Quote:
* fixed Z3/RTG RAM leaking when restarting
* D3D scanlines can be (finally) enabled on the fly
* blank right border after approximate hsync position instead of filling it with border color that extends far outside of displayable area
* ECS Agnus bitplane DMA state machine update, state machine advances on OCS when DDFSTRT matches hpos and display window is open. ECS only requires DDFSTRT match. (yaqube)
* triple/double/single buffer option was never saved to configuration file
* autovsync was broken
* "old" overlays are now masks in pluginsmasks, "new" overlays are in pluginsoverlays. (rename overlays -> masks and create new overlays directory)
* both overlays and masks can be selected using GUI (overlay position can be only modified by editing configuration file)
* added Automatic center option to autoscale select menu, centers image using autoscale parameters
* audio state machine emulation rewrite, now it is fully cycle-exact, including correct DMAL DMA request line delays. (DMAL logic examined by yaqube, I was too slow and lazy..)
* disk emulation DMA accesses are now fully cycle-exact, emulates DMAL and Paula 3-word FIFO
* fixed ancient bug in disk emulation that emulated remaining cycles in current line twice when disk event triggered (index, word sync), basically made disk rotate twice the speed for about 20 or so bits of data from disk
* above disk events didn´t work correctly if multiple drives were selected at the same time (very rare situation)
* added CDROM drives to quickstart CD32/CDTV "autodetect" select menu (works on the fly too)
* cdimage0=: at startup didn´t work (only worked if changed on the fly)
* rawkeyboard hardcoded F12 check ignored focus
* drive sound emulation crashed if specific sound files were missing
* non-accurate sound enabled -option removed, difference compared to 100% accurate was practically non-existing, also now totally useless chipset sound options removed, disabled but emulated is now also 100% accurate
* files inside plugins-directory will be checked from 3 locations (until found): plugin_path, data_path and exe_path. (fixes some path issues introduced in 2.2)
In cycle-exact mode DMA fetches happen in exactly correct positions (need more CPU power), in non-cycle-exact modes all DMA fetches use audio channel 0 position (faster, less timing events needed). Complete custom chipset should be fully cycle exact now (in theory at least)
All "too fast cpu" audio hacks removed except audio.device AUDxDAT write with AUDxPER=8 and then waiting for interrupt, in JIT modes interrupt code may run fast enough to enable DMA before audio state machine has returned back to idle state. This hack forces AUDxDAT writes with AUDxPER less than 10 to finish immediately.
