Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!mordor!lll-tis!ames!ucbcad!pasteur!ucbvax!cbosgd!ihnp4!homxb!largo
From: largo@homxb.UUCP (J.BAKER)
Newsgroups: comp.sys.att
Subject: Re: 6300 graphics and EGA cards
Message-ID: <1392@homxb.UUCP>
Date: 20 Jan 88 16:19:32 GMT
References: <1037@bakerst.UUCP> <9300048@bradley> <835@ur-tut.UUCP> <2870@homxc.UUCP>
Organization: AT&T Bell Laboratories, Holmdel
Lines: 54

In article <2870@homxc.UUCP>, rps@homxc.UUCP (R.SHARPLES) writes:
>
> 	The new card is essentially a multifrequency graphics card
> 	that supports CGA, EGA and AT&T's 640 x 400 modes.  Installing
> 	it in a 6300 is just as problematic as installing any other
> 	EGA card - you still have the 16bit word order problem (when
> 	transmitting a 16 graphics instruction to the card on the 8 
> 	bit bus, it must be sent as two 8 bit words.  Problem is
> 	which to send first?  We send the 8 high bits then the 8
> 	low bits, IBM sends the the low bits then the high bits.  For
> 	some reason, fixing this on the 6300 was "not possible".  
> 	The 6300 Plus uses the same word order as the IBM standard.)
> 

On the IBM, byte order is determined by the INTEL 8088 which does
the muxing for you.  It writes all 16-bit words (1 byte at a time)
in double write cycles because of the 8-bit bus.  The 6300, of
course, uses a 8086 which can take advantage of its 16-bit bus when
writing a word to an even address--it uses one write cycle.  Now,
in this case we've got to get our 16-bit word onto the bus
converter's 8-bit bus so we can be compatible with the wealth of
IBM XT standard expansion boards.  The problem, which byte to send
first, is answered by *looking* at the 8088 bus timing.  Strangely
enough, the pc6300 bus converter does it in descending address
order.  First the high byte goes to the word address + 1, then the
low byte follows to the word address.  I believe the design is
Olivetti's as is the decision to fix or not to fix.

The reason EGA loves this is because of its 'hybrid' register
design.  EGA has pairs of 'ports' (select, value) which allows the
software to write to a number of EGA registers through the same
port pair.  First, write a number to 'select' to select the desired
register, then write the desired number to 'value'.

Software wizards (I guess there are a lot of them at Microsoft)
noticed that a few ASM instruction could be saved by loading AL
with 'select' and AH with 'value' and using a 16-bit OUT
instruction.  Of course they are assuming, and rightly so, that the
low byte is sent during the first cycle (as 8088).

Since the EGA port pairs start at even addresses, when the 16-bit
OUT instruction is executed on the 6300, the bus converter sends
the 'value' byte during the first cycle and the 'select' byte
follows.  Guess what, it doesn't work.

For now software with this problem can be patched by replacing OUTW
with INT 3 and writing an interrupt handler to output the bytes in
the correct order.

If all software developers would stop using OUTW in EGA S/W then
life would be bliss.

j baker

All technical corrections are appreciated.