--- gforth/prims2x.fs	2001/02/06 16:53:06	1.77
+++ gforth/prims2x.fs	2002/08/28 21:46:58	1.112
@@ -40,26 +40,29 @@
 \ add the store optimization for doubles
 \ regarding problem 1 above: It would be better (for over) to implement
 \ 	the alternative
+\ store optimization for combined instructions.
 
-warnings off
+\ Design Uglyness:
 
-[IFUNDEF] vocabulary	\ we are executed just with kernel image
-			\ load the rest that is needed
-			\ (require fails because this file is needed from a
-			\ different directory with the wordlibraries)
-include ./search.fs			
-include ./extend.fs
-[THEN]
-include ./stuff.fs
+\ - global state (values, variables) in connection with combined instructions.
+
+\ - index computation is different for instruction-stream and the
+\ stacks; there are two mechanisms for dealing with that
+\ (stack-in-index-xt and a test for stack==instruction-stream); there
+\ should be only one.
 
-[IFUNDEF] environment?
-include ./environ.fs
+warnings off
+
+[IFUNDEF] try
+include startup.fs
 [THEN]
 
 : struct% struct ; \ struct is redefined in gray
 
-include ./gray.fs
+warnings off
+\ warnings on
 
+include ./gray.fs
 32 constant max-effect \ number of things on one side of a stack effect
 4 constant max-stacks  \ the max. number of stacks (including inst-stream).
 255 constant maxchar
@@ -75,10 +78,17 @@ variable line-start \ pointer to start o
 0 line !
 2variable filename \ filename of original input file
 0 0 filename 2!
+2variable out-filename \ filename of the output file (for sync lines)
+0 0 out-filename 2!
 2variable f-comment
 0 0 f-comment 2!
 variable skipsynclines \ are sync lines ("#line ...") invisible to the parser?
-skipsynclines on 
+skipsynclines on
+variable out-nls \ newlines in output (for output sync lines)
+0 out-nls !
+variable store-optimization \ use store optimization?
+store-optimization off
+
 
 : th ( addr1 n -- addr2 )
     cells + ;
@@ -90,6 +100,13 @@ skipsynclines on
     loop
     drop ;
 
+: insert-wordlist { c-addr u wordlist xt -- }
+    \ adds name "addr u" to wordlist using defining word xt
+    \ xt may cause additional stack effects
+    get-current >r wordlist set-current
+    c-addr u nextname xt execute
+    r> set-current ;
+
 : start ( -- addr )
  cookedinput @ ;
 
@@ -111,23 +128,31 @@ skipsynclines on
 	    0
 	recover endtry
 	r> to outfile-id throw
-	abort
+	1 (bye) \ abort
     endif ;
 
 : quote ( -- )
     [char] " emit ;
 
+\ count output lines to generate sync lines for output
+
+: count-nls ( addr u -- )
+    bounds u+do
+	i c@ nl-char = negate out-nls +!
+    loop ;
+
+:noname ( addr u -- )
+    2dup count-nls
+    defers type ;
+is type
+
 variable output          \ xt ( -- ) of output word for simple primitives
 variable output-combined \ xt ( -- ) of output word for combined primitives
 
-: printprim ( -- )
- output @ execute ;
-
 struct%
     cell%    field stack-number \ the number of this stack
     cell% 2* field stack-pointer \ stackpointer name
     cell%    field stack-type \ name for default type of stack items
-    cell% 2* field stack-cast \ cast string for assignments to stack elements
     cell%    field stack-in-index-xt \ ( in-size item -- in-index )
 end-struct stack%
 
@@ -156,15 +181,22 @@ create stacks max-stacks cells allot \ a
 : inst-in-index ( in-size item -- in-index )
     nip dup item-offset @ swap item-type @ type-size @ + 1- ;
 
-: make-stack ( addr-ptr u1 type addr-cast u2 "stack-name" -- )
+: make-stack ( addr-ptr u1 type "stack-name" -- )
+    next-stack-number @ max-stacks < s" too many stacks" ?print-error
     create stack% %allot >r
     r@ stacks next-stack-number @ th !
-    next-stack-number @ r@ stack-number !  1 next-stack-number +!
-    save-mem r@ stack-cast 2!
+    next-stack-number @ r@ stack-number !
+    1 next-stack-number +!
     r@ stack-type !
     save-mem r@ stack-pointer 2! 
     ['] stack-in-index r> stack-in-index-xt ! ;
 
+: map-stacks { xt -- }
+    \ perform xt for all stacks except inst-stream
+    next-stack-number @ 1 +do
+	stacks i th @ xt execute
+    loop ;
+
 \ stack items
 
 : init-item ( addr u addr1 -- )
@@ -194,6 +226,7 @@ struct%
     cell% 2* field prim-c-code
     cell% 2* field prim-forth-code
     cell% 2* field prim-stack-string
+    cell%    field prim-num            \ ordinal number
     cell%    field prim-items-wordlist \ unique items
     item% max-effect * field prim-effect-in
     item% max-effect * field prim-effect-out
@@ -208,15 +241,26 @@ end-struct prim%
     s" " p prim-doc 2! s" " p prim-forth-code 2! s" " p prim-wordset 2!
     p ;
 
-0 value prim
+0 value prim     \ in combined prims either combined or a part
+0 value combined \ in combined prims the combined prim
+variable in-part \ true if processing a part
+ in-part off
+
+1000 constant max-combined
+create combined-prims max-combined cells allot
+variable num-combined
+
+table constant combinations
+  \ the keys are the sequences of pointers to primitives
+
+create current-depth max-stacks cells allot
+create max-depth     max-stacks cells allot
+create min-depth     max-stacks cells allot
 
 wordlist constant primitives
 
 : create-prim ( prim -- )
-    get-current >r
-    primitives set-current
-    dup prim-name 2@ nextname constant
-    r> set-current ;
+    dup prim-name 2@ primitives ['] constant insert-wordlist ;
 
 : stack-in ( stack -- addr )
     \ address of number of stack items in effect in
@@ -242,10 +286,9 @@ Variable function-number 0 function-numb
 : complement ( set1 -- set2 )
  empty ['] bit-equivalent binary-set-operation ;
 
-\ types
+\ stack access stuff
 
-: stack-access ( n stack -- )
-    \ print a stack access at index n of stack
+: normal-stack-access ( n stack -- )
     stack-pointer 2@ type
     dup
     if
@@ -254,30 +297,60 @@ Variable function-number 0 function-numb
 	drop ." TOS"
     endif ;
 
+\ forward declaration for inst-stream (breaks cycle in definitions)
+defer inst-stream-f ( -- stack )
+
+: part-stack-access { n stack -- }
+    \ print _<stack><x>, x=inst-stream? n : maxdepth-currentdepth-n-1
+    ." _" stack stack-pointer 2@ type
+    stack stack-number @ { stack# }
+    current-depth stack# th @ n + { access-depth }
+    stack inst-stream-f = if
+	access-depth
+    else
+	combined prim-stacks-in stack# th @
+	assert( dup max-depth stack# th @ = )
+	access-depth - 1-
+    endif
+    0 .r ;
+
+: stack-access ( n stack -- )
+    \ print a stack access at index n of stack
+    in-part @ if
+	part-stack-access
+    else
+	normal-stack-access
+    endif ;
+
 : item-in-index { item -- n }
     \ n is the index of item (in the in-effect)
     item item-stack @ dup >r stack-in @ ( in-size r:stack )
     item r> stack-in-index-xt @ execute ;
 
+: item-stack-type-name ( item -- addr u )
+    item-stack @ stack-type @ type-c-name 2@ ;
+
 : fetch-single ( item -- )
- \ fetch a single stack item from its stack
- >r
- r@ item-name 2@ type
- ."  = vm_" r@ item-stack @ stack-type @ type-c-name 2@ type
- ." 2" r@ item-type @ print-type-prefix ." ("
- r@ item-in-index r@ item-stack @ stack-access
- ." );" cr
- rdrop ; 
+    \ fetch a single stack item from its stack
+    >r
+    ." vm_" r@ item-stack-type-name type
+    ." 2" r@ item-type @ print-type-prefix ." ("
+    r@ item-in-index r@ item-stack @ stack-access ." ,"
+    r@ item-name 2@ type
+    ." );" cr
+    rdrop ; 
 
 : fetch-double ( item -- )
- \ fetch a double stack item from its stack
- >r
- ." FETCH_DCELL("
- r@ item-name 2@ type ." , "
- r@ item-in-index r@ item-stack @ 2dup ." (Cell)" stack-access
- ." , "                      -1 under+ ." (Cell)" stack-access
- ." );" cr
- rdrop ;
+    \ fetch a double stack item from its stack
+    >r
+    ." vm_two"
+    r@ item-stack-type-name type ." 2"
+    r@ item-type @ print-type-prefix ." ("
+    r@ item-in-index r@ item-stack @ 2dup ." (Cell)" stack-access
+    ." , "                      -1 under+ ." (Cell)" stack-access
+    ." , " r@ item-name 2@ type
+    ." )" cr
+    rdrop ;
 
 : same-as-in? ( item -- f )
  \ f is true iff the offset and stack of item is the same as on input
@@ -301,34 +374,36 @@ Variable function-number 0 function-numb
     >r r@ item-stack @ stack-out @ r> item-offset @ - 1- ;
 
 : really-store-single ( item -- )
- >r
- r@ item-out-index r@ item-stack @ stack-access ."  = vm_"
- r@ item-type @ print-type-prefix ." 2"
- r@ item-stack @ stack-type @ type-c-name 2@ type ." ("
- r@ item-name 2@ type ." );"
- rdrop ;
+    >r
+    ." vm_"
+    r@ item-type @ print-type-prefix ." 2"
+    r@ item-stack-type-name type ." ("
+    r@ item-name 2@ type ." ,"
+    r@ item-out-index r@ item-stack @ stack-access ." );"
+    rdrop ;
 
 : store-single ( item -- )
- >r
- r@ same-as-in?
- if
-   r@ item-in-index 0= r@ item-out-index 0= xor
-   if
-       ." IF_" r@ item-stack @ stack-pointer 2@ type
-       ." TOS(" r@ really-store-single ." );" cr
-   endif
- else
-   r@ really-store-single cr
- endif
- rdrop ;
+    >r
+    store-optimization @ r@ same-as-in? and if
+	r@ item-in-index 0= r@ item-out-index 0= xor if
+	    ." IF_" r@ item-stack @ stack-pointer 2@ type
+	    ." TOS(" r@ really-store-single ." );" cr
+	endif
+    else
+	r@ really-store-single cr
+    endif
+    rdrop ;
 
 : store-double ( item -- )
 \ !! store optimization is not performed, because it is not yet needed
  >r
- ." STORE_DCELL(" r@ item-name 2@ type ." , "
+ ." vm_"
+ r@ item-type @ print-type-prefix ." 2two"
+ r@ item-stack-type-name type ." ("
+ r@ item-name 2@ type ." , "
  r@ item-out-index r@ item-stack @ 2dup stack-access
  ." , "                       -1 under+ stack-access
- ." );" cr
+ ." )" cr
  rdrop ;
 
 : single ( -- xt1 xt2 n )
@@ -366,8 +441,9 @@ wordlist constant prefixes
     stack r@ type-stack !
     rdrop ;
 
-: type-prefix ( xt1 xt2 n stack "prefix" -- )
-    create-type
+: type-prefix ( addr u xt1 xt2 n stack "prefix" -- )
+    get-current >r prefixes set-current
+    create-type r> set-current
 does> ( item -- )
     \ initialize item
     { item typ }
@@ -393,7 +469,7 @@ does> ( item -- )
 	endif
 	-1 s+loop
     \ we did not find a type, abort
-    true abort" unknown prefix" ;
+    false s" unknown prefix" ?print-error ;
 
 : declaration ( item -- )
     dup item-name 2@ execute-prefix ;
@@ -417,7 +493,9 @@ does> ( item -- )
     prim prim-effect-out prim prim-effect-out-end @ ['] print-declaration map-items ;
     
 : stack-prefix ( stack "prefix" -- )
+    get-current >r prefixes set-current
     name tuck nextname create ( stack length ) 2,
+    r> set-current
 does> ( item -- )
     2@ { item stack prefix-length }
     item item-name 2@ prefix-length /string item item-name 2!
@@ -425,14 +503,26 @@ does> ( item -- )
     item declaration ;
 
 \ types pointed to by stacks for use in combined prims
-s" Cell"  single 0 create-type cell-type
-s" Float" single 0 create-type float-type
+\ !! output-c-combined shouldn't use these names!
+: stack-type-name ( addr u "name" -- )
+    single 0 create-type ;
+
+wordlist constant type-names \ this is here just to meet the requirement
+                    \ that a type be a word; it is never used for lookup
+
+: stack ( "name" "stack-pointer" "type" -- )
+    \ define stack
+    name { d: stack-name }
+    name { d: stack-pointer }
+    name { d: stack-type }
+    get-current type-names set-current
+    stack-type 2dup nextname stack-type-name
+    set-current
+    stack-pointer lastxt >body stack-name nextname make-stack ;
 
-s" sp" save-mem cell-type  s" (Cell)" make-stack data-stack 
-s" fp" save-mem float-type s" "       make-stack fp-stack
-s" rp" save-mem cell-type  s" (Cell)" make-stack return-stack
-s" IP" save-mem cell-type  s" error don't use # on results" make-stack inst-stream
+stack inst-stream IP Cell
 ' inst-in-index inst-stream stack-in-index-xt !
+' inst-stream <is> inst-stream-f
 \ !! initialize stack-in and stack-out
 
 \ offset computation
@@ -455,11 +545,17 @@ s" IP" save-mem cell-type  s" error don'
     dup stack-in off stack-out off ;
 
 : compute-offsets ( -- )
-    data-stack clear-stack  fp-stack clear-stack return-stack clear-stack
+    ['] clear-stack map-stacks
     inst-stream clear-stack
     prim prim-effect-in  prim prim-effect-in-end  @ ['] compute-offset-in  map-items
     prim prim-effect-out prim prim-effect-out-end @ ['] compute-offset-out map-items
-    inst-stream stack-out @ 0<> abort" # can only be on the input side" ;
+    inst-stream stack-out @ 0= s" # can only be on the input side" ?print-error ;
+
+: process-simple ( -- )
+    prim prim { W^ key } key cell
+    combinations ['] constant insert-wordlist
+    declarations compute-offsets
+    output @ execute ;
 
 : flush-a-tos { stack -- }
     stack stack-out @ 0<> stack stack-in @ 0= and
@@ -469,9 +565,7 @@ s" IP" save-mem cell-type  s" error don'
     endif ;
 
 : flush-tos ( -- )
-    data-stack   flush-a-tos
-    fp-stack     flush-a-tos
-    return-stack flush-a-tos ;
+    ['] flush-a-tos map-stacks ;
 
 : fill-a-tos { stack -- }
     stack stack-out @ 0= stack stack-in @ 0<> and
@@ -482,9 +576,7 @@ s" IP" save-mem cell-type  s" error don'
 
 : fill-tos ( -- )
     \ !! inst-stream for prefetching?
-    fp-stack     fill-a-tos
-    data-stack   fill-a-tos
-    return-stack fill-a-tos ;
+    ['] fill-a-tos map-stacks ;
 
 : fetch ( addr -- )
     dup item-type @ type-fetch @ execute ;
@@ -506,9 +598,7 @@ s" IP" save-mem cell-type  s" error don'
 
 : stack-pointer-updates ( -- )
     inst-pointer-update
-    data-stack   stack-pointer-update
-    fp-stack     stack-pointer-update
-    return-stack stack-pointer-update ;
+    ['] stack-pointer-update map-stacks ;
 
 : store ( item -- )
 \ f is true if the item should be stored
@@ -518,25 +608,6 @@ s" IP" save-mem cell-type  s" error don'
 : stores ( -- )
     prim prim-effect-out prim prim-effect-out-end @ ['] store map-items ;
 
-: output-c-tail ( -- )
-    \ the final part of the generated C code
-    ." NEXT_P1;" cr
-    stores
-    fill-tos
-    ." NEXT_P2;" cr ;
-
-: type-c ( c-addr u -- )
-    \ like TYPE, but replaces "TAIL;" with tail code
-    begin ( c-addr1 u1 )
-	2dup s" TAIL;" search
-    while ( c-addr1 u1 c-addr3 u3 )
-	2dup 2>r drop nip over - type
-	output-c-tail
-	2r> 5 /string
-	\ !! resync #line missing
-    repeat
-    2drop type ;
-
 : print-debug-arg { item -- }
     ." fputs(" quote space item item-name 2@ type ." =" quote ." , vm_out); "
     ." printarg_" item item-type @ print-type-prefix
@@ -546,40 +617,95 @@ s" IP" save-mem cell-type  s" error don'
     ." #ifdef VM_DEBUG" cr
     ." if (vm_debug) {" cr
     prim prim-effect-in prim prim-effect-in-end @ ['] print-debug-arg map-items
+\    ." fputc('\n', vm_out);" cr
+    ." }" cr
+    ." #endif" cr ;
+
+: print-debug-result { item -- }
+    item item-first @ if
+	item print-debug-arg
+    endif ;
+
+: print-debug-results ( -- )
+    cr
+    ." #ifdef VM_DEBUG" cr
+    ." if (vm_debug) {" cr
+    ." fputs(" quote ."  -- " quote ." , vm_out); "
+    prim prim-effect-out prim prim-effect-out-end @ ['] print-debug-result map-items
     ." fputc('\n', vm_out);" cr
     ." }" cr
     ." #endif" cr ;
 
+: output-super-end ( -- )
+    prim prim-c-code 2@ s" SET_IP" search if
+	." SUPER_END;" cr
+    endif
+    2drop ;
+
+: output-c-tail1 ( -- )
+    \ the final part of the generated C code except LABEL2 and NEXT_P2
+    output-super-end
+    print-debug-results
+    ." NEXT_P1;" cr
+    stores
+    fill-tos ;
+
+: output-c-tail ( -- )
+    \ the final part of the generated C code, without LABEL2
+    output-c-tail1
+    ." NEXT_P2;" ;
+
+: output-c-tail2 ( -- )
+    \ the final part of the generated C code, including LABEL2
+    output-c-tail1
+    ." LABEL2(" prim prim-c-name 2@ type ." )" cr
+    ." NEXT_P2;" cr ;
+
+: type-c-code ( c-addr u xt -- )
+    \ like TYPE, but replaces "INST_TAIL;" with tail code produced by xt
+    { xt }
+    ." {" cr
+    ." #line " c-line @ . quote c-filename 2@ type quote cr
+    begin ( c-addr1 u1 )
+	2dup s" INST_TAIL;" search
+    while ( c-addr1 u1 c-addr3 u3 )
+	2dup 2>r drop nip over - type
+	xt execute
+	2r> 10 /string
+	\ !! resync #line missing
+    repeat
+    2drop type
+    ." #line " out-nls @ 2 + . quote out-filename 2@ type quote cr
+    ." }" cr ;
+
 : print-entry ( -- )
-    ." I_" prim prim-c-name 2@ type ." :" ;
+    ." LABEL(" prim prim-c-name 2@ type ." )" ;
     
 : output-c ( -- ) 
- print-entry ."  /* " prim prim-name 2@ type ."  ( " prim prim-stack-string 2@ type ." ) */" cr
- ." /* " prim prim-doc 2@ type ."  */" cr
- ." NAME(" quote prim prim-name 2@ type quote ." )" cr \ debugging
- ." {" cr
- ." DEF_CA" cr
- print-declarations
- ." NEXT_P0;" cr
- flush-tos
- fetches
- print-debug-args
- stack-pointer-updates
- ." {" cr
- ." #line " c-line @ . quote c-filename 2@ type quote cr
- prim prim-c-code 2@ type-c
- ." }" cr
- output-c-tail
- ." }" cr
- cr
+    print-entry ."  /* " prim prim-name 2@ type ."  ( " prim prim-stack-string 2@ type ." ) */" cr
+    ." /* " prim prim-doc 2@ type ."  */" cr
+    ." NAME(" quote prim prim-name 2@ type quote ." )" cr \ debugging
+    ." {" cr
+    ." DEF_CA" cr
+    print-declarations
+    ." NEXT_P0;" cr
+    flush-tos
+    fetches
+    print-debug-args
+    stack-pointer-updates
+    prim prim-c-code 2@ ['] output-c-tail type-c-code
+    output-c-tail2
+    ." }" cr
+    cr
 ;
 
 : disasm-arg { item -- }
     item item-stack @ inst-stream = if
-	."   fputc(' ', vm_out); "
-	." printarg_" item item-type @ print-type-prefix
-	." ((" item item-type @ type-c-name 2@ type ." )"
-	." ip[" item item-offset @ 1+ 0 .r ." ]);" cr
+	." {" cr
+	item print-declaration
+	item fetch
+	item print-debug-arg
+	." }" cr
     endif ;
 
 : disasm-args ( -- )
@@ -587,11 +713,49 @@ s" IP" save-mem cell-type  s" error don'
 
 : output-disasm ( -- )
     \ generate code for disassembling VM instructions
-    ." if (ip[0] == prim[" function-number @ 0 .r ." ]) {" cr
+    ." if (VM_IS_INST(*ip, " function-number @ 0 .r ." )) {" cr
     ."   fputs(" quote prim prim-name 2@ type quote ." , vm_out);" cr
     disasm-args
     ."   ip += " inst-stream stack-in @ 1+ 0 .r ." ;" cr
-    ." } else " ;
+    ."   goto _endif_;" cr
+    ." }" cr ;
+
+: output-profile ( -- )
+    \ generate code for postprocessing the VM block profile stuff
+    ." if (VM_IS_INST(*ip, " function-number @ 0 .r ." )) {" cr
+    ."   add_inst(b, " quote prim prim-name 2@ type quote ." );" cr
+    ."   ip += " inst-stream stack-in @ 1+ 0 .r ." ;" cr
+    prim prim-c-code 2@  s" SET_IP"    search nip nip
+    prim prim-c-code 2@  s" SUPER_END" search nip nip or if
+	."   return;" cr
+    else
+	."   goto _endif_;" cr
+    endif
+    ." }" cr ;
+
+: output-profile-combined ( -- )
+    \ generate code for postprocessing the VM block profile stuff
+    ." if (VM_IS_INST(*ip, " function-number @ 0 .r ." )) {" cr
+    num-combined @ 0 +do
+	."   add_inst(b, " quote
+	combined-prims i th @ prim-name 2@ type
+	quote ." );" cr
+    loop
+    ."   ip += " inst-stream stack-in @ 1+ 0 .r ." ;" cr
+    combined-prims num-combined @ 1- th @ prim-c-code 2@  s" SET_IP"    search nip nip
+    combined-prims num-combined @ 1- th @ prim-c-code 2@  s" SUPER_END" search nip nip or if
+	."   return;" cr
+    else
+	."   goto _endif_;" cr
+    endif
+    ." }" cr ;
+
+: output-superend ( -- )
+    \ output flag specifying whether the current word ends a dynamic superinst
+    prim prim-c-code 2@  s" SET_IP"    search nip nip
+    prim prim-c-code 2@  s" SUPER_END" search nip nip or 0<>
+    prim prim-c-code 2@  s" SUPER_CONTINUE" search nip nip 0= and
+    negate 0 .r ." , /* " prim prim-name 2@ type ."  */" cr ;
 
 : gen-arg-parm { item -- }
     item item-stack @ inst-stream = if
@@ -628,35 +792,36 @@ s" IP" save-mem cell-type  s" error don'
 : output-forthname ( -- )
   '" emit prim prim-name 2@ type '" emit ." ," cr ;
 
-: output-c-func ( -- )
-\ used for word libraries
-    ." Cell * I_" prim prim-c-name 2@ type ." (Cell *SP, Cell **FP)      /* " prim prim-name 2@ type
-    ."  ( " prim prim-stack-string 2@ type ."  ) */" cr
-    ." /* " prim prim-doc 2@ type ."  */" cr
-    ." NAME(" quote prim prim-name 2@ type quote ." )" cr
-    \ debugging
-    ." {" cr
-    print-declarations
-    inst-stream  stack-used? IF ." Cell *ip=IP;" cr THEN
-    data-stack   stack-used? IF ." Cell *sp=SP;" cr THEN
-    fp-stack     stack-used? IF ." Cell *fp=*FP;" cr THEN
-    return-stack stack-used? IF ." Cell *rp=*RP;" cr THEN
-    flush-tos
-    fetches
-    stack-pointer-updates
-    fp-stack   stack-used? IF ." *FP=fp;" cr THEN
-    ." {" cr
-    ." #line " c-line @ . quote c-filename 2@ type quote cr
-    prim prim-c-code 2@ type
-    ." }" cr
-    stores
-    fill-tos
-    ." return (sp);" cr
-    ." }" cr
-    cr ;
+\  : output-c-func ( -- )
+\  \ used for word libraries
+\      ." Cell * I_" prim prim-c-name 2@ type ." (Cell *SP, Cell **FP)      /* " prim prim-name 2@ type
+\      ."  ( " prim prim-stack-string 2@ type ."  ) */" cr
+\      ." /* " prim prim-doc 2@ type ."  */" cr
+\      ." NAME(" quote prim prim-name 2@ type quote ." )" cr
+\      \ debugging
+\      ." {" cr
+\      print-declarations
+\      \ !! don't know what to do about that
+\      inst-stream  stack-used? IF ." Cell *ip=IP;" cr THEN
+\      data-stack   stack-used? IF ." Cell *sp=SP;" cr THEN
+\      fp-stack     stack-used? IF ." Cell *fp=*FP;" cr THEN
+\      return-stack stack-used? IF ." Cell *rp=*RP;" cr THEN
+\      flush-tos
+\      fetches
+\      stack-pointer-updates
+\      fp-stack   stack-used? IF ." *FP=fp;" cr THEN
+\      ." {" cr
+\      ." #line " c-line @ . quote c-filename 2@ type quote cr
+\      prim prim-c-code 2@ type
+\      ." }" cr
+\      stores
+\      fill-tos
+\      ." return (sp);" cr
+\      ." }" cr
+\      cr ;
 
 : output-label ( -- )  
-    ." (Label)&&I_" prim prim-c-name 2@ type ." ," cr ;
+    ." INST_ADDR(" prim prim-c-name 2@ type ." )," cr ;
 
 : output-alias ( -- ) 
     ( primitive-number @ . ." alias " ) ." Primitive " prim prim-name 2@ type cr ;
@@ -688,16 +853,19 @@ s" IP" save-mem cell-type  s" error don'
     name-line @ 0 .r
     ." ,0" cr ;
 
+: output-vi-tag ( -- )
+    name-filename 2@ type #tab emit
+    prim prim-name 2@ type #tab emit
+    ." /^" prim prim-name 2@ type ."  *(/" cr ;
+
 [IFDEF] documentation
 : register-doc ( -- )
-    get-current documentation set-current
-    prim prim-name 2@ nextname create
+    prim prim-name 2@ documentation ['] create insert-wordlist
     prim prim-name 2@ 2,
     prim prim-stack-string 2@ condition-stack-effect 2,
     prim prim-wordset 2@ 2,
     prim prim-c-name 2@ condition-pronounciation 2,
-    prim prim-doc 2@ 2,
-    set-current ;
+    prim prim-doc 2@ 2, ;
 [THEN]
 
 
@@ -759,23 +927,14 @@ s" IP" save-mem cell-type  s" error don'
 \  #line 516 "./prim"
 \  n = n1+n2;
 \  }
-\  NEXT_P1;
 \  _x_sp0 = (Cell)n;
-\  NEXT_P2;
 \  }
 \  NEXT_P1;
 \  spTOS = (Cell)_x_sp0;
 \  NEXT_P2;
 
-1000 constant max-combined
-create combined-prims max-combined cells allot
-variable num-combined
-
-create current-depth max-stacks cells allot
-create max-depth     max-stacks cells allot
-create min-depth     max-stacks cells allot
-
 : init-combined ( -- )
+    prim to combined
     0 num-combined !
     current-depth max-stacks cells erase
     max-depth     max-stacks cells erase
@@ -840,6 +999,10 @@ create min-depth     max-stacks cells al
     loop ;
 
 : process-combined ( -- )
+    combined combined-prims num-combined @ cells
+    combinations ['] constant insert-wordlist
+    combined-prims num-combined @ 1- th ( last-part )
+    @ prim-c-code 2@ prim prim-c-code 2! \ used by output-super-end
     prim compute-effects
     prim init-effects
     output-combined perform ;
@@ -848,8 +1011,8 @@ create min-depth     max-stacks cells al
 
 : print-item { n stack -- }
     \ print nth stack item name
-    ." _" stack stack-type @ type-c-name 2@ type space
-    stack stack-pointer 2@ type n 0 .r ;
+    stack stack-type @ type-c-name 2@ type space
+    ." _" stack stack-pointer 2@ type n 0 .r ;
 
 : print-declarations-combined ( -- )
     max-stacks 0 ?do
@@ -857,13 +1020,41 @@ create min-depth     max-stacks cells al
 	    i stacks j th @ print-item ." ;" cr
 	loop
     loop ;
-    
+
+: part-fetches ( -- )
+    fetches ;
+
+: part-output-c-tail ( -- )
+    print-debug-results
+    stores ;
+
+: output-combined-tail ( -- )
+    part-output-c-tail
+    prim >r combined to prim
+    in-part @ >r in-part off
+    output-c-tail
+    r> in-part ! r> to prim ;
+
+: output-part ( p -- )
+    to prim
+    ." /* " prim prim-name 2@ type ."  ( " prim prim-stack-string 2@ type ." ) */" cr
+    ." NAME(" quote prim prim-name 2@ type quote ." )" cr \ debugging
+    ." {" cr
+    print-declarations
+    part-fetches
+    print-debug-args
+    prim add-depths \ !! right place?
+    prim prim-c-code 2@ ['] output-combined-tail type-c-code
+    part-output-c-tail
+    ." }" cr ;
+
 : output-parts ( -- )
-    prim >r
+    prim >r in-part on
+    current-depth max-stacks cells erase
     num-combined @ 0 +do
-	combined-prims i th @ to prim
-	output-c
+	combined-prims i th @ output-part
     loop
+    in-part off
     r> to prim ;
 
 : output-c-combined ( -- )
@@ -878,12 +1069,39 @@ create min-depth     max-stacks cells al
     \ print-debug-args
     stack-pointer-updates
     output-parts
-    output-c-tail
+    output-c-tail2
     ." }" cr
     cr ;
 
 : output-forth-combined ( -- )
-    ;
+;
+
+
+\ peephole optimization rules
+
+\ in order for this to work as intended, shorter combinations for each
+\ length must be present, and the longer combinations must follow
+\ shorter ones (this restriction may go away in the future).
+  
+: output-peephole ( -- )
+    combined-prims num-combined @ 1- cells combinations search-wordlist
+    s" the prefix for this combination must be defined earlier" ?print-error
+    ." {"
+    execute prim-num @ 5 .r ." ,"
+    combined-prims num-combined @ 1- th @ prim-num @ 5 .r ." ,"
+    combined prim-num @ 5 .r ." }, /* "
+    combined prim-c-name 2@ type ."  */"
+    cr ;
+
+: output-forth-peephole ( -- )
+    combined-prims num-combined @ 1- cells combinations search-wordlist
+    s" the prefix for this combination must be defined earlier" ?print-error
+    execute prim-num @ 5 .r
+    combined-prims num-combined @ 1- th @ prim-num @ 5 .r
+    combined prim-num @ 5 .r ."  prim, \ "
+    combined prim-c-name 2@ type
+    cr ;
+
 
 \ the parser
 
@@ -918,11 +1136,11 @@ print-token !
     endif
     0. r> 6 chars + 20 >number drop >r drop line ! r> ( c-addr )
     dup c@ bl = if
-	char+ dup c@ [char] " <> abort" sync line syntax"
+	char+ dup c@ [char] " <> 0= s" sync line syntax" ?print-error
 	char+ dup 100 [char] " scan drop swap 2dup - save-mem filename 2!
 	char+
     endif
-    dup c@ nl-char <> abort" sync line syntax"
+    dup c@ nl-char <> 0= s" sync line syntax" ?print-error
     skipsynclines @ if
 	dup char+ rawinput !
 	rawinput @ c@ cookedinput @ c!
@@ -962,18 +1180,20 @@ bl singleton tab-char over add-member
 nl-char singleton eof-char over add-member complement	charclass nonl
 nl-char singleton eof-char over add-member
     char : over add-member complement                   charclass nocolonnl
+nl-char singleton eof-char over add-member
+    char } over add-member complement                   charclass nobracenl
 bl 1+ maxchar .. char \ singleton complement intersection
                                                         charclass nowhitebq
 bl 1+ maxchar ..                                        charclass nowhite
 char " singleton eof-char over add-member complement	charclass noquote
 nl-char singleton					charclass nl
 eof-char singleton					charclass eof
-
+nl-char singleton eof-char over add-member		charclass nleof
 
 (( letter (( letter || digit )) **
 )) <- c-ident ( -- )
 
-(( ` # ?? (( letter || digit || ` : )) **
+(( ` # ?? (( letter || digit || ` : )) ++
 )) <- stack-ident ( -- )
 
 (( nowhitebq nowhite ** ))
@@ -1012,9 +1232,15 @@ Variable c-flag
 	THEN }}
 )) <- if-comment
 
-(( (( eval-comment || forth-comment || c-comment || else-comment || if-comment )) ?? nonl ** )) <- comment-body
+(( (( ` g || ` G )) {{ start }} nonl **
+   {{ end
+      forth-flag @ IF  ." group " type cr  THEN
+      c-flag @     IF  ." GROUP(" type ." )" cr  THEN }}
+)) <- group-comment
+
+(( (( eval-comment || forth-comment || c-comment || else-comment || if-comment || group-comment )) ?? nonl ** )) <- comment-body
 
-(( ` \ comment-body nl )) <- comment ( -- )
+(( ` \ comment-body nleof )) <- comment ( -- )
 
 (( {{ start }} stack-ident {{ end 2 pick init-item item% %size + }} white ** )) **
 <- stack-items
@@ -1028,24 +1254,30 @@ Variable c-flag
    ` ( white ** {{ start }} stack-effect {{ end prim prim-stack-string 2! }} ` ) white **
    (( {{ start }} forth-ident {{ end prim prim-wordset 2! }} white **
       (( {{ start }}  c-ident {{ end prim prim-c-name 2! }} )) ??
-   )) ??  nl
-   (( ` " ` "  {{ start }} (( noquote ++ ` " )) ++ {{ end 1- prim prim-doc 2! }} ` " white ** nl )) ??
-   {{ skipsynclines off line @ c-line ! filename 2@ c-filename 2! start }} (( nocolonnl nonl **  nl white ** )) ** {{ end prim prim-c-code 2! skipsynclines on }}
-   (( ` :  white ** nl
-      {{ start }} (( nonl ++  nl white ** )) ++ {{ end prim prim-forth-code 2! }}
-   )) ?? {{  declarations compute-offsets printprim 1 function-number +! }}
-   (( nl || eof ))
+   )) ??  nleof
+   (( ` " ` "  {{ start }} (( noquote ++ ` " )) ++ {{ end 1- prim prim-doc 2! }} ` " white ** nleof )) ??
+   {{ skipsynclines off line @ c-line ! filename 2@ c-filename 2! start }}
+   (( (( ` { nonl ** nleof (( (( nobracenl {{ line @ drop }} nonl ** )) ?? nleof )) ** ` } white ** nleof white ** ))
+   || (( nocolonnl nonl **  nleof white ** )) ** ))
+   {{ end prim prim-c-code 2! skipsynclines on }}
+   (( ` :  white ** nleof
+      {{ start }} (( nonl ++  nleof white ** )) ++ {{ end prim prim-forth-code 2! }}
+   )) ?? {{ process-simple }}
+   nleof
 )) <- simple-primitive ( -- )
 
 (( {{ init-combined }}
-   ` = (( white ++ {{ start }} forth-ident {{ end add-prim }} )) ++
-   (( nl || eof )) {{ process-combined }}
+   ` = white ** (( {{ start }} forth-ident {{ end add-prim }} white ** )) ++
+   nleof {{ process-combined }}
 )) <- combined-primitive
 
-(( {{ make-prim to prim
+(( {{ make-prim to prim 0 to combined
       line @ name-line ! filename 2@ name-filename 2!
-      start }} forth-ident {{ end 2dup prim prim-name 2! prim prim-c-name 2! }}  white ++
-   (( simple-primitive || combined-primitive ))
+      function-number @ prim prim-num !
+      start }} [ifdef] vmgen c-ident [else] forth-ident [then] {{ end
+      2dup prim prim-name 2! prim prim-c-name 2! }}  white **
+   (( ` / white ** {{ start }} c-ident {{ end prim prim-c-name 2! }} white ** )) ??
+   (( simple-primitive || combined-primitive )) {{ 1 function-number +! }}
 )) <- primitive ( -- )
 
 (( (( comment || primitive || nl white ** )) ** eof ))
@@ -1054,6 +1286,18 @@ warnings @ [IF]
 .( parser generated ok ) cr
 [THEN]
 
+
+\ run with gforth-0.5.0 (slurp-file is missing)
+[IFUNDEF] slurp-file
+: slurp-file ( c-addr1 u1 -- c-addr2 u2 )
+    \ c-addr1 u1 is the filename, c-addr2 u2 is the file's contents
+    r/o bin open-file throw >r
+    r@ file-size throw abort" file too large"
+    dup allocate throw swap
+    2dup r@ read-file throw over <> abort" could not read whole file"
+    r> close-file throw ;
+[THEN]
+
 : primfilter ( addr u -- )
     \ process the string at addr u
     over dup rawinput ! dup line-start ! cookedinput !