TF Profiler Instrumentation

tensorflow · Apr 15, 2022 · eb97b19 · eb97b19
1 parent 263043b
commit eb97b19
Show file tree

Hide file tree

Showing 2 changed files with 71 additions and 38 deletions.
diff --git a/tftrt/examples/benchmark_args.py b/tftrt/examples/benchmark_args.py
@@ -245,6 +245,14 @@ def __init__(self):
             "to the set location in JSON format for further processing."
         )
 
+        self._parser.add_argument(
+            "--tf_profile_export_path",
+            type=str,
+            default=None,
+            help="If set, the script will export tf.profile files for further "
+            "performance analysis."
+        )
+
         self._add_bool_argument(
             name="debug",
             default=False,

diff --git a/tftrt/examples/benchmark_runner.py b/tftrt/examples/benchmark_runner.py
@@ -5,6 +5,7 @@
 import os
 
 import abc
+import contextlib
 import copy
 import json
 import logging
@@ -390,7 +391,12 @@ def log_step(step_idx, display_every, iter_time, memcpyHtoD_time, dequeue_time):
             )
 
             @force_gpu_resync
-            @tf.function()
+            @tf.function(jit_compile=self._args.use_xla)
+            def dequeue_batch(ds_iter):
+                return next(ds_iter)
+
+            @force_gpu_resync
+            @tf.function(jit_compile=self._args.use_xla)
             def force_data_on_gpu(data, device="/gpu:0"):
                 with tf.device(device):
                     if isinstance(data, (list, tuple)):
@@ -403,58 +409,77 @@ def force_data_on_gpu(data, device="/gpu:0"):
                             output_data[k] = tf.identity(v)
                     else:
                         output_data = tf.identity(data)
+
                 return output_data
 
+            if self._args.tf_profile_export_path:
+                profiling_ctx = tf.profiler.experimental.Profile(
+                    self._args.tf_profile_export_path
+                )
+                tracing_ctx = tf.profiler.experimental.Trace
+            else:
+                profiling_ctx = contextlib.nullcontext()
+                tracing_ctx = lambda *a, **kw: contextlib.nullcontext()
+
             step_idx = 0
             ds_iter = iter(dataset)
 
-            while True:
+            with profiling_ctx:
 
-                try:
-                    start_time = time.time()
-                    data_batch = next(ds_iter)
-                    dequeue_times.append(time.time() - start_time)
-                except:
-                    break
-
-                start_time = time.time()
-                data_batch = force_data_on_gpu(data_batch)
-                memcopy_times.append(time.time() - start_time)
-
-                x, y = self.preprocess_model_inputs(data_batch)
-
-                start_time = time.time()
-                y_pred = infer_batch(x)
-                iter_times.append(time.time() - start_time)
-
-                if not self._args.debug_performance:
-                    log_step(
-                        step_idx + 1,
-                        display_every=self._args.display_every,
-                        iter_time=np.mean(iter_times[-self._args.display_every:]) * 1000,
-                        memcpyHtoD_time=np.mean(memcopy_times[-self._args.display_every:]) * 1000,
-                        dequeue_time=np.mean(dequeue_times[-self._args.display_every:]) * 1000
-                    )
-                else:
-                    print(f"{'GPU Iteration Time':18s}: {iter_times[-1]:08.4f}s")
-                    print(f"{'Data MemCopyHtoD Time':18s}: {memcpyHtoD_time[-1]:08.4f}s")
-                    print(f"{'Data Dequeue Time':18s}: {dequeue_times[-1]:08.4f}s")
+                while True:
 
-                if not self._args.use_synthetic_data:
-                    data_aggregator.aggregate_data(y_pred, y)
+                    step_idx += 1
 
-                if (self._args.num_iterations is not None and
-                        step_idx + 1 >= self._args.num_iterations):
-                    break
+                    if (self._args.num_iterations is not None and
+                            step_idx >= self._args.num_iterations):
+                        break
+
+                    with tracing_ctx('Inference Step', step_num=step_idx, _r=1):
+
+                        with tracing_ctx('Input Dequeueing', step_num=step_idx, _r=1):
+                            try:
+                                start_time = time.time()
+                                data_batch = dequeue_batch(ds_iter)
+                                dequeue_times.append(time.time() - start_time)
+                            except:
+                                print("[Exiting] Reached end of dataset ...")
+                                break
+
+                        with tracing_ctx('Inputs MemcpyHtoD', step_num=step_idx, _r=1):
+                            start_time = time.time()
+                            data_batch = force_data_on_gpu(data_batch)
+                            memcopy_times.append(time.time() - start_time)
+
+                        with tracing_ctx('Inputs Preprocessing', step_num=step_idx, _r=1):
+                            x, y = self.preprocess_model_inputs(data_batch)
+
+                        with tracing_ctx('GPU Inference', step_num=step_idx, _r=1):
+                            start_time = time.time()
+                            y_pred = infer_batch(x)
+                            iter_times.append(time.time() - start_time)
+
+                    if not self._args.debug_performance:
+                        log_step(
+                            step_idx,
+                            display_every=self._args.display_every,
+                            iter_time=np.mean(iter_times[-self._args.display_every:]) * 1000,
+                            memcpyHtoD_time=np.mean(memcopy_times[-self._args.display_every:]) * 1000,
+                            dequeue_time=np.mean(dequeue_times[-self._args.display_every:]) * 1000
+                        )
+                    else:
+                        print(f"{'GPU Iteration Time':18s}: {iter_times[-1]:08.4f}s")
+                        print(f"{'Data MemCopyHtoD Time':18s}: {memcpyHtoD_time[-1]:08.4f}s")
+                        print(f"{'Data Dequeue Time':18s}: {dequeue_times[-1]:08.4f}s")
 
-                step_idx += 1
+                    if not self._args.use_synthetic_data:
+                        data_aggregator.aggregate_data(y_pred, y)
 
             if (
                 not self._args.debug_performance and
                 step_idx % self._args.display_every != 0
             ):  # avoids double printing
                 log_step(
-                    step_idx + 1,
+                    step_idx,
                     display_every=1,  # force print
                     iter_time=np.mean(iter_times[-self._args.display_every:]) * 1000,
                     memcpyHtoD_time=np.mean(memcopy_times[-self._args.display_every:]) * 1000,