Run CI unittests in parallel (#3445)

* enable parallel tests

* disable parallelism for GPU tests

* [test] limit maximum processes on linux

* [debug] limit max processes even further

* [test] use subprocesses over threads

* [test] limit intra-op threads

* only limit intra op threads for CPU tests

* [poc] use low timeout for showcasing

* [poc] fix syntax

* set timeout to 5 minutes

* fix timeout on windows
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

.circleci/unittest/linux/scripts/environment.yml

@@ -5,6 +5,8 @@ channels:
   - conda-forge
 dependencies:
   - pytest
+  - pytest-xdist
+  - pytest-timeout
   - pytest-cov
   - codecov
   - pip

.circleci/unittest/linux/scripts/run_test.sh

@@ -6,5 +6,20 @@ eval "$(./conda/bin/conda shell.bash hook)"
 conda activate ./env
 
 export PYTORCH_TEST_WITH_SLOW='1'
+if [ "${CU_VERSION:-}" == cpu ] ; then
+    NUMPROCESSES="auto"
+    export OMP_NUM_THREADS="1"
+else
+    NUMPROCESSES="1"
+fi
+
 python -m torch.utils.collect_env
-pytest --cov=torchvision --junitxml=test-results/junit.xml -v --durations 20 test --ignore=test/test_datasets_download.py
+pytest \
+    --numprocesses=$NUMPROCESSES \
+    --timeout=300 \
+    --cov=torchvision \
+    --junitxml=test-results/junit.xml \
+    --verbose \
+    --durations 20 \
+    --ignore=test/test_datasets_download.py \
+    test

.circleci/unittest/windows/scripts/environment.yml

@@ -5,6 +5,8 @@ channels:
   - conda-forge
 dependencies:
   - pytest
+  - pytest-xdist
+  - pytest-timeout
   - pytest-cov
   - codecov
   - pip

.circleci/unittest/windows/scripts/run_test.sh

@@ -6,5 +6,20 @@ eval "$(./conda/Scripts/conda.exe 'shell.bash' 'hook')"
 conda activate ./env
 
 export PYTORCH_TEST_WITH_SLOW='1'
+if [ "${CU_VERSION:-}" == cpu ] ; then
+    NUMPROCESSES="auto"
+    export OMP_NUM_THREADS="1"
+else
+    NUMPROCESSES="1"
+fi
+
 python -m torch.utils.collect_env
-pytest --cov=torchvision --junitxml=test-results/junit.xml -v --durations 20 test --ignore=test/test_datasets_download.py
+pytest \
+    --numprocesses=$NUMPROCESSES \
+    --timeout=300 \
+    --cov=torchvision \
+    --junitxml=test-results/junit.xml \
+    --verbose \
+    --durations 20 \
+    --ignore=test/test_datasets_download.py \
+    test