Revamp testing.

.github/scripts/container_build.sh

@@ -36,9 +36,10 @@ if [ "$CONTAINER_RUNTIME" = "apptainer" ]; then
 
 elif [ "$CONTAINER_RUNTIME" = "docker" ]; then
     echo "[INFO] Checking Docker images..."
-    IMAGE_NAME="iris-dev-triton-aafec41"
+    # Use GitHub variable if set, otherwise default to iris-dev
+    IMAGE_NAME=${DOCKER_IMAGE_NAME:-"iris-dev"}
 
-    # Check if the triton image exists
+    # Check if the image exists
     if docker image inspect "$IMAGE_NAME" &> /dev/null; then
         echo "[INFO] Using existing Docker image: $IMAGE_NAME"
     else

.github/scripts/container_exec.sh

@@ -86,7 +86,9 @@ if [ "$CONTAINER_RUNTIME" = "apptainer" ]; then
     exit $EXIT_CODE
 
 elif [ "$CONTAINER_RUNTIME" = "docker" ]; then
-    IMAGE_NAME=${CUSTOM_IMAGE:-${DOCKER_IMAGE_NAME:-"iris-dev-triton-aafec41"}}
+    # Use custom image if provided, otherwise use GitHub variable or default
+    # GitHub Actions sets DOCKER_IMAGE_NAME, locally defaults to iris-dev
+    IMAGE_NAME=${CUSTOM_IMAGE:-${DOCKER_IMAGE_NAME:-"iris-dev"}}
 
     if ! docker image inspect "$IMAGE_NAME" &> /dev/null; then
         echo "[ERROR] Docker image $IMAGE_NAME not found" >&2

.github/scripts/container_run.sh

@@ -25,7 +25,8 @@ if [ "$CONTAINER_RUNTIME" = "apptainer" ]; then
     bash apptainer/run.sh "$@"
 elif [ "$CONTAINER_RUNTIME" = "docker" ]; then
     echo "[INFO] Running with Docker..."
-    IMAGE_NAME=${1:-"iris-dev-triton-aafec41"}
+    # Use GitHub variable if set, otherwise default to iris-dev
+    IMAGE_NAME=${1:-${DOCKER_IMAGE_NAME:-"iris-dev"}}
     WORKSPACE_DIR=${2:-"$(pwd)"}
     bash docker/run.sh "$IMAGE_NAME" "$WORKSPACE_DIR"
 fi

.github/scripts/run_tests.sh

@@ -3,16 +3,41 @@
 # Copyright (c) 2025 Advanced Micro Devices, Inc. All rights reserved.
 #
 # Run Iris tests in a container
-# Usage: run_tests.sh <num_ranks> [gpu_devices]
+# Usage: run_tests.sh <test_dir> <num_ranks> [gpu_devices] [install_method]
+#   test_dir: subdirectory under tests/ (e.g., examples, unittests, ccl)
+#   num_ranks: number of GPU ranks (1, 2, 4, or 8)
+#   gpu_devices: comma-separated GPU device IDs (optional)
+#   install_method: pip install method - "git", "editable", or "install" (optional, default: "editable")
+#     - "git": pip install git+https://github.com/${{ github.repository }}.git@${{ github.sha }}
+#     - "editable": pip install -e .
+#     - "install": pip install .
 
 set -e
 
-NUM_RANKS=$1
-GPU_DEVICES=${2:-""}
+TEST_DIR=$1
+NUM_RANKS=$2
+GPU_DEVICES=${3:-""}
+INSTALL_METHOD=${4:-"editable"}
 
-if [ -z "$NUM_RANKS" ]; then
-    echo "[ERROR] NUM_RANKS not provided"
-    echo "Usage: $0 <num_ranks> [gpu_devices]"
+if [ -z "$TEST_DIR" ] || [ -z "$NUM_RANKS" ]; then
+    echo "[ERROR] Missing required arguments"
+    echo "Usage: $0 <test_dir> <num_ranks> [gpu_devices] [install_method]"
+    echo "  test_dir: examples, unittests, or ccl"
+    echo "  num_ranks: 1, 2, 4, or 8"
+    echo "  install_method: git, editable, or install (default: editable)"
+    exit 1
+fi
+
+# Validate test directory
+if [ ! -d "tests/$TEST_DIR" ]; then
+    echo "[ERROR] Test directory tests/$TEST_DIR does not exist"
+    exit 1
+fi
+
+# Validate install method
+if [ "$INSTALL_METHOD" != "git" ] && [ "$INSTALL_METHOD" != "editable" ] && [ "$INSTALL_METHOD" != "install" ]; then
+    echo "[ERROR] Invalid install_method: $INSTALL_METHOD"
+    echo "  Must be one of: git, editable, install"
     exit 1
 fi
 
@@ -24,28 +49,31 @@ if [ -n "$GPU_DEVICES" ]; then
     GPU_ARG="--gpus $GPU_DEVICES"
 fi
 
+# Build install command based on method
+INSTALL_CMD=""
+if [ "$INSTALL_METHOD" = "git" ]; then
+    # For git install, we need the repository and SHA from environment or use defaults
+    REPO=${GITHUB_REPOSITORY:-"ROCm/iris"}
+    SHA=${GITHUB_SHA:-"HEAD"}
+    INSTALL_CMD="pip install git+https://github.com/${REPO}.git@${SHA}"
+elif [ "$INSTALL_METHOD" = "editable" ]; then
+    INSTALL_CMD="pip install -e ."
+elif [ "$INSTALL_METHOD" = "install" ]; then
+    INSTALL_CMD="pip install ."
+fi
+
 # Run tests in container
 "$SCRIPT_DIR/container_exec.sh" $GPU_ARG "
     set -e
-    pip install -e .
+    echo \"Installing iris using method: $INSTALL_METHOD\"
+    $INSTALL_CMD
 
-    # Run examples tests
-    for test_file in tests/examples/test_*.py; do
-        echo \"Testing: \$test_file with $NUM_RANKS ranks\"
-        python tests/run_tests_distributed.py --num_ranks $NUM_RANKS \"\$test_file\" -v --tb=short --durations=10
+    # Run tests in the specified directory
+    for test_file in tests/$TEST_DIR/test_*.py; do
+        if [ -f \"\$test_file\" ]; then
+            echo \"Testing: \$test_file with $NUM_RANKS ranks (install: $INSTALL_METHOD)\"
+            python tests/run_tests_distributed.py --num_ranks $NUM_RANKS \"\$test_file\" -v --tb=short --durations=10
+        fi
     done
-
-    # Run unit tests
-    for test_file in tests/unittests/test_*.py; do
-        echo \"Testing: \$test_file with $NUM_RANKS ranks\"
-        python tests/run_tests_distributed.py --num_ranks $NUM_RANKS \"\$test_file\" -v --tb=short --durations=10
-    done
-
-    # Run ccl tests
-    # DISABLED: CCL host-side APIs have issues for some data types/algorithms
-    # for test_file in tests/ccl/test_*.py; do
-    #     echo \"Testing: \$test_file with $NUM_RANKS ranks\"
-    #     python tests/run_tests_distributed.py --num_ranks $NUM_RANKS \"\$test_file\" -v --tb=short --durations=10
-    # done
 "
 

.github/workflows/iris-external-validation-test.yml

@@ -11,6 +11,9 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
+env:
+  DOCKER_IMAGE_NAME: ${{ vars.DOCKER_IMAGE_NAME || 'iris-dev-triton-aafec41' }}
+
 jobs:
   build-container-image:
     runs-on: [self-hosted, mi3008x]

.github/workflows/iris-performance-regression-test.yml

@@ -11,6 +11,9 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.ref }}
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
+env:
+  DOCKER_IMAGE_NAME: ${{ vars.DOCKER_IMAGE_NAME || 'iris-dev-triton-aafec41' }}
+
 jobs:
   build-container-image:
     runs-on: [self-hosted, mi3008x]