Add benchmark to compare multi-stream vs single-stream encoding

Author: NicolasHug

benchmarks/encoders/bench_multi_stream_vs_single_stream.py

@@ -0,0 +1,136 @@
+"""
+Basic benchmark aimed at validating that the new Encoder class isn't slower than
+the existing VideoEncoder and AudioEncoder.
+"""
+
+import os
+import tempfile
+from time import perf_counter_ns
+
+import torch
+
+from torchcodec.encoders import AudioEncoder, Encoder, VideoEncoder
+
+
+def bench(f, *args, num_exp=100, warmup=0, **kwargs):
+    for _ in range(warmup):
+        f(*args, **kwargs)
+
+    times = []
+    for _ in range(num_exp):
+        start = perf_counter_ns()
+        f(*args, **kwargs)
+        end = perf_counter_ns()
+        times.append(end - start)
+    return torch.tensor(times).float()
+
+
+def report_stats(times, unit="ms"):
+    mul = {
+        "ns": 1,
+        "µs": 1e-3,
+        "ms": 1e-6,
+        "s": 1e-9,
+    }[unit]
+    times = times * mul
+    std = times.std().item()
+    med = times.median().item()
+    print(f"{med = :.2f}{unit} +- {std:.2f}")
+    return med
+
+
+# --- Config ---
+height, width = 256, 256
+frame_rate = 30
+num_channels = 2
+sample_rate = 48000
+durations_s = [3, 10]
+
+
+def encode_video_with_video_encoder(frames, dest):
+    encoder = VideoEncoder(frames, frame_rate=frame_rate)
+    encoder.to_file(dest)
+
+
+def encode_video_with_streaming_encoder(frames, dest):
+    encoder = Encoder()
+    video_stream = encoder.add_video(
+        height=frames.shape[2],
+        width=frames.shape[3],
+        frame_rate=frame_rate,
+        device=str(frames.device),
+    )
+    with encoder.open_file(dest):
+        video_stream.add_frames(frames)
+
+
+def encode_audio_with_audio_encoder(samples, dest):
+    encoder = AudioEncoder(samples, sample_rate=sample_rate)
+    encoder.to_file(dest)
+
+
+def encode_audio_with_streaming_encoder(samples, dest):
+    encoder = Encoder()
+    audio_stream = encoder.add_audio(sample_rate=sample_rate, num_channels=num_channels)
+    with encoder.open_file(dest):
+        audio_stream.add_samples(samples)
+
+
+if __name__ == "__main__":
+    tmpdir = tempfile.mkdtemp()
+
+    devices = ["cpu"]
+    if torch.cuda.is_available():
+        devices.append("cuda")
+
+    for duration_s in durations_s:
+        print(f"\n{'=' * 50}")
+        print(f"Duration: {duration_s}s")
+        print(f"{'=' * 50}")
+
+        num_frames = frame_rate * duration_s
+        video_frames = torch.randint(
+            0, 256, (num_frames, 3, height, width), dtype=torch.uint8
+        )
+
+        num_samples = sample_rate * duration_s
+        audio_samples = torch.randn(num_channels, num_samples)
+
+        # --- Video benchmarks ---
+        for device in devices:
+            frames = video_frames.to(device)
+            dest = os.path.join(tmpdir, "video.mp4")
+
+            print(f"\n--- Video encoding ({device}) ---")
+
+            print("VideoEncoder:    ", end="")
+            times = bench(
+                encode_video_with_video_encoder, frames, dest, num_exp=10, warmup=2
+            )
+            report_stats(times)
+
+            print("StreamingEncoder:", end="")
+            times = bench(
+                encode_video_with_streaming_encoder, frames, dest, num_exp=10, warmup=2
+            )
+            report_stats(times)
+
+        # --- Audio benchmarks ---
+        print("\n--- Audio encoding (cpu) ---")
+        dest = os.path.join(tmpdir, "audio.wav")
+
+        print("AudioEncoder:    ", end="")
+        times = bench(
+            encode_audio_with_audio_encoder, audio_samples, dest, num_exp=10, warmup=2
+        )
+        report_stats(times)
+
+        print("StreamingEncoder:", end="")
+        times = bench(
+            encode_audio_with_streaming_encoder,
+            audio_samples,
+            dest,
+            num_exp=10,
+            warmup=2,
+        )
+        report_stats(times)