Path: utzoo!utgpu!jarvis.csri.toronto.edu!clyde.concordia.ca!uunet!willett!ForthNet
From: ForthNet@willett.UUCP (ForthNet articles from GEnie)
Newsgroups: comp.lang.forth
Subject: Graphics
Message-ID: <292.UUL1.3#5129@willett.UUCP>
Date: 19 Jan 90 23:18:46 GMT
Organization: Latest Link in ForthNet Chain (Pittsburgh, PA)
Lines: 67


 Date: 01-17-90 (23:48)              Number: 1716 (Echo)
   To: MARK SMILEY                   Refer#: 1710
 From: MIKE SPERL                      Read: NO
 Subj: GRAPHICS                      Status: PUBLIC MESSAGE

 Mark,

 > LABEL DODEFER   ( addr -- )    <-- expects a number on TOS
 old diagram; should be ( -- ) if my suggestion is adopted.
 >                MOV BX, AX      <-- what's in AX?
 see discussion below
 >                MOV AX,3 [BX]
 >                JMP AX      END-CODE  <-- how was addr used?
 it wasn't there to be used, actually.

      My one track mind neglected to change the stack diagram when 
 pasting it into the last message; the new one is just ( -- ).
 Lets look at the senario of a word calling a deferred word.  

 defer mark
 : t1 ." I'm Mark." cr ;     ' t1 is mark
 : t2  ." I'm testing mark, and " mark ; \ run T2 then

 ' mark 1 dump (dump cfa's in code space)

             81fa = mark     81ff = t1       8204 = t2
 UMMMMMMMMMQMMMMMMMMMMMMMMMQMMMMMMMMMMMMMMMQMMMMMMMMMMMMMMMM
 3 SEG:OFF 3  A  B  C  D  E3  F  0  1  2  33  4  5  6  7  8
 FMMMMMMMMMXMMMMMMMMMMMMMMMXMMMMMMMMMMMMMMMXMMMMMMMMMMMMMMMM
 34BDF:81FA3 E8 7B 81 FF 813 E9 57 80 C9 143 E9 52 80 CA 14
 TMMMMMMMMMOMMMMMMMMMMMMMMMOMMMMMMMMMMMMMMMOMMMMMMMMMMMMMMMM
             cfa of mark     cfa of t1       cfa of t2


 E8 7B 81 == CALL DODEFER (call $0398, you can set a breakpoint
                            at 081fa in D86 to see it work)
 81FF is placed on the stack by CALL as the "return address", 
 explaining the PRESENT stack diagram.  Notice that it is the 
 address of T1, the word that will be run by DODEFER.  This is the 
 same trick used in F-PC by VARIABLE to push the address of its 
 body. 

     Now, to NEXT.  NEXT does ES: LODSW to get the cfa from list 
 space into ax, and jumps to the code with JMP AX.  In the deferred 
 word, the code jumped to is CALL DODEFER, compiled in MARK by 
 DEFER. The CALL places the address of the calling word's body on 
 the stack (it thinks we'll return, and that's where the next 
 instruction should be).  AX isn't changed, so the cfa of the 
 deferred word is still in ax when DODEFER runs; my suggestion is 
 simply to move it to bx and MOV AX, 3 [BX] (which takes the same 
 number of clocks as the present MOV AX, 0 [BX]) thus saving the 
 overhead of a CALL and POP, which will make the whole process 
 twice as fast.  For this to work properly DEFER also must be 
 modified as was indicated previously to change the CALL compiled 
 in the deferred word by DEFER to JMP (or the stack will accumulate 
 all those return addresses!).  Clear as mud?  Use D86 to step 
 through it.

     Setting a breakpoint at $0398 reveals the remarkable number of 
 deferred words in the keyboard input loop - almost twenty in F-PC. 

     - Mike -

-----
This message came from GEnie via willett through a semi-automated process.
Report problems to: 'uunet!willett!dwp' or 'willett!dwp@gateway.sei.cmu.edu'