improves nvfuser import check

.github/workflows/integration.yml

@@ -6,10 +6,10 @@ on:
     type: [integration-test-command]
 
 jobs:
-  integration-py3:
+  integration-auto3dseg:
     container:
       image: nvcr.io/nvidia/pytorch:22.04-py3  # CUDA 11.6 py38
-      options: --gpus "device=0" --ipc host  # shm-size 4g works fine
+      options: --gpus "device=1" --ipc host  # shm-size 4g works fine
     runs-on: [self-hosted, linux, x64, command]
     steps:
     # checkout the pull request branch
@@ -20,8 +20,7 @@ jobs:
         ref: ${{ github.event.client_payload.pull_request.head.ref }}
     - name: cache weekly timestamp
       id: pip-cache
-      run: |
-        echo "datew=$(date '+%Y-%V')" >> $GITHUB_OUTPUT
+      run: echo "datew=$(date '+%Y-%V')" >> $GITHUB_OUTPUT
     - name: cache for pip
       uses: actions/cache@v3
       id: cache
@@ -32,20 +31,20 @@ jobs:
         key: docker-py3-pip-${{ steps.pip-cache.outputs.datew }}
     - name: Install the dependencies
       run: |
-        which python
+        pwd && git log -1 && which python
         python -m pip install --upgrade pip wheel
         pip uninstall -y monai
         pip uninstall -y monai
         pip uninstall -y monai-weekly
         pip uninstall -y monai-weekly
-        python -m pip install --upgrade torch torchvision torchaudio
+        python -m pip install --upgrade torch torchvision torchaudio torchtext
         python -m pip install -r requirements-dev.txt
         rm -rf /github/home/.cache/torch/hub/mmars/
     - name: Clean directory
       run: |
         python -m pip list
         git config --global --add safe.directory /__w/MONAI/MONAI
-        git clean -ffdx
+        git clean -ffdx && git reset --hard HEAD
         nvidia-smi
         export CUDA_VISIBLE_DEVICES=$(python -m tests.utils -c 1 | tail -n 1)
         echo $CUDA_VISIBLE_DEVICES
@@ -57,19 +56,79 @@ jobs:
       env:
         BUILD_MONAI: 0
       run: |
-        BUILD_MONAI=0 ./runtests.sh --build
+        pwd && git log -1 && which python
+        ./runtests.sh -b
         python -m tests.test_auto3dseg_ensemble
         python -m tests.test_auto3dseg_hpo
         python -m tests.test_integration_autorunner
         python -m tests.test_integration_gpu_customization
+    - name: Integration tests
+      shell: bash
+      env:
+        BUILD_MONAI: 1
+      run: ./runtests.sh --build --net
+
+    - name: Add reaction
+      uses: peter-evans/create-or-update-comment@v2
+      if: github.event.pull_request.number != ''
+      with:
+        token: ${{ secrets.PR_MAINTAIN }}
+        repository: ${{ github.event.client_payload.github.payload.repository.full_name }}
+        comment-id: ${{ github.event.client_payload.github.payload.comment.id }}
+        reaction-type: rocket
+
+
+  integration-unit:
+    container:
+      image: nvcr.io/nvidia/pytorch:22.04-py3  # CUDA 11.6 py38
+      options: --gpus "device=2" --ipc host  # shm-size 4g works fine
+    runs-on: [self-hosted, linux, x64, command1]
+    steps:
+    # checkout the pull request branch
+    - uses: actions/checkout@v3
+      with:
+        token: ${{ secrets.PR_MAINTAIN }}
+        repository: ${{ github.event.client_payload.pull_request.head.repo.full_name }}
+        ref: ${{ github.event.client_payload.pull_request.head.ref }}
+    - name: cache weekly timestamp
+      id: pip-cache
+      run: echo "datew=$(date '+%Y-%V')" >> $GITHUB_OUTPUT
+    - name: cache for pip
+      uses: actions/cache@v3
+      id: cache
+      with:
+        path: |
+          ~/.cache/pip
+          ~/.cache/torch
+        key: docker-py3-pip-${{ steps.pip-cache.outputs.datew }}
+    - name: Install the dependencies
+      run: |
+        pwd && git log -1 && which python
+        python -m pip install --upgrade pip wheel
+        pip uninstall -y monai
+        pip uninstall -y monai
+        pip uninstall -y monai-weekly
+        pip uninstall -y monai-weekly
+        python -m pip install --upgrade torch torchvision torchaudio torchtext
+        python -m pip install -r requirements-dev.txt
+        rm -rf /github/home/.cache/torch/hub/mmars/
+    - name: Clean directory
+      run: |
+        python -m pip list
+        git config --global --add safe.directory /__w/MONAI/MONAI
+        git clean -ffdx && git reset --hard HEAD
+        nvidia-smi
+        export CUDA_VISIBLE_DEVICES=$(python -m tests.utils -c 1 | tail -n 1)
+        echo $CUDA_VISIBLE_DEVICES
+        python -c "import torch; print(torch.__version__); print('{} of GPUs available'.format(torch.cuda.device_count()))"
+        python -c 'import torch; print(torch.rand(5,3, device=torch.device("cuda:0")))'
 
     - name: Auto3dseg latest algo
       shell: bash
       env:
         BUILD_MONAI: 0
       run: |
-        # test latest template
-        echo "test latest algo"
+        pwd
         cd ../
         rm -rf research-contributions
         rm -rf algorithm_templates
@@ -82,28 +141,18 @@ jobs:
         export OMP_NUM_THREADS=4
         export MKL_NUM_THREADS=4
         export MONAI_TESTING_ALGO_TEMPLATE=algorithm_templates
+        pwd && git log -1 && which python
+        ./runtests.sh -b
         python -m tests.test_auto3dseg_ensemble
         python -m tests.test_auto3dseg_hpo
         python -m tests.test_integration_autorunner
         python -m tests.test_integration_gpu_customization
-
-    - name: Integration tests
-      shell: bash
-      env:
-        BUILD_MONAI: 1
-      run: |
-        pwd
-        ls -ll
-        ./runtests.sh --build --net
-
     - name: Unit tests
       shell: bash
       env:
         BUILD_MONAI: 1
-      run: |
-        pwd
-        ls -ll
-        ./runtests.sh --unittests
+        QUICKTEST: True
+      run: ./runtests.sh --build --unittests
 
     - name: Add reaction
       uses: peter-evans/create-or-update-comment@v2
@@ -112,4 +161,4 @@ jobs:
         token: ${{ secrets.PR_MAINTAIN }}
         repository: ${{ github.event.client_payload.github.payload.repository.full_name }}
         comment-id: ${{ github.event.client_payload.github.payload.comment.id }}
-        reaction-type: rocket
+        reaction-type: +1

monai/apps/auto3dseg/bundle_gen.py

@@ -36,7 +36,7 @@
 from monai.utils.enums import AlgoKeys
 
 logger = get_logger(module_name=__name__)
-ALGO_HASH = os.environ.get("MONAI_ALGO_HASH", "5b4e379")
+ALGO_HASH = os.environ.get("MONAI_ALGO_HASH", "e619d26")
 
 __all__ = ["BundleAlgo", "BundleGen"]
 

monai/apps/detection/transforms/box_ops.py

@@ -21,12 +21,10 @@
 from monai.data.box_utils import COMPUTE_DTYPE, TO_REMOVE, get_spatial_dims
 from monai.transforms import Resize
 from monai.transforms.utils import create_scale
-from monai.utils import look_up_option, optional_import
+from monai.utils import look_up_option
 from monai.utils.misc import ensure_tuple, ensure_tuple_rep
 from monai.utils.type_conversion import convert_data_type, convert_to_dst_type
 
-scipy, _ = optional_import("scipy")
-
 
 def _apply_affine_to_points(points: torch.Tensor, affine: torch.Tensor, include_shift: bool = True) -> torch.Tensor:
     """

monai/apps/pathology/transforms/post/dictionary.py

@@ -35,7 +35,6 @@
 from monai.utils.enums import HoVerNetBranch
 
 find_contours, _ = optional_import("skimage.measure", name="find_contours")
-moments, _ = optional_import("skimage.measure", name="moments")
 
 __all__ = [
     "WatershedD",

monai/inferers/utils.py

@@ -280,7 +280,7 @@ def sliding_window_inference(
             else:
                 sw_device_buffer[ss] *= w_t
                 sw_device_buffer[ss] = sw_device_buffer[ss].to(device)
-                _compute_coords(sw_batch_size, unravel_slice, z_scale, output_image_list[ss], sw_device_buffer[ss])
+                _compute_coords(unravel_slice, z_scale, output_image_list[ss], sw_device_buffer[ss])
         sw_device_buffer = []
         if buffered:
             b_s += 1
@@ -342,7 +342,7 @@ def _create_buffered_slices(slices, batch_size, sw_batch_size, buffer_dim, buffe
     return slices, n_per_batch, b_slices, windows_range
 
 
-def _compute_coords(sw, coords, z_scale, out, patch):
+def _compute_coords(coords, z_scale, out, patch):
     """sliding window batch spatial scaling indexing for multi-resolution outputs."""
     for original_idx, p in zip(coords, patch):
         idx_zm = list(original_idx)  # 4D for 2D image, 5D for 3D image

monai/networks/layers/factories.py

@@ -67,10 +67,9 @@ def use_factory(fact_args):
 
 import torch.nn as nn
 
+from monai.networks.utils import has_nvfuser_instance_norm
 from monai.utils import look_up_option, optional_import
 
-InstanceNorm3dNVFuser, has_nvfuser = optional_import("apex.normalization", name="InstanceNorm3dNVFuser")
-
 __all__ = ["LayerFactory", "Dropout", "Norm", "Act", "Conv", "Pool", "Pad", "split_args"]
 
 
@@ -267,12 +266,12 @@ def instance_nvfuser_factory(dim):
         warnings.warn(f"`InstanceNorm3dNVFuser` only supports 3d cases, use {types[dim - 1]} instead.")
         return types[dim - 1]
 
-    if not has_nvfuser:
+    if not has_nvfuser_instance_norm():
         warnings.warn(
             "`apex.normalization.InstanceNorm3dNVFuser` is not installed properly, use nn.InstanceNorm3d instead."
         )
         return nn.InstanceNorm3d
-    return InstanceNorm3dNVFuser
+    return optional_import("apex.normalization", name="InstanceNorm3dNVFuser")[0]
 
 
 Act.add_factory_callable("elu", lambda: nn.modules.ELU)

monai/networks/utils.py

@@ -27,10 +27,10 @@
 import torch
 import torch.nn as nn
 
-from monai.apps.utils import get_logger, optional_import
+from monai.apps.utils import get_logger
 from monai.config import PathLike
 from monai.utils.misc import ensure_tuple, save_obj, set_determinism
-from monai.utils.module import look_up_option, pytorch_after
+from monai.utils.module import look_up_option, optional_import, pytorch_after
 from monai.utils.type_conversion import convert_to_dst_type, convert_to_tensor
 
 onnx, _ = optional_import("onnx")
@@ -59,10 +59,34 @@
     "replace_modules_temp",
     "look_up_named_module",
     "set_named_module",
+    "has_nvfuser_instance_norm",
 ]
 
 logger = get_logger(module_name=__name__)
 
+global _has_nvfuser
+_has_nvfuser = None
+
+
+def has_nvfuser_instance_norm():
+    """whether the current environment has InstanceNorm3dNVFuser
+    https://github.com/NVIDIA/apex/blob/23.05-devel/apex/normalization/instance_norm.py#L15-L16
+    """
+    global _has_nvfuser
+    if _has_nvfuser is not None:
+        return _has_nvfuser
+
+    _, _has_nvfuser = optional_import("apex.normalization", name="InstanceNorm3dNVFuser")
+    if not _has_nvfuser:
+        return False
+    try:
+        import importlib
+
+        importlib.import_module("instance_norm_nvfuser_cuda")
+    except ImportError:
+        _has_nvfuser = False
+    return _has_nvfuser
+
 
 def look_up_named_module(name: str, mod, print_all_options=False):
     """

tests/test_dynunet.py

@@ -19,11 +19,8 @@
 
 from monai.networks import eval_mode
 from monai.networks.nets import DynUNet
-from monai.utils import optional_import
 from tests.utils import assert_allclose, skip_if_no_cuda, skip_if_windows, test_script_save
 
-InstanceNorm3dNVFuser, _ = optional_import("apex.normalization", name="InstanceNorm3dNVFuser")
-
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
 strides: Sequence[Sequence[int] | int]
@@ -128,15 +125,6 @@ def test_script(self):
 @skip_if_no_cuda
 @skip_if_windows
 class TestDynUNetWithInstanceNorm3dNVFuser(unittest.TestCase):
-    def setUp(self):
-        try:
-            layer = InstanceNorm3dNVFuser(num_features=1, affine=False).to("cuda:0")
-            inp = torch.randn([1, 1, 1, 1, 1]).to("cuda:0")
-            out = layer(inp)
-            del inp, out, layer
-        except Exception:
-            self.skipTest("NVFuser not available")
-
     @parameterized.expand([TEST_CASE_DYNUNET_3D[0]])
     def test_consistency(self, input_param, input_shape, _):
         for eps in [1e-4, 1e-5]:

tests/test_lmdbdataset.py

@@ -12,7 +12,6 @@
 from __future__ import annotations
 
 import os
-import shutil
 import tempfile
 import unittest
 
@@ -22,7 +21,7 @@
 
 from monai.data import LMDBDataset, json_hashing
 from monai.transforms import Compose, LoadImaged, SimulateDelayd, Transform
-from tests.utils import DistCall, DistTestCase, skip_if_windows
+from tests.utils import skip_if_windows
 
 TEST_CASE_1 = [
     Compose(
@@ -204,44 +203,5 @@ def test_shape(self, transform, expected_shape, kwargs=None):
                     dataset_postcached.set_data(data=test_data_new)  # filename list updated, files do not exist
 
 
-@skip_if_windows
-class TestMPLMDBDataset(DistTestCase):
-    def setUp(self):
-        self.tempdir = tempfile.mkdtemp()
-
-    def tearDown(self):
-        shutil.rmtree(self.tempdir)
-
-    @DistCall(nnodes=1, nproc_per_node=1)
-    def test_mp_cache(self):
-        items = [[list(range(i))] for i in range(5)]
-
-        ds = LMDBDataset(items, transform=_InplaceXform(), cache_dir=self.tempdir, lmdb_kwargs={"map_size": 10 * 1024})
-        self.assertEqual(items, [[[]], [[0]], [[0, 1]], [[0, 1, 2]], [[0, 1, 2, 3]]])
-        ds1 = LMDBDataset(items, transform=_InplaceXform(), cache_dir=self.tempdir, lmdb_kwargs={"map_size": 10 * 1024})
-        self.assertEqual(list(ds1), list(ds))
-        self.assertEqual(items, [[[]], [[0]], [[0, 1]], [[0, 1, 2]], [[0, 1, 2, 3]]])
-
-        ds = LMDBDataset(
-            items,
-            transform=_InplaceXform(),
-            cache_dir=self.tempdir,
-            lmdb_kwargs={"map_size": 10 * 1024},
-            hash_func=json_hashing,
-        )
-        self.assertEqual(items, [[[]], [[0]], [[0, 1]], [[0, 1, 2]], [[0, 1, 2, 3]]])
-        ds1 = LMDBDataset(
-            items,
-            transform=_InplaceXform(),
-            cache_dir=self.tempdir,
-            lmdb_kwargs={"map_size": 10 * 1024},
-            hash_func=json_hashing,
-        )
-        self.assertEqual(list(ds1), list(ds))
-        self.assertEqual(items, [[[]], [[0]], [[0, 1]], [[0, 1, 2]], [[0, 1, 2, 3]]])
-
-        self.assertTrue(isinstance(ds1.info(), dict))
-
-
 if __name__ == "__main__":
     unittest.main()