!!!

!!! This file is no longer updated (and hasn't been for a while).

!!! Instead we prefer to put changelog messages in the git commit

!!! messages. The only reason to keep this file around is that in

!!! the old days the messages in this file were more detailed than

!!! the commit messages, so this file may still contain useful info

!!! when you're investigating old changes.

!!!

2010-06-06 Manuel Bilderbeek <[email protected]>

	* Released openMSX 0.8.0.

	* NOTE: this is probably the last entry in this ChangeLog. From now

	  on, just use the SVN commit log to keep track of changes.

2010-01-10 Manuel Bilderbeek <[email protected]>

	* Added Generation MSX id to software info topic. Use this to link

	  openMSX games to the Generation MSX web site.

2009-12-20 Manuel Bilderbeek <[email protected]>

	* Made softwaredb information available from Tcl, using sha1sum as

	  key. First step in making better use of the softwaredb, but mostly

	  meant to make it possible to replace the automatic ROM info printing

	  with a command to query that information on demand.

2009-11-08 Manuel Bilderbeek <[email protected]>

	* Added saving and loading of replays (in reverse feature). Thanks to

	  Wouter for doing the difficult work.

2009-09-27 Maarten ter Huurne <[email protected]>

	* Corrected palette colors on V9958:

	  The V9958 uses 5-bit DACs for R/G/B, shared by the YJK colors and

	  the V9938-compatible palette colors. The R/G/B palette colors are

	  converted from 3 bits to 5 bits in the same way the V9990 does.

	  Thanks to n_n for capturing the output of my test program a real

	  turbo R.

2009-08-02 Patrick van Arkel [email protected]

	 * Made some major changes to the ROM database

	   and added <text> tags to the <remark> field.

	   Included are also Generation MSX-ids to have

	   a common 'key' to bind databases.

2009-07-16 Manuel Bilderbeek <[email protected]>

	* Added session management script. When you do:

	  set enable_session_management true

	  your session (all machines) will be saved when exiting openMSX and

	  loaded again when starting openMSX (ignoring the machine/extensions

	  you put on the command line). You can also save, load, list and

	  delete sessions manually from the console. Default, this whole

	  feature is disabled.

2009-07-15 Sean Young <[email protected]>

	* First commit of Laserdisc code. Some of the outstanding issues

	  are listed in doc/laserdisc-todo.txt.

2009-07-09 Manuel Bilderbeek <[email protected]>

	* Added tabbed_machine_view script, visualizing multiply running

	  machines in a kind of tab widget.

2009-07-08 Manuel Bilderbeek <[email protected]>

	* Renamed 'msxonly' option to 'raw', which fits much better to what it

	  does.

2009-07-02 Wouter Vermaelen <[email protected]>

	* Major reductions in executable size for serialization code:

	   See the commit messages for details, here I'll only list one line

	   per commit and the size of the executable in each step (stripped

	   devel flavour on linux x86_64 gcc-4.3):

	      7200kB   [initial version]

	      7140kB   MemBuffer iso std::vector

	      7060kB   made serialize without ID default

	      6920kB   string="" -> string*=NULL

	      6918kB   XXX::serialize() tweaks

	      6899kB   outline OutputBuffer::insert()

	      6861kB   __builtin_constant_p

	      6590kB   beginTag(const char*)

	      6292kB   move error handling outline

	      6171kB   moved more of loadVersion() outline

	      6120kB   use const char* in PolymorphicSaver

	      6109kB   moved more of PolymorphicXXXRegistry outline

	      5781kB   give ArchiveXXX classes a non-templatized base

	      5780kB   moved ArchiveXXX::load/save() methods outline

	      5726kB   pass attrbute names as const char*

	      5710kB   outline attribute(...) calls

	   So combined this is an executable size reduction of over 1.4MB.

	   Apart from this, compilation is also noticeably faster: went from

	   8'00" to 6'23" on my machine (user time).

2009-06-07 Manuel Bilderbeek <[email protected]>

	* Apart from full screen (which was already implemented), also hide

	  the mouse pointer always when input grab is enabled.

2009-06-04 Manuel Bilderbeek <[email protected]>

	* Added Sanyo MPC-25FD, MSX2 with diskdrive and one cartridge slot

	  only. Thanks to jltursan!

2009-06-04 Wouter Vermaelen <[email protected]>

	* Fixed IE0 related VDP bug:

	   When there is a pending VBLANK IRQ (bit 7 in S#00 is set) and you

	   switch IE0 (bit 5 in R#1) on, there is immediately a IRQ send to

	   the Z80. Before this fix openMSX would only generate one on the

	   next VBLANK.

	   This fixes several long-standing bugs:

	    - [1849488] Game "Adonis" plays sound way too slowly

	    - [ 739150] Zanac: title corrupted

	    - [1847275] Penguin Adventure doesn't work on some machines

	    - [ 658430] Galaga/Bosconian slows down when you die

	   Thanks a lot to 'n_n' for providing a z80 trace of a Adonis run in

	   meisei. Comparing this trace with a openMSX trace allowed to

	   pin-point this bug.

2009-05-23 Maarten ter Huurne <[email protected]>

	* Always build application folder on Mac OS X.

	  This is the way we release and supporting both app folder and

	  systemwide installation only adds to the  maintaince effort.

2009-05-05 Max Feingold <[email protected]>

	* Added a Win32 implementation of LocalFile:mmap().

	  This should make file handling on Windows imperceptibly faster.

	* Implemented correct shutdown logic for PipeConnection thread

	  using async I/O and events.

2009-04-24 Manuel Bilderbeek <[email protected]>

	* Final step in un-uglifying Tcl scripts using Tcl namespaces :)

2009-04-19 Manuel Bilderbeek <[email protected]>

	* Added Sharp-Epcom HB-3600 dual 5.25" external disk drives, thanks to

	  FRS for contributing it.

2009-04-09 Wouter Vermaelen <[email protected]>

	* Added support for taking screenshots without OSD elements:

	  Restructured OutputSurface and child classes to make it possible

	  to render to an off-screen surface, so that it becomes possible

	  to re-render the current frame without the OSD layers.

2009-04-04 Manuel Bilderbeek <[email protected]>

	* Implemented enough of the Arkanoid pad to make it work. Use the

	  mouse to control it.

2009-04-03 Manuel Bilderbeek <[email protected]>

	* Added pointer_hide_delay setting to control visibility of the mouse

	  pointer. If you set it to -1, it will be always shown. 0 (the default)

	  means it is never shown. Positive values is the amount of milliseconds

	  it takes before it will be automatically hidden after it got shown due

	  to mouse activity. Docs are TODO.

2009-03-30 Manuel Bilderbeek <[email protected]>

	* Added a fun OSD VU-meter script, see script for details.

	  Thanks to Wouter and BiFi for their support.

2009-03-26 Max Feingold <[email protected]>

	* Secured sockets used by CliServer on Windows by requiring authentication

	  and authorizing only the user that launches the openMSX process.

2009-03-24 David Heremans <[email protected]>

	* Added new emu-state icons for set1 and replaced the default

	  ones in share/skins.

2009-03-20 Manuel Bilderbeek <[email protected]>

	* Added throttle setting to the new OSD icons (thanks Wouter).

2009-03-11 Manuel Bilderbeek <[email protected]>

	* Convert ancient history_size and remove_doubles settings to modern

	  settings. They influence how openMSX handles console history.

2009-03-09 Manuel Bilderbeek <[email protected]>

	* Added pause_on_lost_focus setting. After enabling it, openMSX will

	  pause when your cursor leaves the openMSX window. Useful in some

	  cases. This was default in fMSX, but in openMSX it is off by default.

2009-03-03 Wouter Vermaelen <[email protected]>

	* Split Y8950Adpcm sample generation part in two:

	  - There is now an emulation and an audio part. See comments in

	    Y8950Adpcm.cc for details. The emulation state is now no longer

	    dependant on the audio-thread. So in UR, the scope part keeps

	    on working when the audio is muted.

	  - Also made some minor adjustments to the rest of the adpcm code.

2009-02-24 Max Feingold <[email protected]>

	* Updated VC++ projects with a Python-based script that builds config

	  headers using mth's Python scripts. This automates one of the more

	  annoying manual steps in the VC++ build process.

2009-02-24 Wouter Vermaelen <[email protected]>

	* ZMBVEncoder motion vector search heuristic tweaks:

	  - try vectors with smallest euclidean size first (was Chebyshev size)

	    also give preference to purely horizontal/vertical motions and to a

	    lesser degree purely diagonal motions

	  - start search at best motion vector of previous block (before we

	    always started from the center)

	  - use different set of possible motion vectors, search further in

	    pure hor/ver/diag directions, less far in other (random) directions

	  Combined these tweaks give measurable smaller video files (5% - 25%)

	  and also run faster (2% - 6%). The 2nd tweak (start from previous

	  best) has the largest effect.

2009-02-23 Max Feingold <[email protected]>

	* Created a WiX-based installer for openMSX on Windows. The output is

	  an MSI file with the usual wizard-based installation UI.

2009-02-21 Wouter Vermaelen <[email protected]>

	* Refactored disk classes:

	  - simplified class hierarchy:

	     - removed virtual inheritance

	     - merged WriteProtectableDisk into SectorAccessibleDisk

	     - Disk now inherits from SectorAccessibleDisk,

	       ATM we don't support disk image formats that are lower level

	       than sector level. But even for those images it makes sense

	       to support reading/writing sectors.

	     - moved patch functionality up in the class hierarchy,

	       to SectorAccessibleDisk

	* Added syntax to insert a partition of a hard disk image:

	      <hd-image>:<num>

	  This is a preparartion to support disk partitions in the nowind

	  interface.

	* Implemented hard disk support for nowind interface

2009-02-11 Wouter Vermaelen <[email protected]>

	* Initial nowind implementation:

	  still need to work on the nowind openmsx commands

2009-02-09 Max Feingold <[email protected]>

	* Ported most significant inline assembly code to VC++ format,

	  where significance is defined as 'being noticeable in kernrate'.

	* VC++ binaries now benchmark 15% faster than mingw32 binaries on my system

	  openMSX svn VC++ x64 - 10.72 seconds

	  openMSX svn VC++ x86 - 11.19 seconds

	  openMSX svn mingw x86 - 12.62 seconds

	  openMSX 0.7.0 mingw x86 - 13.96 seconds

2009-02-07 Max Feingold <[email protected]>

	* Reduced VC++ W3 warning load to ~250, with help from Vampier and Wouter

2009-02-05 Max Feingold <[email protected]>

	* Rationalized the following definitions throughout the codebase:

	  - ASM_X86 = build assembly code on x86 and x64 platforms

	    (gcc's inline asm can target both; VC++ doesn't support x64 inline asm)

	  - ASM_X86_32 = build assembly code on x86 platforms (assumes ASM_X86)

	  - ASM_X86_64 = build assembly code on x64 platforms (assumes ASM_X86)

	  - __x86_64 = x64 target is specified

	  - _WIN64 = Windows x64 target is specified (same as _WIN32 && __x86_64)

	* These definitions should act that same for both mingw32 and VC++.

2009-02-01 Wouter Vermaelen <[email protected]>

	* Implemented 'ld a,i' Z80 quirk:

	  See comments in CPUCore.cc for details. Thanks to n_n for reporting

	  this issue (first discovered by GuyveR800) and writing a test

	  program for it.

2009-01-29 Max Feingold <[email protected]>

	* Unicode filename support on Windows platforms

	  - On Windows, standard CRT functions and 'ANSI' Win32 API calls

	    expect and return ANSI encodings, not UTF8. Unicode support on

	    Windows is generally achieved by calling CRT functions that accept

	    wchar_t and 'wide char' xxxW Win32 API calls.

	  - openMSX uses std::string and UTF8-encoded strings throughout.

	    Consequently, on Windows we wrap all system calls that might

	    touch the filesystem in UTF8-to-UTF16 conversions before calling

	    the 'wide char' API and (if necessary) converting back to UTF8.

	* As a side effect, this checkin also deprecates openMSX support for

	  Windows 95-based platforms, as well as for Windows NT 4.0.

	  At the time of writing, only Windows 2000 and above are supported.

2009-01-25 Maarten ter Huurne <[email protected]>

	* Use Python to generate "Version.ii".

	* Use Python to generate "build-info.hh".

2009-01-18 Max Feingold <[email protected]>

	* Fixed bug [ 1206036 ] Windows Control Menu while using ALT

	  With this fix in place, the ALT+SPACE combination no longer

	  pops up an annoying context menu on Windows.

	* The solution was to register our own winproc, filter out the

	  offending WM_SYSCOMMAND events, and delegate the rest to SDL.

2009-01-20 Wouter Vermaelen <[email protected]>

	* Console refactoring:

	  - Share much more code between SDL and GL console variant

	  - Use TTF fonts instead of a .png file with the ASCII symbols:

	    console can now also show unicode chars (if the ttf font has them)

	  - Added new setting: consolefontsize

	  - Removed (now unused) classes:

	       Font, DummyFont, SDLFont, GLFont, SDLConsole, GLConsole

	* Reverted part of last CPU refactoring (2008-01-15):

	  - Omitting needExit() check on some instructions can causes

	    acceptance of an IRQ to be a couple of instructions late,

	    see comments in src/cpu/CPUCore.cc for details.

	  - But even without this part, the threaded interpreter model is a

	    big win (more than 10% faster) over the old model.

2009-01-23 Max Feingold <[email protected]>

	* Checked in VC++ project files that build dependencies and openMSX itself.

	  The project files are in VC9 (Visual C++ 2008) format and should be usable

	  with the Windows SDK's msbuild tool, as well as with the free (as in beer)

	  Visual C++ 2008 Express Edition.

	* This checkin also brings the first 64-bit Windows build of openMSX.

	  x64 binaries are fully functional and I'm pleased to note that I found no

	  significant 64-bit portability issues in the codebase (inline asm aside).

2009-01-17 Wouter Vermaelen <[email protected]>

	* Initial implementation of the VDP sprite collision coordinate status

	  registers:

	  Not yet 100% correct (see top of src/video/SpriteChecker.cc for some

	  TODOs). But at least 'Ace in Space level 3' works now. So far this

	  seems to be the only game that actually uses this feature.

	  Fixes [2510696].

2009-01-15 Wouter Vermaelen <[email protected]>

	* Major changes in CPU emulation core:

	  I've worked on this patch on a local (private) branch for several

	  weeks. Actually it are several related patches and ideally I should

	  also commit it as seperate patches. However developing on a private

	  branch means (at least for me) do a couple of commits, detect a

	  mistake in one of the earlier commits, make a new commit that

	  corrects that mistake and so on. Usually I do the extra effort to

	  extract a clean patch-series from my local branches before committing

	  them. However this time that would require quite some extra time, and

	  as you guessed already I was too lazy to spend that time. Though as

	  compensation I wrote some extra doc about the inner workings of our

	  CPU emulation (see src/cpu/CPUCore.cc)

	  These are more or less the steps that are combined in this commit:

	  * Put instruction selection into one big function:

	     this allows to 'goto' between all part of this code (required

	     in later steps)

	  * Move opcode fetching (initial byte) and time accounting into

	    this function (preparation for next step)

	  * Allow to execute multiple instructions inside this function

	  * Use gcc extension 'computed goto' to implement faster dispatch

	  * Added fallback for compilers that don't have this extension (VC++)

	  * Simplified 'exit-test', see doc in CPUCore.cc for details

	  * Some smaller time bookkeeping optimizations:

	    - IX/IY instructions updated clock in two steps -> merged into one

	      step

	    - Some helper routines took a 'extra-cycles' function parameter.

	      Turned this into a template parameter to force the compiler to

	      treat it as a compile-time-constant. (Actually gcc already did

	      this optimization by itself most of the time when the function

	      got inlined, but now the optimization is guaranteed.)

2009-01-10 Wouter Vermaelen <[email protected]>

	* Implemented difference between AY8910/YM2149 reading of registers:

	   - thanks to enen for doing most of the investigation on real MSX

	     machines

2009-01-08 Wouter Vermaelen <[email protected]>

	* Merged (part of) mfeingol's work to make openMSX compile with VC++:

	  - code still compiles with gcc, but i couldn't test myself with VC++,

	    so probably this work is not finished yet ;)

	  - build system is still TODO, these are only changed in the C++

	    source files

2009-01-07 Wouter Vermaelen <[email protected]>

	* Released openmsx-0.7.0

2008-12-20 Wouter Vermaelen <[email protected]>

	* Updated to cbios 0.22

2008-12-20 Wouter Vermaelen <[email protected]>

	* For carta command, check for errors (rom and ips files)

	  before removing the currently inserted cartridge.

2008-11-25 Wouter Vermaelen <[email protected]>

	* Added '-repeat' option to 'bind' command:

	   - When the '-repeat' options is given and the key remains pressed,

	     the command will be repeat (just like regular key repeat while

	     typing).

	   - Used this feature in the OSD menu.

2008-11-01 Wouter Vermaelen <[email protected]>

	* Added possibility to take screenshots of only the msx screen (so no

	  OSD stuff and console):

	   For the moment this is done by passing the option -msx to the

	   screenshot command, this will take a screenshot of size 640x480.

	   There will also be no scaler or any other effects in the screenshot.

	   You can use -msxsmall to take a screenshot of size 320x240 (-msxbig

	   is a synonym for -msx).

	   TODO is the syntax of this command ok? Also still need to document

	   it.

2008-10-29 Wouter Vermaelen <[email protected]>

	* Optimized VDPCmdEngine:

	  The core of most VDP commands goes like this:

	     1 transform x,y coordinate(s) to vram address(es)

	     2 do some operation on that address(es)

	     3 increase (or decrease) coordinates to loop over the whole

	       rectangle (mostly only the x-coordinate changes)

	     4 repeat if there is still time left

	  The coordinate to address transformation requires quite some shift

	  and logical operations. In the next iteration all these operations

	  are repeated.

	  In this patch the vram address is calculated incrementally: we

	  calculate it once at the beginning of the loop and then adjust the

	  address when the x-coordinate changes. We have to take care of vram

	  interleaving (screen 7 and 8) and that several pixels map to the same

	  address (screen 5, 6, 7).

	  A second optimization is to simplify the stop-condition for the

	  innermost loop. Before we broke out of this loop either when

	  destination time was reached or when we reached the end of a

	  horizontal line (implemented as two seperate tests). Now we calculate

	  up-front which of these two condition will occur first and iterate

	  that many times (with only one test).

	  I measured between 4% and 8% speedup on total emulation time for the

	  Space Manbow intro demo (renderer=none, sounddriver=null).

2008-10-25 Manuel Bilderbeek <[email protected]>

	* Added Sony HB-10P and Sony HB-55P, thanks to Rudi Westerhof. Also

	  added Sony HB-20P a few days ago, thanks to MSXKun.

2008-10-16 Manuel Bilderbeek <[email protected]>

	* Implemented rough, but measurement-based timing for V9990 for

	  commands LMMV, LMMM, BMXL, BMLX and BMLL. Code by Wouter. Original

	  measurement programs by Peter Mastijn. Measurements by Wouter and

	  me, using Wouter's more generic measurement program. Fixes [421861]!

2008-09-25 David Heremans <[email protected]>

	* Vram debuggable changed:

	  First step towards creating a debuggable that provides a logical

	  view upon the VRAM based upon the current videomode. This is the

	  behavior an ML-coder expects when dealing with the VDP.

	  The old chipbased view upon the vram is now called "physical VRAM"

	  instead.

2008-09-21 Manuel Bilderbeek <[email protected]>

	* Added Brazillian OPL3Cartridge (MoonSound without wavetable).

	  Still mostly untested, although MSX-Audio games (e.g. from Compile)

	  seem to work with it. See RFE tracker item 1815450.

2008-08-28 Wouter Vermaelen <[email protected]>

	* Added 'probe' debug infrastructure:

	   see 'help debug probe' or (doc/manual/command.html)

2008-08-20 Wouter Vermaelen <[email protected]>

	* Made Disk and SectorAccessibleDisk write-protectable:

	   Added WriteProtectable interface and refactor code so that really

	   all write methods check this write protected flag.

2008-08-20 Wouter Vermaelen <[email protected]>

	* Split MSXtar class into two parts:

	    - DiskImageUtils: this part knows about partitions and format

	    - MSXtar:         this part knows about files and directories

	  Rational: before this change, the interface of the MSXtar class was

	  a bit confusing, especially the usePartition() method. Depending on

	  which MSXtar functionality you plan to use, you either must call it,

	  were not allowed to call it or must call it but ignore exceptions

	  thrown by it.

	  To simplify this, I split the functionality of the MSXtar class in

	  two parts. Now MSXtar doesn't know anything anymore about partitions

	  and formatting (of course it's still possible to work on partitions,

	  but that's now abstracted via the new DiskPartition class).

2008-08-16 Wouter Vermaelen <[email protected]>

	* Fixed some UMRs triggered during savestate:

	  Before savestates this was not a problem because those value _are_

	  initialized before they are used in actual msx emulation code. Though

	  savestates can read these values earlier. And if the variable was an

	  enum, it could even lead to an assert.

	  Thanks to Vampier for reporting this bug.

2008-08-16 Wouter Vermaelen <[email protected]>

	* Diskmanipulator changes:

	  - Teach diskmanipulator about multiple active msx machines.

	    Before this change 'diska' always refered to the 1st drive of

	    the oldest machine. Now it refers to the active machine. It's

	    also possible to explicitly qualify the wanted machine with

	    syntax like 'machine2::diska'.

	  - Improved check on whether the specified partition exists or

	    not.

2008-07-24 Wouter Vermaelen <[email protected]>

	* Changed our Unicode code with UTF8-CPP code:

	       http://utfcpp.sourceforge.net/

	    This fixes the TODO's left in our Unicode code. But it's mainly

	    dome as preparation to serialize the Keyboard class (because a

	    'type' command in progress doesn't survive a savestate ATM). The

	    old code in the Keyboard class contained a UTF16 encoded string,

	    but the serialization framework can't (directly) serialize that.

2008-07-24 Wouter Vermaelen <[email protected]>

	* Optimized de-serialization of arrays:

	   Directly deserialize the elements in the correct position in

	   memory. Before the element was loaded in a local variable and

	   then copied to the correct position in the array.

	   This also fixes a bug in AY8910 when detune/vibrato was used,

	   because copying a AY8910::ToneGenerator object is not allowed.

2008-07-20 Manuel Bilderbeek <[email protected]>

	* Renamed Midi{In,Out}Native to Midi{In,Out}Windows (untested!)

	* Made save/loadstate commands a bit more comfortable, so we can

	  easier test this stuff :)

2008-07-19 Wouter Vermaelen <[email protected]>

	* Enabled savestates to memory (as opposed to XML files):

	   - Currently available via the 'mem_savestate' and 'mem_loadstate'

	     commands, but this will still certainly change.

	   - Speed seems to be quite good:

	       savestate MSX1:          125us

	                 MSX2:          310us

	                 TurboR:        734us

	                 boosted MSX2: 3770us

	       This between 200 and 400 times faster than the XML savestates!

	       So it seems to be fast enough to implement a 'reverse-time'

	       feature.

2008-07-14 Wouter Vermaelen <[email protected]>

	* Serialize pluggables (WIP):

	   Current code serializes all pluggables even those that are not

	   plugged in. The available pluggables depends on the platform (e.g.

	   the Midi{In,Out}Native pluggables on windows), the openmsx

	   configure options (e.g. CassetteJack), the openmsx version (more

	   pluggables in newer versions) and even the state of the host

	   machine (connected a USB joystick). ATM loading a savestate only

	   works when the exact same list of pluggables is available.

	   Maybe we should only serialize those pluggables that are actually

	   plugged in in the emulated machine (TODO verify that pluggables

	   don't keep state over unplug-plug cycle).

	   ATM only the state of the pluggables itself is serialized. We don't

	   yet restore the actual plug state (which pluggable is plugged in

	   which connector).

2008-07-05 Wouter Vermaelen <[email protected]>

	* Merged Rom{4,8,16}kBBlocks classes into one templatized class

2008-07-01 David Heremans <[email protected]>

	* Fixed two bugs in the diskmanipulator:

	   - one preventing correct 360KB disk creation on some machines

	   - one valgrind warning about variable initialization

2008-06-30 Wouter Vermaelen <[email protected]>

	* More serialization work:

	   - Philips_NMS_8250 can be serialized now, but still very

	     few other machines or extensions are supported

2008-06-28 Wouter Vermaelen <[email protected]>

	* Added HIGHLY EXPERIMENTAL savestates:

	   - Only very few things work yet, but I was able to resume

	     'Road Fighter' running in a MSX1 (actually that's about the

	     most complex setup that already works).

	   - There are two new commands (still VERY primitive)

	         savestate <machineID>

	         loadstate

	     'savestate' will create a openmsx.xml.gz file in the current

	     directory. 'loadstate' will read this file and create a new

	     machine based on this state. You can use 'activate_machine

	     <machineID>' to switch to that machine.

2008-06-18 Wouter Vermaelen <[email protected]>

	* Require extension hardwareconfig.xml files to list devices inside a

	  <devices> tag. We already updated all our device configs for this

	  over two years ago.

2008-06-17 Wouter Vermaelen <[email protected]>

	* Implemented TMS99x8 VRAM remapping (reg#1, bit 7):

	   thanks to: - enen for providing the technical info

	              - Quibus for testing it on both real and emulated

	                machines

2008-06-14 Wouter Vermaelen <[email protected]>

	* Implemented 'tabbed' machines:

	  It's now possible to have multiple msx machines in memory and freely

	  switch between them. This is done with the already existing

	  'activate_machine' command. Only the active machine is running, the

	  other machines are paused.

	  This may not be the most useful feature, but we already had the

	  'activate_machine' command for other reasons (e.g. machine

	  switching) and you could argue that it's a bug that it didn't

	  already work like it does now. Basically we already had all the

	  infrastructure in place. Actually enabling session support required

	  relatively few changes (mostly mechanical). Also having this feature

	  working gives more confidence that the design is right.

2008-06-12 Wouter Vermaelen <[email protected]>

	* Simplified machine handling stuff in Reactor:

	  Before we had some (complicated) logic in the Reactor class to

	  postpone the actually machine switch till the main reactor loop has

	  had a chance to run. The reason for this was that the 'machine'

	  command (or other related commands) can get executed via a code path

	  that runs through the current machine, meaning there are stackframes

	  active that refer to the current machine. Only in the main reactor

	  loop we're sure there are no such stackframes. A consequence of this

	  approach is that for example in this sequence of Tcl commands:

	     machine turbor ; diska mydisk.dsk

	  The 'diska' command is still executed in the old machine, because

	  the reactor loop has not run yet. If there is 'some pause' inbetween

	  the two commands, the 'diska' command is executed in the new

	  machine. At the very least this is not very intuitive.

	  Taking a step back, switching the active machine (which has active

	  stackframes) was not the problem, but deletion of that machine is.

	  So instead of postponing the whole switch, we now only postpone the

	  deletion of the old machine. The Tcl commands now work as expected.

	  Also did some other refactoring to make the code easier to

	  understand.

2008-06-12 David Heremans <[email protected]>

	* Fixed a small bug in DirAsDisk:

	  Read-only host-OS files became truncated in the MSX dsk image.

2008-06-05 Wouter Vermaelen <[email protected]>

	* Fixed compilation of CPU code with gcc-3.4:

	   Compilation with gcc-3.4 was broken because of a compiler bug,

	   gcc-4.x works just fine. Worked around it. Though I still like

	   to demote gcc-3.4 support to second class.

2008-06-04 Alex Wulms

	* Added French version of VG8010/VG8020 and corresponding unicodemap files

2008-06-04 Wouter Vermaelen <[email protected]>

	* Fixed V9990 vertical scroll behaviour:

	  Based on MRC forum post by GhostwriterP (page 3, post 12)

	      http://www.msx.org/forumtopicl8574.html

	  1) Writing to high byte of scroll register only takes effect at the

	     start of the next frame. This is for both scroll A and B

	     registers.

	  2) Writing to low byte has effect the next line. The next line that

	     will be displayed is the line number just written to the register

	  3) point 2) has no effect on the sprite vertical sprite position

	     (this was wrong in the initial openmsx implementation)

2008-06-03 Wouter Vermaelen <[email protected]>

	* Added -clip option to OSD widgets:

	   - If this option is true, then child widgets are clipped to the

	     bounding box of this widget.

	   - Used in the OSD menu script to make sure (long) text lines don't

	     gun over the end of the menu window.

	* V9990 display enable only takes effect at the next frame

2008-05-28 Wouter Vermaelen <[email protected]>

	* Reduced code duplication in CPUCore by templatizing various stuff:

	  - For example, a lot of instruction methods only differ in the

	    registers they act upon (e.g. inc_hl() vs inc_de). By making

	    templatized helper functions to read/write registers, a lot of

	    these instruction methods can be combined into a single

	    templatized instruction method.

	  - This reduced code size by almost 2000(!) lines.

	  - The generated code is also about 100kB smaller. I'm not sure

	    why exactly, but I think because it doesn't need to keep a

	    non-inlined version of the templatized instruction methods in

	    the final executable.

	  - Speed of the CPU code has not changed.

2008-05-27 Wouter Vermaelen <[email protected]>

	* Tuned R800 refresh (still WIP)

2008-05-26 Wouter Vermaelen <[email protected]>

	* Implemented dynamic IO port allocation for Panasonic MSX-AUDIO:

	   fixes: [1231256] Panasonic MSX-AUDIO not detected for games

2008-05-24 Maarten ter Huurne <[email protected]>

	* Fixed AVI recording (ZMBV encoding, to be exact) on big endian CPUs.

2008-05-24 Wouter Vermaelen <[email protected]>

	* Fixed R800 page-break behaviour:

	   - did a lot of measurements on a real turbor, see

	       doc/r800test.txt

	   - refactored CPU code to match this (WIP)

2008-05-21 Wouter Vermaelen <[email protected]>

	* Reading sectors from a disk image via the SectorAccessibleDisk

	  interface didn't apply the IPS patches:

	    This had as effect that accessing the disk in the emulated

	    machine did show the effect of the IPS patches, but using

	    the diskmanipulator on the same disk ignored the IPS patches.

2008-05-19 Manuel Bilderbeek <[email protected]>

	* Final fixes on Nettou Yakyuu mapper, including enhanced sample

	  player.

2008-05-17 Manuel Bilderbeek <[email protected]>

	* Added WIP version of Nettou Yakyuu mapper, used by the game "Moero!!

	  Nettou Yakyuu '88". Thanks to hap/enen for the research and blueMSX

	  implementation and Wouter for teaching me about this stuff.

2008-05-16 Wouter Vermaelen <[email protected]>

	* Fixed crash in DirAsDsk / DiskManipulator:

	   Occured when a diskimage used a different end-of-fat marker than

	   0xFFF. In fact the whole range 0xFF8-0xFFF should be interpreted

	   as EOF. See:

	      http://en.wikipedia.org/wiki/File_Allocation_Table#File_Allocation_Table

2008-05-03 Wouter Vermaelen <[email protected]>

	* CPU time keeping speedup:

	  - We want to have cycle-accurate bus transactions (memory or IO-port

	    read/writes). So for example if in one instruction there are

	    multiple transactions, they should happen at slightly different

	    moments in time. Also the time of the first transaction is not

	    necessarily the same time as the start of the instruction.

	    Initially we implemented this by incrementing the 'current time' by

	    very small steps at each sub-instruction.

	    We actually only need this time for very few transactions. The vast

	    majority of the transactions goes to ROM or RAM (IOW to 'cachable'

	    memory regions), and there the exact time is not important at all.

	    To speedup this time keeping, we now only increment the 'current

	    time' with the whole instruction cycle count. In case we do

	    perform a transaction to non-cachable memory we still first have

	    to calculate the current time. To be able to do this all

	    transactions have been annotated with 'cycles since start of the

	    instruction' information.

	    I measured between 5% and 10% speedup on the ZEXALL benchmark with

	    this change.

2008-04-30 Wouter Vermaelen <[email protected]>

	* Optimized check for pause in CPU code

2008-04-29 Wouter Vermaelen <[email protected]>

	* Use switch statement instead of jump table in CPUCore:

	  - Is about 10%(!) faster when using gcc-4.2, more when using

	    recent gcc snapshot. With gcc-4.1 it's performance neutral,

	    gcc-3.4 was a big slowdown (but we care less about such old

	    compilers).

	  - Some history:

	     I had already tried this before, but at that time it wasn't a

	     win. Recent gcc versions are able to eliminate the check for the

	     'default' case when it can prove all cases are enumerated in the

	     switch statement (here we switch on a byte type and provide all

	     256 cases). With this optimization the switch approach gave a

	     small speed gain. Then I tried some additional tweaking (force

	     inlining some core functions). And this gave a much bigger gain,

	     even on older compilers.

	  - Because we no longer use a jumptable there is no advantage anymore

	    in using static functions (with an explicit 'this' argument) over

	    regular member functions. This makes the code more readable again.

2008-04-08 David Heremans <[email protected]>

	* Fixed a bug in the load_icons script

2008-04-07 David Heremans <[email protected]>

	* Added a new 'fancy' skin

2008-04-06 Alex Wulms

	* Review unicodemaps for MSX models with 'international' font:

	   - There were some errors in the maps. Especially for

	     graph and shift-graph key combinations.

2008-04-06 David Heremans <[email protected]>

	* New load_icons script:

	   - Added suport for frame.png

	   - Added time-before-fade-starts

	* Some extra comments in the DirAsDisk

2008-04-05 Wouter Vermaelen <[email protected]>

	* Added share/skins/Vera.ttf.gz:

	   - also added support for compressed font files

2008-03-30 Alex Wulms

	* Add support for Philips VG8010:

	   - Add new unicodemap unicode.proto_int; the VG8010 has a keyboard

	     matrix that differs completely from all other MSX models. It seems

	     to be some kind of prototype of the keyboard matrix

	   - Add 'GRAPH locks' processing logic to the copy&paste routine in

	     the keyboard driver (GRAPH has a locking behaviour on VG8010)

	* Fix DEAD key behaviour in keyboard driver in KEY mapping mode:

	   - In KEY mapping mode, RCTRL key must be mapped to row 5 col 2 in the

	     keyboard matrix (DEAD key on 'international' MSX keyboard) or to

	     whatever other position if user has loaded another KeyToKey mapping

	     file. Either way, it has to follow the 'Key to Key' mapping logic

	   - In CHARACTER mapping mode, RCTRL key must be mapped according to

	     the DEADKEY position in the unicode map files (for the MSX models

	     that do have a DEAD key)

2008-03-26 Wouter Vermaelen <[email protected]>

	* Fixed Z80 timing of accepting a IRQ:

	   - Accepting a IRQ in IM0 or IM1 should take 13 cycles, there is no

	     extra wait cycle. For IM2 it takes 19 cycles (also no extra wait

	     cycle).

	   - Thanks to enen for testing this

2008-03-25 Wouter Vermaelen <[email protected]>

	* Added -fadeTarget and -fadePeriod properties for OSD widgets:

	   - This allows to smoothly fade the alpha value of a widget without

	     further interaction with this widget. (I mean you don't have to

	     manually change the alpha value in small steps.)

	   - Allows to greatly speedup the OSD icons Tcl script. (In the GP2X

	     port, it took a lot a time)

2008-03-16 Manuel Bilderbeek <[email protected]>

	* Some small improvements in the cassette player:

	  - clearer errors when both a wav and a cas image fail to work

	  - properly flush the recording when switching to play mode

	  Still needs more work to properly handle 'zero-recordings' after

	  switching to play mode, see comments in the class.

2008-03-12 Wouter Vermaelen <[email protected]>

	* Replaced C++ IconLayer with a Tcl script

2008-03-12 Wouter Vermaelen <[email protected]>

	* Added 'user_setting' command:

	   This allows to create openMSX settings from Tcl scripts. The main

	   advantage of settings over regular Tcl variables is that their

	   value is preserved over different openMSX sessions.

	   See 'help user_setting' for more details.

2008-03-08 Manuel Bilderbeek <[email protected]>

	* Added -doublesize switch to the 'record start' command, to record at a

	  size of 640x480. Note: this is usually a lot slower...

2008-03-07 Wouter Vermaelen <[email protected]>

	* Fixed bug in OSD rectangle drawing:

	   As an optimization rectangles with alpha=0 (completely

	   transparent), were not allocated in memory. This caused problems

	   when we needed the dimensions of this rectangle to calculate the

	   position of child widgets. Fixed by refactoring the transformation

	   code.

2008-03-05 Wouter Vermaelen <[email protected]>

	* Added 'openmsx_info realtime':

	   It was already possible to query the emutime of a machine in

	   Tcl, now it's also possible to query the realtime.

	* Added '-scaled' OSD property:

	   OSD widgets with this property (or child widgets whose parent has

	   this property) are automatically scaled according to the

	   'scale_factor' setting. This makes it possible to design a GUI

	   independent of the host window resolution.

2008-03-03 Wouter Vermaelen <[email protected]>

	* Extended OSD stuff:

	   - widgets are now hierarchically structured (the 'osd create'

	     subcommand has been changed to allow this)

	   - child widgets are placed relative to their parent widget

	   - there can be an offset relative to parent in pixels (in the

	     future probably also in 'scaled' pixels)

	   - or offset can be as a percentage of the bounding box of the

	     parent widget

2008-03-02 Alex Wulms

	* Several updates to keyboard driver:

	  - Rename all keyboard settings to start with kbd_ so that they

	    are grouped together when listed in alphabetic order, like in

	    the documents.html file and in the tab expansion of the

	    various commands in the console

	  - Add support for a few keys that are specific to the jp106

	    keyboard model (the most used keyboard in Japan)

2008-02-25 Wouter Vermaelen <[email protected]>

	* Added 'MatraShockware' mapper type:

	   This is a non-bankswitched mapper that uses a AMD flash ROM. The

	   flash is wired-up in such a way that writes end up in a different

	   region as reads. The effect is that the flash' autoselect function

	   tells you that a region is not write-protected, but when you

	   actually try to reprogram that region it has no effect. The game

	   'INK' actually uses this as a sort of copy protection (if this

	   check fails, stage 2 goes back to stage 1).

	   See also this forum thread:

	     http://www.msx.org/forumtopic8329.html

2008-02-25 Wouter Vermaelen <[email protected]>

	* Added new mapper type: plainflash

	   It's plain mapper type (no bankswitching), but implemented

	   with a Flash Rom, so it reacts to certain write-sequences.

	   According to this forum topic

	      http://www.msx.org/forumtopicl8329.html

	   the Matra INK game uses this as a sort of copy protection.

2008-02-25 Alex Wulms

	* Several updates to keyboard driver:

	   - Implemented auto-toggle of KANA-lock (and CODE-lock on VG8010)

	     based on the character that the user enters. This is only done on

	     MSX models on which CODE/KANA has a locking behaviour. Though,

	     user can switch it off via auto_toggle_code_kana_lock setting.

	   - Implemented auto-toggle of CODE/KANA-lock when pasting characters

	     (either through catapult or via the type command in the console).

	     This is always done on MSX models on which the CODE/KANA locks.

	   - Implemented auto-toggle of CAPSLOCK when pasting characters. This

	     is especially required for japanese models, which use CAPSLOCK to

	     differentiate between hiragana and katakana entry mode when

	     CANALOCK is on.

	   - Added a mnemomic DEADKEY to the unicode maps, to indicate the

	     position of the dead key in the MSX matrix (it is different on

	     german MSX models)

	   - Added setting keyboard_mapping_mode, so that user can select

	     between KEY to key mapping mode and CHARACTER to key-combination

	     mapping mode.

	   - Refactored keyboard driver; unicode mapping data is now in its own

	     class and it uses a std::map to store the mapping data in stead of

	     a (sparse) 64k entries matrix.

2008-02-23 Wouter Vermaelen <[email protected]>

	* Added low level OSD Tcl commands:

	   - It's now possible to draw OnScreen filled-rectangles, images

	     and text. See 'help osd' for more details.

	   - Font rendering is done with SDL_ttf. This is a new library

	     dependency.

	   - This is still very much work in progress.

2008-02-13 Wouter Vermaelen <[email protected]>

	* Added 'horizontal_stretch' setting:

	   - amount of horizontal stretch can now be configured (some

	     users repeatedly asked for this ;-)

	   - removed horizontal_stretch enum from display_deform setting

	   - as a bonus horizontal stretch can now also be used in

	     combination with the 3D deform mode

2008-01-28 Alex Wulms

	* Initialized key_ghosting_sgc_protected parameter in all

	  hardwareconfig.xml file:

	    Initial value of the parameter is derived from keyboard

	    code at ROM BIOS address 0x2c; for Japanese machines,

	    it is set to false and for all others to true (this is based

	    on tests with some real machines)

2008-01-24 Wouter Vermaelen <[email protected]>

	* Introduced PLATFORM_GP2X #define,

	  don't call getpwuid() if this #define is set

2008-01-23 Wouter Vermaelen <[email protected]>

	* Conditionally compile 16/32bpp and scaler code:

	   GP2X port only needs 16bpp and scale factor 1

2008-01-21 Wouter Vermaelen <[email protected]>

	* Removed workaround for pre gcc-3.4 compilers

2008-01-18 Alex Wulms

	* Added key-ghost protection for SHIFT, GRAPH and CODE keys:

	    This exists on several real models; these three keys are connected

	    to row 6 via a diode (one diode per key)

	    Have made it configurable through hardware.xml. Default is that

	    the protection is on

2008-01-17 Wouter Vermaelen <[email protected]>

	* Refactored OutputSurface class:

	   - Made all members private (was protected) and added protected

	     getter/setters. This makes the interface between base and sub

	     classes explicit.

	   - Renamed getFormat() to getSDLFormat() to make it clear that this

	     method still works dircetly on SDL data structures.

	   - Added lock(), unlock() and isLocked() methods. These wrap the

	     SDL_LockSurface() and SDL_UnlockSurface() functions. It for

	     example allows to check that getLinePtrDirect() is only called on

	     a locked surface.

	   The first two changes allow for easier refactorings. The last change

	   is a step towards enabling the use of HW accelerated SDL-blit

	   functions on the GP2X port.

	* Move calls to OutputSurface::lock() / unlock() to a lower level:

	   Every function that wants to access pixels directly should now make

	   sure the surface is locked. Similarly a function that wants to use

	   blit operations must unlock the surface. This allows for the

	   minimal amount of locked<->unlocked transitions. (Locking while

	   already locked or unlocking while already unlocked is very fast.)

2008-01-16 Alex Wulms

	* Added German keyboard map (unicodemap.de):

	    It is needed for Sony HB-F700D. Please note that currently it gives

	    a key-ghosting effect when entering { character; on german MSX this

	    character must be entered by pressing SHIFT, CODE and ( keys

	    SHIFT and CODE are on same ROW

	    SHIFT and ( are on same COLUMN

	    Hence the < character gets 'ghost pressed' (in same column as CODE)

	    Need to figure out how this works on the real machine. Does it contain

	    hardware to prevent key ghosting?

2008-01-12 Wouter Vermaelen <[email protected]>

	* Optimized SCC::generateChannels()

2008-01-11 Wouter Vermaelen <[email protected]>

	* Optimized CheckedRam for the case there is no umrcallback set:

	   In some cases (e.g. in manbow2) it still caused some overhead:

	   Internally openMSX divides memory in 256 byte chunks (see

	   ChacheLine.hh) as soon as all 256 bytes in a chunk have been

	   written to, read/writes to this chunk happen without overhead.

	   Usually this happens early in the MSX boot process or startup code

	   of a game. However manbow2 has 2 or 3 chunks that are read very

	   frequently but only partly initialized.

	   Solved by treating the memory completely initialized in case the

	   'umrcallback' settings is empty (no callback function registered).

2008-01-08 Wouter Vermaelen <[email protected]>

	* Fixed reading of wav cassette images on Windows:

	   Some versions of(?) windows can't open the same file multiple

	   times in read-write mode. But this is exactly what happened by

	   using the (now removed) File::getLocalFile() method combined with

	   SDL_LoadWav(). In the past (2007-08-29) we already solved a similar

	   problem with IMG_Load(). We solved this by deleting the File object

	   before using the path returned by getLocalFile(). This works for

	   regular files, but not for compressed files. I like to gzip my wav

	   images so we need a different solution.

	   To solve this I removed the old getLocalFile() method and created a

	   new class LocalFileReference. This class makes sure there is no

	   open file reference anymore after its constructor finishes and that

	   there is a local uncompressed version of the file for at least as

	   long as the lifetime of the corresponding objects this class.

2008-01-07 Wouter Vermaelen <[email protected]>

	* in CPUCore prefer to use 'unsigned int' over 'word' datatype:

	   Results in smaller and faster code. On x86, instructions that work

	   on 32-bit are shorter than those for 16-bit. Code size reduced

	   about 10kB, code is slighter faster (depends a lot on compiler

	   version and optimization options). ARM CPUs don't even have 16-bit

	   operations, only (limited) 16-bit load/store instructions. Here code

	   size reduced about 20kB and is about 4% faster.

2008-01-05 Wouter Vermaelen <[email protected]>

	* changed SoundDevice::updateBuffer() method:

	   As parameters this method (still) receives a number of buffers,

	   each buffer should be filled with the audio-output of one channel.

	   Before this change the existing data in the buffer should simply be

	   overwritten. Now the new data should be added to the existing data

	   in the buffer.

	   This new approach allows to let several channel buffers actually

	   point to the same underlying buffer (in fact only a muted or

	   recorded channel will still get it's own private buffer). This

	   allows to (mostly) skip the mix-channels step.

	   I measured on my core2duo CPU and this new approach is slightly

	   slower (a few percent) than the old one. If I disable SSE2

	   optimizations in mixChannels, it's about the same speed. However on

	   older CPUs without SSE2 and with smaller caches (like ARM) it's

	   definitely faster. I guess because this new approach is more

	   cache-friendly.

2008-01-03 Maarten ter Huurne <[email protected]>

	* Removed execution of ~/.openMSX/share/init.tcl on openMSX startup.

	  We have a "scripts" dir which can be used instead.

	  This avoids an infinite loop if the user copies the openMSX internal

	  version of "init.tcl" into the user dir.

2008-01-03 Wouter Vermaelen <[email protected]>

	* Created 5 new commands:

	     - create_machine   (returns machine ID)

	     - delete_machine <machine-ID>

	     - list_machines    (return list of machine IDs)

	     - activate_machine <machine-ID>

	     - <machine-ID>::load_machine <config-name>

	  - These commands allow to implement the existing commands 'machine'

	    and 'test_machine' as Tcl procs (not yet done).

	  - This also allows to test extensions with the existing 'ext'

	    command.

	  - It also allows to have multiple running msx machine in memory,

	    though ATM we always show the video/sound of the oldest machine.

2008-01-02 Wouter Vermaelen <[email protected]>

	* Added DivModByConst and DivModBySame utility classes:

	   - these allow to replace a 64-bit by 32-bit division or modulo

	     operation by a multiplication/addition/shift. This helps if

	     the divider is a constant, especially on CPUs without division

	     instructions (like ARM)

	* Always use 512kb flash for Manbow2 and MegaFlashRomScc mapper types:

	   Flash RAM saving now works in 'Lotus F3'.

2008-01-01 Patrick van Arkel ([email protected])

	* Added Original ROM tag to database and added tag handeling to

	  RomInfo.cc.

	* Changed list.php on http://romdb.vampier.net to include original

	  release tags.

2007-12-31 Wouter Vermaelen <[email protected]>

	* Fixes for videosource setting:

	   - after switching machines, the list of allowed values could be

	     wrong

	   - doing 'unset videosource' crashed while MSX was not powered on

2007-12-26 Wouter Vermaelen <[email protected]>

	* Restructured MemoryOps::memset implementation:

	   should make it easier to also add non-x86 asm routines

2007-12-24 Wouter Vermaelen <[email protected]>

	* Optimized sprite drawing in mode 2:

	   I still like this code to be reviewed. I believe it's correct,

	   but sprite mode 2 is tricky sometimes.

	* Fixed error handling while parsing setting files

2007-12-19 Wouter Vermaelen <[email protected]>

	* More YM2413 stuff:

	   One of the previous optimizations triggered an assert, so I had

	   to revert (part of) it. Now I reimplemented it in a correct way,

	   it's even slightly faster than the previous implementation.

	* Made some YM2413 methods ALWAYS_INLINE:

	   Gcc already did this for x86, but not for ARM. And it is measurably

	   faster (also on ARM) if they are inlined.

2007-12-17 Wouter Vermaelen <[email protected]>

	* Send out an update event for newly created settings:

	   - This way catapult is informed of value changes for

	     destroyed/recreated setting (happens for some settings during a

	     machine switch).

2007-12-16 Wouter Vermaelen <[email protected]>

	* Refactored BooleanSetting:

	   Don't inherit from EnumSetting<bool> anymore. This requires

	   slightly more code but allows to better tune the code:

	    - openmsx_info setting now returns 'boolean' (was 'enumeration')

	    - querying the value now returns true/false (was on/false)

	    - as a bonus the new implementation has less #include dependencies

2007-12-14 Wouter Vermaelen <[email protected]>

	* More YM2413 optimizations:

	   - swapped inner/outer loop, now generate all samples for a channel

	     before moving on the next channel.

	   - made a special case for channels without AM or FM modulation

2007-12-13 Wouter Vermaelen <[email protected]>

	* More YM2413 optimizations:

	   - only update noise value in rhythm mode

	   - made Channel::global a function parameter iso a member variable

	   - moved 'fnum' and 'block' from Slot to Channel, merge them into one

	     new 'freq' variable

	* General SoundDevice/MSXMixer optimization:

	   - added SoundDevice::getAmplificationFactor(), this allows to move

	     a multiplication (or shift) from the sounddevice to the mixer. In

	     the mixer there already was a multiplication (for per device

	     volume setting) so those two can be merged.

2007-12-12 Wouter Vermaelen <[email protected]>

	* Various YM2413 optimizations:

	   - made lfo_pm and lfo_am local variables iso member variables

	   - we used to store the last two Slot output values, but actually

	     one the last one was needed

	   - inlined advance() and calcSample() methods, this allows the next

	     two items:

	   - calc_phase() and calc_envelope() are not required when channel is

	     inactive (there is an exception for two rhythm channels)

	   - moved calc_phase() and calc_envelope() into the different

	     calc_slot_xxx() methods, this in turn enables the next two items

	   - the egout member variable can become a local variable

	   - rhythm channels don't use FM or AM, so we can use simpler

	     calc_phase() and calc_envelope() variants

	   - small simplifications, removed dead code

	* Made keyboard event parsing (for event recorder/replayer) slightly

	  more robust

2007-12-10 Alex Wulms

	* Added unicode support to event recorder/replayer logic

2007-12-09 Aernold Metselaar <[email protected]>

	* Fixed small bug in diskmanipulator:

	  Extracted files timestamp had most signifiocant bit of minutes set

	  to zero.

2007-12-09 David Heremans <[email protected]>

	* Fixed small bug in diskmanipulator:

	   Extracted files timestamp had an offset of one month.

2007-12-08 Wouter Vermaelen <[email protected]>

	* Another YM2413 optimization:

	   Before the Slot objects kept a pointer to a Patch object, possibly

	   they can share the same Patch. One of the patches is used for the

	   custom instrument, so this one can change.

	   After this patch we keep the Patch object by value in Slot. A

	   disadvantage is that if the custom instrument changes, we have to

	   update it in every Slot that has this instrument selected (though

	   we already had to check this for other reasons).

	   This change removes one pointer indirection from frequently used

	   code, in total I measured about 2% speedup.

2007-12-06 Wouter Vermaelen <[email protected]>

	* Various optimizations in YM2413:

	   Combined I could measure upto 10% speedup.

2007-12-05 Wouter Vermaelen <[email protected]>

	* Optimized BitmapConverter::renderGraphic4() and renderGraphic6():

	   Created a 16x16 table to lookup two pixels at once. This helps

	   a little on 64-bit machines in 32bpp mode. On 32-bit I couldn't

	   measure a difference (positive nor negative). It also helps on

	   32-bit machines in 16bpp mode, as is the case for GP2X. There I

	   could measure a clear speed gain.

	* Greatly simplified ResampleLQ:

	   New implementation can only do downsampling, but that's good

	   enough, even when output freq is 48kHz.

2007-12-01 Maarten ter Huurne <[email protected]>

	* Dropped support for GCC 3.3.

	  This was triggered by link errors when using std::basic_string<unsigned>

	  in the new keyboard mapping code. Working around this would require a

	  large amount of extra code to be written, which is not worth it.

2007-12-01 Wouter Vermaelen <[email protected]>

	* Fixed problem in BliBuffer:

	   In some cases our mute-optimization kicked in too soon,

	   (low frequency, low volume, short audio sync intervals triggers it

	   more often). This was audible as low frequency, low volume noise.

	* Previous changes in keyboard handling turned OPENMSX_FOCUS_EVENT and

	  OPENMSX_BOOT_EVENT (also) into msx events, this is wrong.

	  Event recording/replaying is still wrong because the logs don't

	  contain unicode information. Need to discuss this with Alex. Maybe

	  we need to log msx keyboard matrix info instead?

2007-11-25 Alex Wulms

	* Initialized keyboard_type and has_keypad parameter in all

	  hardwareconfig.xml file:

	    Initial value of keyboard_type is derived from keyboard

	    code at ROM BIOS address 0x2c.

	    Initial value of has_keypad is guessed from MSX version via a

	    simple heuristical rule; it is set to false for MSX1 and to

	    true for MSX2 and higher.

2007-11-24 Manuel Bilderbeek <[email protected]>

	* Added connector update events, to keep track of which connectors

	  are available.

2007-11-23 Wouter Vermaelen <[email protected]>

	* Removed EmuTime parameter from MSXDevice constructor (was not used).

2007-11-22 Wouter Vermaelen <[email protected]>

	* Optimization in SpriteChecker::checkSprites1() and

	  SpriteChecker::checkSprites2():

	   Before we checked line-by-line which sprites are visible for that

	   line. Now we check it for lines in a whole range at the same time.

2007-11-21 Alex Wulms

	* Added docs/keyboard.readme:

	   This document explains which new parameters have been

	   introduced in hardwareconfig.xml and in settings.xml

2007-11-21 Wouter Vermaelen <[email protected]>

	* Optimized MSXMixer::generate():

	   see comments in that method for more details

2007-11-20 Wouter Vermaelen <[email protected]>

	* Big optimization in AY8910::generateChannels():

	   Because the PSG generates mostly square waves and because we

	   emulate the PSG at a high frequency (224kHz), lots of consecutive

	   output samples have the same value (typically 100 or more!). We

	   used to emulate the PSG step-by-step, instead now we calculate the

	   moment in time when the output will change and output with that

	   many equal samples.

	   Because of this change PSG emulation itself became 3x-4x faster. On

	   total emulation time this is upto 10% faster.

	   This change was actually done as a preparation to use BlipBuffer

	   more directly in the PSG code (because BlipBuffer only needs to

	   know about changes in the sample stream). But on it's own it was

	   already a nice speedup. This also means there is still potential

	   for another big speedup.

2007-11-19 Alex Wulms

	* Removed conversion program (CVP) marker from russian keymap and

	  removed few duplicate entries from spanish keymap

2007-11-18 Alex Wulms

	* Fixed unicode map jp_jis, introduced jp_ansi, ru, es :

	   - The unicode maps are now derived from the MSX key

	     matrixes on the MAP. See following URL:

	     http://map.tni.nl/articles/keymatrix.php

	   - Updated harwareconfig.xml of Sony_HB-F9P_Russian and

	     of Panasonic_FS-A1 to test maps ru and jp_ansi

2007-11-16 Wouter Vermaelen <[email protected]>

	* Optimization in CPUClock:

	   Before this change we had two member variables 'limit' and 'extra'.

	   Now they are replaced by 'limit' and 'remaining'. The variable

	   'limit' still has the same meaning, the variable 'remaining' is equal

	   to the old 'limit - extra'. So the same information is represented

	   in a slightly different way.  The advantage is that the expression

	   'limit <= extra' now becomes 'remaining <= 0'. This is slightly

	   faster but it's used in the innermost CPU loop.

2007-11-14 Alex Wulms

	* Processed review comments of Wouter on Keyboard class

	  and made several other small improvements

2007-11-14 Wouter Vermaelen <[email protected]>

	* Optimized cacheline handling in CPUCore:

	   - before it was used like this:

	       const byte* line = readCacheLine[address >> 8];

	       byte result = line[address & 0xFF];

	     now it's like this:

	       const byte* line = readCacheLine[address >> 8];

	       byte result = line[address];

	     So the cacheline doesn't store directly the address of the

	     buffer, but the buffer minus (address & 0xFF00).

2007-11-13 Wouter Vermaelen <[email protected]>

	* tiny speedup:

	   Added non-virtual method RawFrame::getLinePtrDirect(), the base

	   class already had a virtual method FrameSource::getLinePtr() that

	   does the (on this level). In some places (like SDLRasterizer) we

	   know we're working on a RawFrame, so using getLinePtrDirect is

	   slightly faster.

	   For consistency I also renamed the (non-virtual) method in

	   OutputSurface to getLinePtrDirect().

2007-11-12 Wouter Vermaelen <[email protected]>

	* Optimized ResampleBlip:

	   This is not the blip algorithm itself, but the preprocess step that

	   transforms a sequence of samples into a sequence of 'sample value

	   changed' events.

2007-11-08 Wouter Vermaelen <[email protected]>

	* In various places performed the following two optimizations:

	  - Prefer to use int or unsigned int over shorter types (char,

	    short). Results in both faster and shorter code (both on x86

	    and ARM).

	  - Prefer to use unsigned over signed types. Especially when the

	    var is used in division or module operations (more important on

	    ARM than on x86). But even a simple divide by 2 (or any power of

	    2) is slower on signed types! This is because divide rounds

	    toward zero, while a shift-right operation rounds down.

2007-11-07 Wouter Vermaelen <[email protected]>

	* Added 'machine_info device [<devicename>]' info topic:

	   - Allows to query the device type of a given devicename

	   - Preparation to be able to query rom types

	* Extended 'machine_info device' to also return the romtype (only

	  for roms of course)

2007-11-06 David Heremans <[email protected]>

	* DirAsDisk has a new sector cache format:

	  - Forced little endianess of numbers in the cache.

	  - Made sure that files are mapped on the same sectors.

	  - Invalidate the cache if files have changed size

	  - New files are added afterwards and do not conflict with the cache.

2007-11-06 Wouter Vermaelen <[email protected]>

	* Fixed manbow2 mapper type:

	   First 7 sectors (1 sector is a 64kb block) are write-protected,

	   only the last sector is writable.

	   Created a seperate mapper type 'MegaFlashRomScc' (existed already

	   as an alias) which has the complete range writable.

2007-11-04 Alex Wulms

	* Introduced unicode character handling in Keyboard class:

	   - Before, SDL key-presses where mapped to MSX key-presses.

	     Now, unicode characters get mapped to MSX key-combinations

	     when SDL returns a unicode character. Otherwise, SDL key

	     still gets mapped to MSX key

2007-11-03 Wouter Vermaelen <[email protected]>

	* Optimized AY8910::generateChannels():

	   Swapped inner/outer loop. Now the outer loop goes over all (=3)

	   channels and the inner loop over the samples. We (potentially)

	   duplicate the calculations for the 'shared resources' between the

	   channels (envelope and noise), but the inner loops are much

	   simpler. In the cases I measured there was a significant net gain:

	   it only takes 60% of the original time.

	* Additional AY8910::generateChannels() optimization:

	   - Also use the 'silent channel' optimization for channels that use

	     envelope volume and envelope has reached zero (and stays there).

	   - Simplified the initial tests in generateChannels()

2007-11-02 Wouter Vermaelen <[email protected]>

	* Optimized MSXCPUInterface::setPrimarySlot():

	   more efficient way of updating the not-allowed-to-be-cached

	   region information

2007-11-01 Wouter Vermaelen <[email protected]>

	* Various CPUCore tweaks:

	   - mostly to eliminate branches or to produce shorter code

	   - I couldn't measure a speed difference, but the code is 2% shorter

2007-10-31 Wouter Vermaelen <[email protected]>

	* Made normal and '_fast' variant of Clock::getTicksTill() and

	  Clock::advance(). Use the _fast variant on places where speed is

	  important and where it's safe to use.

	  (Since 2007-10-03 we always used the fast variant, but occasionally

	  that's wrong).

2007-10-28 David Heremans <[email protected]>

	* DirAsDisk now only saves 'CACHED' type sectors and the

	  final sector of 'MIXED' file caches in the '.sector.cache'

	  when the disk is ejected.

2007-10-26 David Heremans <[email protected]>

	* DirAsDisk now dedects new files on the Host OS

2007-10-24 David Heremans <[email protected]>

	* Fixed DirAsDisk detecting moved/erased files on the Host OS

2007-10-24 David Heremans <[email protected]>

	* Extended the DirAsDisk functionality:

	   Due to popular demand....

	   Got sync from MSX disk to Host OS correct for create,

	   delete, rename, grow and shrinking of files.

	   Tested with Philips NMS 8250 and FS A1-GT who both have a

	   completely different approach/sequence to handling these actions.

	   Still a lot to do and test but we are getting somewhere at least.

2007-10-23 Wouter Vermaelen <[email protected]>

	* Fixed regression in SCC (introduced in 6430):

	    update output after a freq register write

2007-10-21 Wouter Vermaelen <[email protected]>

	* Fixed detail in SCC rotation mode:

	   if deform bits 7-6 are 10, wave form 4 rotates at freq5 speed

	   see: http://www.msx.org/forumtopicl7875.html

2007-10-15 Wouter Vermaelen <[email protected]>

	* Initial version of Manbow2 mapper:

	   this is work in progress, possibly the way the flash is saved

	   will still change

	* Added MegaFlashRomScc extension

2007-10-08 Manuel Bilderbeek <[email protected]>

	* Added Sony HB-F1XDJ config, thanks to msxrestarter. Function and use

	  of the big firmware ROM is not clear yet, but it should work, as it

	  uses the Halnote mapper.

2007-10-03 Wouter Vermaelen <[email protected]>

	* Optimized 64-bit divide-by 32-bit operations:

	   - the x86 'divl' instruction already does a 64-bit by 32-bit

	     division, though it generates a trap when the result does not

	     fit in 32-bit. That's why it's not (directly) used by gcc.

	     In certain (time critical) places in openmsx we know the result

	     fits, so we can directly use this instruction.

2007-10-03 David Heremans <[email protected]>

	* Extended the DiskManipulator:

	   It is now possible to export a single file or subdir from the msx

	   image

2007-09-29 Wouter Vermaelen <[email protected]>

	* Fixed Halnote mapper, based on info from blueMSX

2007-09-19 Wouter Vermaelen <[email protected]>

	* Fixed bug in stereo blip resampler:

	   in case only one of the two channels was active, we left the other

	   channel undefined (now we fill it with zeros)

	* Fixed bug in CPU code (probably introduced by me on 2007-08-28):

	   - limitReached() bookkeeping was wrong when the advanceTime()

	     method was used. This only happens when switching between

	     Z80/R800. Fixed now.

	   - This new implementation is also faster. Before the CPU polled

	     the scheduler for the earliest sync-point (after every IO and

	     memory mapped access). Now the Scheduler pushes new sync-point

	     data to the CPU.

2007-09-14 Wouter Vermaelen <[email protected]>

	* Fixed TR status bit in V9990 (was inverted):

	   - TR=1 means ready for transfer, thus V9990 is *not* ready to

	     execute (part of) the command, it's waiting for data

	   - Apparently most SW doesn't bother to check this bit (since

	     the V9990 is fast enough). However Symbos does check it!

	* Initial implementation of V9990 LMCM command:

	   - maybe not correct yet, but certainly better than no

	     implementation

	* Fixed odd/even pixel problem in LMMC and LMCM commands

2007-09-13 Wouter Vermaelen <[email protected]>

	* Mute individual PSG channels:

	   - already speeds up the PSG code itself (though i only added code)

	   - also enables speedups in the following stages of the audio flow

	* In SoundDevice::mixChannels(), special case mixing of 0, 1, and

	   odd or even number of channels, also unroll 2x

	* Added SSE2 optimized version of SoundDevice::mixChannels()

2007-09-12 Wouter Vermaelen <[email protected]>

	* Use resampler for SamplePlayer

	* Process samples per 4 in SoundDevice::mixChannels()

	    On itself this is already faster, but it also paves the way for

	    future SSE2 optimizations.

2007-09-08 Wouter Vermaelen <[email protected]>

	* Fixed harmless(?) UMR in Y8950Adpcm

	* Don't take sqrt() of negative numbers (in case of rounding errors)

	* Fixed crash while using breakpoints

2007-09-06 Wouter Vermaelen <[email protected]>

	* SpriteChecker improvements:

	   - get rid of expensive division in SpriteChecker fast-path

	   - duplicate SpriteChecker::checkSprites2() for planar and

	     non-planar modes

2007-09-06 Wouter Vermaelen <[email protected]>

	* CPU code tweak:

	   - slightly better signed overflow calculation

	   - use RD_WORD_PC to read two consecutive opcode bytes

2007-09-05 Wouter Vermaelen <[email protected]>

	* Big optimization in CPU code:

	   - Before we used function-to-member-pointers, now we use

	     function-pointers to static functions that take an explicit

	     'this' argument. This doesn't look as nice, but gcc can

	     generate faster code for it.

	   - In the 'zexall' test with renderer=none speed improved from about

	     106s to 85s on my system! In a 'Space Manbow' test (also with

	     renderer=none) speed improved from 54s to 48s!

2007-08-30 Wouter Vermaelen <[email protected]>

	* Optimized YM2413 and YM2413_2 idle mode:

	    After profiling we found that even a completely idle YM2413 was

	    at the top of a CPU profile (while running PSG games). The only

	    thing the YM2413 was doing was updating the internal noise

	    generator and AM/PM/evelope counters.

	    Because this is so expensive we disable that after some idle

	    time (for now 200ms). This means the noise sequence and AM/PM

	    phase will not be exactly the same as on real HW when the music

	    resumes, but that's most likely not audible.

2007-08-29 Maarten ter Huurne <[email protected]>

	* Destruct File object before loading a file with IMG_Load from

	  SDL_image. This should fix sharing violations on Windows.

2007-08-28 Wouter Vermaelen <[email protected]>

	* Split CPUClock::setLimit() in inline/non-inline part:

	    gives a nice speedup in IO-port intensive applications

	* Tweaked calculation of current CPU time

2007-08-27 Maarten ter Huurne <[email protected]>

	* Do not cache the result of native system detection.

	  This makes it easy to use the same source tree on different systems,

	  for example when the sources are on a network drive or if you do

	  both 32-bit and 64-bit builds on an x86_64 system.

2007-08-27 Wouter Vermaelen <[email protected]>

	* Created release branch for openMSX 0.6.3.

	* Various optimizations in CPU:

	   - wrote specialized (simpler) code for instructions like

	        xor a / cp a / or a / add a,a / sbc hl,hl

	   - split cpuTracePost in a small inlineable part and a non-inline

	     rarely executed part

	   - optimized time book-keeping in the read/write memory paths

	* Small optimization in SDLRasterizer (+ cleanup):

	   - don't render the same VRAM line twice in case of (single page)

	     hardware horizontal scroll

2007-08-20 Wouter Vermaelen <[email protected]>

	* Optimization in SDLRasterizer:

	   - if possible directly render to destination (instead of to local

	     buffer followed by memcpy)

2007-08-11 Wouter Vermaelen <[email protected]>

	* Made diskmanipulator command also available on machines without

	  disk drives

2007-08-07 Wouter Vermaelen <[email protected]>

	* Added -nopbo command line option to disable use of

	  OpenGL Pixel Buffer Objects

2007-08-07 Wouter Vermaelen <[email protected]>

	* Show build flavour in the version string (not for release builds)

	* Fixed crash on exit while removing an extension that created extra

	  external slot(s) and occupied (some of) those itself.

2007-08-06 Wouter Vermaelen <[email protected]>

	* Compiled with -Wpadded and changed the order of some members so that

	  the enclosing struct/class requires less internal padding:

	   - I did the same thing on 2007-04-24, however this time I'm

	     compiling on x86_64, so pointers require 8 bytes (iso 4). There

	     were also a few new classes that needed some shuffling.

2007-08-04 Wouter Vermaelen <[email protected]>

	* Enable -Wold-style-cast warnings by default

	  and fixed all warnings triggered by this

	* In CPUCore only do unaligned memory accesses if we know

	  the host CPU supports it

2007-08-01 Wouter Vermaelen <[email protected]>

	* Added psg_profile Tcl script

2007-07-31 Wouter Vermaelen <[email protected]>

	* Optimized cassetteplayer audio part, also use resampler

2007-07-30 Wouter Vermaelen <[email protected]>

	* Some fixes in ESE_SCC

2007-07-29 Wouter Vermaelen <[email protected]>

	* Commented out win32 specific sleep/timer code, it gives problems on

	  dual core CPUs

2007-07-28 Wouter Vermaelen <[email protected]>

	* Fixed [ 1622018 ] New persistant directories created for every

	  run-time switch

	* Use resampler for YMF278 (MoonSound wave part):

	   also added a trivial resample mode for the case that input and

	   output sample rate are the same (as will often be the case for

	   YMF278, as its native freq is 44100Hz)

2007-07-27 Wouter Vermaelen <[email protected]>

	* Disabled OpenGL SaI shader, see comments in GLScalerFactory

2007-07-26 Wouter Vermaelen <[email protected]>

	* Fixed [ 1730148 ] config checker creates big HD images:

	   image is now only create on first use (read/write), so

	   not yet when the machine is instantiated (=config check)

	* Added experimental PSG detune option, similar to PSG vibrato:

	   - control with the PSG_detune_frequency and PSG_detune_percent

	     settings

	   - next step: experiment with the PSG detune/vibrato and hopefully

	     remove one of the two again. Maybe also remove the

	     PSG_*_frequency setting(s).

2007-07-21 Patrick van Arkel ([email protected])

	* Updated to Version 2.1 of the cheat defenitions after

	  adding a ton of new cheats provided by Mars2000you.

2007-07-16 Wouter Vermaelen <[email protected]>

	* Added proper MSXDOS2 mapper type (before we used generic16kb):

	   - implemented by BiFiMSX, I only fixed some bugs

2007-07-14 Wouter Vermaelen <[email protected]>

	* Implemented 'Super Lode Runner' mapper:

	   This mapper is special because it ignores the SLOT-SELECT signal;

	   it reacts to writes to address 0x0000 in _any_ slot. This violates

	   the MSX standard.

	   For a long time we didn't implement this mapper because it would

	   add a (small?) performance hit on all memory writes.

	   However, now, since we implemented the watchpoint infrastructure,

	   the mapper can be implemented without any performance impact (no

	   performance impact if the mapper is not inserted).

2007-07-05 Maarten ter Huurne <[email protected]>

	* Added vibrato effect for PSG.

	  You can control it with the settings "PSG_vibrato_percent" and

	  "PSG_vibrato_frequency". Good values are around 1.0% and 4.5Hz.

	  By default "PSG_vibrato_percent" is 0, which turns off the effect.

	  Thanks to Wolf for the idea and his feedback on experiments.

2007-07-02 Wouter Vermaelen <[email protected]>

	* Use pimple idiom to move implementation details from hh to cc for

	   YM2413 YM2413_2 YMF262 YMF278

	* Fixed regression in YM2413_2 core:

	   Was introduced in revision 6372. We used a 8.24 FixedPoint

	   datatype, but this goes wrong because our FixedPoint types are

	   signed. Solved by using much smaller resolution FixedPoint

	   datatypes (this is even a simplification and a small speedup).

2007-06-30 Wouter Vermaelen <[email protected]>

	* allow to use bindings on joysticks that are not plugged in (into the

	  emulated MSX)

2007-06-27 Wouter Vermaelen <[email protected]>

	* Created new directory 'utils' and moved several files into it

2007-06-26 Manuel Bilderbeek <[email protected]>

	* Added Software and Creation Date info to AVI files.

	* Added Creation Time info to PNG files (Software was already there).

2007-06-23 Wouter Vermaelen <[email protected]>

	* Added experimental BlipBuffer support for DACSound based sound

	  devices:

	   - heavily based on

	       http://slack.net/~ant/libs/audio.html#Blip_Buffer

	* Added BlipBuffer based resampler:

	   - Selectable via 'set resampler blip'

	   - I'm not sure we want to keep this global or if we want to make

	     this resampler default for specific devices. But for now it's an

	     easy way to experiment

2007-06-19 Manuel Bilderbeek <[email protected]>

	* Voice ROM of Keyboard Master is now also searched in ROM pools (sha1

	  is hardcoded).

2007-06-11 Wouter Vermaelen <[email protected]>

	* Fixed bug in SCC peek:

	   - peek of deform register also triggered a reset of that

	     register (like a regular read does)

	* Implemented peekMem() for ESE_SCC / MSXSCCPlusCart

	* Implemented read/write cache line stuff for ESE_SCC

2007-06-11 Wouter Vermaelen <[email protected]>

	* Added (optimized) clipping routines in Math.hh and use them

	  in several places:

	   - idea for int->short clipping is taken from Blip_Buffer

	       http://slack.net/~ant/libs/audio.html#Blip_Buffer

2007-06-09 Maarten ter Huurne <[email protected]>

	* Use resampler for AY8910 (PSG).

2007-06-06 Manuel Bilderbeek <[email protected]>

	* Added a dummy port of blueMSX's YM2148. TODO: actually fill in the

	  code. It's harmless, so I checked it in to have a base for a real

	  port.

2007-06-06 Wouter Vermaelen <[email protected]>

	* Access all CPU registers via getter/setters:

	   - as preparartion for performance experiments, this change on its

	     own has very little impact on the generated code

	* Alternative implementation for CPURegs:

	   - see comments in the code for details

	   - this version is a few percent faster for me

2007-06-05 Maarten ter Huurne <[email protected]>

	* Added support for DESTDIR in the build system:

	  If you run "make install DESTDIR=/some/path", all installation

	  paths will have that prefix prepended to them.

	  For example, the default location of the executable would become:

	    /some/path/opt/openMSX/bin/openmsx

2007-06-04 Wouter Vermaelen <[email protected]>

	* Another CPUCore optimization:

	  - Made optimized 16-bit read/write routines. Originally this gave a

	    slowdown, caused by gcc no longer inlining some bigger routines.

	    After forcing gcc to inline the memory read/write functions, there

	    was a huge speedup: upto 15% in a CPU only benchmark!! Forced

	    inlining is responsible for the largest part of this speedup (very

	    roughly 3/4 by always inline, 1/4 by new 16-bit routines).

	* Experimental IM2 support for YamahaSfg05 (untested)

2007-06-03 Wouter Vermaelen <[email protected]>

	* CPUCore optimization:

	  - in RDMEM / WRMEM routines, keep cache line address in local

	    variable instead of reading it twice from an array. Apparently gcc

	    was not able to prove that array didn't change during a call to

	    T::POST_MEM().

	  - Gave more than 10% speedup here in a CPU only benchmark (zexall

	    with frameskip=99)

2007-06-02 Wouter Vermaelen <[email protected]>

	* Sound recording (seperate channels or combined) now always records

	  to sound, irregardless of the 'mute' setting

2007-06-02 Albert Beevendorp

	* Added Password Cartridge implementation.

	  The value between the <password> tags in the hardwareconfig.xml

	  may vary from 0x0000 to 0xffff. It's used by the BIOS to read the

	  password from the cartridge if one is set. This implements feature

	  request 1411507.

2007-06-01 Manuel Bilderbeek <[email protected]>

	* Added an LS-120 SCSI device.

	  - Mostly as demonstrator of a removable SCSI device.

	  - Could be used to make a SCSICDROM easier to make.

	  - Can also be used to determine common stuff between SCSI devices.

	  - TODO: change 'lsX' commands to something better.

	  - TODO: remove massive code duplication with SCSIHD/HDCommand.

2007-05-31 Manuel Bilderbeek <[email protected]>

	* Removed all dead code from SCSIHD and attempted to move

	  device-specific stuff inside the class. I'll try to make an

	  LS-120 device soon, to demonstrate a removable device.

	* Note that yesterday we also added ESE SCC and WAVE-SCSI. The latter

	  doesn't seem to work properly, though.

2007-05-28 Manuel Bilderbeek <[email protected]>

	* Added initial MEGA-SCSI emulation.

	  It's not working completely OK yet... Please help debugging :)

	  (Thanks a lot to Wouter and Albert for their help.)

	* It seems to work after all: make sure to do a physical format of the

	  disk first (using SFORM-2.COM).

	* Lots of cleanups, also by Wouter.

	* Added ESE Ramdisk (a.k.a. ESE RAM). This is MEGA-SCSI without SCSI.

2007-05-26 Manuel Bilderbeek <[email protected]>

	* Created common superclass for SCSIHD and IDEHD called... HD.

	  This makes it possible to let HDCommand work with any kind of HD.

	  Now you can use the hdx commands also for SCSIHD's.

	  Note: (off-topic) diskmanipulator can't handle corrupt disks.

	  Note 2: diskmanipulator doesn't support the partition table formats

	  used on SCSI harddisks yet. So, until then, SCSI devices cannot be

	  used from diskmanipulator.

	  Note 3: implementation of HD class could be cleaner... (it's using a

	  protected member "file" now...)

2007-05-24 Manuel Bilderbeek <[email protected]>

	* Read initial size and file name from config XML and create disk

	  image in the persistent directory.

	* GoudaSCSI is now usable!

	* Implemented LED events for SCSIHD.

2007-05-23 Manuel Bilderbeek <[email protected]>

	* (GoudaSCSI is basically working, yay! But please do not use it yet.)

	* Introduced abstract base class for SCSIDevices and added

	  DummySCSIDevice (and made use of it).

2007-05-21 Manuel Bilderbeek <[email protected]>

	* Added initial GoudaSCSI support, ported from blueMSX.

	  NOTE: DO NOT USE YET!

	  This is very much WIP! It's not working yet and contains lots of

	  shortcuts and hardcoded stuff. So, again: don't use it. But please

	  help improve it! :) (You are warned!)

2007-05-19 Wouter Vermaelen <[email protected]>

	* Removed old (unoptimized) implementation of the high-quality

	  resampler: sound quality is the same

2007-05-18 Wouter Vermaelen <[email protected]>

	* Added fast/low-quality resample algorithm

	* Added 'resample' setting with possible values 'fast' and 'hq' to

	  switch between the different resample algorithm

2007-05-17 Wouter Vermaelen <[email protected]>

	* Moved int<->float conversion from sounddevices to Resampler:

	   - preparation for fast/low-quality resampler

2007-05-14 Wouter Vermaelen <[email protected]>

	* Share resample coeff table between Resample instances with the same

	  resample ratio

2007-05-12 Wouter Vermaelen <[email protected]>

	* Additional speedup for the Resampler:

	   - unrolled the loop (+ break dependencies)

	   - added SSE optimized routine (mono only for the moment)

	   - also added SSE optimized stereo resampler

	* Fixed memset4_2_MMX():

	   - we used to pass a pointer to a struct on the stack, but

	     newer gcc versions optimize this struct away

2007-05-11 Wouter Vermaelen <[email protected]>

	* Experimental speedup for the Resampler:

	   - code is very quick-and-dirty

	   - still contains both old and new implementation to make

	     sound quality comparison between the two possible

2007-05-10 Wouter Vermaelen <[email protected]>

	* Use 'SCC' instead of rom-name as name for the SCC sounddevice

2007-05-09 Wouter Vermaelen <[email protected]>

	* Optimized Resampler a bit:

	   - used 'x - trunc(x)' instead of 'fmod(x, 1.0)'

	   - but it's not used in the innermost loop so the speed gain is not

	     that big (it is measurable though)

2007-05-08 Wouter Vermaelen <[email protected]>

	* Added SSE2 optimized noise routine in FBPostProcessor

	* Restored parsing of the <volume> and <mode> tags in the

	  soundsection of hardwareconfig files. There can also

	  be a <balance> tag now (that overrules <mode>)

	* Removed volume handling from sounddevice code:

	   - it's now handled (together with balance) inside MSXMixer

2007-05-07 Wouter Vermaelen <[email protected]>

	* Removed <sounddevice>_mode setting:

	   - will be replaced by a balance setting soon

	   - TODO what about backwards compatibilty?

	* Added <sounddevice>_balance setting

2007-05-04 Wouter Vermaelen <[email protected]>

	* Added support for SSE2 in HostCPU

	* Added SSE2 optimized routine in ScanLine class

2007-05-01 Wouter Vermaelen <[email protected]>

	* Reworked internal sounddevice mute code:

	   - As an optimization is was possible for a sounddevice to indicate

	     that it won't produce sound in it's current state. In that case

	     the mixer wouldn't bother to call the updateBuffer() method

	     for such devices.

	     It was also already possible for a sounddevice to indicate that

	     a specific channel is completely silent.

	     Now these two approaches are unified: sounddevice should only

	     inidicate which individual channels are muted. If all of the

	     channels are muted, the rest of the sound code detects this and

	     the same optimization as before happens.

	     In this new approach the updateBuffer() is called in any case,

	     so it's now possible to both correctly update internal state and

	     benefit from the muted optimization.

	   - The glitches that happened after unmute on resampled devices are

	     now gone. But that because muting is effectively disabled for

	     these devices for now. Although this new approach does make it

	     possible to solve it and have the mute optimization. It's just

	     not implemeted yet.

2007-04-30 Maarten ter Huurne <[email protected]>

	* Fixed MSX reading from SCC: frequency and volume block is write-only.

	* Fixed SCC reset behaviour: only deform and channel enable are changed.

2007-04-24 Wouter Vermaelen <[email protected]>

	* Compiled with -Wpadded and fixed some of the warnings:

	   - only fixed the cases that were solvable by rearranging the

	     order of the fields locally in one class

	* Fixed cassette input bit (PSG R#14 bit 7) on turboR (thanks BiFiMSX)

2007-04-23 Wouter Vermaelen <[email protected]>

	* Extended mute_channel related procs:

	    - mute_channels without any arguments lists which channels are

	      currently muted

	    - unmute_channels without arguments now unmutes all channels

	    - new command 'solo' mutes everything except for the specified

	      channels

	* Small optimization in SDLGL-PP simple scaler:

	   - pack several scalar uniform variables in one uniform vector

	     variable. Removes one MOV instruction from the compiled fragment

	     shader.

2007-04-21 Wouter Vermaelen <[email protected]>

	* Recording/muting of individual sound device channels works:

	   - it's still quick-and-dirty, but i wanted to have something

	     working soon

	   - use either via the <sounddevice>_ch<n>_record and

	                        <sounddevice>_ch<n>_mute settings

	     or via the record_channels Tcl script (no mute yet)

	     examples:   record_channels PSG

	                 record_channels SCC 1,3-5

	                 record_channels stop

	                 record_channels stop PSG 2

	                 record_channels list

	   - TODO: documentation

	           make recording stutterless

	           record silence when complete sounddevice is muted

	              (muted manually or implicitly by the device itself)

	* Improved channel recording:

	   - Channel recording is now also stutterless.

	   - Recording continues when sounddevice is internally muted.

	     Unfortunately explicitly muting the device doesn't work now

	     anymore when some channels of the device are beiing recorded.

	     Fixing this requires some more work.

	* Added tabcompletion + help for the record_channels script

	* Extended script with (un)mute_channels procs

2007-04-19 Wouter Vermaelen <[email protected]>

	* Converted sounddevices to generate output as seperate channels:

	   - This is very much work in progress. There is no code yet to

	     take advantage of this split. The code in the converted

	     sounddevices should still be cleaned-up and optimized.

	* Refactored SoundDevice and ChannelMixer:

	   - merged SoundDevice and ChannelMixer classes

	   - made SoundDevice aware of input and output sample rates

	   - added methods in SoundDevice to record and mute individual

	     channels. Next step is to actually use these methods.

2007-04-18 Joost Yervante Damad <[email protected]>

	* fix build with gcc snapshot 20070326

2007-04-16 Wouter Vermaelen <[email protected]>

	* Implemented SDLGL-PP RGBTriplet scaler

	* Implemented SDLGL-PP SaI scaler:

	    This algorithm is not well suited for a HW implementation (it

	    has a lot of branches). So this scaler will probably run a bit

	    slower than the other HW scalers.

2007-04-15 Manuel Bilderbeek <[email protected]>

	* Added emulation of Yamaha YM2151 and partly the Yamaha SFG-05

	  sound module. For the latter, I included the config of the Yamaha

	  CX5MII/128. TODO: MIDI and keyboard, so only few software titles

	  for it work. Thanks to Daniel Vik for his support! The code is

	  taken from blueMSX and MAME.

	* Added emulation of the VLM5030 sound device, which is used in the

	  unreleased Konami game Keyboard Master. For now, the sample ROM

	  should be located in share/keyboardmaster with the name voice.rom.

	  Thanks to Daniel Vik for his support. Also here, the code is

	  taken from blueMSX and MAME.

2007-04-15 Wouter Vermaelen <[email protected]>

	* Extended resampler class for stereo devices

	* Speedup SDLGL-PP simple scaler

	* Use (stereo) resampler for YMF262 (FM-part of OPL4)

	* Use resampler for (both) YM2413 cores (MSX-MUSIC)

	* Use resampler for Y8950 (MSX-AUDIO)

2007-04-15 Maarten ter Huurne <[email protected]>

	* Released openmsx-0.6.2.

2007-04-12 Manuel Bilderbeek <[email protected]>

	* Fix for MSX1 VDP's: on these VDP's the VRAM pointer is always

	  updated when writing to the VDP registers. Fixes some MSX1 demos

	  like Planet of the Epas.

2007-03-23 Manuel Bilderbeek <[email protected]>