Fix multithreaded decoding to significantly improve Tegra 3 decoding performance. It's not as smooth as Tegra 4 CPU decoding or Snapdragon hardware decoding, but it's much better than it was before.

jni/nv_avc_dec/nv_avc_dec.c

@@ -41,12 +41,16 @@ int nv_avc_init(int width, int height) {
 		return -1;
 	}
 
-	// We're a latency-sensitive application
-	decoder_ctx->flags |= CODEC_FLAG_LOW_DELAY;
-
 	// Show frames even before a reference frame
 	decoder_ctx->flags2 |= CODEC_FLAG2_SHOW_ALL;
 
+	// Skip the loop filter for performance reasons
+	decoder_ctx->skip_loop_filter = AVDISCARD_ALL;
+
+	// Run 2 threads for decoding
+	decoder_ctx->thread_count = 2;
+	decoder_ctx->thread_type = FF_THREAD_FRAME;
+
 	decoder_ctx->width = width;
 	decoder_ctx->height = height;
 	decoder_ctx->pix_fmt = PIX_FMT_YUV420P;