Update on "Add OpInfo test to check that floating point inputs in OpInfos have requires_grad set to True"

This test detected a number of sampling methods that were not generating
the samples as expected, e.g. `index_put`, `cosine_embedding`, `stft`, but
perhaps most notably the generator for `BinOps`.

It also detected that `reminder` and `fmod` did not have implemented the
backward formula for the second input. I added this in the previous PR.

[ghstack-poisoned]

Author: lezcano

.circleci/docker/common/install_conda.sh

@@ -120,7 +120,7 @@ if [ -n "$ANACONDA_PYTHON_VERSION" ]; then
   # Install numba only on python-3.8 or below
   # For numba issue see https://github.com/pytorch/pytorch/issues/51511
   if [[ $(python -c "import sys; print(int(sys.version_info < (3, 9)))") == "1" ]]; then
-    as_jenkins pip install --progress-bar off numba librosa>=0.6.2
+    as_jenkins pip install --progress-bar off numba==0.54.1 librosa>=0.6.2
   else
     as_jenkins pip install --progress-bar off numba==0.49.0 librosa>=0.6.2
   fi

.circleci/scripts/binary_populate_env.sh

@@ -6,14 +6,15 @@ tagged_version() {
   # Grabs version from either the env variable CIRCLE_TAG
   # or the pytorch git described version
   if [[ "$OSTYPE" == "msys" ]]; then
-    GIT_DESCRIBE="git --git-dir ${workdir}/p/.git describe"
+    GIT_DIR="${workdir}/p/.git"
   else
-    GIT_DESCRIBE="git --git-dir ${workdir}/pytorch/.git describe"
+    GIT_DIR="${workdir}/pytorch/.git"
   fi
+  GIT_DESCRIBE="git --git-dir ${GIT_DIR} describe --tags --match v[0-9]*.[0-9]*.[0-9]*"
   if [[ -n "${CIRCLE_TAG:-}" ]]; then
     echo "${CIRCLE_TAG}"
-  elif ${GIT_DESCRIBE} --exact --tags >/dev/null; then
-    ${GIT_DESCRIBE} --tags
+  elif ${GIT_DESCRIBE} --exact >/dev/null; then
+    ${GIT_DESCRIBE}
   else
     return 1
   fi

.github/generated-ciflow-ruleset.json

@@ -0,0 +1,20 @@
+[
+   {
+    "name": "ONNX exporter",
+    "patterns": ["torch/onnx/**", "torch/csrc/jit/passes/onnx/**", "torch/csrc/jit/passes/onnx.*", "test/onnx/**", "docs/source/onnx.rst"],
+    "approved_by": ["garymm"],
+    "mandatory_app_id": 12274
+   },
+   {
+    "name": "NVFuser",
+    "patterns": ["torch/csrc/jit/codegen/fuser/cuda/**", "torch/csrc/jit/codegen/cuda/**", "benchmarks/cpp/nvfuser/**"],
+    "approved_by": ["csarofeen", "ngimel"],
+    "mandatory_app_id": 12274
+   },
+   {
+    "name": "OSS CI",
+    "patterns": [".github/**", ".circleci/**", ".jenkins/**"],
+    "approved_by": ["seemethere", "malfet", "suo"],
+    "mandatory_app_id": 12274
+   }
+]

.github/merge_rules.json

@@ -16,7 +16,7 @@
 CUDA_ARCHES = ["10.2", "11.1", "11.3", "11.5"]
 
 
-ROCM_ARCHES = ["4.2", "4.3.1"]
+ROCM_ARCHES = ["4.3.1", "4.5.2"]
 
 
 def arch_type(arch_version: str) -> str:
@@ -65,6 +65,7 @@ def arch_type(arch_version: str) -> str:
     "3.7",
     "3.8",
     "3.9",
+    "3.10"
 ]
 
 

.github/scripts/generate_binary_build_matrix.py

@@ -29,9 +29,22 @@
 LINUX_CPU_TEST_RUNNER = "linux.2xlarge"
 # contains 1 gpu
 LINUX_CUDA_TEST_RUNNER = "linux.4xlarge.nvidia.gpu"
+# contains 4 gpus
+LINUX_ROCM_TEST_RUNNER = "linux.rocm.gpu"
 LINUX_RUNNERS = {
     LINUX_CPU_TEST_RUNNER,
     LINUX_CUDA_TEST_RUNNER,
+    LINUX_ROCM_TEST_RUNNER,
+}
+
+LINUX_DISTRIBUTED_GPU_RUNNERS = {
+    LINUX_CUDA_TEST_RUNNER : "linux.8xlarge.nvidia.gpu",
+    LINUX_ROCM_TEST_RUNNER : LINUX_ROCM_TEST_RUNNER,
+}
+
+LINUX_MULTIGPU_RUNNERS = {
+    LINUX_CUDA_TEST_RUNNER : "linux.16xlarge.nvidia.gpu",
+    LINUX_ROCM_TEST_RUNNER : LINUX_ROCM_TEST_RUNNER,
 }
 
 MACOS_TEST_RUNNER_10_15 = "macos-10.15"
@@ -46,6 +59,9 @@
     WINDOWS_CUDA_TEST_RUNNER,
     LINUX_CUDA_TEST_RUNNER,
 }
+ROCM_RUNNERS = {
+    LINUX_ROCM_TEST_RUNNER,
+}
 CPU_RUNNERS = {
     WINDOWS_CPU_TEST_RUNNER,
     LINUX_CPU_TEST_RUNNER,
@@ -55,6 +71,7 @@
 LABEL_CIFLOW_BAZEL = "ciflow/bazel"
 LABEL_CIFLOW_CPU = "ciflow/cpu"
 LABEL_CIFLOW_CUDA = "ciflow/cuda"
+LABEL_CIFLOW_ROCM = "ciflow/rocm"
 LABEL_CIFLOW_DOCS = "ciflow/docs"
 LABEL_CIFLOW_DEFAULT = "ciflow/default"
 LABEL_CIFLOW_LIBTORCH = "ciflow/libtorch"
@@ -76,56 +93,25 @@
 LABEL_CIFLOW_MACOS = "ciflow/macos"
 LABEL_CIFLOW_TRUNK = "ciflow/trunk"
 LABEL_CIFLOW_BINARIES = "ciflow/binaries"
-LABEL_CIFLOW_BINARIES_WHEEL = "ciflow/binaries/wheel"
-LABEL_CIFLOW_BINARIES_CONDA = "ciflow/binaries/conda"
-LABEL_CIFLOW_BINARIES_LIBTORCH = "ciflow/binaries/libtorch"
+LABEL_CIFLOW_BINARIES_WHEEL = "ciflow/binaries_wheel"
+LABEL_CIFLOW_BINARIES_CONDA = "ciflow/binaries_conda"
+LABEL_CIFLOW_BINARIES_LIBTORCH = "ciflow/binaries_libtorch"
 
 
 @dataclass
 class CIFlowConfig:
     # For use to enable workflows to run on pytorch/pytorch-canary
     run_on_canary: bool = False
     labels: Set[str] = field(default_factory=set)
-    trigger_action: str = 'unassigned'
-    trigger_actor: str = 'pytorchbot'
-    root_job_condition: str = ''
-    label_conditions: str = ''
     # Certain jobs might not want to be part of the ciflow/[all,trunk] workflow
     isolated_workflow: bool = False
 
-    def gen_root_job_condition(self) -> None:
-        # CIFlow conditions:
-        #  - Workflow should always run on push
-        #  - CIFLOW_DEFAULT workflows should run on PRs even if no `ciflow/` labels on PR
-        #  - Otherwise workflow should be scheduled on all qualifying events
-        label_conditions = [f"contains(github.event.pull_request.labels.*.name, '{label}')" for label in sorted(self.labels)]
-        self.label_conditions = ' || '.join(label_conditions)
-        repo_condition = "github.repository_owner == 'pytorch'" if self.run_on_canary else "github.repository == 'pytorch/pytorch'"
-        push_event = "github.event_name == 'push'"
-        scheduled_event = "github.event_name == 'schedule'"
-        pr_updated_event = f"github.event_name == 'pull_request' && github.event.action != '{self.trigger_action}'"
-        if LABEL_CIFLOW_DEFAULT in self.labels:
-            run_with_no_labels = f"({pr_updated_event}) && " \
-                                 f"!contains(join(github.event.pull_request.labels.*.name), '{LABEL_CIFLOW_PREFIX}')"
-        else:
-            run_with_no_labels = "false"
-        self.root_job_condition = f"${{{{ ({repo_condition}) && (\n" \
-                                  f"            ({push_event}) ||\n" \
-                                  f"            ({scheduled_event}) ||\n" \
-                                  f"            ({self.label_conditions}) ||\n" \
-                                  f"            ({run_with_no_labels}))\n"\
-                                  f"         }}}}"
-
-    def reset_root_job(self) -> None:
-        self.root_job_condition = ''
-
     def __post_init__(self) -> None:
         if not self.isolated_workflow:
             self.labels.add(LABEL_CIFLOW_ALL)
             if LABEL_CIFLOW_SCHEDULED not in self.labels:
                 self.labels.add(LABEL_CIFLOW_TRUNK)
         assert all(label.startswith(LABEL_CIFLOW_PREFIX) for label in self.labels)
-        self.gen_root_job_condition()
 
 
 @dataclass
@@ -164,6 +150,8 @@ class CIWorkflow:
 
     # Optional fields
     test_runner_type: str = ''
+    multigpu_runner_type: str = ''
+    distributed_gpu_runner_type: str = ''
     ciflow_config: CIFlowConfig = field(default_factory=CIFlowConfig)
     cuda_version: str = ''
     docker_image_base: str = ''
@@ -172,13 +160,15 @@ class CIWorkflow:
     build_generates_artifacts: bool = True
     build_with_debug: bool = False
     is_scheduled: str = ''
+    is_default: bool = False
     num_test_shards: int = 1
     only_run_smoke_tests_on_pull_request: bool = False
     num_test_shards_on_pull_request: int = -1
     distributed_test: bool = True
     fx2trt_test: bool = False
     timeout_after: int = 240
     xcode_version: str = ''
+    only_on_pr: bool = False
 
     # The following variables will be set as environment variables,
     # so it's easier for both shell and Python scripts to consume it if false is represented as the empty string.
@@ -205,6 +195,12 @@ def __post_init__(self) -> None:
         if self.fx2trt_test:
             self.enable_fx2trt_test = 1
 
+        self.multigpu_runner_type = LINUX_MULTIGPU_RUNNERS.get(self.test_runner_type, "linux.16xlarge.nvidia.gpu")
+        self.distributed_gpu_runner_type = LINUX_DISTRIBUTED_GPU_RUNNERS.get(self.test_runner_type, "linux.8xlarge.nvidia.gpu")
+
+        if LABEL_CIFLOW_DEFAULT in self.ciflow_config.labels:
+            self.is_default = True
+
         # If num_test_shards_on_pull_request is not user-defined, default to num_test_shards unless we are
         # only running smoke tests on the pull request.
         if self.num_test_shards_on_pull_request == -1:
@@ -223,7 +219,6 @@ def assert_valid(self) -> None:
             assert self.test_runner_type in WINDOWS_RUNNERS, err_message
 
         assert LABEL_CIFLOW_ALL in self.ciflow_config.labels
-        assert LABEL_CIFLOW_ALL in self.ciflow_config.label_conditions
         if self.arch == 'linux':
             assert LABEL_CIFLOW_LINUX in self.ciflow_config.labels
         if self.arch == 'windows':
@@ -235,6 +230,8 @@ def assert_valid(self) -> None:
             assert self.test_runner_type != ''
         if self.test_runner_type in CUDA_RUNNERS:
             assert LABEL_CIFLOW_CUDA in self.ciflow_config.labels
+        if self.test_runner_type in ROCM_RUNNERS:
+            assert LABEL_CIFLOW_ROCM in self.ciflow_config.labels
         if self.test_runner_type in CPU_RUNNERS and not self.exclude_test:
             assert LABEL_CIFLOW_CPU in self.ciflow_config.labels
         if self.is_scheduled:
@@ -576,6 +573,17 @@ def generate_workflow_file(self, workflow_template: jinja2.Template) -> None:
             labels=set([LABEL_CIFLOW_DEFAULT, LABEL_CIFLOW_LINUX, LABEL_CIFLOW_CPU]),
         ),
     ),
+    CIWorkflow(
+        arch="linux",
+        build_environment="linux-bionic-rocm4.5-py3.7",
+        docker_image_base=f"{DOCKER_REGISTRY}/pytorch/pytorch-linux-bionic-rocm4.5-py3.7",
+        test_runner_type=LINUX_ROCM_TEST_RUNNER,
+        num_test_shards=2,
+        only_on_pr=True,
+        ciflow_config=CIFlowConfig(
+            labels=set([LABEL_CIFLOW_LINUX, LABEL_CIFLOW_ROCM]),
+        ),
+    ),
     CIWorkflow(
         arch="linux",
         build_environment="libtorch-linux-xenial-cuda11.3-py3.7-gcc7",
@@ -619,8 +627,10 @@ def generate_workflow_file(self, workflow_template: jinja2.Template) -> None:
         num_test_shards=2,
         distributed_test=False,
         enable_noarch_test=1,
+        # Disabled, see https://github.com/pytorch/pytorch/issues/71732
+        # enable_xla_test=1,
         ciflow_config=CIFlowConfig(
-            labels={LABEL_CIFLOW_DEFAULT, LABEL_CIFLOW_LINUX, LABEL_CIFLOW_CPU, LABEL_CIFLOW_NOARCH},
+            labels={LABEL_CIFLOW_DEFAULT, LABEL_CIFLOW_LINUX, LABEL_CIFLOW_CPU, LABEL_CIFLOW_XLA, LABEL_CIFLOW_NOARCH},
         ),
     ),
     CIWorkflow(
@@ -824,8 +834,8 @@ def generate_workflow_file(self, workflow_template: jinja2.Template) -> None:
     DockerWorkflow(
         build_environment="docker-builds",
         docker_images=sorted(DOCKER_IMAGES),
-        # Run every Wednesday to ensure they can build
-        is_scheduled="1 * * * 3",
+        # Run every Wednesday at 3:01am to ensure they can build
+        is_scheduled="1 3 * * 3",
     ),
 ]
 
@@ -835,7 +845,7 @@ def generate_workflow_file(self, workflow_template: jinja2.Template) -> None:
         package_type="manywheel",
         build_configs=generate_binary_build_matrix.generate_wheels_matrix(),
         ciflow_config=CIFlowConfig(
-            labels={LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_WHEEL},
+            labels={LABEL_CIFLOW_DEFAULT, LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_WHEEL},
             isolated_workflow=True,
         ),
     ),
@@ -844,7 +854,7 @@ def generate_workflow_file(self, workflow_template: jinja2.Template) -> None:
         package_type="conda",
         build_configs=generate_binary_build_matrix.generate_conda_matrix(),
         ciflow_config=CIFlowConfig(
-            labels={LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_CONDA},
+            labels={LABEL_CIFLOW_DEFAULT, LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_CONDA},
             isolated_workflow=True,
         ),
     ),
@@ -856,7 +866,7 @@ def generate_workflow_file(self, workflow_template: jinja2.Template) -> None:
             generate_binary_build_matrix.CXX11_ABI
         ),
         ciflow_config=CIFlowConfig(
-            labels={LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_LIBTORCH},
+            labels={LABEL_CIFLOW_DEFAULT, LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_LIBTORCH},
             isolated_workflow=True,
         ),
     ),
@@ -868,7 +878,7 @@ def generate_workflow_file(self, workflow_template: jinja2.Template) -> None:
             generate_binary_build_matrix.PRE_CXX11_ABI
         ),
         ciflow_config=CIFlowConfig(
-            labels={LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_LIBTORCH},
+            labels={LABEL_CIFLOW_DEFAULT, LABEL_CIFLOW_BINARIES, LABEL_CIFLOW_BINARIES_LIBTORCH},
             isolated_workflow=True,
         ),
     ),