--- gforth/prims2x.fs	2001/10/26 12:00:15	1.99
+++ gforth/prims2x.fs	2002/09/22 14:21:30	1.116
@@ -60,9 +60,9 @@ include startup.fs
 : struct% struct ; \ struct is redefined in gray
 
 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
@@ -78,10 +78,22 @@ 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
+
+variable include-skipped-insts
+\ does the threaded code for a combined instruction include the cells
+\ for the component instructions (true) or only the cells for the
+\ inline arguments (false)
+include-skipped-insts off
 
 : th ( addr1 n -- addr2 )
     cells + ;
@@ -121,12 +133,24 @@ 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
 
@@ -231,6 +255,12 @@ variable in-part \ true if processing a
 create combined-prims max-combined cells allot
 variable num-combined
 
+: map-combined { xt -- }
+    \ perform xt for all components of the current combined instruction
+    num-combined @ 0 +do
+	combined-prims i th @ xt execute
+    loop ;
+
 table constant combinations
   \ the keys are the sequences of pointers to primitives
 
@@ -312,26 +342,26 @@ defer inst-stream-f ( -- stack )
     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-type-name 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
- ." 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
- ." );" 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
@@ -355,26 +385,25 @@ defer inst-stream-f ( -- stack )
     >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-type-name 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
@@ -385,7 +414,7 @@ defer inst-stream-f ( -- stack )
  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 )
@@ -423,7 +452,7 @@ wordlist constant prefixes
     stack r@ type-stack !
     rdrop ;
 
-: type-prefix ( xt1 xt2 n stack "prefix" -- )
+: type-prefix ( addr u xt1 xt2 n stack "prefix" -- )
     get-current >r prefixes set-current
     create-type r> set-current
 does> ( item -- )
@@ -624,58 +653,70 @@ stack inst-stream IP Cell
     endif
     2drop ;
 
-: output-c-tail ( -- )
-    \ the final part of the generated C code
+: 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
+    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 "TAIL;" with tail code produced by 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" TAIL;" search
+	2dup s" INST_TAIL;" search
     while ( c-addr1 u1 c-addr3 u3 )
 	2dup 2>r drop nip over - type
 	xt execute
-	2r> 5 /string
+	2r> 10 /string
 	\ !! resync #line missing
     repeat
-    2drop type ;
+    2drop type
+    ." #line " out-nls @ 2 + . quote out-filename 2@ type quote cr
+    ." }" cr ;
 
 : print-entry ( -- )
-    ." CASE 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@ ['] output-c-tail type-c-code
- ." }" 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 ( -- )
@@ -683,7 +724,7 @@ stack inst-stream IP Cell
 
 : 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
@@ -693,7 +734,7 @@ stack inst-stream IP Cell
 : 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
+    ."   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
@@ -703,6 +744,31 @@ stack inst-stream IP Cell
     endif
     ." }" cr ;
 
+: output-profile-part ( p )
+    ."   add_inst(b, " quote
+    prim-name 2@ type
+    quote ." );" cr ;
+    
+: output-profile-combined ( -- )
+    \ generate code for postprocessing the VM block profile stuff
+    ." if (VM_IS_INST(*ip, " function-number @ 0 .r ." )) {" cr
+    ['] output-profile-part map-combined
+    ."   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
 	." , " item item-type @ type-c-name 2@ type space
@@ -767,11 +833,14 @@ stack inst-stream IP Cell
 \      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 ;
 
+: output-prim-num ( -- )
+    prim prim-num @ 8 + 4 .r space prim prim-name 2@ type cr ;
+
 : output-forth ( -- )  
     prim prim-forth-code @ 0=
     IF    	\ output-alias
@@ -783,7 +852,7 @@ stack inst-stream IP Cell
     THEN ;
 
 : output-tag-file ( -- )
-    name-filename 2@ last-name-filename 2@ compare if
+    name-filename 2@ last-name-filename 2@ str= 0= if
 	name-filename 2@ last-name-filename 2!
 	#ff emit cr
 	name-filename 2@ type
@@ -799,6 +868,11 @@ stack inst-stream IP Cell
     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 ( -- )
     prim prim-name 2@ documentation ['] create insert-wordlist
@@ -868,9 +942,7 @@ stack inst-stream IP Cell
 \  #line 516 "./prim"
 \  n = n1+n2;
 \  }
-\  NEXT_P1;
 \  _x_sp0 = (Cell)n;
-\  NEXT_P2;
 \  }
 \  NEXT_P1;
 \  spTOS = (Cell)_x_sp0;
@@ -880,6 +952,7 @@ stack inst-stream IP Cell
     prim to combined
     0 num-combined !
     current-depth max-stacks cells erase
+    include-skipped-insts @ current-depth 0 th !
     max-depth     max-stacks cells erase
     min-depth     max-stacks cells erase
     prim prim-effect-in  prim prim-effect-in-end  !
@@ -891,11 +964,16 @@ stack inst-stream IP Cell
 : min! ( n addr -- )
     tuck @ min swap ! ;
 
+: inst-stream-correction ( nin1 nstack -- nin2 )
+    0= if
+	include-skipped-insts @ -
+    endif ;
+
 : add-depths { p -- }
     \ combine stack effect of p with *-depths
     max-stacks 0 ?do
 	current-depth i th @
-	p prim-stacks-in  i th @ +
+	p prim-stacks-in  i th @ + i inst-stream-correction
 	dup max-depth i th max!
 	p prim-stacks-out i th @ -
 	dup min-depth i th min!
@@ -987,19 +1065,14 @@ stack inst-stream IP Cell
     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
+    ['] output-part map-combined
     in-part off
     r> to prim ;
 
@@ -1015,7 +1088,7 @@ stack inst-stream IP Cell
     \ print-debug-args
     stack-pointer-updates
     output-parts
-    output-c-tail
+    output-c-tail2
     ." }" cr
     cr ;
 
@@ -1025,13 +1098,16 @@ stack inst-stream IP Cell
 
 \ peephole optimization rules
 
+\ data for a simple peephole optimizer that always tries to combine
+\ the currently compiled instruction with the last one.
+
 \ 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
+    s" the prefix for this superinstruction must be defined earlier" ?print-error
     ." {"
     execute prim-num @ 5 .r ." ,"
     combined-prims num-combined @ 1- th @ prim-num @ 5 .r ." ,"
@@ -1039,15 +1115,53 @@ stack inst-stream IP Cell
     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 ;
 
+\ cost and superinstruction data for a sophisticated combiner (e.g.,
+\ shortest path)
+
+\ This is intended as initializer for a structure like this
+
+\  struct cost {
+\    int loads;       /* number of stack loads */
+\    int stores;      /* number of stack stores */
+\    int updates;     /* number of stack pointer updates */
+\    int length;      /* number of components */
+\    int *components; /* array of vm_prim indexes of components */
+\  };
+
+\ How do you know which primitive or combined instruction this
+\ structure refers to?  By the order of cost structures, as in most
+\ other cases.
+
+: compute-costs { p -- nloads nstores nupdates }
+    \ compute the number of loads, stores, and stack pointer updates
+    \ of a primitive or combined instruction; does not take TOS
+    \ caching into account, nor that IP updates are combined with
+    \ other stuff
+    0 max-stacks 0 +do
+	p prim-stacks-in i th @ +
+    loop
+    0 max-stacks 0 +do
+	p prim-stacks-out i th @ +
+    loop
+    0 max-stacks 0 +do
+	p prim-stacks-in i th @ p prim-stacks-out i th @ <> -
+    loop ;
+
+: output-num-part ( p -- )
+    prim-num @ 4 .r ." ," ;
+
+: output-costs ( -- )
+    ." {" prim compute-costs
+    rot 2 .r ." ," swap 2 .r ." ," 2 .r ." ,"
+    combined if
+	num-combined @ 2 .r
+	." , ((int []){" ['] output-num-part map-combined ." })}, /* "
+    else
+	."  1, ((int []){" prim prim-num @ 4 .r ." })}, /* "
+    endif
+    prim prim-name 2@ type ."  */"
+    cr ;
 
 \ the parser
 
@@ -1077,7 +1191,7 @@ print-token !
     \ when input points to a newline, check if the next line is a
     \ sync line.  If it is, perform the appropriate actions.
     rawinput @ >r
-    s" #line " r@ over compare 0<> if
+    s" #line " r@ over str= 0= if
 	rdrop 1 line +! EXIT
     endif
     0. r> 6 chars + 20 >number drop >r drop line ! r> ( c-addr )
@@ -1126,6 +1240,8 @@ 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
@@ -1137,7 +1253,7 @@ nl-char singleton eof-char over add-memb
 (( letter (( letter || digit )) **
 )) <- c-ident ( -- )
 
-(( ` # ?? (( letter || digit || ` : )) **
+(( ` # ?? (( letter || digit || ` : )) ++
 )) <- stack-ident ( -- )
 
 (( nowhitebq nowhite ** ))
@@ -1200,7 +1316,10 @@ Variable c-flag
       (( {{ 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 }}
+   {{ 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 }}
@@ -1215,7 +1334,9 @@ Variable c-flag
 (( {{ 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 ++
+      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 ( -- )