relnotes20B 13.1 KB

Nov 20 release RELEASE NOTES
============================

SUMMARY:
--------

	This release of Ultra 64 (version 2.0B for RCP 2.0 and version 1.0F for 
	RCP 1.0) contains various bug fixes and enhancements. Please note 
	that some applications (such as bumpmap, print, and spbench) can only 
	run on RCP 2.0.

NEW FEATURES:
-------------

	1) IMPORTANT: Flexible support for boot address
           ---------
	The memory map has changed. The final machine will have 4 MB
	of 9-bit RDRAM memory. The color and z framebuffers can reside 
	anywhere in memory. The default boot address is 0x80000400 and can be 
	changed by adding the statement "address 0x80XXXXXX" to the BOOT 
	segment in the spec file. See the spec files in simple and onetri for 
	more info.

	2) 3D Turbo graphics ucode
	This new microcode version is feature-limited, precision-
	reduced, but faster. It is intended for "characters"
	that remain in the center of the viewing area.

	There is documentation about the turbo graphics in the
	man pages, the "Performance Tuning" guide, and elsewhere
	in the developer documentation.	

	Here are the new, turbo-related man pages:

		gspTurbo3D.3p
		gDPClearOtherMode.3p
		gDPEndDisplayList.3p
		gtStateSetOthermode.3p
		guMtxCatF.3p
		guMtxCatL.3p
		guMtxF2L.3p
		guMtxIdent.3p
		guMtxL2F.3p
		guMtxXFMF.3p
		guMtxXFML.3p

	plus changes to all other guMtx* man pages

	3) gspFast3D_fifo microcode. (Performance improvement)
	This new microcode is very similar to gspFast3D and gspFast3D_dram
	microcode.  It uses the DRAM to buffer data flowing from the RSP to
	the RDP (like the gspFast3D_dram microcode), but it allows the DP to 
	read the data as it is being generated (like the gspFast3D microcode).
	The data buffer (which the developer must supply in DRAM) is used as
	a circular buffer for RSP output (RDP input).  The microcode sets up
	all neccessary DMA's so that you do not need to call osDpSetNextBuffer
	(which is neccessary with the gspFast3D_dram code).  You can use it
	exactly the way you use gspFast3D code except that you must supply the
	pointer and length of the buffer in the task header.  Make
	task->output_buff point to the beginning of the buffer and make
	task->output_buff_size be the size of the buffer in bytes. 
	IMPORTANT NOTE: In gspFast3D_dram the output_buff_size is a POINTER
	to a u64 where the microcode will store the length of the buffer.  In
	gspFast3D_fifo the output_buff_size is a u32 which has been cast to
	to a (u64 *), and represents the length og the buffer, NOT a pointer.
	The length of the buffer must be at least 0x100 bytes long.  However
	the advantage of the gspFast3D_fifo microcode is that you can use a
	larger buffer than the gspFast3D microcode uses and therefore gain
	performance through better parallelism.  The gspFast3D buffer is 1024
	bytes long, so the buffer you use with gspFast3D_fifo should be
	significantly larger (several times larger or more reccommended).
	See the man page for osSpTaskLoad.


NEW DEMO APPLICATIONS:
----------------------

	1) autofill - performance benchmark, see README_DEMOS file.
	2) ci8fb    - performance benchmark, see README_DEMOS file.
	3) gtdemo   - turbo ucode demo, see README_DEMOS file.
	4) hostio   - U64-Indy interface demo, see README_DEMOS file.
	5) spbench  - sprite benchmark, see README_DEMOS file.
	6) turbomonkey - turbo performance tool, see README_DEMOS file.

BUG FIXES:
----------

	1) Due to bug fixes for floating point exception handling, the
	thread structure has been changed. You need to "make clobber" your
	application before using this new release.

	2) There are bug fixes in the boot code to correct the power-up hang 
	problem and the DMA size from ROM.

	3) The thread initialization bug is fixed. There is no need to 
	initialize the next and queue fields of a thread structure before
	calling osCreateThread.

	4) There are bug fixes in the VI manager to set the display to black
	upon the calling of osCreateViManager. To resolve the problem of
	having uninitialized frame buffer displayed on the TV screen,
	osViBlack (i.e. osViBlack(TRUE)) can be called immediately after 
	osViSetMode to set the display to black. And, when frame buffer is 
	ready for display, osViBlack can be called again (i.e. osViBlack(FALSE))
	to turn off the black mode so that VI manager can display the real 
	frame buffer. Please see the man page on osViBlack for more info.

	5) Applications using audio must make several small but important 
	changes to operate successfully. These changes, implemented with 
	releases from 10/16/95 forward are listed below in the audio section.

	6) There was a bug with the gsSPFogPosition and gsSPFogFactor macros.
	They used to take 3 parameters (n, min, max) (n was ignored).  They
	now take only 2 parameters (min, max).  Note that gSPFogPosition and
	gSPFogFactor are unchanged; it is only gsSPFogPosition and
	gsSPFogFactor which had the extra unneccessary parameter.

	7) A bug in the scheduler that caused the RDP to hang (not complete)
	in rare situations has been fixed.

FEATURES ENHANCED:
------------------

	1) The application blockmonkey has been enhanced to allow the VI mode
	(Lpn1) to dynamically change the width (or x-resolution); this feature
	can be used to assess the best video resolution for a game with a 
	given performance requirement. Global transformations, flat shading,
	lighting and mip-mapped textures have also been added. Some performance
	improvements have been added.

	2) A new gbi macro: gSPLightColor has been added for changing light
	colors.  The advantage of this macro over using gSPLight or
	gSPSetLights is speed and memory savings.  This command allows you to
	specify an immediate color in the display list so no separate light
	structure is needed to store various different colors for different 
	objects.  It is also fast because the lights do not need to be
	recalculated in the RSP the way they are when you change the direction 
	with gSPLight or gSPSetLights.

FEATURES NO LONGER SUPPORT:
---------------------------

	1) There is no longer any support for osAsyncPrintf. Please use 
	osSyncPrintf to write to the debug port.


THINGS TO WATCH OUT FOR:
------------------------

	1) ROM images sent to Nintendo don't work
	Symptom: ROM images sent to Nintendo don't work on the real hardware
	while images run on the development board do.

	Fix: Compile with the non-debug library (libultra.a), remove all
	printfs and argument parsing code from your game.

	2) RDP hangs
	Symptom: The scheduler hangs waiting for an RDP interrupt which never
	occurs. Audio still runs.

	Workaround: If you are running audio, the first thing to try is to set
	the flags field for *audio* tasks as follows

	    t->list.t.flags  = OS_TASK_DP_WAIT; /* do not set this to 0 */

	Note: The RDP will also hang if the display list is corrupted, if the
	RDP message queue is too small (and the RDP message is dropped), or if
	code overwrites the exception table.

	3) Clicks in the audio
	Symptom: Audio occasionally clicks or crashes.

	Fix: Make sure that the sound bank data is cache aligned. The easiest
	way to do this is to allocate from the audio heap with alHeapAlloc().

	4) Initialization of Bss in each segment
	Symptom: Variables in Bss segments contain garbage.

	Fix: Make sure that Bss of each segment is initialized to zero. See the
	app overlay to see how this can be accomplished by using bzero().
	The system only initializes the Bss of the boot segment.

CAVEATS---------------------------------------------------------------------

Graphics

	---------------------------------------------------------------------

Audio
	Fixed bug in alSynRemovePlayer that would cause players to not be 
	successfully removed.

	Fixed bug in osScRemoveClient that would cause corruption of the client
	list.

	A bug in the synthesizer that would occasionally hang the system
	for about 30 seconds when both the Sound Player and Sequence Player
	were in use has been fixed.

	Changes effective 11/12/95

	AL_MAX_PRIORITY has been changed from 16 to 127.
	
	midiDmon has a couple of bug fixes, and now uses the sample rate given
	in the bank.inst file, instead of a default of 44100.

	A sound player bug which caused sounds which were stopped and restarted
	to play incorrectly has been fixed.

	A bug in the synthesizer which caused noise when playing sounds at very 
	low pitch values has been fixed. There is still a potential problem with
   	pitch values below 0.01 which we are investigating.

	Changes needed after 10/16/95:
	Both the compact sequence player and the regular sequence player now 
	implement a method of vibrato and tremelo, using callbacks to routines
	provided by the application. This new system of handling tremelo and
	vibrato is demonstrated in the playseq example. If the application 
	chooses not to use this feature it must set the callback pointers to 
	zero in the ALSeqpConfig structure, before calling alSeqpNew or 
	alCSPNew.

	The following code demonstrates how you would init the config struct to
	not use vibrato or tremelo: 
   
	    seqc.maxVoices      = MAX_VOICES;
	    seqc.maxEvents      = EVT_COUNT;
	    seqc.maxChannels    = 16;
	    seqc.heap           = &hp;
	    seqc.initOsc        = 0;
	    seqc.updateOsc      = 0;
	    seqc.stopOsc        = 0;

	Failure to set initOsc, updateOsc and stopOsc to zero when not 
	implementing vibrato and tremelo will cause the CPU to jump to garbage 
	addresses. (If you implement the vibrato and tremelo then set these 
	pointers to your routines. See playseq.c for an example.)

	The above change, adds eight bytes to every instrument structure in 
	a bank .ctl file. For this reason, you must be sure you are using the
	new version of ic, and that you remake any banks you have. If you are 
	using your own tools to build banks, you should be aware of the changes
	in the ALInstrument structure, (listed in libaudio.h) and adjust your 
	tools accordingly.

	AL_MAX_PRIORITY has been changed to 127 to allow for a greater range
	of priorities. If you have hardwired priority values you should adjust
	them to reflect the new max.

	The following routines have been implemented.
	
	ALMicroTime  alSeqpGetTime(ALSeqPlayer *seqp);
	void         alSeqpSetTime(ALSeqPlayer *seqp, ALMicroTime time);
	u8           alSeqpGetChannelPriority(ALSeqPlayer *seqp, u8 chan);
	void         alSeqpSetChannelPriority(ALSeqPlayer *seqp, u8 chan, 
			u8 priority);

	ALMicroTime  alCSPGetTime(ALCSPlayer *seqp);
	void         alCSPSetTime(ALCSPlayer *seqp, ALMicroTime time);
	u8           alCSPGetChannelPriority(ALCSPlayer *seqp, u8 chan);
	void         alCSPSetChannelPriority(ALCSPlayer *seqp, u8 chan, 
			u8 priority);

	void         alCSeqNewMarker(ALCSeq *seq, ALCSeqMarker *m, u32 ticks);
	void         alCSeqSetLoc(ALCSeq *seq, ALCSeqMarker *marker);
	void         alCSeqGetLoc(ALCSeq *seq, ALCSeqMarker *marker); 


	Changes prior to 10/16/95:
	A new application called midiDmon allows you to download your audio
	banks and playback on the Ultra64 with midi from keyboard or sequencer.
	To use this application, the ultra inst image depends on installation
	of the dmedia_eoe (version 5.5) image. The 5.5 dmedia_eoe is included
	on the inst tape sent to you.

	---------------------------------------------------------------------

	audio performance has dramatically improved on this release. We have
	rearranged code to minimize the ICache misses. The cost for audio
	is 1% per channel at 30Hz update for 22KHz output.

	---------------------------------------------------------------------

Operating System

	---------------------------------------------------------------------

Demos
	To build applications for RCP 1.0, make sure that the environment
	variable HW_FLAGS is set to "-D_HW_VERSION_1".

	---------------------------------------------------------------------

Tools

	---------------------------------------------------------------------

Debugger

	gvd sometimes can't find the source because it gets confused on the
	path to the source. A definite occurance is when you login the graphics
	console as root. If you login as guest and su - root, everything seems
	to work.

	---------------------------------------------------------------------

	There has been a bug fix for the multi-thread window in gvd.
	After user has switched to a specific thread, he can then invoke
	the multi-thread window to obtain a complete list of all active threads
	running on the target machine. There is still a bug in showing the
	actual state of the threads in this window; that is, a stopped thread 
	can still be listed as running on the window. Therefore, user should 
	use the threadstatus utility to obtain accurate state information of 
	a specific thread.
	
	---------------------------------------------------------------------

	Currently, the debugger cannot support background mode debugging.
	That is, if a breakpoint is encountered, all user threads are stopped.
	gvd does not support private breakpoints per thread; all breakpoints
	are shared globally. Furthermore, the Continue button on the main gvd
	window only resumes execution of the thread being debugged. To resume
	or stop execution of all threads, user should use the Continue or Stop
	button in the multi-thread window.

	---------------------------------------------------------------------