Add --verify-data CLI flag, Tensor.data property, parallel DataExecutor

- CLI: --verify-data flag enables Phase 2 data verification (ADR-0020)
- Tensor.data: returns actual numpy values (verify-data) or zeros placeholder
- Tensor.__repr__: shows value summary or data=N/A (placeholder)
- DataExecutor: ThreadPoolExecutor for same-timestamp parallel op execution
- BenchResult.engine: exposes op_log/memory_store for Phase 2 access

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

tests/test_cli_verify_data.py

@@ -0,0 +1,49 @@
+"""Tests for --verify-data CLI flag (Phase 1 verification)."""
+import kernbench.cli.main as cli_main
+
+
+def test_cli_verify_data_flag_parsed(monkeypatch):
+    """--verify-data flag is parsed and stored as True."""
+
+    def fake_cmd_run(args) -> int:
+        assert args.verify_data is True
+        return 0
+
+    monkeypatch.setattr(cli_main, "cmd_run", fake_cmd_run)
+    rc = cli_main.main([
+        "run", "--topology", "topology.yaml", "--bench", "qkv_gemm",
+        "--verify-data",
+    ])
+    assert rc == 0
+
+
+def test_cli_verify_data_flag_default(monkeypatch):
+    """Without --verify-data, flag defaults to False."""
+
+    def fake_cmd_run(args) -> int:
+        assert args.verify_data is False
+        return 0
+
+    monkeypatch.setattr(cli_main, "cmd_run", fake_cmd_run)
+    rc = cli_main.main([
+        "run", "--topology", "topology.yaml", "--bench", "qkv_gemm",
+    ])
+    assert rc == 0
+
+
+def test_cmd_run_verify_data_enables_engine():
+    """--verify-data runs full pipeline with enable_data=True and DataExecutor."""
+    rc = cli_main.main([
+        "run", "--topology", "topology.yaml", "--bench", "qkv_gemm",
+        "--device", "sip:0", "--verify-data",
+    ])
+    assert rc == 0
+
+
+def test_cmd_run_without_verify_data_no_op_log():
+    """Without --verify-data, engine runs in timing-only mode (no op_log)."""
+    rc = cli_main.main([
+        "run", "--topology", "topology.yaml", "--bench", "qkv_gemm",
+        "--device", "sip:0",
+    ])
+    assert rc == 0

tests/test_data_executor.py

@@ -186,3 +186,41 @@ def test_sequential_gemm_then_math():
 
     exp_result = store.read("tcm", 0x300)
     assert np.allclose(exp_result, np.exp(expected_gemm))
+
+
+def test_parallel_same_timestamp_ops():
+    """Multiple independent ops at the same t_start produce correct results
+    when executed in parallel (ThreadPoolExecutor)."""
+    store = MemoryStore()
+    n_ops = 8
+    # Each op: independent GEMM writing to a different address
+    for i in range(n_ops):
+        a = np.full((4, 4), float(i + 1), dtype=np.float16)
+        b = np.eye(4, dtype=np.float16)
+        store.write("tcm", 0x1000 * i, a)
+        store.write("tcm", 0x1000 * i + 0x800, b)
+
+    ops = [
+        OpRecord(
+            t_start=0.0, t_end=100.0,
+            component_id=f"pe{i}.pe_gemm",
+            op_kind="gemm", op_name="gemm_f16",
+            params={
+                "src_a_addr": 0x1000 * i,
+                "src_b_addr": 0x1000 * i + 0x800,
+                "dst_addr": 0x80000 + 0x1000 * i,
+                "shape_a": (4, 4), "shape_b": (4, 4), "shape_out": (4, 4),
+                "dtype_in": "f16", "dtype_acc": "f32", "dtype_out": "f16",
+                "addr_space": "tcm",
+            },
+        )
+        for i in range(n_ops)
+    ]
+
+    executor = DataExecutor(ops, store)
+    executor.run()
+
+    for i in range(n_ops):
+        result = store.read("tcm", 0x80000 + 0x1000 * i)
+        expected = np.full((4, 4), float(i + 1), dtype=np.float16)
+        assert np.allclose(result, expected), f"op {i}: expected {expected}, got {result}"