{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":597024300,"defaultBranch":"amd-trunk-dev","name":"llvm-project-rocm","ownerLogin":"skatrak","currentUserCanPush":false,"isFork":true,"isEmpty":false,"createdAt":"2023-02-03T13:09:44.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/2736073?v=4","public":true,"private":false,"isOrgOwned":false},"refInfo":{"name":"","listCacheKey":"v0:1704463238.0","currentOid":""},"activityList":{"items":[{"before":"29528ed9a6ff92e5bbde305f820e33c1c8030567","after":null,"ref":"refs/heads/teams-clauses-fix","pushedAt":"2024-01-05T14:00:38.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"}},{"before":null,"after":"f46296f22cf133858edef413859e6dd61b9b3243","ref":"refs/heads/gpu-unittest-fix","pushedAt":"2024-01-03T11:40:17.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[NFC] Fix failing Lit unit tests\n\nThis fixes failures in unit tests for check-flang, check-mlir and check-llvm.\nIt does not address check-openmp failures.","shortMessageHtmlLink":"[NFC] Fix failing Lit unit tests"}},{"before":"d82d354041fafe528a2790b2faab3874a2f3287c","after":"29528ed9a6ff92e5bbde305f820e33c1c8030567","ref":"refs/heads/teams-clauses-fix","pushedAt":"2023-12-07T13:49:24.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Fix ICE on stream test app","shortMessageHtmlLink":"Fix ICE on stream test app"}},{"before":"b4f8076b13f5ecc4d90fda8958dc61970a3f3d2e","after":"d82d354041fafe528a2790b2faab3874a2f3287c","ref":"refs/heads/teams-clauses-fix","pushedAt":"2023-12-04T11:31:39.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[WIP] Device support for num_teams, thread_limit and num_threads clauses\n\nThis patch brings back support for the use of `num_teams`, `thread_limit` and\n`num_threads` clauses of the TEAMS, TEAMS+TARGET and PARALLEL OpenMP directives\nwhen generating target device code.\n\nThese changes are needed due to recent upstream modifications to the\n`OpenMPIRBuilder`, which now stores default/constant values for these clauses\nin a global structure that is created early in the translation process (when\nthe outlined kernel function is created, before processing its contents). The\ninitial conflict resolution that was done when these changes were merged to the\nATD branch seemed to work, but it was duplicating some code and yielding\ndifferent values than clang for equivalent codes.\n\nAlso, the addition of the `IsolatedFromAbove` trait to the `omp.target` MLIR\noperation broke previous support for these clauses as well, since they were\ndefined outside the `omp.target` operation but then passed as arguments to\nnested `omp.teams` without any mapping clauses.\n\nThis is a summary of the changes introduced in this patch, which enable again\nsupport for the aforementioned clauses:\n\n  - Split the `OpenMPIRBuilder::TargetRegionInfo` structure into a structure\n  holding runtime values and another structure holding default/constant values.\n  Also, rather than having it introduce state about the current target region\n  being lowered to the `OpenMPIRBuilder` class, these structures are populated\n  and passed to the relevant builder functions from clang/flang.\n  - Remove duplicated `OpenMPIRBuilder` code introduced during a merge from\n  main, related to the handling of kernel attributes. Fixed mismatches between\n  clang and flang relating to this.\n  - Extract evaluation of clauses of TEAMS and PARALLEL directives to outside\n  the target region they are in and pass these values to new `omp.target`\n  arguments. This approach is temporary, and alternatives are being discussed\n  in the [LLVM-Flang Slack channel](https://flang-compiler.slack.com/archives/C01PY03PP9P/p1700672266767559).\n  - Runtime values for the trip count, number of teams and number of threads\n  are casted to the type expected by the corresponding RTL function, so using\n  different types won't result in a compilation error.\n  - Prevent `kmpc_push_num_threads` calls being generated for the target\n  device.\n  - Improve MLIR pattern checks for the `omp.target` operation to tell whether\n  it represents an SPMD loop.","shortMessageHtmlLink":"[WIP] Device support for num_teams, thread_limit and num_threads clauses"}},{"before":null,"after":"d472daffb343033859e71278624499b52f310fc8","ref":"refs/heads/remove-cse-interfaces","pushedAt":"2023-12-01T11:37:58.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Remove CSE and constant folding interfaces used for omp.target\n\nThese interfaces are now not necessary after adding the IsolatedFromAbove trait\nto the omp.target operation.","shortMessageHtmlLink":"Remove CSE and constant folding interfaces used for omp.target"}},{"before":null,"after":"b4f8076b13f5ecc4d90fda8958dc61970a3f3d2e","ref":"refs/heads/teams-clauses-fix","pushedAt":"2023-11-29T13:00:53.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[WIP] Device support for num_teams, thread_limit and num_threads clauses\n\nThis patch brings back support for the use of `num_teams`, `thread_limit` and\n`num_threads` clauses of the TEAMS, TEAMS+TARGET and PARALLEL OpenMP directives\nwhen generating target device code.\n\nThese changes are needed due to recent upstream modifications to the\n`OpenMPIRBuilder`, which now stores default/constant values for these clauses\nin a global structure that is created early in the translation process (when\nthe outlined kernel function is created, before processing its contents). The\ninitial conflict resolution that was done when these changes were merged to the\nATD branch seemed to work, but it was duplicating some code and yielding\ndifferent values than clang for equivalent codes.\n\nAlso, the addition of the `IsolatedFromAbove` trait to the `omp.target` MLIR\noperation broke previous support for these clauses as well, since they were\ndefined outside the `omp.target` operation but then passed as arguments to\nnested `omp.teams` without any mapping clauses.\n\nThis is a summary of the changes introduced in this patch, which enable again\nsupport for the aforementioned clauses:\n\n  - Split the `OpenMPIRBuilder::TargetRegionInfo` structure into a structure\n  holding runtime values and another structure holding default/constant values.\n  Also, rather than having it introduce state about the current target region\n  being lowered to the `OpenMPIRBuilder` class, these structures are populated\n  and passed to the relevant builder functions from clang/flang.\n  - Remove duplicated `OpenMPIRBuilder` code introduced during a merge from\n  main, related to the handling of kernel attributes. Fixed mismatches between\n  clang and flang relating to this.\n  - Extract evaluation of clauses of TEAMS and PARALLEL directives to outside\n  the target region they are in and pass these values to new `omp.target`\n  arguments. This approach is temporary, and alternatives are being discussed\n  in the [LLVM-Flang Slack channel](https://flang-compiler.slack.com/archives/C01PY03PP9P/p1700672266767559).\n  - Runtime values for the trip count, number of teams and number of threads\n  are casted to the type expected by the corresponding RTL function, so using\n  different types won't result in a compilation error.\n  - Prevent `kmpc_push_num_threads` calls being generated for the target\n  device.\n  - Improve MLIR pattern checks for the `omp.target` operation to tell whether\n  it represents an SPMD loop.","shortMessageHtmlLink":"[WIP] Device support for num_teams, thread_limit and num_threads clauses"}},{"before":"dcca9fe6519575efe00e627db9ed7d78120ed811","after":"1222d8b3b93d573d9ad9b668e01190b433ff65d9","ref":"refs/heads/fix-num-teams-threads","pushedAt":"2023-10-17T11:57:14.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Fix broken test","shortMessageHtmlLink":"Fix broken test"}},{"before":"7acb89628ea66eeb2ea1edfff195c66ce979abe1","after":"dcca9fe6519575efe00e627db9ed7d78120ed811","ref":"refs/heads/fix-num-teams-threads","pushedAt":"2023-10-17T10:52:15.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[OMPIRBuilder] Fix mismatches on default num_teams and num_threads between clang and flang-new\n\nSet `IsSPMD` flag to get clang and flang-new to produce the same default number\nof teams and threads for an equivalent SPMD loop.","shortMessageHtmlLink":"[OMPIRBuilder] Fix mismatches on default num_teams and num_threads be…"}},{"before":"6f8ac42abe8cb41191157ee4fa0d04b36cbb73a3","after":"7acb89628ea66eeb2ea1edfff195c66ce979abe1","ref":"refs/heads/fix-num-teams-threads","pushedAt":"2023-10-17T10:44:06.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[OMPIRBuilder] Set IsSPMD flag to get clang and flang-new to produce the same default number of teams and threads","shortMessageHtmlLink":"[OMPIRBuilder] Set IsSPMD flag to get clang and flang-new to produce …"}},{"before":null,"after":"6f8ac42abe8cb41191157ee4fa0d04b36cbb73a3","ref":"refs/heads/fix-num-teams-threads","pushedAt":"2023-10-17T10:37:57.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[OMPIRBuilder] Set IsSPMD flag to get clang and flang-new to produce the same default number of teams and threads","shortMessageHtmlLink":"[OMPIRBuilder] Set IsSPMD flag to get clang and flang-new to produce …"}},{"before":"97501bc546eeb9e4a4bcf0485e88ada40c818be7","after":"9d6fe479245edfd279730379d3769fb85d5cbfa0","ref":"refs/heads/num-teams-init","pushedAt":"2023-10-13T12:26:43.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"1/2 Support for variable trip count","shortMessageHtmlLink":"1/2 Support for variable trip count"}},{"before":"a86cf6fa0ff7b3151b48dd0a4fb9c19adb7adb95","after":"97501bc546eeb9e4a4bcf0485e88ada40c818be7","ref":"refs/heads/num-teams-init","pushedAt":"2023-10-12T14:43:25.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"1/2 Support for variable trip count","shortMessageHtmlLink":"1/2 Support for variable trip count"}},{"before":"f8dafce50daa171a2b72094b88d08b09445da826","after":"a86cf6fa0ff7b3151b48dd0a4fb9c19adb7adb95","ref":"refs/heads/num-teams-init","pushedAt":"2023-10-10T13:41:55.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Add target-cpu and target-features kernel attributes.","shortMessageHtmlLink":"Add target-cpu and target-features kernel attributes."}},{"before":"5da9f41c61750d5f9677d8ed5c56573c88da12c9","after":"f8dafce50daa171a2b72094b88d08b09445da826","ref":"refs/heads/num-teams-init","pushedAt":"2023-10-10T12:46:52.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Add target-cpu and target-features kernel attributes.","shortMessageHtmlLink":"Add target-cpu and target-features kernel attributes."}},{"before":"cf392d0fa28a478d6541fc10d79cea9a96e2e9ec","after":"5da9f41c61750d5f9677d8ed5c56573c88da12c9","ref":"refs/heads/num-teams-init","pushedAt":"2023-10-10T11:01:26.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[OpenMP][OMPIRBuilder] Set num_teams, thread_limit and trip_count arguments in call to `__tgt_target_kernel`\n\nThis patch gathers the information during translation to LLVM IR about the\nvalues for num_teams, thread_limit and trip count to later use them to set up\nthe RTL call to `__tgt_target_kernel`. It takes into account different allowed\ncombinations of the teams, distribute, parallel and do directives inside a\ntarget region and it does not impact device codegen.\n\nHowever, this approach has some known issues:\n  - If the thread_limit clause is set for the omp.target operation, it will\n  result in a compiler error due to an attempt to remove the operation from\n  which it is initialized. It complains about removing it while there are uses\n  remaining. This seems likely to be related to the implicit mapping, but\n  we can ignore this limitation for milestone2.\n  - The LLVM values for the num_teams and thread_limit clauses of the omp.teams\n  operation must be compile-time constants, as well as the calculated trip\n  count of the omp.wsloop at the core of the kernel. Otherwise, it will result\n  in a compiler crash. The reason for this is that these values are defined in\n  an inner scope, so they are not visible to the parent omp.target that tries\n  to access them after. This results in the num_teams and thread_limit clauses\n  being defined after their use, as well as the trip count produced by the\n  CanonicalLoopInfo class. In addition, the definition of many of these values\n  ends up outlined to a different function.\n\nThe parallel_test application can be compiled with these limitations, but it's\nlikely real applications won't due to the need for a constant trip count (i.e.\n`do i=1,n` will crash whereas `do i=1,10` will not).\n\nEven though the arguments passed to `__tgt_target_kernel` now match between\nflang-new and clang, switching on libomptarget traces shows that the number of\nteams and threads with which the kernel is launched does not match. It's likely\nthat some kernel attributes that are only set by clang are impacting the\ndefaults for num_teams and thread_limit. As a workaround, the num_teams and\nthread_limit clauses of the teams directive can be used to set these values by\nhand.","shortMessageHtmlLink":"[OpenMP][OMPIRBuilder] Set num_teams, thread_limit and trip_count arg…"}},{"before":null,"after":"cf392d0fa28a478d6541fc10d79cea9a96e2e9ec","ref":"refs/heads/num-teams-init","pushedAt":"2023-10-10T11:00:29.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[WIP][OpenMP][OMPIRBuilder] Set num_teams, thread_limit and trip_count arguments in call to `__tgt_target_kernel`\n\nThis patch gathers the information during translation to LLVM IR about the\nvalues for num_teams, thread_limit and trip count to later use them to set up\nthe RTL call to `__tgt_target_kernel`. It takes into account different allowed\ncombinations of the teams, distribute, parallel and do directives inside a\ntarget region and it does not impact device codegen.\n\nHowever, this approach has some known issues:\n  - If the thread_limit clause is set for the omp.target operation, it will\n  result in a compiler error due to an attempt to remove the operation from\n  which it is initialized. It complains about removing it while there are uses\n  remaining. This seems likely to be related to the implicit mapping, but\n  we can ignore this limitation for milestone2.\n  - The LLVM values for the num_teams and thread_limit clauses of the omp.teams\n  operation must be compile-time constants, as well as the calculated trip\n  count of the omp.wsloop at the core of the kernel. Otherwise, it will result\n  in a compiler crash. The reason for this is that these values are defined in\n  an inner scope, so they are not visible to the parent omp.target that tries\n  to access them after. This results in the num_teams and thread_limit clauses\n  being defined after their use, as well as the trip count produced by the\n  CanonicalLoopInfo class. In addition, the definition of many of these values\n  ends up outlined to a different function.\n\nThe parallel_test application can be compiled with these limitations, but it's\nlikely real applications won't due to the need for a constant trip count (i.e.\n`do i=1,n` will crash whereas `do i=1,10` will not).\n\nEven though the arguments passed to `__tgt_target_kernel` now match between\nflang-new and clang, switching on libomptarget traces shows that the number of\nteams and threads with which the kernel is launched does not match. It's likely\nthat some kernel attributes that are only set by clang are impacting the\ndefaults for num_teams and thread_limit. As a workaround, the num_teams and\nthread_limit clauses of the teams directive can be used to set these values by\nhand.","shortMessageHtmlLink":"[WIP][OpenMP][OMPIRBuilder] Set num_teams, thread_limit and trip_coun…"}},{"before":null,"after":"2b16dcf4e46c46b682ea3f3f60bc2eb08df1703d","ref":"refs/heads/kmpc-for-static-loop","pushedAt":"2023-10-02T13:22:56.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[OMPIRBuilder] Improve type handling for kmpc_for_static_loop\n\nThere are 4 variants to this new RTL function:\n  - kmpc_for_static_loop_4: 4-byte, signed\n  - kmpc_for_static_loop_4u: 4-byte, unsigned\n  - kmpc_for_static_loop_8: 8-byte, signed\n  - kmpc_for_static_loop_8u: 8-byte, unsigned\n\nThis patch aims at using the type of the trip count variable to select the\nappropriate variant of this RTL function, rather than always calling the signed\n4-byte version and possibly truncating the trip count. It's based on\n`getKmpcForDynamicInitForType()` and other similar helper functions, which\nresult always in codegen for the unsigned variants of the function.\n\nA small bug, which produces a compiler crash when a target worksharing loop\nconstruct does not capture outside variables, is also addressed. This example\nwould trigger this bug:\n\n```\nsubroutine foo()\n  !$omp target parallel do\n  do i = 1, 10\n    call bar(i)\n  end do\n  !$omp end target parallel do\nend subroutine\n```","shortMessageHtmlLink":"[OMPIRBuilder] Improve type handling for kmpc_for_static_loop"}},{"before":"f1d4e7cdb941fdde2385c2dff9bfbc6180d4b2a5","after":null,"ref":"refs/heads/lifetime_alloca_loop","pushedAt":"2023-09-29T14:32:56.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"}},{"before":"ced8645f8a3969bd2e7adb708124405dea3ec4e3","after":null,"ref":"refs/heads/cse-fix","pushedAt":"2023-09-29T14:32:52.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"}},{"before":"ff9c6dbb9af36a1da0095c2fefc4c0a3715e03fe","after":"f1d4e7cdb941fdde2385c2dff9bfbc6180d4b2a5","ref":"refs/heads/lifetime_alloca_loop","pushedAt":"2023-09-29T10:35:38.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[Flang][OpenMP][MLIR] Create lifetime markers for allocations only used within OpenMP loop regions\n\nBy creating `llvm.lifetime.start` and `llvm.lifetime.end` markers for outside\nallocations around the body of the loop produced for `omp.wsloop` and\n`omp.simdloop` MLIR operations, later uses of the LLVM `CodeExtractor` class to\npotentially outline the body of these loops into independent functions will be\nable to see the reduced scope of use of these allocations and sink them into\nthe outlined function's body rather than to unnecessarily pass them as\narguments. This can also help later optimization stages to detect cases where\nallocations for loop indices are redundant.","shortMessageHtmlLink":"[Flang][OpenMP][MLIR] Create lifetime markers for allocations only us…"}},{"before":"e8fd4507e3b00781660b7cd96ae25d07952f625f","after":null,"ref":"refs/heads/skatrak/atd/target-combined-parser","pushedAt":"2023-09-28T14:34:26.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"}},{"before":null,"after":"ced8645f8a3969bd2e7adb708124405dea3ec4e3","ref":"refs/heads/cse-fix","pushedAt":"2023-09-28T14:27:06.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Allow CSE to cross omp.{teams, distribute, parallel} regions","shortMessageHtmlLink":"Allow CSE to cross omp.{teams, distribute, parallel} regions"}},{"before":"e36b52ea8296b81126312975fb554701af1c2158","after":null,"ref":"refs/heads/alloca-pass","pushedAt":"2023-09-28T13:18:47.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"}},{"before":null,"after":"ff9c6dbb9af36a1da0095c2fefc4c0a3715e03fe","ref":"refs/heads/lifetime_alloca_loop","pushedAt":"2023-09-27T15:17:50.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[Flang][OpenMP][MLIR] Create lifetime markers for allocations only used within OpenMP loop regions\n\nBy creating `llvm.lifetime.start` and `llvm.lifetime.end` markers for outside\nallocations around the body of the loop produced for `omp.wsloop` and\n`omp.simdloop` MLIR operations, later uses of the LLVM `CodeExtractor` class to\npotentially outline the body of these loops into independent functions will be\nable to see the reduced scope of use of these allocations and sink them into\nthe outlined function's body rather than to unnecessarily pass them as\narguments. This can also help later optimization stages to detect cases where\nallocations for loop indices are redundant.","shortMessageHtmlLink":"[Flang][OpenMP][MLIR] Create lifetime markers for allocations only us…"}},{"before":"0c2dbe5895c7f24bb2d2e6acea61e2516a3712bf","after":"a28425da38726a4417e705c4806c7f2ba9ff56e9","ref":"refs/heads/amd-trunk-dev","pushedAt":"2023-09-21T14:43:11.000Z","pushType":"push","commitsCount":3482,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Merge attempt by ./merge_from_main.sh into branch amd-trunk-dev","shortMessageHtmlLink":"Merge attempt by ./merge_from_main.sh into branch amd-trunk-dev"}},{"before":"06b3f6086c7b46c4caccb868b0f7d09702848222","after":"24161bcc4448d6dad2f6afca55ab0a32451b5ae1","ref":"refs/heads/main","pushedAt":"2023-09-21T13:18:43.000Z","pushType":"push","commitsCount":10000,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[libc++] Move a few XFAILs to UNSUPPORTED for consistency\n\nThe wide stream tests should be marked as `UNSUPPORTED: no-wide-characters`,\nnot XFAIL.","shortMessageHtmlLink":"[libc++] Move a few XFAILs to UNSUPPORTED for consistency"}},{"before":"4ba24bd70905acaf6f73482071ca8d189783fafb","after":"e36b52ea8296b81126312975fb554701af1c2158","ref":"refs/heads/alloca-pass","pushedAt":"2023-09-21T10:44:31.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Update unit tests","shortMessageHtmlLink":"Update unit tests"}},{"before":"59208aefbbf4e2d5cb0710b30fadee4bdea41170","after":"4ba24bd70905acaf6f73482071ca8d189783fafb","ref":"refs/heads/alloca-pass","pushedAt":"2023-09-20T16:36:12.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Added support for HLFIR and omp.simdloop","shortMessageHtmlLink":"Added support for HLFIR and omp.simdloop"}},{"before":"58d7eabf5d44c4a511faab5bfd33becf891b58ef","after":"59208aefbbf4e2d5cb0710b30fadee4bdea41170","ref":"refs/heads/alloca-pass","pushedAt":"2023-09-13T10:50:51.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"Push index fir.alloca inside of omp.wsloop if only used there","shortMessageHtmlLink":"Push index fir.alloca inside of omp.wsloop if only used there"}},{"before":"66ce4b9073453e0d4ab555aba59075db55711727","after":"e8fd4507e3b00781660b7cd96ae25d07952f625f","ref":"refs/heads/skatrak/atd/target-combined-parser","pushedAt":"2023-09-11T09:53:54.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"skatrak","name":"Sergio Afonso","path":"/skatrak","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2736073?s=80&v=4"},"commit":{"message":"[flang][mlir] Initial lowering support for \"target teams distribute parallel do\"","shortMessageHtmlLink":"[flang][mlir] Initial lowering support for \"target teams distribute p…"}}],"hasNextPage":true,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"Y3Vyc29yOnYyOpK7MjAyNC0wMS0wNVQxNDowMDozOC4wMDAwMDBazwAAAAPYnIBQ","startCursor":"Y3Vyc29yOnYyOpK7MjAyNC0wMS0wNVQxNDowMDozOC4wMDAwMDBazwAAAAPYnIBQ","endCursor":"Y3Vyc29yOnYyOpK7MjAyMy0wOS0xMVQwOTo1Mzo1NC4wMDAwMDBazwAAAAN9tqef"}},"title":"Activity · skatrak/llvm-project-rocm"}