Cleanup Comments

Util/FFmpeg/FFmpegNativeDecoder.cs

@@ -23,7 +23,7 @@ namespace ADOFAIRenderer.Util.FFmpeg;
 ///   2. Codec & Hardware Device Discovery:
 ///      a) avcodec_find_decoder: Locate codec implementation by codec ID.
 ///      b) avcodec_get_hw_config: Query supported hardware configurations for the decoder.
-///      c) Prioritized HW device creation (D3D11VA → DXVA2 → QSV → CUDA → D3D12VA → VAAPI → VideoToolbox).
+///      c) HW device creation for the reported configurations, in the order of PreferredHwDeviceTypes.
 ///      d) Attach hw_device_ctx and register custom get_format callback.
 ///      e) If HW initialization fails or is unsupported, fall back to CPU software decoding
 ///         with multi-threading enabled (thread_count = 0, FF_THREAD_FRAME).
@@ -35,8 +35,10 @@ namespace ADOFAIRenderer.Util.FFmpeg;
 ///      c) (HW Transfer) For hardware frames, av_hwframe_transfer_data downloads GPU surface data
 ///         into system RAM as NV12 (av_frame_copy_props preserves timestamps).
 ///      d) (Output Dispatch):
-///         - TryReadNV12Frame: Zero-copy direct memory transfer for NV12 frames, sws_scale fallback.
-///         - TryReadYUV420PFrame: Zero-copy planar copy for YUV420P/YUVJ420P, sws_scale fallback.
+///         - TryReadNV12Frame: NV12 frames are copied row by row (Buffer.MemoryCopy) into the output buffers,
+///           other formats are converted with sws_scale.
+///         - TryReadYUV420PFrame: YUV420P/YUVJ420P planes are copied row by row into the output buffers,
+///           other formats are converted with sws_scale.
 ///         - TryReadFrame: sws_scale conversion to RGBA32 with optional vertical flip (vflip).
 ///
 ///   4. Seeking & Scrubbing (SeekToTime / SeekToFrame):
@@ -76,9 +78,12 @@ internal sealed unsafe class FFmpegNativeDecoder : IDisposable
 
     /// <summary>
     /// Preferred hardware device types in priority order.
-    /// Modern Windows platforms prioritize D3D11VA for wide driver support and direct integration,
-    /// followed by DXVA2 (DirectX 9 / legacy), Intel QSV, NVIDIA CUDA, and DirectX 12 D3D12VA.
-    /// Non-Windows systems fall back to VAAPI or VideoToolbox.
+    /// The candidates are the hardware configurations the selected decoder reports (avcodec_get_hw_config),
+    /// sorted by this list; configurations not in the list follow in reported order.
+    /// Windows: D3D11VA, then DXVA2 (DirectX 9), CUDA and D3D12VA. Other platforms: VAAPI or VideoToolbox.
+    /// The QSV entry matches only decoders that report a QSV configuration. The native decoders returned by
+    /// avcodec_find_decoder report DXVA2/D3D11VA/D3D12VA/CUDA/VAAPI/Vulkan in the bundled FFmpeg 8 build;
+    /// the QSV decoders (h264_qsv, etc.) are separate decoders selected by name.
     /// </summary>
     private static readonly AVHWDeviceType[] PreferredHwDeviceTypes = new[]
     {
@@ -661,9 +666,11 @@ internal sealed unsafe class FFmpegNativeDecoder : IDisposable
             _getFormatDelegate = GetFormat;
             _pCodecContext->get_format = _getFormatDelegate;
 
-            // Hardware decoding requires single-threaded decoding (thread_count = 1).
-            // Multi-threading (thread_count = 0 / FF_THREAD_FRAME) is incompatible with fixed-function
-            // GPU decoders and causes surface pool allocation exhaustion or crashes in get_format.
+            // Hardware decoding runs with one decoder thread (thread_count = 1).
+            // The GPU performs the decoding; the decoder thread only parses packets and submits them.
+            // With frame threading, libavcodec adds one hardware surface per thread to the frame pool
+            // (avcodec_get_hw_frames_parameters). The software fallback in GetFormat keeps this thread count.
+            // Reference: https://github.com/FFmpeg/FFmpeg/blob/release/8.0/libavcodec/decode.c
             _pCodecContext->thread_count = 1;
 
             // Attempt opening codec context with hardware configuration