Add github benchmark action

.github/workflows/bench.yml

@@ -10,6 +10,8 @@ on:
 jobs:
   bench:
     runs-on: self-hosted-rpi4
+    permissions:
+      contents: write
     if: github.repository_owner == 'pq-code-package' && (github.event.label.name == 'benchmark' || github.ref == 'refs/heads/main')
     steps:
       - uses: actions/checkout@v4
@@ -34,4 +36,13 @@ jobs:
       - name: Run benchmark
         shell: nix develop .#ci -c bash -e {0}
         run: |
-          tests bench -c PMU --cflags -mcpu=cortex-a72 -v
+          tests bench -c PMU --cflags -mcpu=cortex-a72 -v --output output.json
+      - name: Store benchmark result
+        if: github.repository_owner == 'pq-code-package' && github.ref == 'refs/heads/main'
+        uses: benchmark-action/github-action-benchmark@v1
+        with:
+          name: Arm Cortex-A72 (Raspberry Pi 4) benchmarks
+          tool: 'customSmallerIsBetter'
+          output-file-path: output.json
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          auto-push: true

scripts/tests

@@ -345,7 +345,13 @@ def kat(force_qemu, verbose, cflags, arch_flags):
     default="NO",
     help="Method for counting clock cycles. PMU requires (user-space) access to the Arm Performance Monitor Unit (PMU). PERF requires a kernel with perf support.",
 )
-def bench(force_qemu, verbose, cycles, cflags, arch_flags):
+@click.option(
+    "-o",
+    "--output",
+    nargs=1,
+    help="Path to output file in json format",
+)
+def bench(force_qemu, verbose, cycles, cflags, arch_flags, output):
     config_logger(verbose)
 
     results = test_schemes(
@@ -360,6 +366,34 @@ def bench(force_qemu, verbose, cycles, cflags, arch_flags):
         [f"CYCLES={cycles}"],
     )
 
+    if output:
+        import json
+
+        with open(output, "w") as f:
+            v = []
+            for scheme in results:
+                schemeStr = str(scheme)
+                r = results[scheme]
+
+                # The first 3 lines of the output are expected to be
+                # keypair cycles=X
+                # encaps cycles=X
+                # decaps cycles=X
+
+                d = {
+                    k: int(v)
+                    for k, v in (l.decode().split("=") for l in r.splitlines()[:3])
+                }
+                for primitive in ["keypair", "encaps", "decaps"]:
+                    v.append(
+                        {
+                            "name": f"{schemeStr} {primitive}",
+                            "unit": "cycles",
+                            "value": d[f"{primitive} cycles"],
+                        }
+                    )
+            f.write(json.dumps(v))
+
 
 @click.group(invoke_without_command=True)
 def cli():