changes.txt

著名SFC模拟器Snes9x的源代码。

  the code was ignoring the screen flip value and the fact that X render
  direction reversed if the screen was flipped horizontally. Was causing a
  crash on the whale boss screen of Kirby Superstar.
- Mortal Kombat 3 now auto-adjusts emulated cycles per scan-line work-around
  a speech sample being cut short.
- Added sample data register reading support to the sound DSP - somehow I
  seem to have missed implementing this. Not sure if any ROM actually reads
  the value.
- Followed Sumire Kinoshita's suggestion and stopped clearing the ENDX flags
  when the value is read, against my better judgement, and it does actually
  improve speech samples in several games. Ooops! The Mortal Kombat series,
  Magical Drop 2 and Metal Combat are the ones I've discovered so far.
- WWF Arcade now auto-adjusts the cycles per scan-line value to work-around
  a sound sample repeat problem.
- Hmm. There's something about offset-per-tile mode I don't understand - WWF
  Wrestlemania Arcade is getting corrupt graphics; not sure what effect the
  ROM is trying to produce. Disabled offset-per-tile mode for the game for now.
- Fixed Street Racer player 1 wobble problem during the soccer game by auto-
  adjusting the cycles per scan-line value slightly.
- Made Power Rangers Fight auto-adjust emulated cycles per scan-line to work
  around a slight timing problem that causes an NMI to corrupt register
  values that an IRQ handler is trying to update. Without it the scrolling
  back-drop and fighter graphics are corrupt.
- Illusion of Gaia seems to need the mode 7 centre X & Y values to be mod 1024
  if the screen repeat flag is set. Fixes the island fly-over bug right at
  the end of the intro but breaks a few other games. Hmm. Made it auto-switch
  on for this game only.
- Added memory map support for Radical Dreamers. Thanks to satellite hut master
  for the information.
- Made updates to the top bit of the sprite write address register be ignored
  unless the low byte had been written to first. A ROM coding bug in
  James Pond II requires this, otherwise it writes a junk byte value into the
  main character's X position and Robocod wobbles around all over the place.
- Reverted back to pre 1.31 way of initialising unknown register values -
  Rock and Roll Racing was reading a junk register value and using the value
  to set up DMA, which in turn was causing corruption on the player select
  screen.
- Added Star Ocean memory map - thanks zsKnight! The original ROM I was testing
  was corrupt, no wonder I couldn't figure out the memory map myself! The game
  still isn't playable, though, due to missing S-DD1 graphics decompression
  (+ encryption?) emulation.
- Started to dump some compressed data values from Street Fighter 2 Alpha in
  the hope that one day someone will be able to crack the S-DD1's compression
  algorithm.
1.33a
- C4 emulation wasn't being automatically enabled for Rockman X2 / X3 - the
  Japanese versions of Megaman X2 / X3.
- Fixed the Super FX plot table pointer that I accidentally broke while saving
  1Mb of workspace RAM - it was stopping all Super FX games from working.
1.33
- Noticed another problem with the CPU_SHUTDOWN code - Chrono Trigger locked
  up during the intro but only when using the asm code CPU core. Found the
  algorithm difference between the code and made the CPU match what the C
  version was doing. Still not sure why it caused a problem in the first place.
- Changed colour subtraction code to use Lindsey Dubb's newer version he sent
  me some time ago but I 'forgot' to include. I say forgot, but I really put
  off including it because, although it improves most games that use the
  effect, it does result in one or two slight visual glitches.
- Hacked in zsKnight's C4 emulation asm code - now both Megaman X2 and X3 are
  playable. Still got to complete the reverse engineering of the i386 asm code
  to C so other, non-Intel ports can have C4 emulation.
- Shuffled the keyboard mapping a bit on the Linux port so now Tab key acts as
  an emulation speed turbo button, `, # and ~ act as superscope turbo and
  / acts as the superscope pause button. 
- Fixed asm CPU_SHUTDOWN code that I accidentally broke while trying to
  optimise it! Thanks to all the people who noticed Snes9x's frame skipping
  had changed between releases. Frames rates should be improved again for more
  than 50% of games.
- Re-enabled in-lining of the C SNES memory access routines, improves frame
  rate by one or two on slower machines.
- Optimised the asm 65c816 addressing mode emulation code a little.
- Included some code changes making life easier for the Mac porter, John Stiles.
- Added memory map support for Sufami Turbo using information supplied by
  Nose0000. No idea if it works because I don't have the ROM.
- Spent a few minutes trying to figure out the Star Ocean memory map so at
  least the sound effects could be heard. But gave up after a couple of hours
  due to laziness. If anyone knows the memory map details, let me know please!
1.32a
- The delay loading of the OpenGL DLLs on the Windows port was causing the
  OpenGL initialisation code to fail. Reverted back to normal DDL loading but
  with the side effect that Windows 95 users must visit the Microsoft web site
  and download the OpenGL add-on before Snes9x will work for them.
- Corrected the OpenGL bump-map display option - my attempt to get the 
  bi-linear OpenGL display option to work with Voodoo card's limited texture
  size had broken the bump-map mode.
1.32
- Changed the Windows port to delay load the two OpenGL DLLs, so now they're
  only loaded if you switch to OpenGL mode. The original version of Windows 95
  didn't include the OpenGL DDLs, so Snes9x wouldn't even start on that
  platform; now it should.
- Added yet another sound buffer option to the Windows port - this time the
  block size of sound data to mix. Some DirectSound sound card drivers only
  report the play position moving in steps rather than continuous amounts and
  Snes9x's default mix block size turned out to be smaller than this step
  value on several cards.
  Snes9x couldn't work out out where the true play position was accurately
  enough resulting in broken, noisy sound output.
- Modified the Windows frame timer code to use semaphores rather than events -
  they should make Snes9x more reliable at not missing frame sync pulses when
  Windows is busy doing background tasks.
- Added SA-1 shutdown code - basically, Snes9x now stops emulating SA-1 CPU
  instructions when the SA-1 enters an idle loop waiting for the main SNES
  CPU to give it something to do. All SA-1 run much faster and smoother now.
- Added multi-axis joystick/game controller support to the Windows port and
  tweaked the dead-zone threshold position a little.
- It looks like the SNES PPU was designed to support 128K of V-RAM but only
  64K was fitted; Snes9x wasn't wrapping all V-RAM address to stay within the
  64K limit causing a corrupt title screen on ReX Ronan - there will be others.
- Added amend functionality to the Windows Cheat Entry dialog and added extra
  text boxes for direct address and cheat value input rather than only being
  able to type in a Game Genie or Pro-Action Reply code.
- BS Suttehakkun2 was crashing just before start of play - the ROM was
  performing a junk DMA that was corrupting RAM, crashing the game when it
  went searching for a particular value.
- F-1 Grand Prix requires IRQ triggering when IRQ scan-line register set to
  current scan line, but Chuck Rock objects. Hmm. Chuck Rock seems to indicate
  the CPU emulation is running too fast, but I can't see where the mistake is.
  Special-cased Chuck Rock for now.
- Optimised SNES DMA handling slightly - copying data to SNES V-RAM is now
  significantly faster.
- Windows Cheat search dialog was ignoring data type parameter in various
  places which was causing problems when larger numbers were being searched
  for.
- Forced unknown PPU register reads to always return 0 - a coding bug in
  Equinox shows that this is required. An earlier fix didn't work.
- Puya Puya 2 & remix were objecting to an NMI being triggered when enabling
  NMIs after scan-line 226, but Ys 5 seems to require this. Hmm. Added a hack
  to support both games.
1.31
- Snes9x DirectSound code modified - the mixing block size is now always 10ms
  for Windows 95/98/2000 and 20ms for NT 4.x, now there should be no need to
  enable Sync Sound when a large sound buffer is required (helps emulation
  speed). The maximum sound buffer length values have been updated to reflect
  the smaller mixing block size.
- Changed the DirectSound code back to use an offset from the play position
  as the place to write new sample data into the sound buffer - on NT 4.x the
  write position seems to vary randomly rather than being a fixed distance
  in front of the play position as documented. Now I know why I used the play
  position originally!
- Changed the DirectSound code to fill the sound buffer at the write position
  supplied by DirectSound, rather than just before the current play position -
  should help reduce latency.
- Added an auto-detect method for interleaved mode 2 Super FX ROM images -
  well, not really auto-detect: if the game crashes and its a Super FX game,
  Snes9x assumes its in interleaved mode 2, de-mangles the ROM image and tries
  to run the game again.
- Had to update the Snes9x Windows registry version number as the additional
  diagonal settings make old registry settings incompatible.
- Added diagonal keyboard controls to the Windows port, as requested by
  several users.
- Changed PPU code to return zero when reading non-existent registers - the
  game Equinox relies on this due to an original game coding bug.
- Included FMOD sound driver support to Windows port - people experiencing
  broken sound or delayed sound, etc, might want to give it a try.
- Tales of Phantasia - un-interleaved format ROM memory map changes to match
  odd ZSNES format, now the hacked ROM works.
- Changed NMI again. Made reading or writing to PPU register 0x4210
  clear NMI pending flag again, without this Super Tennis does not work.
- Changed NMI timing back to be the same as several versions ago and just
  special cased Cacoma Knight instead - although kept the code to prevent
  the re-triggering of an NNI more than once in the same frame.
1.30
- Forgot to force GUI surface to be displayed when some dialogs where popped
  up - problem only happened on full-screen mode with triple or double
  buffering enabled, or when using 3dfx mode. It appeared as if Snes9x had
  locked up, but pressing Esc would pop down the hidden dialog.
- Added a couple of options to the Settings dialog. Now its possible to
  disable S-RAM auto-save which was causing Snes9x to write to the hard disk
  every 30 seconds on some games, causing the occasional skipped frame.
- Fixed Reset option which was accidentally broken when Netplay support was
  added.
- Added support for Dirt Racer - it leaves the Super FX chip running all the
  time, so the default CPU emulation method never allocated any time to other
  CPUs and the emulation seemed to lock up.
- NMI timing changed again. Now an NMI can only be triggered once per
  frame and enabling an NMI after the normal trigger scan line triggers
  an NMI immediately. This fixes display glitches in Ys 5, Stargate and
  Daffy Duck.
- Fixed the WAI instruction to only 'wake up' once an actual NMI has
  triggered, rather than just waking up when it should have triggered.
  This fixes Battletoads, broken since version 1.29(ish).
- Changed NMI again. Made reading or writing to PPU register 0x4210 not
  clear NMI pending flag. Seems to allow all the NMI timing sensitive ROMs
  I had on my list to now work without any special hacks. Illusion of
  Gaia now works again.
- Another NMI fix - cleared the CPU pending NMI flag at start of frame;
  Battletoads intro was crashing without this. A long DMA was stopping the
  SNES CPU so it couldn't and shouldn't respond to the NMI signal from the PPU.
- Fixed Netplay problem when game didn't have any S-RAM and Sync Using Reset
  was being used. An error dialog was displayed and the client would disconnect
  from the server.
1.30b#1
- The Windows auto-frame skip code was broken - badly. It didn't re-sync a
  timer value with timer events being generated, causing Snes9x to deliberately
  stop and wait for an event when it didn't need to, slowing down the overall
  emulation speed and increasing the number of frames skipped.
- Improved the Windows cheat search dialog - its now possible to compare
  against a value and more comparison functions are available.
- Finally worked out why Voodoo 3 support was so buggy in Snes9x - the Voodoo 3
  card generates a WM_DISPLAYCHANGE message when switching to Voodoo mode (the
  Voodoo 1 and 2 cards don't); Snes9x thought that some other application had
  changed the screen depth or resolution and tried to adjust its window to
  match - triggering another WM_DISPLAYCHANGE message. No idea how the code
  worked at all; it must have been only by chance and very dependant on the
  driver version you were using!
- Implemented Netplay on the Windows port - but its buggy as hell. I seem to
  be having major Windows multi-threading problems. Comments I've seen seem to
  suggest that Windows 95/98 don't implement true multi-threading; hmm...
- Not happy with the current Netplay, so I scrapped it and tried again;
  the protocol is much improved and not using select to control game timing
  seems to have removed lots of the threading-type problems I was having.