Better audio timing test: compensate for audio server latency

2 files changed, 8 insertions, 1 deletions

diff --git a/TODO b/TODO
index 9b107d2..86911c3 100644
--- a/TODO
+++ b/TODO
@@ -113,4 +113,7 @@ Modify ffmpeg to accept opengl texture for nvenc encoding. Removes extra buffers
 
 When vulkan encode is added, mention minimum nvidia driver required. (550.54.14?).
 
-Support drm plane rotation. Neither X11 nor any Wayland compositor currently rotates drm planes so this might not be needed.
\ No newline at end of file
+Support drm plane rotation. Neither X11 nor any Wayland compositor currently rotates drm planes so this might not be needed.
+
+Investigate if there is a way to do gpu->gpu copy directly without touching system ram to enable video encoding on a different gpu. On nvidia this is possible with cudaMemcpyPeer, but how about from an intel/amd gpu to an nvidia gpu or the other way around or any combination of iGPU and dedicated GPU?
+    Maybe something with clEnqueueMigrateMemObjects? on AMD something with DirectGMA maybe?
diff --git a/src/main.cpp b/src/main.cpp
index 167f8a5..bb54ca2 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -2388,6 +2388,10 @@ int main(int argc, char **argv) {
                 double received_audio_time = clock_get_monotonic_seconds();
                 const int64_t timeout_ms = std::round((1000.0 / (double)audio_track.codec_context->sample_rate) * 1000.0);
 
+                // Move audio back by around 252 ms. This is just a shitty way to handle audio latency but pulseaudio latency calculation
+                // returns much lower value which isn't helpful.
+                audio_device.frame->pts = audio_track.codec_context->frame_size * 12;
+
                 while(running) {
                     void *sound_buffer;
                     int sound_buffer_size = -1;