--- gforth/prims2x.fs	2001/01/18 19:43:02	1.68
+++ gforth/prims2x.fs	2001/06/16 22:53:36	1.98
@@ -19,7 +19,8 @@
 \ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
 
 
-\ This is not very nice (hard limits, no checking, assumes 1 chars = 1)
+\ This is not very nice (hard limits, no checking, assumes 1 chars = 1).
+\ And it grew even worse when it aged.
 
 \ Optimizations:
 \ superfluous stores are removed. GCC removes the superfluous loads by itself
@@ -39,26 +40,31 @@
 \ 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]
+\ - 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.
+
+warnings off
 
-[IFUNDEF] environment?
-include ./environ.fs
+[IFUNDEF] try
+include startup.fs
 [THEN]
 
 : struct% struct ; \ struct is redefined in gray
 
+warnings off
+
 include ./gray.fs
 
-100 constant max-effect \ number of things on one side of a stack effect
+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
 maxchar 1+ constant eof-char
 #tab constant tab-char
@@ -77,26 +83,58 @@ variable line-start \ pointer to start o
 variable skipsynclines \ are sync lines ("#line ...") invisible to the parser?
 skipsynclines on 
 
+: th ( addr1 n -- addr2 )
+    cells + ;
+
+: holds ( addr u -- )
+    \ like HOLD, but for a string
+    tuck + swap 0 +do
+	1- dup c@ hold
+    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 @ ;
 
 : end ( addr -- addr u )
  cookedinput @ over - ;
 
+: print-error-line ( -- )
+    \ print the current line and position
+    line-start @ endrawinput @ over - 2dup nl-char scan drop nip ( start end )
+    over - type cr
+    line-start @ rawinput @ over - typewhite ." ^" cr ;
+
+: ?print-error { f addr u -- }
+    f ?not? if
+	outfile-id >r try
+	    stderr to outfile-id
+	    filename 2@ type ." :" line @ 0 .r ." : " addr u type cr
+	    print-error-line
+	    0
+	recover endtry
+	r> to outfile-id throw
+	abort
+    endif ;
+
 : quote ( -- )
     [char] " emit ;
 
-variable output \ xt ( -- ) of output word
-
-: printprim ( -- )
- output @ execute ;
+variable output          \ xt ( -- ) of output word for simple primitives
+variable output-combined \ xt ( -- ) of output word for combined primitives
 
 struct%
+    cell%    field stack-number \ the number of this stack
     cell% 2* field stack-pointer \ stackpointer name
-    cell% 2* field stack-cast \ cast string for assignments to stack elements
+    cell%    field stack-type \ name for default type of stack items
     cell%    field stack-in-index-xt \ ( in-size item -- in-index )
-    cell%    field stack-in  \ number of stack items in effect in
-    cell%    field stack-out \ number of stack items in effect out
 end-struct stack%
 
 struct%
@@ -115,24 +153,30 @@ struct%
     cell%    field type-store \ xt of store code generator ( item -- )
 end-struct type%
 
+variable next-stack-number 0 next-stack-number !
+create stacks max-stacks cells allot \ array of stacks
+
 : stack-in-index ( in-size item -- in-index )
     item-offset @ - 1- ;
 
 : inst-in-index ( in-size item -- in-index )
     nip dup item-offset @ swap item-type @ type-size @ + 1- ;
 
-: make-stack ( addr-ptr u1 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
-    save-mem r@ stack-cast 2!
+    r@ stacks next-stack-number @ th !
+    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 ! ;
 
-s" sp" save-mem s" (Cell)" make-stack data-stack 
-s" fp" save-mem s" "       make-stack fp-stack
-s" rp" save-mem s" (Cell)" make-stack return-stack
-s" IP" save-mem s" error don't use # on results" make-stack inst-stream
-' inst-in-index inst-stream stack-in-index-xt !
-\ !! initialize stack-in and stack-out
+: map-stacks { xt -- }
+    \ perform xt for all stacks except inst-stream
+    next-stack-number @ 1 +do
+	stacks i th @ xt execute
+    loop ;
 
 \ stack items
 
@@ -148,32 +192,73 @@ s" IP" save-mem s" error don't use # on
 	i xt execute
     item% %size +loop ;
 
+\ types
+
+: print-type-prefix ( type -- )
+    body> >head name>string type ;
+
 \ various variables for storing stuff of one primitive
 
-2variable forth-name
-2variable wordset
-2variable c-name
-2variable doc
-2variable c-code
-2variable forth-code
-2variable stack-string
-create effect-in  max-effect item% %size * allot
-create effect-out max-effect item% %size * allot
-variable effect-in-end ( pointer )
-variable effect-out-end ( pointer )
+struct%
+    cell% 2* field prim-name
+    cell% 2* field prim-wordset
+    cell% 2* field prim-c-name
+    cell% 2* field prim-doc
+    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
+    cell%    field prim-effect-in-end
+    cell%    field prim-effect-out-end
+    cell% max-stacks * field prim-stacks-in  \ number of in items per stack
+    cell% max-stacks * field prim-stacks-out \ number of out items per stack
+end-struct prim%
+
+: make-prim ( -- prim )
+    prim% %alloc { p }
+    s" " p prim-doc 2! s" " p prim-forth-code 2! s" " p prim-wordset 2!
+    p ;
+
+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 -- )
+    dup prim-name 2@ primitives ['] constant insert-wordlist ;
+
+: stack-in ( stack -- addr )
+    \ address of number of stack items in effect in
+    stack-number @ cells prim prim-stacks-in + ;
+
+: stack-out ( stack -- addr )
+    \ address of number of stack items in effect out
+    stack-number @ cells prim prim-stacks-out + ;
+
+\ global vars
 variable c-line
 2variable c-filename
 variable name-line
 2variable name-filename
 2variable last-name-filename
-
 Variable function-number 0 function-number !
 
-\ for several reasons stack items of a word are stored in a wordlist
-\ since neither forget nor marker are implemented yet, we make a new
-\ wordlist for every word and store it in the variable items
-variable items
-
 \ a few more set ops
 
 : bit-equivalent ( w1 w2 -- w3 )
@@ -182,10 +267,9 @@ variable items
 : 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
@@ -194,25 +278,55 @@ variable items
 	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
- ."  = (" 
- r@ item-type @ type-c-name 2@ type ." ) "
+ ."  = vm_" r@ item-stack-type-name type
+ ." 2" r@ item-type @ print-type-prefix ." ("
  r@ item-in-index r@ item-stack @ stack-access
- ." ;" cr
+ ." );" cr
  rdrop ; 
 
 : fetch-double ( item -- )
  \ fetch a double stack item from its stack
  >r
- ." FETCH_DCELL("
+ ." vm_two"
+ r@ item-stack-type-name type ." 2"
+ r@ item-type @ print-type-prefix ." ("
  r@ item-name 2@ type ." , "
  r@ item-in-index r@ item-stack @ 2dup ." (Cell)" stack-access
  ." , "                      -1 under+ ." (Cell)" stack-access
@@ -222,8 +336,10 @@ variable items
 : same-as-in? ( item -- f )
  \ f is true iff the offset and stack of item is the same as on input
  >r
- r@ item-name 2@ items @ search-wordlist 0=
- abort" bug"
+ r@ item-first @ if
+     rdrop false exit
+ endif
+ r@ item-name 2@ prim prim-items-wordlist @ search-wordlist 0= abort" bug"
  execute @
  dup r@ =
  if \ item first appeared in output
@@ -240,9 +356,10 @@ variable items
 
 : really-store-single ( item -- )
  >r
- r@ item-out-index r@ item-stack @ stack-access ."  = "
- r@ item-stack @ stack-cast 2@ type
- r@ item-name 2@ type ." ;"
+ r@ item-out-index r@ item-stack @ stack-access ."  = vm_"
+ r@ item-type @ print-type-prefix ." 2"
+ r@ item-stack-type-name type ." ("
+ r@ item-name 2@ type ." );"
  rdrop ;
 
 : store-single ( item -- )
@@ -262,7 +379,10 @@ variable items
 : 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
@@ -304,13 +424,14 @@ wordlist constant prefixes
     rdrop ;
 
 : type-prefix ( xt1 xt2 n stack "prefix" -- )
-    create-type
+    get-current >r prefixes set-current
+    create-type r> set-current
 does> ( item -- )
     \ initialize item
     { item typ }
     typ item item-type !
     typ type-stack @ item item-stack !default
-    item item-name 2@ items @ search-wordlist 0= if \ new name
+    item item-name 2@ prim prim-items-wordlist @ search-wordlist 0= if
 	item item-name 2@ nextname item declare
 	item item-first on
 	\ typ type-c-name 2@ type space type  ." ;" cr
@@ -330,7 +451,7 @@ 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 ;
@@ -339,9 +460,9 @@ does> ( item -- )
     ['] declaration map-items ;
 
 : declarations ( -- )
- wordlist dup items ! set-current
- effect-in effect-in-end @ declaration-list
- effect-out effect-out-end @ declaration-list ;
+ wordlist dup prim prim-items-wordlist ! set-current
+ prim prim-effect-in prim prim-effect-in-end @ declaration-list
+ prim prim-effect-out prim prim-effect-out-end @ declaration-list ;
 
 : print-declaration { item -- }
     item item-first @ if
@@ -350,17 +471,42 @@ does> ( item -- )
     endif ;
 
 : print-declarations ( -- )
-    effect-in  effect-in-end  @ ['] print-declaration map-items
-    effect-out effect-out-end @ ['] print-declaration map-items ;
+    prim prim-effect-in  prim prim-effect-in-end  @ ['] print-declaration map-items
+    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!
     stack item item-stack !
     item declaration ;
 
+\ types pointed to by stacks for use in combined prims
+\ !! 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 ;
+
+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
 \ the leftmost (i.e. deepest) item has offset 0
 \ the rightmost item has the highest offset
@@ -381,11 +527,17 @@ does> ( item -- )
     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
-    effect-in  effect-in-end  @ ['] compute-offset-in  map-items
-    effect-out effect-out-end @ ['] compute-offset-out map-items
-    inst-stream stack-out @ 0<> abort" # can only be on the input side" ;
+    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= 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
@@ -395,9 +547,7 @@ does> ( item -- )
     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
@@ -408,15 +558,13 @@ does> ( item -- )
 
 : 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 ;
+    dup item-type @ type-fetch @ execute ;
 
 : fetches ( -- )
-    effect-in effect-in-end @ ['] fetch map-items ;
+    prim prim-effect-in prim prim-effect-in-end @ ['] fetch map-items ;
 
 : stack-pointer-update { stack -- }
     \ stack grow downwards
@@ -432,9 +580,7 @@ does> ( item -- )
 
 : 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
@@ -442,47 +588,71 @@ does> ( item -- )
  dup item-type @ type-store @ execute ;
 
 : stores ( -- )
-    effect-out effect-out-end @ ['] store map-items ;
+    prim prim-effect-out prim prim-effect-out-end @ ['] store map-items ;
+
+: print-debug-arg { item -- }
+    ." fputs(" quote space item item-name 2@ type ." =" quote ." , vm_out); "
+    ." printarg_" item item-type @ print-type-prefix
+    ." (" item item-name 2@ type ." );" cr ;
+    
+: print-debug-args ( -- )
+    ." #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-tail ( -- )
     \ the final part of the generated C code
+    output-super-end
+    print-debug-results
     ." NEXT_P1;" cr
     stores
     fill-tos
-    ." NEXT_P2;" cr ;
+    ." NEXT_P2;" ;
 
-: type-c ( c-addr u -- )
-    \ like TYPE, but replaces "TAIL;" with tail code
+: type-c-code ( c-addr u xt -- )
+    \ like TYPE, but replaces "TAIL;" with tail code produced by xt
+    { xt }
     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
+	xt execute
 	2r> 5 /string
 	\ !! resync #line missing
     repeat
     2drop type ;
 
-: print-type-prefix ( type -- )
-    body> >head .name ;
-
-: print-debug-arg { item -- }
-    ." fputs(" quote space item item-name 2@ type ." =" quote ." , vm_out); "
-    ." printarg_" item item-type @ print-type-prefix
-    ." (" item item-name 2@ type ." );" cr ;
-    
-: print-debug-args ( -- )
-    ." #ifdef VM_DEBUG" cr
-    ." if (vm_debug) {" cr
-    effect-in effect-in-end @ ['] print-debug-arg map-items
-    ." fputc('\n', vm_out);" cr
-    ." }" cr
-    ." #endif" cr ;
+: print-entry ( -- )
+    ." I_" prim prim-c-name 2@ type ." :" ;
     
 : output-c ( -- ) 
- ." I_" c-name 2@ type ." :	/* " forth-name 2@ type ."  ( " stack-string 2@ type ." ) */" cr
- ." /* " doc 2@ type ."  */" cr
- ." NAME(" quote forth-name 2@ type quote ." )" cr \ debugging
+ 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
@@ -493,7 +663,7 @@ does> ( item -- )
  stack-pointer-updates
  ." {" cr
  ." #line " c-line @ . quote c-filename 2@ type quote cr
- c-code 2@ type-c
+ prim prim-c-code 2@ ['] output-c-tail type-c-code
  ." }" cr
  output-c-tail
  ." }" cr
@@ -509,15 +679,29 @@ does> ( item -- )
     endif ;
 
 : disasm-args ( -- )
-    effect-in effect-in-end @ ['] disasm-arg map-items ;
+    prim prim-effect-in prim prim-effect-in-end @ ['] disasm-arg map-items ;
 
 : output-disasm ( -- )
     \ generate code for disassembling VM instructions
     ." if (ip[0] == prim[" function-number @ 0 .r ." ]) {" cr
-    ."   fputs(" quote forth-name 2@ type quote ." , vm_out);" 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 ;
 
 : gen-arg-parm { item -- }
     item item-stack @ inst-stream = if
@@ -526,7 +710,7 @@ does> ( item -- )
     endif ;
 
 : gen-args-parm ( -- )
-    effect-in effect-in-end @ ['] gen-arg-parm map-items ;
+    prim prim-effect-in prim prim-effect-in-end @ ['] gen-arg-parm map-items ;
 
 : gen-arg-gen { item -- }
     item item-stack @ inst-stream = if
@@ -535,11 +719,11 @@ does> ( item -- )
     endif ;
 
 : gen-args-gen ( -- )
-    effect-in effect-in-end @ ['] gen-arg-gen map-items ;
+    prim prim-effect-in prim prim-effect-in-end @ ['] gen-arg-gen map-items ;
 
 : output-gen ( -- )
     \ generate C code for generating VM instructions
-    ." void gen_" c-name 2@ type ." (Inst **ctp" gen-args-parm ." )" cr
+    ." void gen_" prim prim-c-name 2@ type ." (Inst **ctp" gen-args-parm ." )" cr
     ." {" cr
     ."   gen_inst(ctp, vm_prim[" function-number @ 0 .r ." ]);" cr
     gen-args-gen
@@ -549,52 +733,53 @@ does> ( item -- )
     stack stack-in @ stack stack-out @ or 0<> ;
 
 : output-funclabel ( -- )
-  ." &I_" c-name 2@ type ." ," cr ;
+  ." &I_" prim prim-c-name 2@ type ." ," cr ;
 
 : output-forthname ( -- )
-  '" emit forth-name 2@ type '" emit ." ," cr ;
+  '" emit prim prim-name 2@ type '" emit ." ," cr ;
 
-: output-c-func ( -- )
-\ used for word libraries
-    ." Cell * I_" c-name 2@ type ." (Cell *SP, Cell **FP)      /* " forth-name 2@ type
-    ."  ( " stack-string 2@ type ."  ) */" cr
-    ." /* " doc 2@ type ."  */" cr
-    ." NAME(" quote forth-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
-    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_" c-name 2@ type ." ," cr ;
+    ." (Label)&&I_" prim prim-c-name 2@ type ." ," cr ;
 
 : output-alias ( -- ) 
-    ( primitive-number @ . ." alias " ) ." Primitive " forth-name 2@ type cr ;
+    ( primitive-number @ . ." alias " ) ." Primitive " prim prim-name 2@ type cr ;
 
 : output-forth ( -- )  
-    forth-code @ 0=
+    prim prim-forth-code @ 0=
     IF    	\ output-alias
 	\ this is bad for ec: an alias is compiled if tho word does not exist!
 	\ JAW
-    ELSE  ." : " forth-name 2@ type ."   ( "
-	stack-string 2@ type ." )" cr
-	forth-code 2@ type cr
+    ELSE  ." : " prim prim-name 2@ type ."   ( "
+	prim prim-stack-string 2@ type ." )" cr
+	prim prim-forth-code 2@ type cr
     THEN ;
 
 : output-tag-file ( -- )
@@ -607,26 +792,263 @@ does> ( item -- )
 
 : output-tag ( -- )
     output-tag-file
-    forth-name 2@ 1+ type
+    prim prim-name 2@ 1+ type
     127 emit
-    space forth-name 2@ type space
+    space prim prim-name 2@ type space
     1 emit
     name-line @ 0 .r
     ." ,0" cr ;
 
 [IFDEF] documentation
 : register-doc ( -- )
-    get-current documentation set-current
-    forth-name 2@ nextname create
-    forth-name 2@ 2,
-    stack-string 2@ condition-stack-effect 2,
-    wordset 2@ 2,
-    c-name 2@ condition-pronounciation 2,
-    doc 2@ 2,
-    set-current ;
+    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, ;
 [THEN]
 
 
+\ combining instructions
+
+\ The input should look like this:
+
+\ lit_+ = lit +
+
+\ The output should look like this:
+
+\  I_lit_+:
+\  {
+\  DEF_CA
+\  Cell _x_ip0;
+\  Cell _x_sp0;
+\  Cell _x_sp1;
+\  NEXT_P0;
+\  _x_ip0 = (Cell) IPTOS;
+\  _x_sp0 = (Cell) spTOS;
+\  INC_IP(1);
+\  /* sp += 0; */
+\  /* lit ( #w -- w ) */
+\  /*  */
+\  NAME("lit")
+\  {
+\  Cell w;
+\  w = (Cell) _x_ip0;
+\  #ifdef VM_DEBUG
+\  if (vm_debug) {
+\  fputs(" w=", vm_out); printarg_w (w);
+\  fputc('\n', vm_out);
+\  }
+\  #endif
+\  {
+\  #line 136 "./prim"
+\  }
+\  _x_sp1 = (Cell)w;
+\  }
+\  I_plus:	/* + ( n1 n2 -- n ) */
+\  /*  */
+\  NAME("+")
+\  {
+\  DEF_CA
+\  Cell n1;
+\  Cell n2;
+\  Cell n;
+\  NEXT_P0;
+\  n1 = (Cell) _x_sp0;
+\  n2 = (Cell) _x_sp1;
+\  #ifdef VM_DEBUG
+\  if (vm_debug) {
+\  fputs(" n1=", vm_out); printarg_n (n1);
+\  fputs(" n2=", vm_out); printarg_n (n2);
+\  fputc('\n', vm_out);
+\  }
+\  #endif
+\  {
+\  #line 516 "./prim"
+\  n = n1+n2;
+\  }
+\  NEXT_P1;
+\  _x_sp0 = (Cell)n;
+\  NEXT_P2;
+\  }
+\  NEXT_P1;
+\  spTOS = (Cell)_x_sp0;
+\  NEXT_P2;
+
+: init-combined ( -- )
+    prim to combined
+    0 num-combined !
+    current-depth max-stacks cells erase
+    max-depth     max-stacks cells erase
+    min-depth     max-stacks cells erase
+    prim prim-effect-in  prim prim-effect-in-end  !
+    prim prim-effect-out prim prim-effect-out-end ! ;
+
+: max! ( n addr -- )
+    tuck @ max swap ! ;
+
+: min! ( n addr -- )
+    tuck @ min swap ! ;
+
+: add-depths { p -- }
+    \ combine stack effect of p with *-depths
+    max-stacks 0 ?do
+	current-depth i th @
+	p prim-stacks-in  i th @ +
+	dup max-depth i th max!
+	p prim-stacks-out i th @ -
+	dup min-depth i th min!
+	current-depth i th !
+    loop ;
+
+: add-prim ( addr u -- )
+    \ add primitive given by "addr u" to combined-prims
+    primitives search-wordlist s" unknown primitive" ?print-error
+    execute { p }
+    p combined-prims num-combined @ th !
+    1 num-combined +!
+    p add-depths ;
+
+: compute-effects { q -- }
+    \ compute the stack effects of q from the depths
+    max-stacks 0 ?do
+	max-depth i th @ dup
+	q prim-stacks-in i th !
+	current-depth i th @ -
+	q prim-stacks-out i th !
+    loop ;
+
+: make-effect-items { stack# items effect-endp -- }
+    \ effect-endp points to a pointer to the end of the current item-array
+    \ and has to be updated
+    stacks stack# th @ { stack }
+    items 0 +do
+	effect-endp @ { item }
+	i 0 <# #s stack stack-pointer 2@ holds [char] _ hold #> save-mem
+	item item-name 2!
+	stack item item-stack !
+	stack stack-type @ item item-type !
+	i item item-offset !
+	item item-first on
+	item% %size effect-endp +!
+    loop ;
+
+: init-effects { q -- }
+    \ initialize effects field for FETCHES and STORES
+    max-stacks 0 ?do
+	i q prim-stacks-in  i th @ q prim-effect-in-end  make-effect-items
+	i q prim-stacks-out i th @ q prim-effect-out-end make-effect-items
+    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 ;
+
+\ C output
+
+: 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 ;
+
+: print-declarations-combined ( -- )
+    max-stacks 0 ?do
+	max-depth i th @ min-depth i th @ - 0 +do
+	    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?
+    ." {" cr
+    ." #line " c-line @ . quote c-filename 2@ type quote cr
+    prim prim-c-code 2@ ['] output-combined-tail type-c-code
+    ." }" cr
+    part-output-c-tail
+    ." }" cr ;
+
+: output-parts ( -- )
+    prim >r in-part on
+    current-depth max-stacks cells erase
+    num-combined @ 0 +do
+	combined-prims i th @ output-part
+    loop
+    in-part off
+    r> to prim ;
+
+: output-c-combined ( -- )
+    print-entry cr
+    \ debugging messages just in parts
+    ." {" cr
+    ." DEF_CA" cr
+    print-declarations-combined
+    ." NEXT_P0;" cr
+    flush-tos
+    fetches
+    \ print-debug-args
+    stack-pointer-updates
+    output-parts
+    output-c-tail
+    ." }" 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
 
 eof-char max-member \ the whole character set + EOF
@@ -660,35 +1082,19 @@ 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!
     endif
     drop ;
 
-: print-error-line ( -- )
-    \ print the current line and position
-    line-start @ endrawinput @ over - 2dup nl-char scan drop nip ( start end )
-    over - type cr
-    line-start @ rawinput @ over - typewhite ." ^" cr ;
-    
 : ?nextchar ( f -- )
-    ?not? if
-	outfile-id >r try
-	    stderr to outfile-id
-	    filename 2@ type ." :" line @ 0 .r ." : syntax error, wrong char:"
-	    getinput . cr
-	    print-error-line
-	    0
-	recover endtry
-	r> to outfile-id throw
-	abort
-    endif
+    s" syntax error, wrong char" ?print-error
     rawinput @ endrawinput @ <> if
 	rawinput @ c@
 	1 chars rawinput +!
@@ -726,7 +1132,7 @@ bl 1+ maxchar ..
 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 ( -- )
@@ -770,32 +1176,47 @@ 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
 
-(( ` \ comment-body nl )) <- comment ( -- )
+(( (( eval-comment || forth-comment || c-comment || else-comment || if-comment || group-comment )) ?? nonl ** )) <- comment-body
+
+(( ` \ comment-body nleof )) <- comment ( -- )
 
 (( {{ start }} stack-ident {{ end 2 pick init-item item% %size + }} white ** )) **
 <- stack-items
 
-(( {{ effect-in }}  stack-items {{ effect-in-end ! }}
+(( {{ prim prim-effect-in }}  stack-items {{ prim prim-effect-in-end ! }}
    ` - ` - white **
-   {{ effect-out }} stack-items {{ effect-out-end ! }}
+   {{ prim prim-effect-out }} stack-items {{ prim prim-effect-out-end ! }}
 )) <- stack-effect ( -- )
 
-(( {{ s" " doc 2! s" " forth-code 2! s" " wordset 2! }}
-   (( {{ line @ name-line ! filename 2@ name-filename 2! }}
-      {{ start }} forth-ident {{ end 2dup forth-name 2! c-name 2! }}  white ++
-      ` ( white ** {{ start }} stack-effect {{ end stack-string 2! }} ` ) white **
-        (( {{ start }} forth-ident {{ end wordset 2! }} white **
-	   (( {{ start }}  c-ident {{ end c-name 2! }} )) ??
-	)) ??  nl
-   ))
-   (( ` " ` "  {{ start }} (( noquote ++ ` " )) ++ {{ end 1- doc 2! }} ` " white ** nl )) ??
-   {{ skipsynclines off line @ c-line ! filename 2@ c-filename 2! start }} (( nocolonnl nonl **  nl white ** )) ** {{ end c-code 2! skipsynclines on }}
-   (( ` :  white ** nl
-      {{ start }} (( nonl ++  nl white ** )) ++ {{ end forth-code 2! }}
-   )) ?? {{  declarations compute-offsets printprim 1 function-number +! }}
-   (( nl || eof ))
+(( {{ prim create-prim }}
+   ` ( 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! }} )) ??
+   )) ??  nleof
+   (( ` " ` "  {{ start }} (( noquote ++ ` " )) ++ {{ end 1- prim prim-doc 2! }} ` " white ** nleof )) ??
+   {{ skipsynclines off line @ c-line ! filename 2@ c-filename 2! start }} (( 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 }} white ** )) ++
+   nleof {{ process-combined }}
+)) <- combined-primitive
+
+(( {{ make-prim to prim 0 to combined
+      line @ name-line ! filename 2@ name-filename 2!
+      function-number @ prim prim-num !
+      start }} forth-ident {{ end 2dup prim prim-name 2! prim prim-c-name 2! }}  white ++
+   (( simple-primitive || combined-primitive )) {{ 1 function-number +! }}
 )) <- primitive ( -- )
 
 (( (( comment || primitive || nl white ** )) ** eof ))
@@ -804,29 +1225,33 @@ warnings @ [IF]
 .( parser generated ok ) cr
 [THEN]
 
-: primfilter ( file-id xt -- )
-\ fileid is for the input file, xt ( -- ) is for the output word
- output !
- here dup rawinput ! dup line-start ! cookedinput !
- here unused rot read-file throw
- dup here + endrawinput !
- allot
- align
- checksyncline
-\ begin
-\     getinput dup eof-char = ?EXIT emit true ?nextchar
-\ again ;
- primitives2something ;
 
-: process-file ( addr u xt -- )
-    >r
+\ 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 !
+    + endrawinput !
+    checksyncline
+    primitives2something ;    
+
+: process-file ( addr u xt-simple x-combined -- )
+    output-combined ! output !
     save-mem 2dup filename 2!
-    0 function-number !
-    r/o open-file abort" cannot open file"
+    slurp-file
     warnings @ if
 	." ------------ CUT HERE -------------" cr  endif
-    r> primfilter ;
+    primfilter ;
 
-: process      ( xt -- )
-    bl word count rot
-    process-file ;
+\  : process      ( xt -- )
+\      bl word count rot
+\      process-file ;