Add fused-pipeline primitive and test for compiled CUDA kernels

test_fused_pipeline.sexp

@@ -0,0 +1,43 @@
+;; Test Fused Pipeline - Should be much faster than interpreted
+;;
+;; This uses the fused-pipeline primitive which compiles all effects
+;; into a single CUDA kernel instead of interpreting them one by one.
+
+(stream "fused_pipeline_test"
+  :fps 30
+  :width 1920
+  :height 1080
+  :seed 42
+
+  ;; Load primitives
+  (require-primitives "streaming_gpu")
+  (require-primitives "image")
+  (require-primitives "math")
+
+  ;; Define the effects pipeline (compiled to single CUDA kernel)
+  (def effects-pipeline
+    [{"op" "rotate" "angle" 0}
+     {"op" "zoom" "amount" 1.0}
+     {"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}])
+
+  ;; Frame pipeline
+  (frame
+    (let [;; Create a gradient image
+          r (+ 0.5 (* 0.5 (math:sin (* t 1))))
+          g (+ 0.5 (* 0.5 (math:sin (* t 1.3))))
+          b (+ 0.5 (* 0.5 (math:sin (* t 1.7))))
+          color [(* r 255) (* g 255) (* b 255)]
+          base (image:make-image 1920 1080 color)
+
+          ;; Dynamic parameters (change per frame)
+          angle (* t 30)
+          zoom (+ 1.0 (* 0.2 (math:sin (* t 0.5))))
+          phase (* t 2)]
+
+      ;; Apply fused pipeline - all effects in ONE CUDA kernel!
+      (streaming_gpu:fused-pipeline base effects-pipeline
+        :rotate_angle angle
+        :zoom_amount zoom
+        :ripple_phase phase))))