Update on "[AOTInductor] Implement autograd eager backend for native …

…triton kernels"

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng

[ghstack-poisoned]

.clang-tidy

@@ -30,8 +30,13 @@ cppcoreguidelines-*,
 -facebook-hte-RelativeInclude,
 hicpp-exception-baseclass,
 hicpp-avoid-goto,
-misc-unused-alias-decls,
-misc-unused-using-decls,
+misc-*,
+-misc-const-correctness,
+-misc-use-anonymous-namespace,
+-misc-unused-parameters,
+-misc-no-recursion,
+-misc-non-private-member-variables-in-classes,
+-misc-confusable-identifiers,
 modernize-*,
 -modernize-concat-nested-namespaces,
 -modernize-macro-to-enum,
@@ -44,7 +49,7 @@ modernize-*,
 performance-*,
 readability-container-size-empty,
 '
-HeaderFilterRegex: '^(c10/(?!test)|torch/csrc/).*$'
+HeaderFilterRegex: '^(c10/|torch/csrc/).*$'
 AnalyzeTemporaryDtors: false
 WarningsAsErrors: '*'
 ...

.github/actionlint.yaml

@@ -23,3 +23,5 @@ self-hosted-runner:
     - macos-12-xl
     - macos-12
     - macos12.3-m1
+    - macos-latest-xlarge
+    - macos-13-xlarge

.github/ci_commit_pins/vision.txt

@@ -1 +1 @@
-3da865859d458673e733b5857a14efc1e02383ef
+48f8473e21b0f3e425aabc60db201b68fedf59b3

.github/requirements/conda-env-iOS → .github/requirements/conda-env-iOS.txt

@@ -1,7 +1,5 @@
 blas=1.0
 cmake=3.22.1
-mkl=2022.1.0
-mkl-include=2022.1.0
 ninja=1.10.2
 numpy=1.23.3
 pyyaml=6.0

.github/scripts/generate_ci_workflows.py

@@ -60,7 +60,7 @@ class BinaryBuildWorkflow:
     branches: str = "nightly"
     # Mainly for macos
     cross_compile_arm64: bool = False
-    xcode_version: str = ""
+    macos_runner: str = "macos-12-xl"
 
     def __post_init__(self) -> None:
         if self.abi_version:
@@ -307,7 +307,8 @@ class OperatingSystem:
         build_configs=generate_binary_build_matrix.generate_wheels_matrix(
             OperatingSystem.MACOS_ARM64
         ),
-        cross_compile_arm64=True,
+        cross_compile_arm64=False,
+        macos_runner="macos-13-xlarge",
         ciflow_config=CIFlowConfig(
             labels={LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_WHEEL},
             isolated_workflow=True,

.github/scripts/test_trymerge.py

@@ -21,7 +21,6 @@
 from trymerge import (
     categorize_checks,
     find_matching_merge_rule,
-    FlakyRule,
     get_classifications,
     get_drci_classifications,
     get_rockset_results,
@@ -226,16 +225,6 @@ def mocked_read_merge_rules_raise(repo: Any, org: str, project: str) -> List[Mer
     raise RuntimeError("testing")
 
 
-def empty_flaky_rules() -> List[FlakyRule]:
-    return []
-
-
-def xla_is_flaky_rules() -> List[FlakyRule]:
-    return [
-        FlakyRule("xla", ["FAILED: Build did NOT complete successfully"]),
-    ]
-
-
 def xla_merge_rules(repo: Any, org: str, project: str) -> List[MergeRule]:
     return [
         MergeRule(
@@ -247,6 +236,7 @@ def xla_merge_rules(repo: Any, org: str, project: str) -> List[MergeRule]:
                 "EasyCLA",
                 "pull / linux-bionic-py3_8-clang8-xla / build",
                 "pull / linux-bionic-py3_8-clang8-xla / test (xla, 1, 1, linux.4xlarge)",
+                "inductor / cuda11.8-py3.10-gcc7-sm86 / test (inductor_torchbench_dynamic, 1, 1, linux.g5.4xlarge.nvidia.gpu)",
             ],
             ignore_flaky_failures=False,
         ),
@@ -268,7 +258,6 @@ def commit_message(self, ref: str) -> str:
         return "super awsome commit message"
 
 
-@mock.patch("trymerge.read_flaky_rules", side_effect=empty_flaky_rules)
 @mock.patch("trymerge.get_rockset_results", side_effect=empty_rockset_results)
 @mock.patch("trymerge.gh_graphql", side_effect=mocked_gh_graphql)
 @mock.patch(
@@ -677,55 +666,74 @@ def test_get_merge_base(self, *args: Any) -> None:
 )
 class TestBypassFailures(TestCase):
     def test_get_classifications(self, *args: Any) -> None:
-        flaky_rules = [
-            # Try a regex rule
-            FlakyRule("distributed", ["##\\[error\\]The operation [wW]as .+"])
-        ]
-        pr = GitHubPR("pytorch", "pytorch", 92863)
+        pr = GitHubPR("pytorch", "pytorch", 109584)
         checks = pr.get_checkrun_conclusions()
         checks = get_classifications(
             pr.pr_num,
             pr.project,
             checks,
             pr.last_commit()["oid"],
             pr.get_merge_base(),
-            flaky_rules,
             [],
         )
         self.assertTrue(
             checks[
-                "pull / linux-bionic-py3_7-clang8-xla / test (xla, 1, 1, linux.4xlarge)"
+                "pull / linux-focal-py3.11-clang10 / test (dynamo, 1, 2, linux.2xlarge)"
+            ].classification
+            == "BROKEN_TRUNK"
+        )
+        self.assertTrue(
+            checks[
+                "trunk / win-vs2019-cpu-py3 / test (default, 2, 3, windows.4xlarge.nonephemeral)"
             ].classification
             == "BROKEN_TRUNK"
         )
         self.assertTrue(
             checks[
-                "pull / linux-focal-py3.7-gcc7 / test (distributed, 1, 2, linux.2xlarge)"
+                "pull / linux-jammy-py3.8-gcc11 / test (distributed, 1, 2, linux.2xlarge)"
+            ].classification
+            == "BROKEN_TRUNK"
+        )
+        self.assertTrue(
+            checks[
+                "pull / linux-focal-cuda11.8-py3.10-gcc9 / test (distributed, 1, 3, linux.8xlarge.nvidia.gpu)"
             ].classification
             == "FLAKY"
         )
+
+        # Set the threshold larger or equal to the number of ok failures
         pending, failed, ignorable = categorize_checks(
-            checks, list(checks.keys()), ok_failed_checks_threshold=2
+            checks, list(checks.keys()), ok_failed_checks_threshold=6
         )
         self.assertTrue(len(pending) == 0)
         self.assertTrue(len(failed) == 0)
-        self.assertTrue(len(ignorable["FLAKY"]) == 1)
-        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 1)
+        self.assertTrue(len(ignorable["FLAKY"]) == 2)
+        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 4)
 
         # Not set any threshold, defaults to -1 to ignore all flaky and broken trunk failures
         pending, failed, ignorable = categorize_checks(checks, list(checks.keys()))
         self.assertTrue(len(pending) == 0)
         self.assertTrue(len(failed) == 0)
-        self.assertTrue(len(ignorable["FLAKY"]) == 1)
-        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 1)
+        self.assertTrue(len(ignorable["FLAKY"]) == 2)
+        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 4)
 
+        # Set the threshold lower than the number of ok failures
         pending, failed, ignorable = categorize_checks(
             checks, list(checks.keys()), ok_failed_checks_threshold=1
         )
         self.assertTrue(len(pending) == 0)
-        self.assertTrue(len(failed) == 2)
-        self.assertTrue(len(ignorable["FLAKY"]) == 1)
-        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 1)
+        self.assertTrue(len(failed) == 6)
+        self.assertTrue(len(ignorable["FLAKY"]) == 2)
+        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 4)
+
+        # Set the threshold to 0 like when ignore_flaky_failures is on
+        pending, failed, ignorable = categorize_checks(
+            checks, list(checks.keys()), ok_failed_checks_threshold=1
+        )
+        self.assertTrue(len(pending) == 0)
+        self.assertTrue(len(failed) == 6)
+        self.assertTrue(len(ignorable["FLAKY"]) == 2)
+        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 4)
 
     def test_get_classifications_similar_failures(self, *args: Any) -> None:
         pr = GitHubPR("pytorch", "pytorch", 109750)
@@ -737,7 +745,6 @@ def test_get_classifications_similar_failures(self, *args: Any) -> None:
             pr.last_commit()["oid"],
             pr.get_merge_base(),
             [],
-            [],
         )
         pending, failed, ignorable = categorize_checks(checks, list(checks.keys()))
         self.assertTrue(len(pending) == 0)
@@ -754,7 +761,6 @@ def test_get_classifications_unstable(self, *args: Any) -> None:
             pr.last_commit()["oid"],
             pr.get_merge_base(),
             [],
-            [],
         )
         workflow_name = "linux-bionic-cuda12.1-py3.10-gcc9-bazel-test"
         job_name = "build-and-test (default, 1, 1, linux.4xlarge.nvidia.gpu, unstable)"
@@ -812,7 +818,6 @@ def test_get_classifications_broken_trunk(self, *args: Any) -> None:
                     pr.last_commit()["oid"],
                     pr.get_merge_base(),
                     [],
-                    [],
                 )
 
                 pending, failed, _ = categorize_checks(checks, list(checks.keys()))
@@ -832,35 +837,28 @@ def test_ignore_current(self, *args: Any) -> None:
         # current checks takes place after other classifications: flaky, unstable,
         # or broken trunk. Only actual new failures should be kept in the list of
         # ignore current checks to use to record force merge with actual failures
-        flaky_rules = [
-            FlakyRule("distributed", ["##\\[error\\]The operation was canceled."])
-        ]
-        flaky = (
-            "pull / linux-focal-py3.7-gcc7 / test (distributed, 1, 2, linux.2xlarge)"
-        )
+        flaky = "pull / linux-focal-cuda11.8-py3.10-gcc9 / test (distributed, 1, 3, linux.8xlarge.nvidia.gpu)"
         broken_trunk = (
-            "pull / linux-bionic-py3_7-clang8-xla / test (xla, 1, 1, linux.4xlarge)"
+            "pull / linux-focal-py3.11-clang10 / test (dynamo, 1, 2, linux.2xlarge)"
         )
 
-        pr = GitHubPR("pytorch", "pytorch", 92863)
+        pr = GitHubPR("pytorch", "pytorch", 109584)
         checks = pr.get_checkrun_conclusions()
 
-        # No broken trunk or flaky rules, then all failures are ignored when ic is used
+        # No broken trunk or flaky as the merge base is not set, these failures are
+        # counted as ignore current when ic is used
         checks = get_classifications(
             pr.pr_num,
             pr.project,
             checks,
             pr.last_commit()["oid"],
             None,
-            [],
             [broken_trunk, flaky],
         )
         self.assertTrue(checks[flaky].classification == "IGNORE_CURRENT_CHECK")
         self.assertTrue(checks[broken_trunk].classification == "IGNORE_CURRENT_CHECK")
-        _, failed, ignorable = categorize_checks(
-            checks, list(checks.keys()), ok_failed_checks_threshold=2
-        )
-        self.assertTrue(len(failed) == 0)
+        _, failed, ignorable = categorize_checks(checks, list(checks.keys()))
+        self.assertTrue(len(failed) == 4)
         self.assertTrue(len(ignorable["IGNORE_CURRENT_CHECK"]) == 2)
         self.assertTrue(len(ignorable["FLAKY"]) == 0)
         self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 0)
@@ -874,46 +872,22 @@ def test_ignore_current(self, *args: Any) -> None:
             checks,
             pr.last_commit()["oid"],
             pr.get_merge_base(),
-            flaky_rules,
             [broken_trunk, flaky],
         )
         self.assertTrue(checks[flaky].classification == "FLAKY")
         self.assertTrue(checks[broken_trunk].classification == "BROKEN_TRUNK")
-        _, failed, ignorable = categorize_checks(
-            checks, list(checks.keys()), ok_failed_checks_threshold=2
-        )
+        _, failed, ignorable = categorize_checks(checks, list(checks.keys()))
         self.assertTrue(len(failed) == 0)
         self.assertTrue(len(ignorable["IGNORE_CURRENT_CHECK"]) == 0)
-        self.assertTrue(len(ignorable["FLAKY"]) == 1)
-        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 1)
-
-        # Broken trunk takes precedence over ignore current (no flaky rule is set here)
-        checks = get_classifications(
-            pr.pr_num,
-            pr.project,
-            checks,
-            pr.last_commit()["oid"],
-            pr.get_merge_base(),
-            [],
-            [broken_trunk, flaky],
-        )
-        self.assertTrue(checks[flaky].classification == "IGNORE_CURRENT_CHECK")
-        self.assertTrue(checks[broken_trunk].classification == "BROKEN_TRUNK")
-        _, failed, ignorable = categorize_checks(
-            checks, list(checks.keys()), ok_failed_checks_threshold=2
-        )
-        self.assertTrue(len(failed) == 0)
-        self.assertTrue(len(ignorable["IGNORE_CURRENT_CHECK"]) == 1)
-        self.assertTrue(len(ignorable["FLAKY"]) == 0)
-        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 1)
+        self.assertTrue(len(ignorable["FLAKY"]) == 2)
+        self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 4)
 
-    @mock.patch("trymerge.read_flaky_rules", side_effect=xla_is_flaky_rules)
     @mock.patch("trymerge.read_merge_rules", side_effect=xla_merge_rules)
     def test_dont_ignore_flaky_failures(self, *args: Any) -> None:
         """
         Regression test for https://github.com/pytorch/test-infra/issues/4126
         """
-        pr = GitHubPR("pytorch", "pytorch", 100369)
+        pr = GitHubPR("pytorch", "pytorch", 105312)
         repo = DummyGitRepo()
         # Check that failure is classified as flaky but still raises exception
         with warnings.catch_warnings(record=True) as w, self.assertRaises(RuntimeError):