", "parents": [ "d299f20ff6e292212da558d177b7b24a545a3e73" ], "author": { "name": "khaki3", "email": "47756807+khaki3@users.noreply.github.com", "time": "Tue Jun 09 08:56:28 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 09:00:49 2026 -0700" }, "message": "[flang][hlfir] Reset lower bounds to one when bufferizing hlfir.as_expr move (#202406)\n\nExample:\n```fortran\nmodule m\n type t\n integer, allocatable :: a(:)\n end type\ncontains\n function pf()\n integer, allocatable :: pf(:)\n allocate(pf(-5:-3)); pf \u003d [1,2,3]\n end function\nend module\n\nprogram p\n use m\n type(t) :: z\n z \u003d t(pf())\n print *, lbound(z%a,1) ! must be 1\nend program\n```\n\nIn this code, the allocatable function result `pf()` is used in the\nstructure constructor `t(...)`. As an expression its lower bound is 1,\nbut lowering wraps the result descriptor in `hlfir.as_expr ... move`,\nand bufferization forwarded that descriptor as-is — so its lower bound\n`-5` leaked into the reallocated component, printing `-5`.\n\nFix: when bufferizing `hlfir.as_expr` with `move`, rebox a descriptor\nthat may carry non-default lower bounds back to lower bound 1\n(descriptor only, no data copy).\nGitOrigin-RevId: 2bbf94d210f73d2e1fd52a4179cd89834198675d\n" }, { "commit": "d299f20ff6e292212da558d177b7b24a545a3e73", "tree": "ee79e64b7baa363da379e943f12133ef73375c6c", "parents": [ "9a62839e8f9bd227ea27ea598f220bd69c6f4923" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Tue Jun 09 10:11:24 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 09 08:15:24 2026 -0700" }, "message": "[flang][OpenMP] More detailed checks for argument list items in clauses (#201334)\n\nFor clauses that take list of variable, locator, and extended list\nitems, perform checks that the actual arguments meet the corresponding\nrequirements. This is version-based, since clause requirements have\nchanged over time.\n\nGitOrigin-RevId: a2ec5c68ba8767c5dc1266946663c8715389da66\n" }, { "commit": "9a62839e8f9bd227ea27ea598f220bd69c6f4923", "tree": "f18d76bc4d059e2eec30def90e12ca44a54cb2f0", "parents": [ "23441fb366ea87adc4667b3ef4bb70be50bc0851" ], "author": { "name": "Kareem Ergawy", "email": "kergawy@nvidia.com", "time": "Tue Jun 09 07:47:19 2026 +0200" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 22:50:31 2026 -0700" }, "message": "[flang][hlfir] Extend InlineHLFIRCopy to inline copy_out with copy-back (#202290)\n\nRename `InlineHLFIRCopyIn` to `InlineHLFIRCopy` and extend it to inline\nthe paired `hlfir.copy_out` operation. The copy_out is inlined at its\noriginal location, after the call, ensuring proper ordering of copy-back\nand deallocation.\n\nOnly inlines when no copy-back is required (intent(in));\nintent(inout/out) pairs are left untransformed.\n\nBased on https://github.com/llvm/llvm-project/pull/179096.\n\nCo-Authored-By: Kazuaki Matsumura \u003ckmatsumura@nvidia.com\u003e (Original\nauthor of the changes).\nCo-Authored-By: Claude Sonnet 4.6 \u003cnoreply@anthropic.com\u003e\n\nCo-authored-by: Kazuaki Matsumura \u003ckmatsumura@nvidia.com\u003e\nCo-authored-by: Claude Sonnet 4.6 \u003cnoreply@anthropic.com\u003e\nGitOrigin-RevId: 32b423ec57d76315cd42b99c0cdd9c63623d8b7b\n" }, { "commit": "23441fb366ea87adc4667b3ef4bb70be50bc0851", "tree": "4a25ee3b3e0008afca60063b3c50d3677101fee0", "parents": [ "c0d159df96160023b5258dd1fabdab65959fdf96" ], "author": { "name": "kwyatt-ext", "email": "kwyatt.external@hpe.com", "time": "Mon Jun 08 16:56:24 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 15:00:37 2026 -0700" }, "message": "Reapply \"[flang] Enumeration Type: (PR 1/5) Foundation types + Parser\" (#202440)\n\nFortranEvaluate referenced DerivedTypeSpec::GetScope(), defined\nout-of-line in FortranSemantics, producing an undefined reference in\nlibFortranEvaluate.so under BUILD_SHARED_LIBS\u003dON. Made GetScope() inline\nin symbol.h so no cross-library symbol is needed.\n\nThis is the fix missing from the original PR (#192651), which was\nreverted in #202408.\n\n---------\n\nCo-authored-by: Kevin Wyatt \u003ckwyatt@hpe.com\u003e\nGitOrigin-RevId: 52ba9fa9732e8dcace2d28561caf1ef0b9631a78\n" }, { "commit": "c0d159df96160023b5258dd1fabdab65959fdf96", "tree": "70db003c754aea37b0d9b032d0fd7405d3cd8572", "parents": [ "21fac3c7db5387cc67acca2808c0fbcd6949ea72" ], "author": { "name": "nvptm", "email": "pmathew@nvidia.com", "time": "Mon Jun 08 13:38:15 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 13:40:21 2026 -0700" }, "message": "[flang][acc] Fix separate compilation for module !$acc declare create on allocatables. (#202409)\n\nWith separate compilation, a module defining `!$acc declare create` on\nan allocatable and a using file that allocates it did not get\ndeclare-action lowering in the using Translation Unit(TU):\n`ACCDeclareActionConversion` could not resolve the post-alloc recipe\n(defined only in the module .o), so no `fir.call` was emitted.\n\nAdd `acc.declare_action` for allocatable/pointer symbols under !$acc\ndeclare.\n* In the defining TU: Export module-global post-alloc/post-dealloc\nrecipes as linkable definitions and mark them with acc.declare_action at\ncreation.\n* In the using TU: When declaring a USE-associated module global, emit\nprivate external recipe stubs so the declare-action conversion pass can\ninsert fir.calls that link to the module definition.\n\nGitOrigin-RevId: a8cb8253e4e53138894f646a0bcb5034866e3a95\n" }, { "commit": "21fac3c7db5387cc67acca2808c0fbcd6949ea72", "tree": "f57726300be143bafda6a37e649e033b7efb0767", "parents": [ "5d8165b880c69b14078998a64b640394174f1060" ], "author": { "name": "Jameson Nash", "email": "vtjnash@gmail.com", "time": "Mon Jun 08 15:28:03 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 12:31:03 2026 -0700" }, "message": "Revert \"[docs] update CI to use latest release of doxygen\" (#202412)\n\nReverts llvm/llvm-project#191501 as cmake CI is having an issue with it:\n\n```\n-- Doxygen enabled (1.14.0).\nCMake Error at /work/as-worker-4/publish-doxygen-docs/llvm-project/cmake/Modules/HandleDoxygen.cmake:29 (add_custom_target):\n add_custom_target cannot create target \"doxygen\" because another target\n with the same name already exists. The existing target is a custom target\n created in source directory\n \"/work/as-worker-4/publish-doxygen-docs/llvm-project/llvm\". See\n documentation for policy CMP0002 for more details.\nCall Stack (most recent call first):\n docs/CMakeLists.txt:59 (include)\n```\n\nGitOrigin-RevId: 50088e227544f2e6fe54dae54c57e3fd51623111\n" }, { "commit": "5d8165b880c69b14078998a64b640394174f1060", "tree": "629f55b3be5f69cfeebd39564087e64206db8094", "parents": [ "5d527e59d932e4a5a38ebbe97e4a1877dcb9e1e9" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Mon Jun 08 13:52:46 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 11:54:10 2026 -0700" }, "message": "Revert \"[flang] Enumeration Type: (PR 1/5) Foundation types + Parser\" (#202408)\n\nReverts llvm/llvm-project#192651\n\nIt seems to be causing linker errors. Verified locally that reverting\nthis commit helps.\n```\n/usr/bin/ld: tools/flang/lib/Evaluate/CMakeFiles/FortranEvaluate.dir/type.cpp.o: in function `Fortran::evaluate::DynamicType::GetAlignment(Fortran::evaluate::TargetCharacteristics const\u0026) const\u0027:\ntype.cpp:(.text._ZNK7Fortran8evaluate11DynamicType12GetAlignmentERKNS0_21TargetCharacteristicsE+0x71): undefined reference to `Fortran::semantics::DerivedTypeSpec::GetScope() const\u0027\n/usr/bin/ld: type.cpp:(.text._ZNK7Fortran8evaluate11DynamicType12GetAlignmentERKNS0_21TargetCharacteristicsE+0x87): undefined reference to `Fortran::semantics::DerivedTypeSpec::GetScope() const\u0027\n/usr/bin/ld: tools/flang/lib/Evaluate/CMakeFiles/FortranEvaluate.dir/type.cpp.o: in function `Fortran::evaluate::DynamicType::MeasureSizeInBytes(Fortran::evaluate::FoldingContext\u0026, bool, std::optional\u003clong\u003e) const\u0027:\ntype.cpp:(.text._ZNK7Fortran8evaluate11DynamicType18MeasureSizeInBytesERNS0_14FoldingContextEbSt8optionalIlE+0x1ed): undefined reference to `Fortran::semantics::DerivedTypeSpec::GetScope() const\u0027\n/usr/bin/ld: type.cpp:(.text._ZNK7Fortran8evaluate11DynamicType18MeasureSizeInBytesERNS0_14FoldingContextEbSt8optionalIlE+0x202): undefined reference to `Fortran::semantics::DerivedTypeSpec::GetScope() const\u0027\n/usr/bin/ld: type.cpp:(.text._ZNK7Fortran8evaluate11DynamicType18MeasureSizeInBytesERNS0_14FoldingContextEbSt8optionalIlE+0x218): undefined reference to `Fortran::semantics::DerivedTypeSpec::GetScope() const\u0027\nclang++: error: linker command failed with exit code 1 (use -v to see invocation)\n```\n\nGitOrigin-RevId: 3ab997597e8a5f2920efdad01cf8c29673826ba0\n" }, { "commit": "5d527e59d932e4a5a38ebbe97e4a1877dcb9e1e9", "tree": "f0311fb50ef1c3a359bfa2936a78c14bf8aec81f", "parents": [ "9ee3959b825d780c71aa483f92c74e07734efc88" ], "author": { "name": "Chi-Chun, Chen", "email": "chichun.chen@hpe.com", "time": "Mon Jun 08 13:48:58 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 11:50:33 2026 -0700" }, "message": "[flang][OpenMP][NFC] Share declare mapper helpers for iterator modifier lowering (#197752)\n\nMove mapper lookup and implicit default mapper creation into reusable\nOpenMP lowering helpers so regular map lowering and iterator-generated\nmap entries (#197757) can use the same resolution path.\n\nThis prepares Flang iterator modifier lowering for map and motion\nclauses\nwithout changing the generated IR for existing non-iterator maps.\n\nGitOrigin-RevId: 17449ef9b7296da8a10d46410c62f16fde98bcd6\n" }, { "commit": "9ee3959b825d780c71aa483f92c74e07734efc88", "tree": "08b74c0ba300ed96e46b404abe394e047588e3e6", "parents": [ "a23cf11c6ee33bb356b22eb32cff5f9afe1384df" ], "author": { "name": "Chi-Chun, Chen", "email": "chichun.chen@hpe.com", "time": "Mon Jun 08 12:53:49 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 10:55:47 2026 -0700" }, "message": "[mlir][OpenMP] Add iterator support to map/motion clause (#197047)\n\nExtend omp.target_data, omp.target_enter_data, omp.target_exit_data,\nand omp.target_update to support `!omp.iterated\u003cTy\u003e`.\n\nThis is part of feature work for llvm#188061\n\nAssisted with copilot\n\nGitOrigin-RevId: e643d7e7a3008e3b97a6d9a1912cd19549b712f2\n" }, { "commit": "a23cf11c6ee33bb356b22eb32cff5f9afe1384df", "tree": "2cdde64a6000b28e177af4f28e69b3b06045b174", "parents": [ "17b2f229280178800cc6cd6f5e6da445566df470" ], "author": { "name": "khaki3", "email": "47756807+khaki3@users.noreply.github.com", "time": "Mon Jun 08 10:38:23 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 10:40:18 2026 -0700" }, "message": "[flang][cuda] Don\u0027t allocate managed descriptors for OpenACC data-clause boxes (#201957)\n\nExample:\n```fortran\nmodule m\n real(8), allocatable :: u(:,:,:,:) ! device-resident allocatable\ncontains\n subroutine core(uf, n)\n integer :: n\n real(8) :: uf(n,n,n,5)\n !$acc kernels loop present(uf) ! descriptor built for uf\n ...\n end subroutine\n subroutine wrap(n)\n call core(u, n)\n end subroutine\nend module\n```\n\nIn this code, the descriptor for `present(uf)` is only used by `acc`\ndata-clause ops (the kernel gets the clause result, not the box), so it\nnever needs to be device-accessible. But `isDeviceAllocation()` is true,\nso it\u0027s allocated via `_FortranACUFAllocDescriptor` in managed memory\nand never freed — leaking it and leaving stale descriptors when the\naddress is reused.\n\nFix: allocate a descriptor in managed memory only when truly\ndevice-accessible — used by a GPU kernel launch\n(`isUsedByGPULaunchFunc`), or device-allocated and not consumed by an\n`acc` data-clause op. Add `isUsedByOpenACCDataClause` and gate the\n`isDeviceAllocation` branch with it.\nGitOrigin-RevId: 8472bc79bf157925c87c3acb7dbb4ac4ddb3070d\n" }, { "commit": "17b2f229280178800cc6cd6f5e6da445566df470", "tree": "a98bcdace4e36347f3cbb8e2cd0d72e7da875eee", "parents": [ "e669c75412aa356147d11c0e0e69e70ba3fbc1cf" ], "author": { "name": "kwyatt-ext", "email": "kwyatt.external@hpe.com", "time": "Mon Jun 08 12:37:52 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 10:40:10 2026 -0700" }, "message": "[flang] Enumeration Type: (PR 1/5) Foundation types + Parser (#192651)\n\nThis PR creates the concept of an ENUMERATION TYPE from the Fortran 2023\nStandard. It is implemented as a special case of a Derived Type\ninternally to the compiler. It adds the parse/unparse support. The\ncompiler will be able to recognize the syntax of an ENUMERATION TYPE and\nreport to users that the feature is not yet implemented.\n\nIt is the 1st of 5 stacked PRs.\n\nAI Usage Disclosure: AI tools (Claude Opus 4.6) were used to assist with\nimplementation of this feature and test code generation. I have\nreviewed, modified, and tested all AI-generated code.\n\n---------\n\nCo-authored-by: Kevin Wyatt \u003ckwyatt@hpe.com\u003e\nGitOrigin-RevId: 85da175778b7ebaeb7e5a4b8275060e5a17b62c6\n" }, { "commit": "e669c75412aa356147d11c0e0e69e70ba3fbc1cf", "tree": "fabb3215b3dd25cfb5672946387fbb802c8fee23", "parents": [ "843805b593d33f2f7e0ef5dc56146f5b2a667b4e" ], "author": { "name": "Jameson Nash", "email": "vtjnash@gmail.com", "time": "Mon Jun 08 13:17:05 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 10:20:38 2026 -0700" }, "message": "[docs] update CI to use latest release of doxygen (#191501)\n\nPreviously this was likely 1.9.8, with the Ubuntu 24.04 worker. Now this\nis 1.17.0. This is a fix for\nhttps://discourse.llvm.org/t/many-doxygen-links-are-broken/83809/2.\n\nFixes 3 significant issues for LLVM:\n\n- `dot` execution performance is very slow (cuts this half hour step\ndown to mere seconds). https://github.com/doxygen/doxygen/pull/12037\n- multi-thread performance is very slow (worse than single threading),\nand now uses all cores for ncpu times speedup (when using version with\nfix, autodetected by cmake).\nhttps://github.com/doxygen/doxygen/pull/12027\n- file links for IR.cpp and similar files were wrong\nhttps://github.com/doxygen/doxygen/pull/11944\n\nAssisted-by: Claude Code\nGitOrigin-RevId: 2eed192c213322a235197286f08f146b94b20c86\n" }, { "commit": "843805b593d33f2f7e0ef5dc56146f5b2a667b4e", "tree": "1d8737e124f8930db76dd653e655d94c76cf41fc", "parents": [ "ff0be97c14f9a1674d2ee506e190261dba5714f8" ], "author": { "name": "Sairudra More", "email": "sairudra60@gmail.com", "time": "Mon Jun 08 21:36:35 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 09:10:53 2026 -0700" }, "message": "[flang] Restrict O0 hlfir.assign scalar-to-array inlining to OpenMP target device (#201774)\n\nFollow-up to #197092.\n\nThat PR fixed the OpenMP target-device case by running scalar-to-array\n`hlfir.assign` inlining at `-O0`, so device code would not lower through\n`_FortranAAssign`.\n\nHowever, running that path for normal host `-g -O0` also changed\ndebugging behavior: a breakpoint on a scalar broadcast like `arr \u003d 11`\ncould be hit once per array element because the assignment became an\ninline loop.\n\nThis patch restricts the `-O0` scalar-RHS-only inlining path to OpenMP\ntarget-device compilation. Host `-O0` keeps the existing\n`_FortranAAssign` path and debugger behavior, while device `-O0` still\navoids the runtime call.\n\nTwo focused tests cover both sides:\n- host `-O0` scalar broadcast keeps `_FortranAAssign`\n- OpenMP target-device `-O0` scalar broadcast avoids `_FortranAAssign`\n\nFixing debug locations for generated inline loops more generally is left\nas separate follow-up work.\nThe broader debug-location issue for generated inline assignment loops\nis tracked separately in #202065.\n\nGitOrigin-RevId: 2befcb4d3224a78e0676c6cac29c37125088643d\n" }, { "commit": "ff0be97c14f9a1674d2ee506e190261dba5714f8", "tree": "18c1afaf85085dac4f170eb46613220ef2b1bdc9", "parents": [ "d2341119c9773d9e009a360ea88a122d5a76d903" ], "author": { "name": "Eugene Epshteyn", "email": "eepshteyn@nvidia.com", "time": "Mon Jun 08 06:10:55 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 03:15:56 2026 -0700" }, "message": "[flang][flang-rt] Treat REAL(2)/COMPLEX(2) as C-interoperable types (#201888)\n\nIEEE-754 binary16 (\"half\") maps cleanly to the C `_Float16` type\n(ISO/IEC TS 18661-3), but flang previously rejected `REAL(KIND\u003d2)` and\n`COMPLEX(KIND\u003d2)` in C-interoperable contexts. Make `REAL(KIND\u003d2)` and\n`COMPLEX(KIND\u003d2)` into actual interoperable types.\n\n`ISO_C_BINDING` now exports the gfortran-compatible named constants\n`c_float16` and `c_float16_complex` (both value 2), the kind parameter\nfor the half-precision C interoperable types.\n\nAssisted-by: AI\nGitOrigin-RevId: cf4f3340ba062cdb3ce95c4a600c61bec2862c5c\n" }, { "commit": "d2341119c9773d9e009a360ea88a122d5a76d903", "tree": "6400d577eaa431202d3ee04114dd94b2d406f34a", "parents": [ "d8bc25d320879c948c29d5bb70184fa746bc3697" ], "author": { "name": "Michael Kruse", "email": "llvm-project@meinersbur.de", "time": "Mon Jun 08 10:22:43 2026 +0200" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 08 01:25:39 2026 -0700" }, "message": "[Flang][CMake] Add missing CMakePushCheckState for standalone build (#202285)\n\nNeeded for the functions cmake_push_check_state/cmake_pop_check_state.\nThe corresponding include command was missing for flang-standalone\nbuilds.\n\nGitOrigin-RevId: da17c1ca11a490e1c54c49868a38c5f5c994b947\n" }, { "commit": "d8bc25d320879c948c29d5bb70184fa746bc3697", "tree": "6a4fab68740801477a286407222f8e4690b3ff2a", "parents": [ "1919685f72d9552dbf099b370bb9da95dd42944e" ], "author": { "name": "Zhen Wang", "email": "zhenw@nvidia.com", "time": "Sat Jun 06 16:38:22 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Sat Jun 06 16:40:05 2026 -0700" }, "message": "[flang][CUDA] Allocate converted kernel descriptors in device-accessible storage (#201950)\n\nFix CUDA descriptor lowering when an `fir.embox` result reaches a\n`gpu.launch_func` through an intermediate `fir.convert`.\n\nCodeGen previously failed to recognize this use chain and could place\nthe descriptor in host stack storage. Since CUDA kernels may dereference\nassumed-shape descriptors on the device, such descriptors must be\nallocated through the CUDA descriptor allocation path. Teach the\nGPU-launch-use check to look through `fir.convert` so these descriptors\nare lowered with `_FortranACUFAllocDescriptor`.\n\nAlso adds a regression test for the `fir.embox -\u003e fir.convert -\u003e\ngpu.launch_func` case.\n\nGitOrigin-RevId: b542c92494b6bd156bacedca2d24fea2215f3792\n" }, { "commit": "1919685f72d9552dbf099b370bb9da95dd42944e", "tree": "299921822fa1a1c631c689d8c8c309760a705297", "parents": [ "f84ff500e68122fc9495dfaefac023298a8f81be" ], "author": { "name": "Fangrui Song", "email": "i@maskray.me", "time": "Sat Jun 06 10:59:27 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Sat Jun 06 11:00:46 2026 -0700" }, "message": "[flang] Remove unused DenseMapInfo::getEmptyKey (#201988)\n\nAfter #201281 DenseMapInfo\u003cT\u003e::getEmptyKey() is no longer used by\nDenseMap. Remove the unused getEmptyKey definitions and dead sentinel\nuses.\n\nGitOrigin-RevId: 42ec0c66867ff23642d5745368191e9f7d2b155b\n" }, { "commit": "f84ff500e68122fc9495dfaefac023298a8f81be", "tree": "91d81480be8866e1c367e1b554c1d27da1f63bab", "parents": [ "869a7dd215d0356d5e26c61a034b8a6675a53ec2" ], "author": { "name": "Michael Kruse", "email": "llvm-project@meinersbur.de", "time": "Sat Jun 06 15:33:11 2026 +0200" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Sat Jun 06 06:35:16 2026 -0700" }, "message": "[Flang][Driver] Fix -foffload-device misspelling (#201857)\n\n#200863 added a new `-foffload-device` argument for informing the\nfrontend that it compiling for the device-side (and as a consequence\nmust not overwrite any module files compiled for the host), but the\ndriver was mistakenly adding `-offload-device`.\n\nAlso fix the condition and add a regression test for the driver.\n\nGitOrigin-RevId: 7564cffdad802e80a595b2b3ce032fab7656df35\n" }, { "commit": "869a7dd215d0356d5e26c61a034b8a6675a53ec2", "tree": "86759aafd51de8a577a646f56a74bfe508a4fb0a", "parents": [ "312cb4151b26e774a624aabbdeb65e603a74447a" ], "author": { "name": "Leandro Lupori", "email": "leandro.lupori@linaro.org", "time": "Fri Jun 05 16:43:56 2026 -0300" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 12:45:30 2026 -0700" }, "message": "[flang][OpenMP] Fix failure in linear-clause01.f90 test (#201920)\n\nApparently the order of some OpenMP checks changed since the pre-commit\nCI tested PR #189170.\nNow the error `The list item \u0027i\u0027 must be a dummy argument` occurs\ninstead of\n`\u0027cc\u0027 is a common block name and must not appear in a LINEAR clause`.\n\nGitOrigin-RevId: 1118fa12a9db10712b38db126807df983f4dda9a\n" }, { "commit": "312cb4151b26e774a624aabbdeb65e603a74447a", "tree": "b06241cd55ea65cee41eec282a6e231472cb8636", "parents": [ "f825e2af8ab389a58383f1ff3eb065e832ab1ff2" ], "author": { "name": "Kelvin Li", "email": "kli@ca.ibm.com", "time": "Fri Jun 05 14:37:53 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 11:40:11 2026 -0700" }, "message": "[flang] Update a test case for AIX (NFC) (#201865)\n\n-fdata-sections is specified by default on AIX. This patch is to adjust\nthe testing for the case that the option is off and on by default.\n\nGitOrigin-RevId: a9f6afded01c7a8a9f949937e89d0d961f7a8423\n" }, { "commit": "f825e2af8ab389a58383f1ff3eb065e832ab1ff2", "tree": "b172c4fd68f1ac722f237d9679d2941e11a9505f", "parents": [ "66025deb755d9a5940bd366fb913f2f0d949d3d1" ], "author": { "name": "Akash Banerjee", "email": "akash.banerjee@amd.com", "time": "Fri Jun 05 18:10:14 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 10:15:33 2026 -0700" }, "message": "[Flang][OpenMP] Heap-allocate GPU dynamic private arrays in distribute parallel do (#200841)\n\nFixes GPU offload crashes for Fortran automatic arrays privatised in\ntarget teams distribute parallel do.\n\nFor delayed privatisation on GPU, dynamically sized boxed array privates\nare now routed through the existing heap-allocation path, with matching\ncleanup emitted in the privatiser dealloc region. This avoids lowering\nsuch arrays to runtime-sized scratch allocas whose descriptors can be\ncaptured across the distribute callback boundary.\n\nFixes [#2419](https://github.com/ROCm/llvm-project/issues/2419).\nCo-authored-by: Codex \u003ccodex@openai.com\u003e\nGitOrigin-RevId: b764858749de56611794a7e92b32ee3008936337\n" }, { "commit": "66025deb755d9a5940bd366fb913f2f0d949d3d1", "tree": "9d7d02d2bea52ef83b99f0691ff058e5344cd53a", "parents": [ "4975bbe7360962902833a521e1b85fa8fd033a37" ], "author": { "name": "SunilKuravinakop", "email": "98882378+SunilKuravinakop@users.noreply.github.com", "time": "Fri Jun 05 22:12:51 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 09:45:35 2026 -0700" }, "message": "[flang][OpenMP] Adding support for weak extended-atomic clause (#201823)\n\nAdding support for \"!$omp atomic compare weak\".\n\n!$omp atomic compare weak\nif (var1 \u003d\u003d num1) var1 \u003d num2\n!$omp end atomic\n\nThis also Fixes\n[#201812](https://github.com/llvm/llvm-project/issues/201812)\n\n---------\n\nCo-authored-by: Sunil Kuravinakop \u003ckuravina@pe31.hpc.amslabs.hpecorp.net\u003e\nGitOrigin-RevId: 92a57848c2ac96021bc8abc1d422c60fe7a01565\n" }, { "commit": "4975bbe7360962902833a521e1b85fa8fd033a37", "tree": "c15111daddf6782c2f0b7efc6780a27d699c78cf", "parents": [ "4a7a6060d3f25d963281aa5b02b1e0569c30b81c" ], "author": { "name": "Aditya Trivedi", "email": "120598696+adit4443ya@users.noreply.github.com", "time": "Fri Jun 05 19:40:20 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 07:15:33 2026 -0700" }, "message": "[Flang][OpenMP] Fix false positive common block error in LINEAR clause (#189170)\n\nFixes #184923\n\nGitOrigin-RevId: 441a7255e97c8198988b66ad7e15ab69051ed358\n" }, { "commit": "4a7a6060d3f25d963281aa5b02b1e0569c30b81c", "tree": "181e8742312168d15130971900d5e2204c68da81", "parents": [ "c25f2d8bdc1aea2f2b11716cbf2f4a45beec56f1" ], "author": { "name": "Abid Qadeer", "email": "haqadeer@amd.com", "time": "Fri Jun 05 11:59:49 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 04:05:32 2026 -0700" }, "message": "[flang][OpenMP] NFC: Move trait-matching helpers to Semantics (#201584)\n\nMove the five trait-matching functions (`mapTraitSet`,\n`mapTraitSelector`, `evaluateUserCondition`, `getTraitScore`,\n`processTraitProperties`) that back metadirective lowering out of\nLower/OpenMP/Utils.cpp and into Semantics/openmp-utils.\n\nThese functions only operate on parsed OpenMP trait selectors and the\nFortran evaluate/folding machinery and carry no MLIR or lowering\ndependency. Placing them in Semantics makes them reusable from the\nsemantic-recording phase of `declare variant` (and any other feature\nthat needs to build a `VariantMatchInfo`).\n\nIn Utils.cpp, the four pure mapping/folding functions are removed\nentirely; their callers now call the Semantics versions directly. The\n`processTraitProperties` is kept as a local wrapper to generate a TODO\nerror on Clause and extension properties.\n\n---------\n\nCo-authored-by: Cursor \u003ccursoragent@cursor.com\u003e\nGitOrigin-RevId: 094bbde3e86c5b7890e162afe6d32957153565b6\n" }, { "commit": "c25f2d8bdc1aea2f2b11716cbf2f4a45beec56f1", "tree": "4384f5da5096b964746e415ba01332a47920179c", "parents": [ "da1236dfbc0d3ced039c0e95c91fb1f3edb485d3" ], "author": { "name": "Ming Yan", "email": "ming.yan@terapines.com", "time": "Fri Jun 05 18:37:16 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 03:40:36 2026 -0700" }, "message": "[mlir][arith] Canonicalize cmpi select to min/max (#201737)\n\nThis PR adds canonicalization patterns that fold `arith.select` +\n`arith.cmpi` into\n`arith.minsi`/`arith.maxsi`/`arith.minui`/`arith.maxui` operations.\n\n---------\n\nCo-authored-by: Claude Opus 4.8 \u003cnoreply@anthropic.com\u003e\nGitOrigin-RevId: e26ef0c943415ec3c3bd88c9c3e9bdca3791d4fd\n" }, { "commit": "da1236dfbc0d3ced039c0e95c91fb1f3edb485d3", "tree": "2e7cfc99b84bbb5ae197dc84f84ff017670baff0", "parents": [ "1b0caeadbf469eb63953d7a4778e51a3fec41fa8" ], "author": { "name": "Abid Qadeer", "email": "haqadeer@amd.com", "time": "Fri Jun 05 11:05:19 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 03:11:01 2026 -0700" }, "message": "[flang][debug] Fix DIStringType size for arrays of assumed-length chars. (#201649)\n\nWhen generating DWARF for an assumed-shape array whose element type is\nan assumed-length character, `convertBoxedSequenceType` called\n`convertType` for the element, which in turn called\n`convertCharacterType` with `hasDescriptor`\u003dfalse. With no descriptor\nand a non-constant length, none of the branches that set `sizeInBits` or\nproduce a length expression were taken, so the resulting\n`DIStringTypeAttr` had `sizeInBits` equal to \u003d0 and no\n`stringLengthExp`, leaving GDB unable to determine the string length or\ndisplay the array elements.\n\nFix this by detecting a non-constant-length character element in\n`convertBoxedSequenceType` and calling `convertCharacterType` directly\nwith `hasDescriptor\u003dtrue`. This generates the correct `stringLengthExp`\nthat reads the element byte-size from the descriptor. A\n`genStringLocation` parameter (default true) is also added to suppress\nthe string location expression for the element type, since the data\nlocation of array elements is already provided by the enclosing array.\n\nFixes https://github.com/llvm/llvm-project/issues/113895\n\nGitOrigin-RevId: 25e94da798b1ddc24c8680e8fe0b6f4099e7e876\n" }, { "commit": "1b0caeadbf469eb63953d7a4778e51a3fec41fa8", "tree": "94d9d158a4636d180354fa5bce436ad9713872f3", "parents": [ "5806c7dc64fdafd37b67517efe60a6b8ca60b928" ], "author": { "name": "jay0x", "email": "90309873+blazie2004@users.noreply.github.com", "time": "Fri Jun 05 14:10:23 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Fri Jun 05 01:45:44 2026 -0700" }, "message": "[Flang][OpenMP]add semantic check for linear clause with statement function variables (#199743)\n\n### **Description**\n\n1. This patch adds a missing semantic check for the LINEAR clause.\n2. OpenMP treats LINEAR variables similarly to PRIVATE variables.\nVariables used inside statement function expressions are not allowed to\nbe privatized, but Flang was not checking this for LINEAR.\n3. The existing privatization check already handled PRIVATE,\nFIRSTPRIVATE, and LASTPRIVATE. This patch extends the same check to\nLINEAR.\n\nFixes : [199660](https://github.com/llvm/llvm-project/issues/199660)\n\n### **Reproducer**\n```\nsubroutine test()\n integer :: pi, r, f, x\n f(r) \u003d pi * r + x\n !$omp parallel do linear(x)\n do r \u003d 1, 10\n pi \u003d f(r)\n end do\n !$omp end parallel do\nend subroutine\n```\nwhen compiled with : `-fopenmp -fopenmp-version\u003d45 -c`\n\n**Observed Behavior**\n\n` No diagnostic emitted\n`\n\n**Expected Behavior**\n\n```\nerror: Variable \u0027x\u0027 in statement function expression cannot be in a LINEAR clause\n !$omp parallel do linear(x)\n```\n\nCo-authored-by: Jay Satish Kumar Patel \u003ckumarpat@pe31.hpc.amslabs.hpecorp.net\u003e\nGitOrigin-RevId: 6ab6b803f552d909341e3c561e84ef9bc6beefe0\n" }, { "commit": "5806c7dc64fdafd37b67517efe60a6b8ca60b928", "tree": "f3b6ad20040bdff400e9a552c9fb9c4d63c776e6", "parents": [ "077f8d636d64cacb75a409abd6b223fc6bdcfce4" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Thu Jun 04 17:37:04 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 04 15:40:18 2026 -0700" }, "message": "[flang][OpenMP] Separate checks for type-parameter inquiry and subobject (#201324)\n\nThis will make it possible to diagnose these situations independently.\nThis isn\u0027t perfect, but will be improved gradually in the future.\n\nGitOrigin-RevId: 48d0cbd339310d965b4602b514c420e5ab54a3d4\n" }, { "commit": "077f8d636d64cacb75a409abd6b223fc6bdcfce4", "tree": "97039eb642545fd41b850f9e95deaaa7cc28bf53", "parents": [ "0eabda6429e924016c1282f31e6f277caf840b60" ], "author": { "name": "Dan Bonachea", "email": "dobonachea@lbl.gov", "time": "Thu Jun 04 15:47:08 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 04 12:50:13 2026 -0700" }, "message": "docs/ParallelMultiImageFortranRuntime: Update link to latest PRIF Specification (#201403)\n\nThe PRIF Committee is pleased to announce the publication of the\nParallel Runtime Interface for Fortran (PRIF) Specification, Revision\n0.8. The latest iteration of this specification represents the efforts\nof a collaborative design process involving multiple individuals across\nseveral institutions.\n\nThe document is available here: https://doi.org/10.25344/S4Z88F\n\nThe PRIF specification is governed by a formal PRIF Committee. For more\ndetails, see: https://go.lbl.gov/prif-governance\n\nThe Committee vote to approve the technical content in this revision\nbegan on 2026-04-30 and concluded successfully on 2026-05-18.\n\nThe 7-day committee comment period for cosmetic feedback began on\n2026-05-21 and concluded on 2026-05-28 with no comments, and only minor\neditorial changes by the Editor.\n\nSee the Change Log in Section 1 of the document for the list of changes\nrelative to the prior revision.\n\nGitOrigin-RevId: 6f5702095457319e5e530baeafb191e4bcd556f3\n" }, { "commit": "0eabda6429e924016c1282f31e6f277caf840b60", "tree": "591f4790127dbc3aa1e309237fc2a3b88aa2b5c5", "parents": [ "0606f01ff6e160bc0eec741bae7ce40462447aed" ], "author": { "name": "Alexis Perry-Holby", "email": "aperry@lanl.gov", "time": "Thu Jun 04 13:43:31 2026 -0600" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 04 12:45:35 2026 -0700" }, "message": "[flang] Adding notes from the Flang Community Call June 3rd (#201458)\n\nGitOrigin-RevId: cac3e7973a5f84473c9a6543cdae5a614114eaa9\n" }, { "commit": "0606f01ff6e160bc0eec741bae7ce40462447aed", "tree": "2fb2319395fd75a1a93a5ab020c75b8620f70a71", "parents": [ "6f56245b1f3aec818b009d0b32ba8f652885b313" ], "author": { "name": "Razvan Lupusoru", "email": "razvan.lupusoru@gmail.com", "time": "Thu Jun 04 10:48:57 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 04 10:50:59 2026 -0700" }, "message": "[flang][acc] Attach FortranObjectViewOpInterface to acc.unwrap_private (#201646)\n\nSince this operation is simply a zero-offset view, attach the\nFortranObjectViewOpInterface to allow FIR AA to walk this if needed.\n\nGitOrigin-RevId: 2851820759fd912c17ac07a492687499aa83906a\n" }, { "commit": "6f56245b1f3aec818b009d0b32ba8f652885b313", "tree": "23ad72111fdf6d5e68a9a9b5fe71280f5d3d7cdf", "parents": [ "333ffe45ab53683a7fd99ef1ee379cdca3ef7545" ], "author": { "name": "Eugene Epshteyn", "email": "eepshteyn@nvidia.com", "time": "Thu Jun 04 13:17:20 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 04 10:20:47 2026 -0700" }, "message": "[flang][docs] Documented `c_float128` and `c_float128_complex` extension (#201614)\n\nflang has supported this for a long time, but it wasn\u0027t documented as an\nextension\n\nGitOrigin-RevId: c193b2df39c711eab8fdd9ad9a5cffb045835c28\n" }, { "commit": "333ffe45ab53683a7fd99ef1ee379cdca3ef7545", "tree": "d0f7fdddb9784d709abeaa64c1f6da7ff1117fe9", "parents": [ "b469e44d57e8e5c5a036301c8628ae1c5a4e6f16" ], "author": { "name": "Eugene Epshteyn", "email": "eepshteyn@nvidia.com", "time": "Thu Jun 04 12:06:26 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 04 09:11:07 2026 -0700" }, "message": "[flang-rt] Extension: accept \u0027!\u0027 as value separator in NAMELIST input (#200441)\n\nTreat \u0027!\u0027 as a self-delimiting value separator when reading NAMELIST\ninput, so that \"name\u003dvalue!comment\" is accepted without an intervening\nblank, comma, slash, or end of record. This matches gfortran, ifx, and\nclassic nvfortran behavior on real-world namelist input files.\n\nF2023 13.11.3.6 p.1 requires a value separator before a \u0027!\u0027 comment\nintroducer in namelist input, so this is a documented extension. The\nchange does not affect \u0027!\u0027 characters inside character literal\nconstants, which continue to be taken literally.\n\nThe extension was documented in flang/docs/Extensions.md.\n\nAssisted-by: AI\nGitOrigin-RevId: 01dbfa0b6e04c09e389bafd7668b2f99bba3ea79\n" }, { "commit": "b469e44d57e8e5c5a036301c8628ae1c5a4e6f16", "tree": "db47bc0ffb4ae2647c8545de26f9dc3d67d27b52", "parents": [ "efd51b258cf6547e3ec15d0735cc2f511a510431" ], "author": { "name": "ShashwathiNavada", "email": "shashwathinavada@gmail.com", "time": "Thu Jun 04 12:37:58 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 04 00:11:11 2026 -0700" }, "message": "[Flang][Semantics] Throw diagnostics for identical entry and result names. (#198500)\n\nFixes https://github.com/llvm/llvm-project/issues/198499\n\nIn flang entry name and result names cannot be same as per\n[C1583](https://j3-fortran.org/doc/year/25/25-007r1.pdf).\nPreviously, Flang did not diagnose cases where the ENTRY name and RESULT\nname were identical, e.g.\n```\n function m1f1()\n integer :: m1f1\n real :: m1f1e1\n m1f1 \u003d 0\n entry m1f1e1() result(m1f1e1)\n m1f1e1 \u003d 0.1\nend function\n```\n\nGitOrigin-RevId: bdd7b4cc2a0c9a8c58e4025edae079ce6af73aa0\n" }, { "commit": "efd51b258cf6547e3ec15d0735cc2f511a510431", "tree": "ba6ee8278f6beede168fad611b8a635961daee31", "parents": [ "ac6c4e9014ec9c969bab37731383465fd9ef9a0d" ], "author": { "name": "ShashwathiNavada", "email": "shashwathinavada@gmail.com", "time": "Thu Jun 04 12:35:57 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Thu Jun 04 00:11:02 2026 -0700" }, "message": "[Flang] Fix -frelaxed-c-loc-checks being ignored when using the driver (#200733)\n\n`-frelaxed-c-loc-checks` worked correctly when passed directly to -fc1,\nbut was silently ignored when using the driver (e.g., flang -c\n-frelaxed-c-loc-checks), causing the flag to go unused. This patch fixes\nit by adding `OPT_relaxed_c_loc` to the `addAllArgs` call in Flang.cpp\nAlso extend the existing test with a driver-mode RUN line to cover this\npath.\n\nGitOrigin-RevId: 364a8837a9d67135d050fe594ba51e319fbab938\n" }, { "commit": "ac6c4e9014ec9c969bab37731383465fd9ef9a0d", "tree": "685c4d9cb03b4a197d742f22f6c10a6b2ae8287d", "parents": [ "28b546917c806d553313b95d391fa2729e1f5b55" ], "author": { "name": "Chi-Chun, Chen", "email": "chichun.chen@hpe.com", "time": "Wed Jun 03 20:39:59 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 20:45:46 2026 -0700" }, "message": "[flang][OpenMP] Support lowering of metadirective (part 2) (#194424)\n\nLower non-constant `user\u003d{condition(expr)}` selectors in OpenMP\nmetadirectives to a runtime `fir.if` / `else` selection cascade.\n\nDynamic user conditions are handled in two separate phases:\n\n- Static applicability uses only selector traits that are known at\ncompile time.\n- Guarded ranking preserves selector specificity and dynamic-condition\nscores for the path where the runtime condition evaluates to true.\n\nLowering first filters candidates using compile-time selector traits,\nthen orders the remaining candidates with the normal OpenMP variant\nranking rules. If the selected candidate has a non-constant user\ncondition, that condition is emitted as a `fir.if` guard. When the\ncondition evaluates to false, the `else` branch continues selection\namong the remaining candidates.\n\nFor example:\n\n```fortran\n!$omp metadirective \u0026\n!$omp \u0026 when(user\u003d{condition(score(100): high)}: barrier) \u0026\n!$omp \u0026 when(user\u003d{condition(low)}: taskwait) \u0026\n!$omp \u0026 otherwise(nothing)\n```\n\nis selected as:\n\n```text\nif (high) barrier\nelse if (low) taskwait\nelse nothing\n```\n\nScores attached to dynamic `condition(score(...): expr)` selectors are\npreserved for the guarded candidate. As a result, those scores affect\nranking only on the runtime path where the condition is true, instead of\nallowing a false runtime condition to influence unconditional selection.\n\nThis also fixes the interactions with `extension(match_any)` and\n`extension(match_none)`:\n\n- `match_any` can be satisfied by a runtime condition when no static\ntrait matches.\n- If static traits already satisfy `match_any`, the dynamic condition\nscore is used only when the condition is true.\n- `match_none` keeps the dynamic condition inactive for static\napplicability while still preserving its score for ranking.\n\nFor example, this `match_any` selector is statically applicable through\n`vendor(llvm)`, but the score on `flag` is used only when `flag`\nevaluates to true:\n\n```fortran\n!$omp metadirective \u0026\n!$omp \u0026 when(implementation\u003d{extension(match_any), vendor(llvm)}, \u0026\n!$omp \u0026 user\u003d{condition(score(100): flag)}: barrier) \u0026\n!$omp \u0026 when(user\u003d{condition(score(10): .true.)}: taskwait) \u0026\n!$omp \u0026 otherwise(nothing)\n```\n\nis selected as:\n\n```text\nif (flag) barrier\nelse taskwait\n```\n\nThe final fallback is the highest-ranked unguarded candidate, an\nexplicit `otherwise` / `default` clause, or implicit `nothing`.\n\nThis patch is part of the feature work for #188820 and is stacked on top\nof [#193664](https://github.com/llvm/llvm-project/pull/193664).\n\nAssisted with Copilot and GPT-5.4.\n\nGitOrigin-RevId: 70256dff97f1393c0bceb6c90c0341215d2ba598\n" }, { "commit": "28b546917c806d553313b95d391fa2729e1f5b55", "tree": "c743cb2f174241f2e4331b4565e302545f6d40aa", "parents": [ "26e2aed8539f77fefd1f9e09789a72c8fe08491a" ], "author": { "name": "Zhen Wang", "email": "zhenw@nvidia.com", "time": "Wed Jun 03 15:17:47 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 15:20:40 2026 -0700" }, "message": "[flang][cuda] Handle Constant and Shared attributes in CUDA generic matching distance (#201451)\n\nThe CUDA generic resolution matching distance function did not handle\nactual arguments with the `Constant` or `Shared` data attribute. These\nattributes represent device memory but were unhandled, causing the\ndistance to fall through to infinity. Under `-gpu\u003dmem:unified`, this led\nto spurious ambiguity errors when multiple specifics (e.g.\nhost-to-device and device-to-device overloads) both became candidates\nwith tied infinite distances.\n\nTreat Constant and Shared actuals the same as Device in the matching\ndistance table, since all three reside in device memory.\n\nGitOrigin-RevId: ad422cd10bfd0be0c069221546de7c7299ca305d\n" }, { "commit": "26e2aed8539f77fefd1f9e09789a72c8fe08491a", "tree": "24ab81c3d7e06548c2c4e1a9bc3116b31599c748", "parents": [ "c93935293cd9ac2a3c1be9586f0dfad392c6c4f0" ], "author": { "name": "Slava Zakharin", "email": "szakharin@nvidia.com", "time": "Wed Jun 03 13:21:24 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 13:25:14 2026 -0700" }, "message": "[flang][OpenACC] Re-declare cached array sections with lower bounds only (#201223)\n\nWhen acc.cache yields a box, rebind the symbol with a fir.shift (lower\nbounds, no extents) instead of a fir.shape_shift, so the rebox preserves\nthe materialized descriptor\u0027s extents and strides. Add lowering tests.\n\nAssisted by Claude\n\nGitOrigin-RevId: 3d830a9c517725bd73e9497cbe4c159c6d04f9ba\n" }, { "commit": "c93935293cd9ac2a3c1be9586f0dfad392c6c4f0", "tree": "9bf379b24873a9cc933122f775a5c62e7c6ae9f6", "parents": [ "b62fdf14a4ccedd9344d828ec4b36dee12d14aad" ], "author": { "name": "Ron Green [NVIDIA]", "email": "rogreen@nvidia.com", "time": "Wed Jun 03 12:45:23 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 12:50:29 2026 -0700" }, "message": "[flang][OpenACC] Add minus operator to reduction clause (#200909)\n\nAdds \u0027-\u0027 as a recognized reduction operator alongside \u0027+\u0027. The new\nReductionOperator::Operator::Minus is parsed from the \u0027-\u0027 token,\naccepted on integer/real/complex reduction variables, and lowered the\nsame as Plus (mlir::acc::ReductionOperator::AccAdd in OpenACC, and\nfir::ReduceOperationEnum::Add in DO CONCURRENT). The unparser emits it\nas \u0027-\u0027.\n\nCo-authored-by: Claude Opus 4.8 (1M context) \u003cnoreply@anthropic.com\u003e\nGitOrigin-RevId: 0fcf63e179486c0a7e72d656c65d2c0a45754a11\n" }, { "commit": "b62fdf14a4ccedd9344d828ec4b36dee12d14aad", "tree": "a3116ec2e89f1033caa21797ed26a779534f0bd1", "parents": [ "6e91bffcd87315dcd5aa84114dcbf2e008d4d54f" ], "author": { "name": "Andre Kuhlenschmidt", "email": "andre.kuhlenschmidt@gmail.com", "time": "Wed Jun 03 09:17:39 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 09:20:35 2026 -0700" }, "message": "[flang][openacc] change option feature flag names to address comments. (#200037)\n\n- Fixes FeatureFlags and CLI flags to use OpenACC instead of ACC.\n- Offline comments further refined the name to be\nOpenACCDefaultNoneScalarsStrict\n- Which changes the semantics requiring the user to opt out of the\ndefault behavior, instead of opt-in to an extension.\n- Also makes the CLI flag OptOut instead of OptIn to match the cli\nbehavior.\n- Gets rid of the unneeded warning flag since FeatureFlags have both a\ndisabled bit and a warning bit.\n- Updates old test for these changes and adds a new test to document all\nthe different cli configurations.\n\nGitOrigin-RevId: 15375d84799ec260d101aa35670f61fa456b6a4e\n" }, { "commit": "6e91bffcd87315dcd5aa84114dcbf2e008d4d54f", "tree": "cb18ed30dc6f94f391641b5362c8e5fffca59d00", "parents": [ "de9cee6609950806f30dd943f4cf052f971859b3" ], "author": { "name": "Andre Kuhlenschmidt", "email": "andre.kuhlenschmidt@gmail.com", "time": "Wed Jun 03 07:50:30 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 07:55:27 2026 -0700" }, "message": "[flang][openacc] add extension which accepts multiple names in a OpenACC routine directive (#200296)\n\nThis PR adds an extension which allows one or more function names in a\nsingle named routine directive. This is treated as multiple named\nroutine directives with the same clauses. The bind clause is forbidden.\nThe empty list of names isn\u0027t excepted. Routine clauses are stable under\nunparsing.\n\nThis PR tests Parsing, Unparsing, Semantics, and Lowering.\n\nGitOrigin-RevId: fdf8fb6c13deb88c2e179a4eed0c66900dcd92d8\n" }, { "commit": "de9cee6609950806f30dd943f4cf052f971859b3", "tree": "213956e8d9fc5ab797e485dbaf51f19c8147e5ff", "parents": [ "16b8a40709a9bc56ac06b3c6d1d84f191ff09a8d" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Wed Jun 03 09:40:42 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 07:43:14 2026 -0700" }, "message": "[flang][OpenMP] Store DECLARE_TARGET information in WithOmpDeclarative (#201103)\n\nThis will be used to emit DECLARE_TARGET directives into module files.\n\nWhen a symbol apperars in DECLARE_TARGET, the OmpDeclareTarget flag will\nbe set on it (this includes procedures containing a DECLARE_TARGET\nwithout arguments or clauses). The set of accompanying clauses will be\nstored in the associated details, in the WithOmpDeclarative mixin. The\nmixin was added to ObjectEntityDetails, ProcEntityDetails, and\nCommonBlockDetails.\n\nThe design goal was to be able to reconstruct the appropriate DECLARE_\nTARGET directive for individual symbols for the purpose of emitting it\nin a module file. Simply storing and then unparsing the AST node may\ninclude symbols that should not be emitted.\n\nAdditionally, refactor the WithOmpDeclarative printing code for reuse in\nsymbol dumping for debugging, and for printing clause sets.\n\nGitOrigin-RevId: 270a6a8f6742b6a0e86d1a6b3ca6b90b39a64e87\n" }, { "commit": "16b8a40709a9bc56ac06b3c6d1d84f191ff09a8d", "tree": "b149d3e9f6ba1b94c599c43e21c70f9454bd221b", "parents": [ "2c34a44dadd8bf9e93c02a1b5767260903614705" ], "author": { "name": "khaki3", "email": "47756807+khaki3@users.noreply.github.com", "time": "Wed Jun 03 06:51:20 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 06:55:41 2026 -0700" }, "message": "[flang][hlfir] Resolve shape_of users when bufferizing eval_in_mem (#201214)\n\nA follow-up to #197814.\n\nExample:\n\n```fortran\nbmat \u003d matmul(mat, mat) ! bmat is allocatable\n```\n\nIn this code, `SeparateAllocatableAssign` sizes the reallocation with an\n`hlfir.shape_of` of the RHS. Once the `matmul` is lowered to\n`hlfir.eval_in_mem`, that `shape_of` is an extra user, so\n`EvaluateIntoMemoryAssignBufferization` erases the `eval_in_mem` while\nit\u0027s still used, hitting a `use-after-erase` assertion at `-O2`.\n\nFix: in `OptimizedBufferization`, redirect a `shape_of` user to the\n`eval_in_mem`\u0027s shape operand before erasing it.\nGitOrigin-RevId: ef4fb183a8d5c0d1910f7b5b6776ba669d6795ef\n" }, { "commit": "2c34a44dadd8bf9e93c02a1b5767260903614705", "tree": "9b057e0df32d3e61e139d0a00f90cc9fb8425428", "parents": [ "366931abe2ff8759f600a24091f284fd4e3b115c" ], "author": { "name": "Michael Kruse", "email": "llvm-project@meinersbur.de", "time": "Wed Jun 03 15:26:50 2026 +0200" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 06:30:25 2026 -0700" }, "message": "[Flang] Fix omp_lib.h location and search path (#201104)\n\nBefore this PR, omp_lib.h is emitted to `${PREFIX}/include` or\n`${PREFIX}/lib/clang/\u003cversion\u003e/include` (install prefix) and\n`${PREFIX}/runtime/src/omp_lib.h` (builddir prefix). It is never found\nthere because the driver only adds `${PREFIX}/include/flang/OpenMP` to\nthe include path.\n\nFix the `omp_lib.h` include by using the same mechanism as the\nomp_lib.mod; that is, move it to\n`${PREFIX}/lib/clang/\u003cversion\u003e/finclude/flang/\u003ctarget-triple\u003e`. The\nsearch path is already added by the driver via\n`-fintrinsics-modules-path` by the driver. Although omp_lib.h currently\ndoes not contain anything target-specific, it could do so in the future\nand I don\u0027t think it is worth the effort to add a mechanism without the\ntarget triple. It should also me consistent with omp_lib.mod.\n\nThe changes in detail consist of:\n1. Move the omp_lib.h output in the builddir to\n`${RUNTIMES_OUTPUT_RESOURCE_MOD_DIR}`. This already takes care of\nwhether it is a bootstrapping build or not.\n3. Move the omp_lib.h install dir to\n`${RUNTIMES_INSTALL_RESOURCE_MOD_PATH}`.\n4. Remove the implicit search path `include/flang/OpenMP` from the\nfrontend. There is nothing in there anyway.\n5. Hardcoding the include search path for testing to the\n`LLVM_RUNTIME_TARGETS\u003ddefault` build becomes unnecessary. This was way\nthe `include-omp-header.f90` test was still passing, although it would\nnot work outside the regression tests. Essentially, it tested\nlit.site.cfg, not the driver.\n6. Replace the old `include-omp-header.f90` test. It created a temporary\n`omp_lib.h` that interferes with any other test that use `include\nomp_lib.h` (there currently isn\u0027t but I originally intended to add the\nreplacement as an additional omp_lib.h test which resulted in a flaky\ntest while the original test was there). Due to how the %flang(_fc1)\nsubstitution works, lookup order may also vary.\n\nGitOrigin-RevId: 911eddeed7ba2359c953396afce934fee269f50e\n" }, { "commit": "366931abe2ff8759f600a24091f284fd4e3b115c", "tree": "a533645f411640d6069b9a8b52eff2aa74c16ebb", "parents": [ "2c778037ad86650563c2f165668a625958356160" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Wed Jun 03 07:30:41 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 05:35:22 2026 -0700" }, "message": "[flang][OpenMP] Refactor interface of WithOmpDeclarative (#200876)\n\nThe two major changes are that:\n1. The clause sets are not optional anymore. In the absence of any\ndeclarative directives (REQUIRES in this case), the set will simply be\nempty.\n2. The optional memory order member will serve as the value of the\nargument to the ATOMIC_DEFAULT_MEM_ORDER clause, and will only be\nmeaningful (and required) when the clause is a member of the clause set.\n\nAdditionally,\n- Rename the RequiredClauses type alias to OmpClauseSet, since it will\nbe reused for other purposes in the future.\n- Remove the has_* functions since they are not necessary, and when more\nmembers are added these functions will only add to the clutter.\n- Add a version_ member for printing directive/clause names.\n\nGitOrigin-RevId: 2ba8bebb270d64a486096b6258f8ea6f764a630c\n" }, { "commit": "2c778037ad86650563c2f165668a625958356160", "tree": "338a52b0af9beaebe7801778966a675afea7a646", "parents": [ "e8227a8c182f37f393b802fc8ff622c78e6e7b8a" ], "author": { "name": "Abid Qadeer", "email": "haqadeer@amd.com", "time": "Wed Jun 03 10:34:09 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Wed Jun 03 02:35:26 2026 -0700" }, "message": "[flang][OpenMP] Rename check-omp-metadirective.cpp (NFC). (#201159)\n\nBoth METADIRECTIVE and DECLARE VARIANT fall into the \"variant\ndirectives\" category, so check-omp-variant.cpp is a more accurate name\nfor the file that hosts their semantic checks.\n\nSuggested in\nhttps://github.com/llvm/llvm-project/pull/198799#issuecomment-4576970335\n\nGitOrigin-RevId: 1023990a57d5824a6fc2f7a899b8f36977bbcf66\n" }, { "commit": "e8227a8c182f37f393b802fc8ff622c78e6e7b8a", "tree": "a3c83feddb2257705d034278bfdcf737b7de3910", "parents": [ "e8adc8da065bbbad8990c9db2c159078ea40790b" ], "author": { "name": "mleair", "email": "leairmark@gmail.com", "time": "Tue Jun 02 14:44:28 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 02 14:45:28 2026 -0700" }, "message": "[flang] Define array named constants from the iso_fortran_env intrins… (#201190)\n\n…ic module\n\ncreateIntrinsicModuleDefinitions() only emitted definitions for array\nnamed constants belonging to the __fortran_ieee_exceptions intrinsic\nmodule. Array constants declared directly in the iso_fortran_env\nintrinsic module -- in practice character_kinds -- were therefore only\nlowered as bodyless `fir.global` external declarations at their use site\nand never defined anywhere, producing an undefined reference at link\ntime.\n\nThis is usually hidden because scalar iso_fortran_env parameters fold to\nimmediates and constant-shape array accesses are folded away, so the\ndangling external symbol is DCE\u0027d before linking. It surfaces when the\naddress of the array genuinely escapes to runtime, e.g.:\n```\n use iso_fortran_env\n integer :: i, x(1)\n do i \u003d 1, size(character_kinds)\n x \u003d findloc(character_kinds, character_kinds(i))\n end do\n```\nwhich fails with:\n\n undefined reference to `_QMiso_fortran_envECcharacter_kinds\u0027\n\nFix by also processing the iso_fortran_env scope in\ncreateIntrinsicModuleDefinitions(), so its array constants are emitted\nas linkonce_odr definitions with initializers.\n\nNote that integer_kinds/real_kinds/logical_kinds are unaffected: they\nare renamed from iso_fortran_env_impl, a non-intrinsic module that is\ncompiled into the runtime, so their definitions already exist there.\n\nAssisted-by: AI\nGitOrigin-RevId: 40ad00beb3e275e1c50d6a718800d197a6b02c4e\n" }, { "commit": "e8adc8da065bbbad8990c9db2c159078ea40790b", "tree": "b898b3feec899199a432072f3c67a259c550c243", "parents": [ "b1ef203f8de250c5cbb7616bf59ce57db533fec4" ], "author": { "name": "Chi-Chun, Chen", "email": "chichun.chen@hpe.com", "time": "Tue Jun 02 10:57:08 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 02 11:00:32 2026 -0700" }, "message": "[flang][OpenMP] Support lowering of metadirective (part 1) (#193664)\n\nAdd lowering support for OpenMP metadirective variants that can be\nselected statically. Selection is performed during lowering with\n`llvm::omp::OMPContext`, and only the selected directive variant is\nlowered.\n\nFor example:\n ```fortran\n subroutine static_vendor()\n !$omp metadirective \u0026\n !$omp \u0026 when(implementation\u003d{vendor(llvm)}: barrier) \u0026\n !$omp \u0026 otherwise(nothing)\n end subroutine\n```\n\nis lowered to a function containing `omp.barrier`, because `vendor(llvm)` is statically applicable when compiling with Flang.\n\nApplicable variants are represented with `llvm::omp::VariantMatchInfo` and ranked using OpenMP context matching and selector scores from the LLVM frontend.\n\nThis patch supports both standalone metadirectives and begin/end\nmetadirectives. For begin/end metadirectives, the variant is selected\nbefore lowering the enclosed region so that the selected directive\ncorrectly wraps the region.\n\nThis initial static-resolution PR supports:\n\n * implementation\u003d{vendor(...)}\n * device\u003d{kind(...), isa(...), arch(...)}\n * user\u003d{condition(\u003cconstant-expr\u003e)}\n * construct\u003d{parallel, target, teams, for}\n * default, nothing, and otherwise clauses\n\nThis patch also preserves `score(...)` on constant user conditions so scored conditions participate correctly in variant ranking. When matching nested construct selectors, it includes constructs introduced by an enclosing selected begin/end metadirective.\n\nDynamic user conditions (#194424), `target_device` selectors, loop-associated directive variants, and complete construct support are deferred to follow-up patches.\n\nThis patch is part of the feature work for #188820.\n\nAssisted by GitHub Copilot and GPT-5.4.\n\nGitOrigin-RevId: f3dcf16115f862f6fa5ea8e520b371f7e57a9c3d\n" }, { "commit": "b1ef203f8de250c5cbb7616bf59ce57db533fec4", "tree": "05fa56ce5e01ca524fe92a1d1beb12c94242c85e", "parents": [ "320ec755c7dd31a069fc4f46cfc5797aa245a16c" ], "author": { "name": "Abid Qadeer", "email": "haqadeer@amd.com", "time": "Tue Jun 02 18:53:30 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 02 10:55:25 2026 -0700" }, "message": "[flang] Support -ffunction-sections and -fdata-sections. (#199731)\n\nWire the flags through the driver, frontend, and TargetMachine, and add\ndriver and codegen lit tests.\n\nFixes https://github.com/llvm/llvm-project/issues/163550\n\n---------\n\nCo-authored-by: Cursor \u003ccursoragent@cursor.com\u003e\nGitOrigin-RevId: fe2f7e40994574dd5d7efa4da8be5d9237905416\n" }, { "commit": "320ec755c7dd31a069fc4f46cfc5797aa245a16c", "tree": "035a3d6c1dd87e4c8073b9fa5e536f81924d8e4c", "parents": [ "4f253e952051591771ab245ad293b2329b11ffc4" ], "author": { "name": "Abid Qadeer", "email": "haqadeer@amd.com", "time": "Tue Jun 02 16:57:43 2026 +0100" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 02 09:00:38 2026 -0700" }, "message": "[flang][OpenMP] Add structure checks for DECLARE VARIANT (#198799)\n\nThis PR adds declare-variant structure checking. Following checks are\nadded:\n\n- Validate [base:]variant arguments (including implicit base for\nsingle-name form).\n- Require exactly one MATCH clause; reject a second MATCH on the same\ndirective.\n- Reject duplicate (base, variant) across multiple declare variant\ndirectives.\n- Reject clauses not allowed on declare variant.\n- Apply shared context-selector checks to MATCH (reuse metadirective\nlogic).\n- Require constant user conditions in MATCH for declare variant (dynamic\nselectors deferred).\n\nRefactor metadirective support:\n\n- Extract CheckContextSelectorSpecification for reuse.\n- Reject SCORE on trait sets that do not allow it (also affects\nmetadirective).\n\nCo-authored-by: Cursor \u003ccursoragent@cursor.com\u003e\nGitOrigin-RevId: cf25fb1e0809aaf4dbb13f2b837d47e329f595b8\n" }, { "commit": "4f253e952051591771ab245ad293b2329b11ffc4", "tree": "b48c82c02946586a76c4548a93103918989d8606", "parents": [ "47963c62e6700ba6593b20d0efd865f5211d3df7" ], "author": { "name": "Kelvin Li", "email": "kli@ca.ibm.com", "time": "Tue Jun 02 10:46:00 2026 -0400" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 02 07:50:41 2026 -0700" }, "message": "[flang] Fix possibly unused variable (NFC) (#200905)\n\nGitOrigin-RevId: 06978156c7ba8a8c8aca60033ea65c8335cca24c\n" }, { "commit": "47963c62e6700ba6593b20d0efd865f5211d3df7", "tree": "79de06ecb640d16aaa63d2c46b327b2ffc8e5609", "parents": [ "aabbae4e94204875df18a90119c3dc6c5930e362" ], "author": { "name": "khaki3", "email": "47756807+khaki3@users.noreply.github.com", "time": "Tue Jun 02 04:55:05 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Tue Jun 02 05:00:28 2026 -0700" }, "message": "[flang][HLFIR] Add SeparateAllocatableAssign pass (#197814)\n\nExample:\n```fortran\n!$acc kernels\nB \u003d A ! A, B allocatable\n!$acc end kernels\n```\n\nIn this code, `B \u003d A` lowers to `hlfir.assign ... realloc`, which\nbecomes a `_FortranAAssign` runtime call inside the compute region — the\nallocation can\u0027t be separated from the copy, and it crashes when `B` is\nunallocated.\n\nFix: add `SeparateAllocatableAssign`, which splits the realloc assign\ninto an explicit conditional [re-]allocation followed by a plain\n`hlfir.assign`, exposing the allocation as plain FIR. For variable RHS\nit skips aliasing cases (`a \u003d a(:n)`) by loading the LHS `fir.box` and\nquerying `fir::AliasAnalysis` on the data. For `hlfir.expr` RHS it\nleaves ordering to bufferization, and fixes\n`ElementalAssignBufferization` to stop fusing an elemental across a\ndeallocation.\nGitOrigin-RevId: 1872f06d60f35133bbc06e047c763b490c2d31be\n" }, { "commit": "aabbae4e94204875df18a90119c3dc6c5930e362", "tree": "5ddb137018779a70f4db9dd30598551417a5f277", "parents": [ "afcc81c2e347b039fd963b23d34bf87ca81d26d1" ], "author": { "name": "Michael Kruse", "email": "llvm-project@meinersbur.de", "time": "Tue Jun 02 01:47:56 2026 +0200" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 01 16:50:03 2026 -0700" }, "message": "[Flang] Fix device-side module lookup (#200863)\n\nWhen invoking flang with device-offloading (eg. `flang modfile.f90\n-fopenmp --offload-arch\u003dgfx90a`), it will invoke the frontend twice:\nonce for the host architecture, and a second time for the architecture\nspecified with `--offload-arch`. However, both frontend invocations are\ngoing to write `modfile.mod` (or whatever the module name in\n`modfile.f90`), and as a result the second one for gfx90a will be what\nthe file contains after the driver invocation returns. Until #171515\nboth version of the file were identical, but now both files are using a\ndifferent set of builtin modules. Since Flang\u0027s mod files store the\nchecksums of used module files in them, this can result in a checksum\nmismatch error. For instance, modfile.mod being the gfx90a version, and\nthen using it to compile with `flang modfile.f90\n--target\u003dx86_64-linux-gnu`) will have a checksum mismath.\n\nflang -fc1 host x86_64 --\u003e modfile.mod --\u003e lib/clang/23/finclude/flang/x86_64-linux-gnu/iso_fortran_env.mod\n / / \\ \\\nflang -fc1 -foffload-device nvptx / \\ lib/clang/23/finclude/flang/nvptx64-nvidia-cuda/iso_fortran_env.mod\n / \\\nflang -fc1 -foffload-device amdgcn lib/clang/23/finclude/flang/amdgcn-amd-amdhsa/iso_fortran_env.mod\n\nWe fix this by\n\n1. Not overwriting the `--target` host module file with the\n `--offload-arch` module; the auxiliary target is the canonical version\n for its contents; and\n\n2. Ignore checksum errors when using an intrinsic module during\n offloading. The device version should be compatible with the host\n version, just with definitions which the .mod file will eventually\n import from the intrinsic module at compile-time.\n\nGitOrigin-RevId: 22e13e71ed46f0d47f90ae5e6cea2956db892be3\n" }, { "commit": "afcc81c2e347b039fd963b23d34bf87ca81d26d1", "tree": "d893b0f70dd5d68c0bad9c46b824569a3051ddf5", "parents": [ "c5ef44ed984041bf8c38398f1e7516778185bb99" ], "author": { "name": "Fangrui Song", "email": "i@maskray.me", "time": "Mon Jun 01 14:27:10 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 01 14:30:31 2026 -0700" }, "message": "[flang] Remove unused DenseMapInfo::getTombstoneKey (#200632)\n\n#200595 changed DenseMap to no longer create tombstone buckets, so\nDenseMapInfo\u003cT\u003e::getTombstoneKey() is never called. Remove dead\ndefinitions and dead tombstone branches.\n\nGitOrigin-RevId: 67139a27775e72b0b199a37946a8b3fa0abdc75e\n" }, { "commit": "c5ef44ed984041bf8c38398f1e7516778185bb99", "tree": "009dd550ec06b874acab36deafe175152ae97200", "parents": [ "cd8e7711cb506629b8fccba06e7a0802db768573" ], "author": { "name": "Krzysztof Parzyszek", "email": "Krzysztof.Parzyszek@amd.com", "time": "Mon Jun 01 13:23:40 2026 -0500" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 01 11:25:28 2026 -0700" }, "message": "[flang][OpenMP] Simplify checks for type-parameter inquiry (#198217)\n\nRemove the no longer needed IsDataRefTypeParamInquiry.\n\nGitOrigin-RevId: 8bae84de90eb5b042cc8a320b88f92357e09975f\n" }, { "commit": "cd8e7711cb506629b8fccba06e7a0802db768573", "tree": "5c6197c5ee02eedb8af54e22c88e2e3b53b20e72", "parents": [ "1b45f5f85ea42477e105fe55e66f43fb15a118e5" ], "author": { "name": "Leandro Lupori", "email": "leandro.lupori@linaro.org", "time": "Mon Jun 01 14:23:50 2026 -0300" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 01 10:25:19 2026 -0700" }, "message": "Reland \"[flang][OpenMP] Fix lowering of LINEAR iteration variables (#188851)\" (#194623)\n\nLinear iteration variables were being treated as private. This fixes\none of the issues reported in #170784.\n\nThe regressions in the OpenMP V\u0026V and Fujitsu testsuites happened\nbecause the users iterator was apparently becoming invalid, after one of\nits uses was replaced. This was fixed by making a copy of the list of\nusers.\n\nGitOrigin-RevId: 754d3d0d966423f9d308c6823deed55b02bac9f2\n" }, { "commit": "1b45f5f85ea42477e105fe55e66f43fb15a118e5", "tree": "6d152cdd5e8621cfb224a3c2c4bd26978875f424", "parents": [ "93455f2f9334ff23f4ffe06601a8667c17a0060b" ], "author": { "name": "Zhen Wang", "email": "zhenw@nvidia.com", "time": "Mon Jun 01 08:45:05 2026 -0700" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 01 08:51:01 2026 -0700" }, "message": "[cuda][flang] Diagnose missing CUDA intrinsic modules in Flang semantics (#200509)\n\n- Replace CUDA intrinsic module `CHECK`s with actionable diagnostics\nwhen `cudadevice` or `__cuda_builtins` cannot be read.\n- Avoid dereferencing missing CUDA module scopes during implicit CUDA\nsymbol import.\n- Add a semantics test covering the missing CUDA intrinsic module\ndiagnostic.\n\nGitOrigin-RevId: 234e46a14635cb1b76e4852fc49a3407f6d87a1b\n" }, { "commit": "93455f2f9334ff23f4ffe06601a8667c17a0060b", "tree": "056ea38aaba4900218b3a489a881f71b62e12530", "parents": [ "21d8bf18d5f4932081491a3ef45392b54ceaf507" ], "author": { "name": "SunilKuravinakop", "email": "98882378+SunilKuravinakop@users.noreply.github.com", "time": "Mon Jun 01 20:49:49 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 01 08:25:52 2026 -0700" }, "message": "[flang][OpenMP] Restricting Integration test atomic-capture-release.f90 run to x86 and aarch64 (#200770)\n\nSimilar to the the change in\n`flang/test/Integration/OpenMP/atomic-compare.f90`, restricting the test\ncase `flang/test/Integration/OpenMP/atomic-capture-release.f90` to run\non\n`x86-registered-target \u0026 aarch64-registered-target`\n\nThis also Fixes\n[#200729](https://github.com/llvm/llvm-project/issues/200729)\n\nA comment has been added to the merge in\n[[Flang][OpenMP]Handling restrictions of using Memory-Order-Clause with\nAtomic-Clause](https://github.com/llvm/llvm-project/pull/199636)\n\n\u003eLLVM Buildbot has detected a new failure on builder\nppc64le-mlir-rhel-clang running on ppc64le-mlir-rhel-test while building\nflang,llvm,mlir at step 4 \"cmake-configure\".\nFull details are available at:\nhttps://lab.llvm.org/buildbot/#/builders/129/builds/45227\n\nHence this PR.\n\nCo-authored-by: Sunil Kuravinakop \u003ckuravina@pe31.hpc.amslabs.hpecorp.net\u003e\nGitOrigin-RevId: 9f1568aca7542dd363a4f5e575311e6c901d8226\n" }, { "commit": "21d8bf18d5f4932081491a3ef45392b54ceaf507", "tree": "f8048241623aaac296a1075f580d0082d98b3ce1", "parents": [ "45d286a59c02a3000e9477c941311262d94fa654" ], "author": { "name": "Ritanya-B-Bharadwaj", "email": "ritanya.b.bharadwaj@gmail.com", "time": "Mon Jun 01 19:06:37 2026 +0530" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 01 06:38:23 2026 -0700" }, "message": "[flang] Add support for TYPEOF and CLASSOF type specifiers (#188804)\n\nImplements parsing, semantics and lowering for the Fortran 2023 TYPEOF and CLASSOF type specifiers (R703). TYPEOF produces the declared type of a data-ref; CLASSOF produces a polymorphic version. Includes constraint\nchecks (C709–C713) and tests.\n\nSemantics resolves TYPEOF/CLASSOF to the concrete underlying type in the symbol table, so no lowering code changes were needed. Added a lowering test to verify FIR/HLFIR generation works correctly for intrinsic types,\nderived types, extended types, and polymorphic CLASSOF with allocatable/pointer.\n\nFixes - https://github.com/llvm/llvm-project/issues/185635\n\nGitOrigin-RevId: 143338131f0e8f1586f7f9aacb7d94dfde169b08\n" }, { "commit": "45d286a59c02a3000e9477c941311262d94fa654", "tree": "5d137942bac48c369fae4aa801c2b18be577a248", "parents": [ "93fc751a76b6f617f554f7c7bae31483056ad0bc" ], "author": { "name": "hcx", "email": "hechenxi@bosc.ac.cn", "time": "Mon Jun 01 20:47:16 2026 +0800" }, "committer": { "name": "Copybara-Service", "email": "copybara-worker@google.com", "time": "Mon Jun 01 05:50:18 2026 -0700" }, "message": "[flang][lowering] Implement component-wise initialization for derived types (#187465)\n\nCurrently, the compiler defaults to a full `memcpy` when initializing\nderived types.\n\nThis patch introduces component-wise initialization for pointer /\nallocatable components, avoiding unnecessary initialization data\ngeneration and redundant copies.\n\nIneligible cases continue to use the existing `memcpy` initialization\npath.\n\nRFC: https://discourse.llvm.org/t/rfc-automatic-static-promotion-of-large-local-variables-in-flang/89539\n\nKey changes:\n\n- In `flang/lib/Lower/ConvertVariable.cpp`:\n- Add `genDerivedTypeComponentInit` for component-wise derived type\ninitialization.\n- Add `isEligibleForComponentWiseInit` to guard the new initialization\npath.\n- Add `genInlinedInitWithMemcpy` to factor out the existing full\n`memcpy` initialization logic.\n- Update `defaultInitializeAtRuntime` to select the appropriate\ninitialization path.\n- Add and update regression tests.\n\n---------\n\nCo-authored-by: ict-ql \u003c168183727+ict-ql@users.noreply.github.com\u003e\nCo-authored-by: buggfg \u003cwangyingying@bosc.ac.cn\u003e\nGitOrigin-RevId: de3529386e1e773f0186755b83bbc12e7884f863\n" } ], "next": "93fc751a76b6f617f554f7c7bae31483056ad0bc" }