[Benchmarks] Add level-zero record-and-replay to set of SubmitGraph benchmarks (#20641)

Level-zero `SubmitGraph` benchmarks support EmulateGraphs=0 (L0
record-and-replay) APIs and EmulateGraphs=1 (submitting command list to
immediate command list). This PR adds the record-and-replay compute
benchmark.

Author: mmichel11

devops/scripts/benchmarks/benches/compute.py

@@ -208,25 +208,17 @@ def benchmarks(self) -> list[Benchmark]:
             measure_completion_time,
             use_events,
         ) in submit_graph_params:
-            # Non-sycl runtimes have to be run with emulated graphs,
-            # see: https://github.com/intel/compute-benchmarks/commit/d81d5d602739482b9070c872a28c0b5ebb41de70
-            emulate_graphs = (
-                0 if runtime in (RUNTIMES.SYCL, RUNTIMES.SYCL_PREVIEW) else 1
-            )
-            benches.append(
-                GraphApiSubmitGraph(
-                    self,
-                    runtime,
-                    in_order_queue,
-                    num_kernels,
-                    measure_completion_time,
-                    use_events,
-                    emulate_graphs,
-                    useHostTasks=0,
-                )
-            )
-            if runtime == RUNTIMES.SYCL:
-                # Create CPU count variant
+            # SYCL only supports graph mode, UR supports only emulation with command buffers,
+            # and L0 supports both modes via graph and command list APIs.
+            if runtime == RUNTIMES.SYCL or runtime == RUNTIMES.SYCL_PREVIEW:
+                emulate_graphs = [0]
+            elif runtime == RUNTIMES.UR:
+                emulate_graphs = [1]
+            else:  # level-zero
+                # SubmitGraph with L0 graph segfaults on PVC
+                device_arch = getattr(options, "device_architecture", "")
+                emulate_graphs = [1] if "pvc" in device_arch else [0, 1]
+            for emulate_graph in emulate_graphs:
                 benches.append(
                     GraphApiSubmitGraph(
                         self,
@@ -235,11 +227,25 @@ def benchmarks(self) -> list[Benchmark]:
                         num_kernels,
                         measure_completion_time,
                         use_events,
-                        emulate_graphs,
+                        emulate_graph,
                         useHostTasks=0,
-                        profiler_type=PROFILERS.CPU_COUNTER,
                     )
                 )
+                if runtime == RUNTIMES.SYCL:
+                    # Create CPU count variant
+                    benches.append(
+                        GraphApiSubmitGraph(
+                            self,
+                            runtime,
+                            in_order_queue,
+                            num_kernels,
+                            measure_completion_time,
+                            use_events,
+                            emulate_graph,
+                            useHostTasks=0,
+                            profiler_type=PROFILERS.CPU_COUNTER,
+                        )
+                    )
 
         # Add other benchmarks
         benches += [
@@ -1178,6 +1184,7 @@ def __init__(
         self._use_events = useEvents
         self._use_host_tasks = useHostTasks
         self._emulate_graphs = emulate_graphs
+        self._emulate_str = " with graph emulation" if self._emulate_graphs else ""
         self._ioq_str = "in order" if self._in_order_queue else "out of order"
         self._measure_str = (
             " with measure completion" if self._measure_completion_time else ""
@@ -1196,10 +1203,10 @@ def __init__(
         )
 
     def name(self):
-        return f"graph_api_benchmark_{self._runtime.value} SubmitGraph{self._use_events_str}{self._host_tasks_str} numKernels:{self._num_kernels} ioq {self._in_order_queue} measureCompletion {self._measure_completion_time}{self._cpu_count_str()}"
+        return f"graph_api_benchmark_{self._runtime.value} SubmitGraph{self._use_events_str}{self._host_tasks_str}{self._emulate_str} numKernels:{self._num_kernels} ioq {self._in_order_queue} measureCompletion {self._measure_completion_time}{self._cpu_count_str()}"
 
     def display_name(self) -> str:
-        return f"{self._runtime.value.upper()} SubmitGraph {self._ioq_str}{self._measure_str}{self._use_events_str}{self._host_tasks_str}, {self._num_kernels} kernels{self._cpu_count_str(separator=',')}"
+        return f"{self._runtime.value.upper()} SubmitGraph {self._ioq_str}{self._measure_str}{self._use_events_str}{self._host_tasks_str}{self._emulate_str}, {self._num_kernels} kernels{self._cpu_count_str(separator=',')}"
 
     def explicit_group(self):
         return f"SubmitGraph {self._ioq_str}{self._measure_str}{self._use_events_str}{self._host_tasks_str}, {self._num_kernels} kernels{self._cpu_count_str(separator=',')}"