bench: add videoscale GstBin path (mirrors _create_appsink, --noscale flag)

tests/benchmark_nv12_decode.py

@@ -91,15 +91,6 @@ for dev in _HW_VPU_DEVICES:
 if not hw_decoders:
     print("[SW] No VPU device or no gst-mpp plugin — using software decode")
 
-# ── Caps: prefer NV12 when HW decode available, same as app ─────────────────
-
-if hw_decoders:
-    caps_str = "video/x-raw,format=NV12;video/x-raw,format=BGRA"
-else:
-    caps_str = "video/x-raw,format=BGRA"
-
-print(f"[caps] appsink caps: {caps_str}")
-
 # ── Pipeline ────────────────────────────────────────────────────────────────
 
 pipeline = Gst.ElementFactory.make("playbin", "player")
@@ -107,14 +98,59 @@ if pipeline is None:
     print("[ERR] playbin unavailable — install gst-plugins-base")
     sys.exit(1)
 
+# Target display size — same default as the app (R36S screen is 640×480).
+SCALE_W, SCALE_H = 640, 480
+
 appsink = Gst.ElementFactory.make("appsink", "vsink")
 appsink.set_property("emit-signals", True)
 appsink.set_property("sync", True)          # keep A/V sync on
 appsink.set_property("max-buffers", 2)
 appsink.set_property("drop", True)
-appsink.set_property("caps", Gst.Caps.from_string(caps_str))
 
-pipeline.set_property("video-sink", appsink)
+# video_sink is what we hand to playbin.  For HW decode we mirror _create_appsink()
+# from the app: wrap videoscale → capsfilter(NV12,640×480) → appsink in a GstBin
+# so the Python callback receives 460 KB per frame instead of 3.1 MB (6.7× smaller).
+# Pass --noscale to disable this and benchmark the unscaled path.
+video_sink = appsink
+
+if hw_decoders and "--noscale" not in sys.argv:
+    scale_el = Gst.ElementFactory.make("videoscale", "vscale")
+    cfilt_el = Gst.ElementFactory.make("capsfilter", "vcaps")
+    if scale_el is not None and cfilt_el is not None:
+        cfilt_el.set_property(
+            "caps",
+            Gst.Caps.from_string(
+                f"video/x-raw,format=NV12,width={SCALE_W},height={SCALE_H}"
+            ),
+        )
+        bin_ = Gst.Bin.new("vscale-bin")
+        bin_.add(scale_el)
+        bin_.add(cfilt_el)
+        bin_.add(appsink)
+        scale_el.link(cfilt_el)
+        cfilt_el.link(appsink)
+        sink_pad = scale_el.get_static_pad("sink")
+        ghost = Gst.GhostPad.new("sink", sink_pad)
+        ghost.set_active(True)
+        bin_.add_pad(ghost)
+        video_sink = bin_
+        print(f"[scale] videoscale → {SCALE_W}×{SCALE_H} NV12 (mirrors app _create_appsink)")
+    else:
+        print("[scale] videoscale element unavailable — falling back to unscaled NV12")
+        appsink.set_property(
+            "caps", Gst.Caps.from_string("video/x-raw,format=NV12;video/x-raw,format=BGRA")
+        )
+elif hw_decoders:
+    print(f"[scale] --noscale: unscaled NV12 ({SCALE_W}×{SCALE_H} disabled)")
+    appsink.set_property(
+        "caps", Gst.Caps.from_string("video/x-raw,format=NV12;video/x-raw,format=BGRA")
+    )
+else:
+    appsink.set_property("caps", Gst.Caps.from_string("video/x-raw,format=BGRA"))
+
+print(f"[caps] video-sink: {'GstBin(videoscale+capsfilter+appsink)' if video_sink is not appsink else 'appsink'}")
+
+pipeline.set_property("video-sink", video_sink)
 pipeline.set_property("uri", url)
 
 # Disable subtitles / visualisations, keep audio+video (same flags as app).
@@ -302,7 +338,9 @@ with stats.lock:
     fmt         = stats.fmt
     w, h        = stats.width, stats.height
 
+scale_active = video_sink is not appsink
 print(f"  URL      : {url}")
+print(f"  Scale    : {'videoscale → %d×%d NV12' % (SCALE_W, SCALE_H) if scale_active else 'none (--noscale or SW)'}")
 print(f"  Format   : {fmt}  {w}x{h}")
 print(f"  Decoder  : {'HW (' + ', '.join(hw_decoders) + ')' if hw_decoders else 'SW (avdec_*)'}")
 print(f"  Total frames decoded : {total}  (excl. {WARMUP_FRAMES} warmup)")