Add autonomous prealloc benchmark

test_autonomous_prealloc.py

@@ -0,0 +1,75 @@
+#!/usr/bin/env python3
+"""
+Test autonomous pipeline with pre-allocated buffer.
+This eliminates ALL Python from the hot path.
+"""
+
+import time
+import sys
+sys.path.insert(0, '/app')
+
+import cupy as cp
+from streaming.sexp_to_cuda import compile_autonomous_pipeline
+
+def test_autonomous_prealloc():
+    width, height = 1920, 1080
+    n_frames = 300
+    fps = 30.0
+
+    print(f"Testing {n_frames} frames at {width}x{height}")
+    print("=" * 60)
+
+    # Pre-allocate frame buffer (stays on GPU)
+    frame = cp.zeros((height, width, 3), dtype=cp.uint8)
+    frame[:, :, 0] = 128  # R
+    frame[:, :, 1] = 100  # G
+    frame[:, :, 2] = 200  # B
+
+    # Define effects
+    effects = [
+        {'op': 'rotate', 'angle': 0},
+        {'op': 'hue_shift', 'degrees': 30},
+        {'op': 'ripple', 'amplitude': 15, 'frequency': 10, 'decay': 2, 'phase': 0, 'center_x': 960, 'center_y': 540},
+        {'op': 'brightness', 'factor': 1.0},
+    ]
+
+    # Dynamic expressions (computed on GPU)
+    dynamic_expressions = {
+        'rotate_angle': 't * 30.0f',
+        'ripple_phase': 't * 2.0f',
+        'brightness_factor': '0.8f + 0.4f * sinf(t * 2.0f)',
+    }
+
+    # Compile autonomous pipeline
+    print("Compiling autonomous pipeline...")
+    pipeline = compile_autonomous_pipeline(effects, width, height, dynamic_expressions)
+
+    # Warmup
+    output = pipeline(frame, 0, fps)
+    cp.cuda.Stream.null.synchronize()
+
+    # Benchmark - ZERO Python in the hot path!
+    print(f"Running {n_frames} frames...")
+    start = time.time()
+    for i in range(n_frames):
+        output = pipeline(frame, i, fps)
+    cp.cuda.Stream.null.synchronize()
+    elapsed = time.time() - start
+
+    ms_per_frame = elapsed / n_frames * 1000
+    actual_fps = n_frames / elapsed
+
+    print("=" * 60)
+    print(f"Time:      {ms_per_frame:.2f}ms per frame")
+    print(f"FPS:       {actual_fps:.0f}")
+    print(f"Real-time: {actual_fps / 30:.1f}x (at 30fps target)")
+    print("=" * 60)
+
+    # Compare with original baseline
+    print(f"\nOriginal Python sexp: ~150ms = 6 fps")
+    print(f"Autonomous GPU:       {ms_per_frame:.2f}ms = {actual_fps:.0f} fps")
+    print(f"Speedup:              {150 / ms_per_frame:.0f}x faster!")
+
+
+if __name__ == '__main__':
+    test_autonomous_prealloc()