{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":23496542,"defaultBranch":"main","name":"triton","ownerLogin":"triton-lang","currentUserCanPush":false,"isFork":false,"isEmpty":false,"createdAt":"2014-08-30T17:07:16.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/169110177?v=4","public":true,"private":false,"isOrgOwned":true},"refInfo":{"name":"","listCacheKey":"v0:1727484875.0","currentOid":""},"activityList":{"items":[{"before":"39e606465ee90fab0bef4f21eb843c94b5f3d252","after":null,"ref":"refs/heads/hoy/tma-dtype","pushedAt":"2024-09-28T00:54:35.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"peterbell10","name":null,"path":"/peterbell10","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/13238737?s=80&v=4"}},{"before":"fe47f98bc73a8c630c87cdde57ce11be26a09b60","after":"6af74b2f4535682abfc0b08958bc2c6831036d29","ref":"refs/heads/main","pushedAt":"2024-09-28T00:54:34.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"peterbell10","name":null,"path":"/peterbell10","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/13238737?s=80&v=4"},"commit":{"message":"[FRONTEND] Support passing dtype as constexpr for tma load (#4821)\n\nFixing an compile error like below when passing dtype through kernel arg\r\nfor `tl._experimental_descriptor_load`:\r\n\r\n AttributeError: 'constexpr' object has no attribute 'to_ir'","shortMessageHtmlLink":"[FRONTEND] Support passing dtype as constexpr for tma load (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2553844756\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4821\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4821/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4821\">#4821</a>)"}},{"before":"900ca56cd394d61dc89cb183fc6af02716062e88","after":"39e606465ee90fab0bef4f21eb843c94b5f3d252","ref":"refs/heads/hoy/tma-dtype","pushedAt":"2024-09-28T00:45:27.000Z","pushType":"push","commitsCount":3,"pusher":{"login":"peterbell10","name":null,"path":"/peterbell10","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/13238737?s=80&v=4"},"commit":{"message":"Merge branch 'main' into hoy/tma-dtype","shortMessageHtmlLink":"Merge branch 'main' into hoy/tma-dtype"}},{"before":"992edb0fe8061e2fd8860dbcb99a22e70e246477","after":"feaa7fa5456c07f0f13700835ffeb209478356a0","ref":"refs/heads/gh-pages","pushedAt":"2024-09-28T00:09:29.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"ptillet","name":"Philippe Tillet","path":"/ptillet","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/614788?s=80&v=4"},"commit":{"message":"[GH-PAGES] Updated website","shortMessageHtmlLink":"[GH-PAGES] Updated website"}},{"before":null,"after":"900ca56cd394d61dc89cb183fc6af02716062e88","ref":"refs/heads/hoy/tma-dtype","pushedAt":"2024-09-27T22:41:35.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"htyu","name":"Hongtao Yu","path":"/htyu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/15899383?s=80&v=4"},"commit":{"message":"[FRONTEND] Support passing dtype as constexpr for tma load","shortMessageHtmlLink":"[FRONTEND] Support passing dtype as constexpr for tma load"}},{"before":"8b498c610406ff64580715f91bc69ed24e38ec24","after":null,"ref":"refs/heads/embg-cublas-sleep","pushedAt":"2024-09-27T21:26:10.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"pawelszczerbuk","name":null,"path":"/pawelszczerbuk","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/153013546?s=80&v=4"}},{"before":"0b4feb77ce54a65897526861333ef8dcd78ef27f","after":"fe47f98bc73a8c630c87cdde57ce11be26a09b60","ref":"refs/heads/main","pushedAt":"2024-09-27T21:26:08.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"pawelszczerbuk","name":null,"path":"/pawelszczerbuk","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/153013546?s=80&v=4"},"commit":{"message":"09-persistent-matmul.py bugfix (#4820)\n\nCurrently we sleep between each rep for Triton kernels, but not for the\r\ncuBLAS kernel. This may improve cuBLAS performance on fp8 due to thermal\r\nissues.","shortMessageHtmlLink":"09-persistent-matmul.py bugfix (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2553448096\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4820\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4820/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4820\">#4820</a>)"}},{"before":null,"after":"8b498c610406ff64580715f91bc69ed24e38ec24","ref":"refs/heads/embg-cublas-sleep","pushedAt":"2024-09-27T17:52:48.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"embg","name":"Elliot Gorokhovsky","path":"/embg","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/12179121?s=80&v=4"},"commit":{"message":"09-persistent-matmul.py bugfix\n\nCurrently we sleep between each rep for Triton kernels, but not for the cuBLAS kernel. This may improve cuBLAS performance on fp8 due to thermal issues.","shortMessageHtmlLink":"09-persistent-matmul.py bugfix"}},{"before":"50f64d19e25b8ed47034a850f37b26331f2026d7","after":null,"ref":"refs/heads/phil/testing-fixup","pushedAt":"2024-09-27T17:02:50.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"ptillet","name":"Philippe Tillet","path":"/ptillet","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/614788?s=80&v=4"}},{"before":"e7ec3fee2983a447a31812da2e56acdc0d7ee6af","after":"0b4feb77ce54a65897526861333ef8dcd78ef27f","ref":"refs/heads/main","pushedAt":"2024-09-27T17:02:49.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"ptillet","name":"Philippe Tillet","path":"/ptillet","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/614788?s=80&v=4"},"commit":{"message":"[testing] moved `di = torch._dynamo.device_interface` into backend (#4818)","shortMessageHtmlLink":"[testing] moved <code>di = torch._dynamo.device_interface</code> into backend (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2553328992\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4818\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4818/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4818\">#…</a>"}},{"before":null,"after":"50f64d19e25b8ed47034a850f37b26331f2026d7","ref":"refs/heads/phil/testing-fixup","pushedAt":"2024-09-27T16:35:52.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"ptillet","name":"Philippe Tillet","path":"/ptillet","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/614788?s=80&v=4"},"commit":{"message":"[testing] moved `di = torch._dynamo.device_interface` into backend","shortMessageHtmlLink":"[testing] moved <code>di = torch._dynamo.device_interface</code> into backend"}},{"before":"2ef33c6c4c39e8c30de71249b6e1cb808f30bc34","after":"e7ec3fee2983a447a31812da2e56acdc0d7ee6af","ref":"refs/heads/main","pushedAt":"2024-09-27T03:35:47.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"manman-ren","name":"Manman Ren","path":"/manman-ren","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/68095600?s=80&v=4"},"commit":{"message":"[SWP] attempt to remove a workaround for a triton llvm codegen bug (#4774)\n\nTriton LLVM codegen has a bug where local_loads from #shared to #mma\r\nlayout can lead to invalid code if the loaded shape is smaller than the\r\nmma tile. Remove the workaround.\r\n\r\nSee https://github.com/triton-lang/triton/pull/3561.\r\n\r\nVerified that with test case: https://pastebin.com/xxP3cFmy (test.mlir),\r\nrunning\r\ntriton-opt test.mlir -tritongpu-pipeline=num-stages=3\r\n--convert-scf-to-cf --allocate-shared-memory\r\n--convert-triton-gpu-to-llvm\r\nhas no issue.\r\n\r\nUnit test case added in https://github.com/triton-lang/triton/pull/4798\r\nalso shows no issue.","shortMessageHtmlLink":"[SWP] attempt to remove a workaround for a triton llvm codegen bug (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2539862596\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4774\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4774/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4774\">#…</a>"}},{"before":"c2107648730402f29cded0665a1dc347d3983e91","after":"2ef33c6c4c39e8c30de71249b6e1cb808f30bc34","ref":"refs/heads/main","pushedAt":"2024-09-27T00:18:18.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"pawelszczerbuk","name":null,"path":"/pawelszczerbuk","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/153013546?s=80&v=4"},"commit":{"message":"[SWP] When num_stages = 2, do not pipeline indirect loads (#4721)\n\nFor indirect loads, we try to assign them to later stages\r\n```\r\nunsigned stagesBetweenLoads =\r\n   ceil<unsigned>(numStages - 2, maxIndirectionLevel + 1);\r\nint stage = (maxIndirectionLevel - indLevel) * stagesBetweenLoads;\r\nschedule.insert(loadOp, stage, loadsClusters[indLevel]);\r\n```\r\nIf numStages is 2, there is no later stage to assign the indirect loads\r\nto. The fix is to not pipeline the indirect loads.\r\nWe also generalize to not pipeline an indirect load if the indirection\r\nlevel >= numStages - 1","shortMessageHtmlLink":"[SWP] When num_stages = 2, do not pipeline indirect loads (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2523375046\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4721\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4721/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4721\">#4721</a>)"}},{"before":"e65dd810411a7c1312637e1ebba0bdf9edb27d80","after":"c2107648730402f29cded0665a1dc347d3983e91","ref":"refs/heads/main","pushedAt":"2024-09-27T00:17:07.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"pawelszczerbuk","name":null,"path":"/pawelszczerbuk","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/153013546?s=80&v=4"},"commit":{"message":"[SWP] split loads to handle incompatible shared encoding (#4784)\n\nSummary: split loads so each group of uses with the same shared encoding\r\nwill have a corresponding load. This enables pipelining loads with\r\nincompatible shared encoding.\r\n\r\nAMD has its own version of assignMemoryLayouts, so the test case\r\nload_two_users_incompatible_layouts will have different results for AMD.","shortMessageHtmlLink":"[SWP] split loads to handle incompatible shared encoding (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2543305414\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4784\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4784/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4784\">#4784</a>)"}},{"before":"0750db32eb329416260c4204d9a8a639c8b1111f","after":"992edb0fe8061e2fd8860dbcb99a22e70e246477","ref":"refs/heads/gh-pages","pushedAt":"2024-09-27T00:10:00.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"ptillet","name":"Philippe Tillet","path":"/ptillet","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/614788?s=80&v=4"},"commit":{"message":"[GH-PAGES] Updated website","shortMessageHtmlLink":"[GH-PAGES] Updated website"}},{"before":"5427206ac3cb2d55d1fedb00b268cc6cb85a199c","after":null,"ref":"refs/heads/keren/fix-tutorial","pushedAt":"2024-09-27T00:00:15.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"Jokeren","name":"Keren Zhou","path":"/Jokeren","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2306281?s=80&v=4"}},{"before":"184fb53b1a573a802e7d4b6a8bab285c44102965","after":"e65dd810411a7c1312637e1ebba0bdf9edb27d80","ref":"refs/heads/main","pushedAt":"2024-09-27T00:00:14.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"Jokeren","name":"Keren Zhou","path":"/Jokeren","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2306281?s=80&v=4"},"commit":{"message":"[TUTORIAL] Multiple improvements to the tutorials, especially to `09-persistent-matmul.py` (#4802)\n\n- Format the introduction section in some tutorials.\r\n- Add instructions for running the persistent matmul tutorial, as well\r\nas instructions for using `proton-viewer`.\r\n- Replace `torch.zeros` with `torch.empty` to remove unnecessary GPU\r\nkernels.\r\n- Add brackets `[` and `]` around shapes to improve the output\r\nformatting.\r\n- Remove redundant metric accumulation, as the Triton hook already\r\nhandles metric accumulation.","shortMessageHtmlLink":"[TUTORIAL] Multiple improvements to the tutorials, especially to `09-…"}},{"before":"1b0f9ea7f1f6e9961251d3a6ad97fd72521a59e5","after":"184fb53b1a573a802e7d4b6a8bab285c44102965","ref":"refs/heads/main","pushedAt":"2024-09-26T21:41:37.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"Jokeren","name":"Keren Zhou","path":"/Jokeren","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2306281?s=80&v=4"},"commit":{"message":"[AUTOTUNER] Fix issue in autotuner which may use the wrong value as the key of config cache. (#4808)\n\nThe autotuner uses the index of the arguments of the Triton kernel\r\nsignature to look up the value to be used as the key for the config\r\ncache.\r\nThere is an issue if the user pass the kernel arguments as keyword args\r\nin arbitrary order.\r\n\r\nThe name of the argument should be used to look up the value of the args\r\npassed by the user instead of the `key_idx`.\r\nThis prevents the autotuner from using a mismatched value as the key for\r\ncaching when the arguments are passed in arbitrary order as keyword\r\nargs.\r\n\r\n- [X] I am not making a trivial change, such as fixing a typo in a\r\ncomment.\r\n\r\n- [X] I have written a PR description following these\r\n  [rules](https://cbea.ms/git-commit/#why-not-how).\r\n\r\n- [X] I have run `pre-commit run --from-ref origin/main --to-ref HEAD`.\r\n\r\n- Select one of the following.\r\n  - [X] I have added tests.\r\n    - `/test` for `lit` tests\r\n    - `/unittest` for C++ tests\r\n    - `/python/test` for end-to-end tests\r\n  - [ ] This PR does not need a test because `FILL THIS IN`.\r\n\r\n- Select one of the following.\r\n  - [X] I have not added any `lit` tests.\r\n- [ ] The `lit` tests I have added follow these [best\r\npractices](https://mlir.llvm.org/getting_started/TestingGuide/#filecheck-best-practices),\r\nincluding the \"tests should be minimal\" section. (Usually running Python\r\ncode\r\n    and using the instructions it generates is not minimal.)","shortMessageHtmlLink":"[AUTOTUNER] Fix issue in autotuner which may use the wrong value as t…"}},{"before":"f4a565ce6678f649e04d5cc8daf34703b825739c","after":null,"ref":"refs/heads/pb/reduction-assert","pushedAt":"2024-09-26T20:56:26.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"peterbell10","name":null,"path":"/peterbell10","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/13238737?s=80&v=4"}},{"before":"a70d5856ff2e8e15bf259c6a0bcbff3061fc15e9","after":"1b0f9ea7f1f6e9961251d3a6ad97fd72521a59e5","ref":"refs/heads/main","pushedAt":"2024-09-26T20:56:25.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"peterbell10","name":null,"path":"/peterbell10","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/13238737?s=80&v=4"},"commit":{"message":"[Backend] Fix device assert inside reduction/scan region (#4811)\n\nCurrently the reduction codegen unconditionally executes the combine\r\nregion which can create problems because we conditionally load from\r\nshared memory, so this uses uninitialized registers.\r\n\r\nGenerally combine regions should be pure, so this shouldn't be\r\nobservable but with the overflow sanitizer the frontend injects\r\nassertions into the combine region.\r\n\r\nThis changes the `accumulate` function to take a predicate and if the\r\ncombine region isn't speculateble we only run it on threads where the\r\npredicate is true. In the common case, the codegen is unchanged.","shortMessageHtmlLink":"[Backend] Fix device assert inside reduction/scan region (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2551325366\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4811\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4811/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4811\">#4811</a>)"}},{"before":"37a4e62544d144d967f99baafadab77fe00c0f36","after":"db3381caf547bca6d68712b78da1a805237a3b35","ref":"refs/heads/keren/local-alloc","pushedAt":"2024-09-26T20:54:21.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Jokeren","name":"Keren Zhou","path":"/Jokeren","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2306281?s=80&v=4"},"commit":{"message":"Update","shortMessageHtmlLink":"Update"}},{"before":"d18e7a23909163321ee14e46fb4ffa67246b9dfd","after":"5427206ac3cb2d55d1fedb00b268cc6cb85a199c","ref":"refs/heads/keren/fix-tutorial","pushedAt":"2024-09-26T20:40:58.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Jokeren","name":"Keren Zhou","path":"/Jokeren","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2306281?s=80&v=4"},"commit":{"message":"Update","shortMessageHtmlLink":"Update"}},{"before":"c5392da26df310e65058e3515c0cedee098866bc","after":"d18e7a23909163321ee14e46fb4ffa67246b9dfd","ref":"refs/heads/keren/fix-tutorial","pushedAt":"2024-09-26T20:39:48.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Jokeren","name":"Keren Zhou","path":"/Jokeren","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/2306281?s=80&v=4"},"commit":{"message":"Update","shortMessageHtmlLink":"Update"}},{"before":"9a732a20bd95be718cd6fda2be0f8d51972ec70c","after":"f4a565ce6678f649e04d5cc8daf34703b825739c","ref":"refs/heads/pb/reduction-assert","pushedAt":"2024-09-26T18:55:12.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"peterbell10","name":null,"path":"/peterbell10","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/13238737?s=80&v=4"},"commit":{"message":"[Backend] Fix device assert inside reduction/scan region\n\nCurrently the reduction codegen unconditionally executes the combine\nregion which can create problems because we conditionally load from\nshared memory, so this uses uninitialized registers.\n\nGenerally combine regions should be pure, so this shouldn't be\nobservable but with the overflow sanitizer the frontend injects\nassertions into the combine region.\n\nThis changes the `accumulate` function to take a predicate and if the\ncombine region isn't speculateble we only run it on threads where the\npredicate is true. In the common case, the codegen is unchanged.","shortMessageHtmlLink":"[Backend] Fix device assert inside reduction/scan region"}},{"before":"dceaa3c786488f43eea54be5d9b37cd3084b7183","after":"9a732a20bd95be718cd6fda2be0f8d51972ec70c","ref":"refs/heads/pb/reduction-assert","pushedAt":"2024-09-26T18:50:22.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"peterbell10","name":null,"path":"/peterbell10","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/13238737?s=80&v=4"},"commit":{"message":"[Backend] Fix device assert inside reduction/scan region\n\nCurrently the reduction codegen unconditionally executes the combine\nregion which can create problems because we conditionally load from\nshared memory, so this uses uninitialized registers.\n\nGenerally combine regions should be pure, so this shouldn't be\nobservable but with the overflow sanitizer the frontend injects\nassertions into the combine region.\n\nThis changes the `accumulate` function to take a predicate and if the\ncombine region isn't speculateble we only run it on threads where the\npredicate is true. In the common case, the codegen is unchanged.","shortMessageHtmlLink":"[Backend] Fix device assert inside reduction/scan region"}},{"before":null,"after":"dceaa3c786488f43eea54be5d9b37cd3084b7183","ref":"refs/heads/pb/reduction-assert","pushedAt":"2024-09-26T18:31:39.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"peterbell10","name":null,"path":"/peterbell10","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/13238737?s=80&v=4"},"commit":{"message":"[Backend] Fix device assert inside reduction/scan region\n\nCurrently the reduction codegen unconditionally executes the combine\nregion which can create problems because we conditionally load from\nshared memory, so this uses uninitialized registers.\n\nGenerally combine regions should be pure, so this shouldn't be\nobservable but with the overflow sanitizer the frontend injects\nassertions into the combine region.\n\nThis changes the `accumulate` function to take a predicate and if the\ncombine region isn't speculateble we only run it on threads where the\npredicate is true. In the common case, the codegen is unchanged.","shortMessageHtmlLink":"[Backend] Fix device assert inside reduction/scan region"}},{"before":"4348109b0a8e1aac748aa9b1bbbcd858e9488940","after":"a70d5856ff2e8e15bf259c6a0bcbff3061fc15e9","ref":"refs/heads/main","pushedAt":"2024-09-26T15:22:44.000Z","pushType":"pr_merge","commitsCount":1,"pusher":{"login":"antiagainst","name":"Lei Zhang","path":"/antiagainst","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/487928?s=80&v=4"},"commit":{"message":"[AMD] Turn stream pipeline v2 as the default (#4665)\n\nThis commit turns on the v2 pipeliner as the default.\r\nWe still keep v1 for some extended time to make\r\nperf debugging easier; but expect to remove it soon.","shortMessageHtmlLink":"[AMD] Turn stream pipeline v2 as the default (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2510740518\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4665\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4665/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4665\">#4665</a>)"}},{"before":"40650b0be75fb937aec6443a2f3ffa53d35ea195","after":"3fe4e6b1ddc2fc53d3ec252f0bb952072cdc21a6","ref":"refs/heads/llvm-head","pushedAt":"2024-09-26T13:39:51.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"karupayun","name":"Pablo Zimmermann","path":"/karupayun","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8610429?s=80&v=4"},"commit":{"message":"[BACKEND] Update LLVM version to https://github.com/llvm/llvm-project/commit/df0864e761107b07e38f5503e0cbee0cebb4c5e8 (#4790)","shortMessageHtmlLink":"[BACKEND] Update LLVM version to <a class=\"commit-link\" data-hovercard-type=\"commit\" data-hovercard-url=\"https://github.com/llvm/llvm-project/commit/df0864e761107b07e38f5503e0cbee0cebb4c5e8/hovercard\" href=\"https://github.com/llvm/llvm-project/commit/df0864e761107b07e38f5503e0cbee0cebb4c5e8\">llvm/llvm-project@<tt>df0864e</tt></a> (<a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2545296698\" data-permission-text=\"Title is private\" data-url=\"https://github.com/triton-lang/triton/issues/4790\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/triton-lang/triton/pull/4790/hovercard\" href=\"https://github.com/triton-lang/triton/pull/4790\">#4790</a>)"}},{"before":"3fe4e6b1ddc2fc53d3ec252f0bb952072cdc21a6","after":"40650b0be75fb937aec6443a2f3ffa53d35ea195","ref":"refs/heads/llvm-head","pushedAt":"2024-09-26T09:54:23.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"karupayun","name":"Pablo Zimmermann","path":"/karupayun","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8610429?s=80&v=4"},"commit":{"message":"Use clang-12 to fix issue with LLVM_ABI_BREAKING_CHECKS\n\nThis commit is fixing the root cause of the issue that appeared during\nan llvm update, and was raised in https://github.com/triton-lang/triton/pull/4212#issuecomment-2234035360.\n\nThe issue was temporary fixed in https://github.com/triton-lang/triton/pull/4512,\nwhere we set LLVM_ABI_BREAKING_CHECKS to FORCE_OFF constant in order to\nnot use a non-deterministic seed inside the hashing function\n(included in this llvm commit: https://github.com/llvm/llvm-project/commit/ce80c80dca45c7b4636a3e143973e2c6cbdb2884).\n\nA further investigation (with chsigg@) found that the issue is that LLVM\nis built with clang11, while Triton use a newer version. The ABI issue\nis brought up here: https://github.com/llvm/llvm-project/pull/96282#discussion_r1657919892,\nbut the consensus seemed to be that this setup is rare.\n\nUpdated the clang version to 12 in the ubuntu build fixed the issue and\ntherefore we can revert setting LLVM_ABI_BREAKING_CHECKS constant. I am\nadditionaly erasing LLVM_ABI_BREAKING_CHECKS in the setup of the other\nhardwares (it seems it was not needed) and I am splitting ubuntu and\nmacOS configurations because it seems cleaner than having a variable\nthat sets the compiler version for each of them.","shortMessageHtmlLink":"Use clang-12 to fix issue with LLVM_ABI_BREAKING_CHECKS"}},{"before":"b9ac1479c5e556339f8026323bf00a066651a4ff","after":"0750db32eb329416260c4204d9a8a639c8b1111f","ref":"refs/heads/gh-pages","pushedAt":"2024-09-26T00:09:35.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"ptillet","name":"Philippe Tillet","path":"/ptillet","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/614788?s=80&v=4"},"commit":{"message":"[GH-PAGES] Updated website","shortMessageHtmlLink":"[GH-PAGES] Updated website"}}],"hasNextPage":true,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"startCursor":"Y3Vyc29yOnYyOpK7MjAyNC0wOS0yOFQwMDo1NDozNS4wMDAwMDBazwAAAATDH5d1","endCursor":"Y3Vyc29yOnYyOpK7MjAyNC0wOS0yNlQwMDowOTozNS4wMDAwMDBazwAAAATA5nH9"}},"title":"Activity · triton-lang/triton"}