freebsd-dev/sys/i386/isa/sound/Readme.v30

VoxWare v3.0
------------

This is a PROTOTYPE of the VoxWare v3.0 to be released late 94.

All features of v2.5 should work as earlier. There could be some
omissions but they are unintentional. I started this version thread
after v2.3 so all features implemented before it are there.

Even this is a prototype, there should not be any fatal bugs. The
prototype just means that I don't have implemented all features
completely. Mainly in the /dev/sequencer2 driver.
For example recording from /dev/sequencer2 won't work
with other cards than a full features MPU-401 or clones. As well the
way how the MIDI controllers are handled will change.

IMPORTANT!!!!!!!!!!!!!!!!!

Don't distribute any binaries compiled with the soundcard.h of this version. 
They will not work together with older drivers.

New features
============

There are now two new device interfaces. The /dev/midi## is a raw
tty like interface to MIDI ports. There is a device file for each MIDI
port on your system. They are named (/dev/midi00 to /dev/midiNN).
The second addition is the /dev/sequencer2 which is higher level interface 
than the old /dev/sequencer. It's intended for writing device independent
applications like sequencers.

/dev/midi##
-----------

This interface should be useful for applications like MIDI sysex librarians.
There are (currently) no timing features so making music could be impossible.

There are as many /dev/midi## devices as there are MIDI ports in the system. 
The /dev/midi00 is connected to the first one, /dev/midi01 to the second etc.

These devices work like tty devices in raw mode. Everything written to them is
sent out to the MIDI port. There is currently an extra delay of at most 
1/100th of sec but it will be removed later.

The reading algorithm is little bit more complicated. There are two different
cases:

1)	There is at least one byte in the input buffer.

The read returns as many bytes as it can without waiting for more bytes.
For example when a process reads 100 bytes and there are 10 bytes in the 
buffer, the read returns just 10 bytes.

2)	The input buffer is empty when the process calls read.

The read waits for the first byte and then continues as in case 1. By
default it waits infinitely but there is an ioctl for setting a timeout
for this. The ioctl(fd, SNDCTL_MIDI_PRETIME, &time) changes the timeout.
The time is given in 1/10th of seconds (10 means one second).

Other ioctl calls:

ioctl(fd, SNDCTL_MIDI_MPUMODE, &mode) is available for full MPU-401
compatible devices such as MPU-IPC-T, MQ PC Midi Card or MQX-32.
It's not available for the so called MPU UART ports of some soundcards
(PAS16, SB16 etc). By default the MIDI port is in UART mode after open. 
If this ioctl is called with mode=1, the interface is put to the intelligent
(coprocessor) mode. NOTE! The MIDI port will be reset when this ioctl is called.
It could have some strange effects if not called immediately after open. This
call returns EINVAL if the midi port doesn't support the MPU-401 intelligent
mode.

ioctl(fd, SNDCTL_MIDI_MPUCMD, &cmdstruct) is valid only if the MIDI port
is put to the coprocessor mode using ioctl(SNDCTL_MIDI_MPUMODE). It's used to
send commands to a MPU-401 compatible MIDI cards. Please refer to the
MPU-401 Technical Reference Manual (or Music Quest Technical Reference
Manual) for descriptions of the commands.

The argument of SNDCTL_MIDI_MPUCOMMAND is of type mpu_command_rec. It
has the following fields:

typedef struct {
		unsigned char cmd;

		char nr_args, nr_returns;
		unsigned char data[30];
	} mpu_command_rec;

where:
	cmd		Contains the command number.
	nr_args		Number of arguments of the command.
			MUST BE INITIALIZED BEFORE CALL
	nr_returns	Number of bytes returned by the command.
			MUST BE INITIALIZED BEFORE CALL
	data		Buffer for the command arguments and returned
			data.

Be extremely careful with the nr_args and nr_returns fields. They
must match the command. An incorrect value will put the card and
the driver out of sync. Refer to the MPU-401/MQX-32M documentation for further
details.



/dev/sequencer2 (if you find a better name, please let me know).
---------------

This device file works much like the /dev/sequencer which has been present
since the beginning. The main differences are the following:

- /dev/sequencer makes the MIDI ports to look like the synth devices. In fact
the result is somewhere between the MIDI specification and the synth devices of
/dev/sequencer. Both kind of devices are accessed using the SEQ_START_NOTE()
like macros. The voice number parameters of the API macros have been redefined
to denote MIDI channels. This means that the driver allocates voices for
the channels automatically (this is a responsibility/right of an application
with /dev/sequencer). The result is that a SEQ_START_NOTE() macro has
similar effects for a synth channel than on a MIDI port. This kind of
solution provides better device independence than the /dev/sequencer. The
drawback is that the new interface doesn't permit so low level access to the
device as the /dev/sequencer does. An application developer must choose between
these two interfaces. I think the old /dev/sequencer is better for applications
like module players while the new one is better for making generic sequencer
programs.

- There are no separate MIDI devices with the /dev/sequencer2. The
ioctl(SNDCTL_SEQ_NRMIDIS) returns always zero. Instead the MIDI ports are
shown as synth devices. ioctl(SNDCTL_SEQ_NRSYNTHS) on /dev/sequencer2 will 
return sum of internal synthesizers (GUS, OPL3) and MIDI ports in the systems.

- The new interface is used much like the ordinary /dev/sequencer. The
event format is new so you have to use the API macros defined in the 
sys/soundcard.h. The interface will probably change before the final 3.0
release but using the API macros should ensure compatibility in source level.
The new event format is not recognized by version 2.X so don't try to 
distribute binaries compiled with soundcard.h of v3.X.

- The basic API usage is similar to the current one. There are some new
macros but the older ones should work as earlier. The most important
incompatibility is that the /dev/sequencer2 driver allocates voices itself.
The other one is that the application must send SEQ_START_TIMER() as it's
first event. Otherwise the timer is not started and the application waits
infinitely.


There are several new features but I don't document them here. There are
some info in the soundcard.h (near the end). I have also included some
sample code in the directory v30. Full documentation will
appear in the Hacker's Guide later.

Don't hesitate to contact me in case you have questions or comments.

Hannu Savolainen
hannu@voxware.pp.fi