Finalize native API host-device benchmark (#82)

1. Use 2D dimension to match memory layout
2. Use default device_get and device_put

Author: linamy85

Ironwood/configs/host_device/host_device.yaml

@@ -0,0 +1,9 @@
+benchmarks:
+- benchmark_name: host_device
+  num_runs: 20
+  benchmark_sweep_params:
+  - {
+      data_size_mib_list: [1, 16, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768]
+    }
+  csv_path: "../microbenchmarks/host_device"
+  trace_dir: "../microbenchmarks/host_device/trace"

Ironwood/configs/host_device/host_device_single_chip.yaml

@@ -24,8 +24,7 @@ spec:
       cd accelerator-microbenchmarks
       pip install -r requirements.txt
 
-      export TPU_VISIBLE_CHIPS=0
-      bash ./Ironwood/scripts/run_host_device_benchmark.sh --config Ironwood/configs/host_device/host_device_single_chip.yaml
+      bash ./Ironwood/scripts/run_host_device_benchmark.sh --config Ironwood/configs/host_device/host_device.yaml
 
     resources:
       requests:

Ironwood/guides/host_device/tpu7x-host-device-benchmark.yaml

@@ -18,43 +18,27 @@
 os.environ["TPU_PREMAPPED_BUFFER_SIZE"] = "68719476736"
 os.environ["TPU_PREMAPPED_BUFFER_TRANSFER_THRESHOLD_BYTES"] = "68719476736"
 
-def get_tpu_devices(num_devices: int):
-    devices = jax.devices()
-    if len(devices) < num_devices:
-        raise RuntimeError(f"Require {num_devices} devices, found {len(devices)}")
-    return devices[:num_devices]
 
 def benchmark_host_device(
-    num_devices: int,
-    data_size_mb: int,
+    data_size_mib: int,
     num_runs: int = 100,
     trace_dir: str = None,
 ) -> Dict[str, Any]:
     """Benchmarks H2D/D2H transfer using simple device_put/device_get."""
-    tpu_devices = get_tpu_devices(num_devices)
 
-    num_elements = 1024 * 1024 * data_size_mb // np.dtype(np.float32).itemsize
+    num_elements = 1024 * 1024 * data_size_mib // np.dtype(np.float32).itemsize
 
     # Allocate Host Source Buffer
-    host_data = np.random.normal(size=(num_elements,)).astype(np.float32)
+    column = 128
+    host_data = np.random.normal(size=(num_elements // column, column)).astype(np.float32)
 
     print(
-        f"Benchmarking (Simple) Transfer with Data Size: {data_size_mb} MB on"
-        f" {num_devices} devices for {num_runs} iterations"
+        f"Benchmarking Transfer with Data Size: {data_size_mib} MB for {num_runs} iterations"
     )
 
-    # Setup Mesh Sharding (1D)
-    mesh = sharding.Mesh(
-        np.array(tpu_devices).reshape((num_devices,)), axis_names=("x",)
-    )
-    # Shard the 1D array across "x"
-    partition_spec = sharding.PartitionSpec("x")
-    
-    data_sharding = sharding.NamedSharding(mesh, partition_spec)
-    
     # Performance Lists
     h2d_perf, d2h_perf = [], []
-        
+
     # Profiling Context
     import contextlib
     if trace_dir:
@@ -65,7 +49,7 @@ def benchmark_host_device(
     with profiler_context:
         # Warmup
         for _ in range(2):
-            device_array = jax.device_put(host_data, data_sharding)
+            device_array = jax.device_put(host_data)
             device_array.block_until_ready()
             host_out = np.array(device_array)
             device_array.delete()
@@ -83,15 +67,14 @@ def benchmark_host_device(
                 t0 = time.perf_counter()
 
                 # Simple device_put
-                device_array = jax.device_put(host_data, data_sharding)
+                device_array = jax.device_put(host_data)
                 device_array.block_until_ready()
 
                 t1 = time.perf_counter()
                 h2d_perf.append((t1 - t0) * 1000)
 
-                # Verify H2D shape/sharding
+                # Verify H2D shape
                 assert device_array.shape == host_data.shape
-                assert device_array.sharding == data_sharding
 
                 # D2H
                 t2 = time.perf_counter()
@@ -111,19 +94,15 @@ def benchmark_host_device(
     }
 
 def benchmark_host_device_calculate_metrics(
-    num_devices: int,
-    data_size_mb: int,
+    data_size_mib: int,
     H2D_Bandwidth_ms: List[float],
     D2H_Bandwidth_ms: List[float],
 ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
     """Calculates metrics for Host-Device transfer."""
     params = locals().items()
 
-    data_size_mib = data_size_mb
-    
     # Filter out list params from metadata to avoid explosion
     metadata_keys = {
-        "num_devices", 
         "data_size_mib", 
     }
     metadata = {k: v for k, v in params if k in metadata_keys}
@@ -134,7 +113,7 @@ def add_metric(name, ms_list):
         # Report Bandwidth (GiB/s)
         # Handle division by zero if ms is 0
         bw_list = [
-            ((data_size_mb / 1024) / (ms / 1000)) if ms > 0 else 0.0 
+            ((data_size_mib / 1024) / (ms / 1000)) if ms > 0 else 0.0 
             for ms in ms_list
         ]
         stats_bw = MetricsStatistics(bw_list, f"{name}_bw (GiB/s)")