2012-05-01, 15:53
@Nimo - it is almost certain that each time you play a movie the wasapi device will be re-initialied. The reason is we default to 2ch 44.1khz pcm when no stream is playing as every windows-compatible audio device *must* support that format. As soon as you play anything different the device is re-initialized to match the format of the stream, whether its a passthrough format or another pcm format with a different bitdepth, channel layout or sample rate.
Unfortunately that is not the same as handshaking, which does not automatically occur when you reinitialize for a new format. This is controlled by the firmware in your hardware, whether receiver or TV, and a major failure of the HDMI specification implementation for many manufacturers. Sometimes the device itself needs to be powered off and on, sometimes just switching the input to another and back is enough. There is no way from the PC end to force this to happen.
@PanDaMan - nice analysis - needs looking into for sure. Haven't run across this yet even with very "hot" tracks unless downmixing, but without having had time to dig deeper certainly a float/double conversion alone can introduce tiny errors - fp math can return unexpected results with very small comparisons, e.g around 1.0 but not quite 1.0. Any soft-clipper will produce distortions by nature, and it is in there as the final blocker to hard-clipping. If you weren't downmixing there should be no values >1.0, so we need to re-check the math there. Thanks for pointing it out as we hadn't run across it at all.
Unfortunately that is not the same as handshaking, which does not automatically occur when you reinitialize for a new format. This is controlled by the firmware in your hardware, whether receiver or TV, and a major failure of the HDMI specification implementation for many manufacturers. Sometimes the device itself needs to be powered off and on, sometimes just switching the input to another and back is enough. There is no way from the PC end to force this to happen.
@PanDaMan - nice analysis - needs looking into for sure. Haven't run across this yet even with very "hot" tracks unless downmixing, but without having had time to dig deeper certainly a float/double conversion alone can introduce tiny errors - fp math can return unexpected results with very small comparisons, e.g around 1.0 but not quite 1.0. Any soft-clipper will produce distortions by nature, and it is in there as the final blocker to hard-clipping. If you weren't downmixing there should be no values >1.0, so we need to re-check the math there. Thanks for pointing it out as we hadn't run across it at all.
