Add environment version reporting to CI workflows (#538)

Co-authored-by: Claude Opus 4 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: 3 people

.github/scripts/collect_versions.py

@@ -0,0 +1,77 @@
+#!/usr/bin/env python3
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2025 Advanced Micro Devices, Inc. All rights reserved.
+"""Collect environment versions and emit JSON to stdout.
+
+Runs inside the CI container. The calling shell script handles
+$GITHUB_STEP_SUMMARY and stdout formatting on the host side.
+"""
+
+import json
+import os
+import subprocess
+import sys
+
+
+def collect():
+    info = {}
+
+    # Driver + ROCm via amd-smi version --json
+    try:
+        raw = subprocess.check_output(
+            ["amd-smi", "version", "--json"],
+            stderr=subprocess.DEVNULL,
+            timeout=10,
+        )
+        smi = json.loads(raw)
+        if isinstance(smi, list):
+            smi = smi[0]
+        info["driver"] = smi.get("amdgpu_version", "")
+        info["rocm"] = smi.get("rocm_version", "")
+    except Exception:
+        pass
+
+    # ROCm fallback from filesystem
+    if not info.get("rocm"):
+        for p in ["/opt/rocm/.info/version", "/opt/rocm/lib/rocm_version"]:
+            if os.path.isfile(p):
+                info["rocm"] = open(p).read().strip()
+                break
+
+    # Python
+    v = sys.version_info
+    info["python"] = f"{v.major}.{v.minor}.{v.micro}"
+
+    # PyTorch, HIP, GPU
+    try:
+        import torch
+
+        info["torch"] = torch.__version__
+        info["hip"] = getattr(torch.version, "hip", None) or getattr(torch.version, "cuda", None) or ""
+        info["gpu_count"] = torch.cuda.device_count()
+        if torch.cuda.device_count() > 0:
+            props = torch.cuda.get_device_properties(0)
+            info["gpu_name"] = props.name
+            info["gpu_arch"] = getattr(props, "gcnArchName", "N/A")
+    except Exception:
+        pass
+
+    # Triton
+    try:
+        import triton
+
+        info["triton"] = triton.__version__
+    except Exception:
+        pass
+
+    # Kernel
+    try:
+        info["kernel"] = os.uname().release
+    except Exception:
+        pass
+
+    return info
+
+
+if __name__ == "__main__":
+    print(json.dumps(collect()))

.github/scripts/report_versions.sh

@@ -0,0 +1,60 @@
+#!/bin/bash
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2025 Advanced Micro Devices, Inc. All rights reserved.
+#
+# Collect environment versions from inside the container and write
+# a GitHub Actions job summary table. Version collection logic lives
+# in collect_versions.py; this script handles container exec + summary.
+
+set -e
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+GPU_ARG=""
+if [ -n "$GPU_DEVICES" ]; then
+    GPU_ARG="--gpus $GPU_DEVICES"
+fi
+
+# Run collect_versions.py inside the container, extract JSON line
+# shellcheck disable=SC2086
+RAW=$("$SCRIPT_DIR/container_exec.sh" $GPU_ARG "python3 /iris_workspace/.github/scripts/collect_versions.py")
+VERSION_JSON=$(echo "$RAW" | grep '^{' | tail -1)
+
+if [ -z "$VERSION_JSON" ]; then
+    echo "[WARN] Failed to collect version info"
+    exit 0
+fi
+
+# Format and print with Python on the host (available on all runners)
+python3 -c "
+import json, sys, os
+
+d = json.loads(sys.argv[1])
+
+def g(k):
+    return str(d.get(k, 'unknown'))
+
+gpu = f\"{g('gpu_name')} ({g('gpu_arch')}) x {g('gpu_count')}\"
+
+print('============================================')
+print('  Iris CI - Environment Report')
+print('============================================')
+for label, key in [('Driver', 'driver'), ('ROCm', 'rocm'), ('Python', 'python'),
+                    ('PyTorch', 'torch'), ('HIP', 'hip'), ('Triton', 'triton')]:
+    print(f'  {label:9s} {g(key)}')
+print(f'  {\"GPU\":9s} {gpu}')
+print(f'  {\"Kernel\":9s} {g(\"kernel\")}')
+print('============================================')
+
+summary_path = os.environ.get('GITHUB_STEP_SUMMARY')
+if summary_path:
+    with open(summary_path, 'a') as f:
+        f.write('### Environment\n\n')
+        f.write('| Component | Version |\n')
+        f.write('|-----------|--------|\n')
+        for label, key in [('Driver', 'driver'), ('ROCm', 'rocm'), ('Python', 'python'),
+                            ('PyTorch', 'torch'), ('HIP', 'hip'), ('Triton', 'triton')]:
+            f.write(f'| {label} | {g(key)} |\n')
+        f.write(f'| GPU | {gpu} |\n')
+        f.write(f'| Kernel | {g(\"kernel\")} |\n\n')
+" "$VERSION_JSON"

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

@@ -43,6 +43,10 @@ jobs:
         run: |
           bash .github/scripts/acquire_gpus.sh 2
 
+      - name: Report environment versions
+        run: |
+          bash .github/scripts/report_versions.sh
+
       - name: Run External Validation Test
         run: |
           set -e
@@ -92,6 +96,10 @@ jobs:
         run: |
           bash .github/scripts/acquire_gpus.sh 2
 
+      - name: Report environment versions
+        run: |
+          bash .github/scripts/report_versions.sh
+
       - name: Run External Gluon Validation Test
         run: |
           set -e

.github/workflows/iris-nightly-triton-test.yml

@@ -105,6 +105,10 @@ jobs:
         run: |
           bash .github/scripts/acquire_gpus.sh "${{ matrix.num_ranks }}"
 
+      - name: Report environment versions
+        run: |
+          bash .github/scripts/report_versions.sh
+
       - name: Run ${{ matrix.test_dir }} tests with ${{ matrix.num_ranks }} ranks (nightly Triton)
         run: |
           set -e

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

@@ -67,6 +67,10 @@ jobs:
         run: |
           bash .github/scripts/acquire_gpus.sh 8
 
+      - name: Report environment versions
+        run: |
+          bash .github/scripts/report_versions.sh
+
       - name: Run ${{ matrix.example_name }} Benchmark (8 ranks)
         run: |
           set -e

.github/workflows/iris-tests.yml

@@ -88,6 +88,10 @@ jobs:
         run: |
           bash .github/scripts/acquire_gpus.sh "${{ matrix.num_ranks }}"
 
+      - name: Report environment versions
+        run: |
+          bash .github/scripts/report_versions.sh
+
       - name: Run ${{ matrix.test_dir }} tests with ${{ matrix.num_ranks }} ranks (git install)
         env:
           GITHUB_REPOSITORY: ${{ github.repository }}