[rocjitsu][ci] Update benchmark workflow for the sim perf dashboard

.github/workflows/rocjitsu-website-test.yml

@@ -6,13 +6,15 @@ on:
       - "emulation/rocjitsu/website/dashboard/**"
       - ".github/workflows/rocjitsu-website-test.yml"
       - ".github/workflows/rocjitsu-publish-website.yml"
+      - ".github/workflows/rocjitsu-benchmarks.yml"
   push:
     branches:
       - develop
     paths:
       - "emulation/rocjitsu/website/dashboard/**"
       - ".github/workflows/rocjitsu-website-test.yml"
       - ".github/workflows/rocjitsu-publish-website.yml"
+      - ".github/workflows/rocjitsu-benchmarks.yml"
   workflow_dispatch:
 
 permissions:
@@ -23,6 +25,35 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
+  dashboard-data:
+    name: Verify dashboard data tools
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    defaults:
+      run:
+        working-directory: emulation/rocjitsu/website/dashboard
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
+        with:
+          python-version: "3.12"
+          cache: pip
+          cache-dependency-path: emulation/rocjitsu/website/dashboard/scripts/dashboard-data/requirements-test.txt
+
+      - name: Install test tools
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y bash git jq
+
+      - name: Install Python test dependencies
+        run: python -m pip install -r scripts/dashboard-data/requirements-test.txt
+
+      - name: Test dashboard data tools
+        run: python -m pytest scripts/dashboard-data/tests
+
   verify:
     name: Verify website
     runs-on: ubuntu-latest

emulation/rocjitsu/website/dashboard/docs/build-and-test.md

@@ -71,6 +71,19 @@ If Chromium's system libraries are already installed, `npx playwright install ch
 installs just the browser without changing OS packages. Rerun browser installation
 after updating Playwright if its required browser version changes.
 
+## Dashboard-data Python tests
+
+Use a Python 3.12 environment with Bash, Git, jq, and coreutils (including
+`realpath`) on `PATH`. From `emulation/rocjitsu/website/dashboard`:
+
+```bash
+python -m pip install -r scripts/dashboard-data/requirements-test.txt
+python -m pytest scripts/dashboard-data/tests
+```
+
+The `dashboard-data` CI job uses the same requirements file. These tests are
+separate from `npm run verify` and do not require the npm dependencies.
+
 ## Run locally
 
 ```bash

emulation/rocjitsu/website/dashboard/docs/website-data-contract.md

@@ -284,6 +284,14 @@ Environments may change between independent historical runs. Runs sharing a
 `comparisonId` must have the same normalized environment because plugin overhead is
 valid only when the compared executions use the same environment.
 
+Benchmark workflow runs include `workflow.commit_sha` (label **Workflow commit**)
+and `tensile.commit_sha` (label **Tensile commit**) for the workflow-tools revision
+and pinned Tensile source revision. These are separate from the measured Rocjitsu
+`source.commit`. Their key/value pairs participate in the existing environment
+identity: different workflow or Tensile revisions intentionally distinguish
+comparison environments. Older runs may omit these details; do not rewrite
+historical run files to add them.
+
 ### `targets` and results
 
 `targets` must contain exactly one group for every target defined by the referenced catalog. Each group contains:
@@ -299,6 +307,12 @@ Every result contains:
 - `durationSeconds`: positive numeric seconds for a completed result; `null` for failed or timed-out results.
 - `status`: required; `completed`, `failed`, or `timeout`.
 - `error`: failure description or `null`.
+- `timing_results_s`: optional array of individual timing samples in seconds, in
+  input measurement order (not sorted). The benchmark converter preserves these
+  samples and uses their median for a completed result's `durationSeconds`. Failed
+  and timed-out results keep a `null` duration, even if samples are present. The
+  current validator accepts this extra field, but normalization omits it from the
+  UI model; the dashboard uses `durationSeconds`, not the sample vector.
 
 Failed and timed-out tests must remain in `results`. Omitting a catalog result invalidates the run instead of silently changing its denominator.
 

emulation/rocjitsu/website/dashboard/scripts/dashboard-data/prepare-benchmark-result.py

@@ -0,0 +1,251 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 Advanced Micro Devices, Inc.
+# SPDX-License-Identifier: MIT
+"""Convert raw benchmark JSON to one default-policy dashboard run JSON.
+
+Existing dashboard catalogs are referenced, not created or published here.
+"""
+
+import argparse
+from datetime import datetime
+import hashlib
+import json
+import math
+from pathlib import Path
+import statistics
+
+
+def require(condition, message):
+    if not condition:
+        raise ValueError(message)
+
+
+def encode_json_bytes(value):
+    return (json.dumps(value, indent=2, sort_keys=True, allow_nan=False) + '\n').encode(
+        'utf-8'
+    )
+
+
+def parse_timestamp(value):
+    return datetime.fromisoformat(value.replace('Z', '+00:00'))
+
+
+def build_published_run(raw, workflow):
+    github = {
+        'run_id': workflow['run_id'],
+        'run_attempt': workflow['run_attempt'],
+        'trigger': workflow['trigger'],
+    }
+    source = {
+        'commit': workflow['source_sha'],
+        'commit_subject': workflow['source_subject'],
+        'repository_url': 'https://github.com/' + workflow['repository'],
+    }
+    summary = raw['execution_summary']
+    require(summary['finished_at'] is not None, 'Cannot convert unfinished checkpoint')
+    require(
+        parse_timestamp(summary['finished_at'])
+        >= parse_timestamp(summary['timestamp']),
+        'Finish precedes start',
+    )
+    require(
+        type(summary['wall_time_s']) in (int, float)
+        and math.isfinite(summary['wall_time_s'])
+        and summary['wall_time_s'] >= 0,
+        'Invalid wall time',
+    )
+    committed = raw['provenance']['rocjitsu']['rocjitsu_commit_timestamp']
+    require(
+        parse_timestamp(committed) <= parse_timestamp(summary['timestamp']),
+        'Source commit is later than benchmark start',
+    )
+    samples = summary['samples']
+    require(
+        type(samples) is int and samples > 0 and samples % 2 == 1,
+        'Invalid sample count',
+    )
+    selected = [
+        result
+        for result in raw['benchmark_results']
+        if result.get('thread_policy') == 'default'
+    ]
+    identities = [(result['id'], result['target']) for result in selected]
+    require(
+        identities and len(identities) == len(set(identities)),
+        'Missing or duplicate default-policy results',
+    )
+    definitions, targets = {}, {}
+    for result in selected:
+        problem = dict(result['problem'])
+        require(
+            'operation' not in problem or problem['operation'] == result['operation'],
+            'Operation metadata collision',
+        )
+        problem['operation'] = result['operation']
+        # The dashboard data contract requires scalar problem values. Serialize
+        # containers and null as JSON strings to preserve their contents.
+        problem = {
+            key: (
+                value
+                if type(value) in (str, int, float, bool)
+                else json.dumps(value, sort_keys=True, allow_nan=False)
+            )
+            for key, value in problem.items()
+        }
+        definition = {
+            'id': result['id'],
+            'suite': result['suite'],
+            'name': result['name'],
+            'problem': problem,
+        }
+        require(
+            result['id'] not in definitions or definitions[result['id']] == definition,
+            'Conflicting workload identity',
+        )
+        definitions[result['id']] = definition
+        timings = result['timing_results_s']
+        require(
+            isinstance(timings, list)
+            and all(
+                type(t) in (int, float) and math.isfinite(t) and t > 0 for t in timings
+            ),
+            'Invalid timing samples',
+        )
+        status = result['status']
+        require(status in ('completed', 'failed', 'timeout'), 'Invalid result status')
+        require(
+            result['error'] is None
+            or isinstance(result['error'], str)
+            and result['error'].strip(),
+            'Invalid result error',
+        )
+        if status == 'completed':
+            require(
+                len(timings) == samples
+                and result['error'] is None
+                and result['exit_code'] in (0, None),
+                'Invalid completed measurement',
+            )
+        targets.setdefault(result['target'], []).append(
+            {
+                'testId': result['id'],
+                'status': status,
+                'durationSeconds': (
+                    statistics.median(timings) if status == 'completed' else None
+                ),
+                'error': result['error'],
+                'timing_results_s': timings,
+            }
+        )
+    catalog = {
+        'tests': [definitions[key] for key in sorted(definitions)],
+        'targets': {
+            target: sorted(result['testId'] for result in results)
+            for target, results in sorted(targets.items())
+        },
+    }
+    catalog['id'] = 'default-' + hashlib.sha256(encode_json_bytes(catalog)).hexdigest()
+    run_id = f"github-{github['run_id']}-attempt-{github['run_attempt']}-default"
+    details = {'thread_policy': 'default'}
+    for section, values in raw['provenance'].items():
+        for key, value in values.items():
+            if section == 'rocjitsu' and key in (
+                'rocjitsu_commit_sha',
+                'rocjitsu_commit_timestamp',
+            ):
+                continue
+            details[f'{section}.{key}'] = (
+                value
+                if type(value) in (str, int, float, bool) and value != ''
+                else json.dumps(value, sort_keys=True, allow_nan=False)
+            )
+    details['workflow.commit_sha'] = workflow['workflow_sha']
+    details['tensile.commit_sha'] = workflow['tensile_sha']
+    for key in ('warmups', 'samples', 'timeout_seconds', 'benchmark_suite'):
+        details['execution.' + key] = summary[key]
+    labels = {
+        'thread_policy': 'Thread policy',
+        'workflow.commit_sha': 'Workflow commit',
+        'tensile.commit_sha': 'Tensile commit',
+        'machine.hostname': 'Hostname',
+        'machine.platform': 'Platform',
+        'machine.kernel': 'Kernel',
+        'machine.cpu': 'CPU',
+        'rocjitsu.rocm_sdk_version': 'ROCm SDK',
+        'corpus.corpus_commit_sha': 'Corpus commit',
+        'corpus.corpus_commit_timestamp': 'Corpus commit timestamp',
+        'auxiliary.python': 'Python',
+        'execution.warmups': 'Warmups',
+        'execution.samples': 'Samples',
+        'execution.timeout_seconds': 'Timeout (seconds)',
+        'execution.benchmark_suite': 'Benchmark suite',
+    }
+    run = {
+        'id': run_id,
+        'comparisonId': f"github-{github['run_id']}-attempt-{github['run_attempt']}",
+        'testCatalog': f"test-catalogs/{catalog['id']}.json",
+        'plugin': {'id': 'vanilla', 'name': 'Vanilla'},
+        'source': {
+            # The workflow requires the measured commit to be in develop history.
+            # This is its source branch, not the workflow dispatch branch.
+            'branch': 'develop',
+            'commit': source['commit'],
+            'committedAt': committed,
+            'message': source['commit_subject'],
+        },
+        'execution': {
+            'completedAt': summary['finished_at'],
+            'trigger': github['trigger'],
+            'machine': workflow['runner'],
+        },
+        'environment': [
+            {
+                'key': key,
+                'label': labels.get(
+                    key,
+                    key.replace('.', ' ').replace('_', ' ').replace('-', ' ').title(),
+                ),
+                'value': value,
+            }
+            for key, value in sorted(details.items())
+        ],
+        'targets': [
+            {'id': target, 'results': sorted(results, key=lambda r: r['testId'])}
+            for target, results in sorted(targets.items())
+        ],
+    }
+    return run
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('--raw-run', required=True, type=Path)
+    parser.add_argument('--output', required=True, type=Path)
+    parser.add_argument('--run-id', required=True)
+    parser.add_argument('--run-attempt', required=True)
+    parser.add_argument('--repository', required=True)
+    parser.add_argument('--commit-message', required=True)
+    parser.add_argument('--machine', required=True)
+    parser.add_argument('--workflow-sha', required=True)
+    parser.add_argument('--tensile-sha', required=True)
+    parser.add_argument('--event', choices=('push', 'workflow_dispatch'), required=True)
+    args = parser.parse_args()
+    raw = json.loads(args.raw_run.read_text())
+    workflow = {
+        'run_id': args.run_id,
+        'run_attempt': args.run_attempt,
+        'repository': args.repository,
+        'source_subject': args.commit_message,
+        'source_sha': raw['provenance']['rocjitsu']['rocjitsu_commit_sha'],
+        'runner': args.machine,
+        'workflow_sha': args.workflow_sha,
+        'tensile_sha': args.tensile_sha,
+        'trigger': 'manual' if args.event == 'workflow_dispatch' else 'auto',
+    }
+    result = build_published_run(raw, workflow)
+    args.output.parent.mkdir(parents=True, exist_ok=True)
+    args.output.write_bytes(encode_json_bytes(result))
+
+
+if __name__ == '__main__':
+    main()

emulation/rocjitsu/website/dashboard/scripts/dashboard-data/requirements-test.txt

@@ -0,0 +1,2 @@
+pytest==9.1.1
+PyYAML==6.0.3