--- gforth/prims2x.fs 2001/01/21 20:36:31 1.71 +++ gforth/prims2x.fs 2002/11/24 13:54:01 1.121 @@ -40,25 +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] +\ - 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 -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 @@ -74,12 +78,42 @@ 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 + +variable immarg \ values for immediate arguments (to be used in IMM_ARG macros) +$12340000 immarg ! + +: th ( addr1 n -- addr2 ) + cells + ; + +: holds ( addr u -- ) + \ like HOLD, but for a string + tuck + swap 0 +do + 1- dup c@ hold + loop + drop ; -variable next-stack-number 0 next-stack-number ! +: 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 @ ; @@ -102,21 +136,31 @@ variable next-stack-number 0 next-stack- 0 recover endtry r> to outfile-id throw - abort + 1 (bye) \ abort endif ; : quote ( -- ) [char] " emit ; -variable output \ xt ( -- ) of output word +\ count output lines to generate sync lines for output -: printprim ( -- ) - output @ execute ; +: 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 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 ) end-struct stack% @@ -136,25 +180,36 @@ 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 - next-stack-number @ r@ stack-number ! 1 next-stack-number +! - 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 + next-stack-number @ 0 +do + stacks i th @ xt execute + loop ; + +: map-stacks1 { xt -- } + \ perform xt for all stacks except inst-stream + next-stack-number @ 1 +do + stacks i th @ xt execute + loop ; \ stack items @@ -170,6 +225,11 @@ 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 struct% @@ -180,6 +240,8 @@ 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 cell% field prim-effect-in-end @@ -193,15 +255,54 @@ 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 + +: prim-context ( ... p xt -- ... ) + \ execute xt with prim set to p + prim >r + swap to prim + catch + r> to prim + throw ; + +1000 constant max-combined +create combined-prims max-combined cells allot +variable num-combined +variable part-num \ current part number during process-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 + +create current-depth max-stacks cells allot +create max-depth max-stacks cells allot +create min-depth max-stacks cells allot + +create sp-update-in max-stacks cells allot +\ where max-depth occured the first time +create max-depths max-stacks max-combined 1+ * cells allot +\ maximum depth at start of each part: array[parts] of array[stack] +create max-back-depths max-stacks max-combined 1+ * cells allot +\ maximun depth from end of the combination to the start of the each part + +: s-c-max-depth ( nstack ncomponent -- addr ) + max-stacks * + cells max-depths + ; + +: s-c-max-back-depth ( nstack ncomponent -- addr ) + max-stacks * + cells max-back-depths + ; 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 @@ -219,11 +320,6 @@ variable name-line 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 ) @@ -232,10 +328,12 @@ variable items : complement ( set1 -- set2 ) empty ['] bit-equivalent binary-set-operation ; -\ types +\ forward declaration for inst-stream (breaks cycle in definitions) +defer inst-stream-f ( -- stack ) -: stack-access ( n stack -- ) - \ print a stack access at index n of stack +\ stack access stuff + +: normal-stack-access1 ( n stack -- ) stack-pointer 2@ type dup if @@ -244,36 +342,108 @@ variable items drop ." TOS" endif ; +: normal-stack-access ( n stack -- ) + dup inst-stream-f = if + ." IMM_ARG(" normal-stack-access1 ." ," immarg ? ." )" + 1 immarg +! + else + normal-stack-access1 + endif ; + +: stack-depth { stack -- n } + current-depth stack stack-number @ th @ ; + +: part-stack-access { n stack -- } + \ print _, x=inst-stream? n : maxdepth-currentdepth-n-1 + ." _" stack stack-pointer 2@ type + stack stack-number @ { stack# } + stack stack-depth 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 ; + +: part-stack-read { n stack -- } + stack stack-depth n + ( ndepth ) + stack stack-number @ part-num @ s-c-max-depth @ +\ max-depth stack stack-number @ th @ ( ndepth nmaxdepth ) + over <= if ( ndepth ) \ load from memory + stack normal-stack-access + else + drop n stack part-stack-access + endif ; + +: stack-diff ( stack -- n ) + \ in-out + dup stack-in @ swap stack-out @ - ; + +: part-stack-write { n stack -- } + stack stack-depth n + + stack stack-number @ part-num @ s-c-max-back-depth @ + over <= if ( ndepth ) + stack combined ['] stack-diff prim-context - + stack normal-stack-access + else + drop n stack part-stack-access + endif ; + +: stack-read ( n stack -- ) + \ print a stack access at index n of stack + in-part @ if + part-stack-read + else + normal-stack-access + endif ; + +: stack-write ( n stack -- ) + \ print a stack access at index n of stack + in-part @ if + part-stack-write + 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 ." ) " - 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-read ." ," + 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-read + ." , " -1 under+ ." (Cell)" stack-read + ." , " 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 >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 @@ -289,33 +459,36 @@ variable items >r r@ item-stack @ stack-out @ r> item-offset @ - 1- ; : 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 ." ;" - 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-write ." );" + 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 @ in-part @ 0= and 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 ." , " - r@ item-out-index r@ item-stack @ 2dup stack-access - ." , " -1 under+ stack-access - ." );" cr + ." 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-write + ." , " -1 under+ stack-write + ." )" cr rdrop ; : single ( -- xt1 xt2 n ) @@ -353,14 +526,15 @@ 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 } 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 @@ -380,7 +554,7 @@ 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 ; @@ -389,7 +563,7 @@ does> ( item -- ) ['] declaration map-items ; : declarations ( -- ) - wordlist dup items ! set-current + 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 ; @@ -404,13 +578,38 @@ 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! 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 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 @@ -427,15 +626,20 @@ does> ( item -- ) : compute-offset-out ( addr1 addr2 -- ) ['] stack-out compute-offset ; -: clear-stack { -- } +: clear-stack ( stack -- ) dup stack-in off stack-out off ; : compute-offsets ( -- ) - data-stack clear-stack fp-stack clear-stack return-stack clear-stack - inst-stream clear-stack + ['] clear-stack map-stacks 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 @@ -445,9 +649,7 @@ does> ( item -- ) endif ; : flush-tos ( -- ) - data-stack flush-a-tos - fp-stack flush-a-tos - return-stack flush-a-tos ; + ['] flush-a-tos map-stacks1 ; : fill-a-tos { stack -- } stack stack-out @ 0= stack stack-in @ 0<> and @@ -458,33 +660,27 @@ 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-stacks1 ; : fetch ( addr -- ) - dup item-type @ type-fetch @ execute ; + dup item-type @ type-fetch @ execute ; : fetches ( -- ) prim prim-effect-in prim prim-effect-in-end @ ['] fetch map-items ; : stack-pointer-update { stack -- } \ stack grow downwards - stack stack-in @ stack stack-out @ - + stack stack-diff ?dup-if \ this check is not necessary, gcc would do this for us - stack stack-pointer 2@ type ." += " 0 .r ." ;" cr - endif ; - -: inst-pointer-update ( -- ) - inst-stream stack-in @ ?dup-if - ." INC_IP(" 0 .r ." );" cr + stack inst-stream = if + ." INC_IP(" 0 .r ." );" cr + else + stack stack-pointer 2@ type ." += " 0 .r ." ;" cr + endif endif ; : 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 @@ -494,28 +690,6 @@ does> ( item -- ) : 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-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 @@ -525,37 +699,108 @@ does> ( item -- ) ." #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-label2 ( -- ) + ." LABEL2(" prim prim-c-name 2@ type ." )" cr + ." NEXT_P2;" cr ; + +: output-c-tail1 { xt -- } + \ the final part of the generated C code, with xt printing LABEL2 or not. + output-super-end + print-debug-results + ." NEXT_P1;" cr + stores + fill-tos + xt execute ; + +: output-c-tail1-no-stores { xt -- } + \ the final part of the generated C code for combinations + output-super-end + ." NEXT_P1;" cr + fill-tos + xt execute ; + +: output-c-tail ( -- ) + ['] noop output-c-tail1 ; + +: output-c-tail2 ( -- ) + ['] output-label2 output-c-tail1 ; + +: output-c-tail-no-stores ( -- ) + ['] noop output-c-tail1-no-stores ; + +: output-c-tail2-no-stores ( -- ) + ['] output-label2 output-c-tail1-no-stores ; + +: 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 ( -- ) + ." LABEL(" prim prim-c-name 2@ type ." )" ; : output-c ( -- ) - ." I_" prim prim-c-name 2@ type ." : /* " 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 ( -- ) @@ -563,11 +808,50 @@ does> ( item -- ) : 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-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 @@ -604,39 +888,43 @@ does> ( item -- ) : 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 ; +: 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 @@ -664,16 +952,19 @@ does> ( item -- ) 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] @@ -735,58 +1026,265 @@ does> ( item -- ) \ #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 - : init-combined ( -- ) + prim to combined 0 num-combined ! current-depth max-stacks cells erase - max-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 ! + prim prim-effect-out prim prim-effect-out-end ! ; : max! ( n addr -- ) tuck @ max swap ! ; +: min! ( n addr -- ) + tuck @ min swap ! ; + +: inst-stream-adjustment ( nstack -- n ) + \ number of stack items to add for each part + 0= include-skipped-insts @ and negate ; + : add-depths { p -- } \ combine stack effect of p with *-depths max-stacks 0 ?do - current-depth i cells + @ - p prim-stacks-in i cells + @ + - dup max-depth i cells + max! - p prim-stacks-out i cells + @ - - current-depth i cells + ! + current-depth i th @ + p prim-stacks-in i th @ + i inst-stream-adjustment + + dup max-depth i th max! + p prim-stacks-out i th @ - + dup min-depth i th min! + current-depth i th ! loop ; +: copy-maxdepths ( n -- ) + max-depth max-depths rot max-stacks * th max-stacks cells move ; + : 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 @ cells + ! + p combined-prims num-combined @ th ! + num-combined @ copy-maxdepths 1 num-combined +! - p add-depths ; + p add-depths + num-combined @ copy-maxdepths ; : compute-effects { q -- } \ compute the stack effects of q from the depths max-stacks 0 ?do - max-depth i cells + @ dup - q prim-stacks-in i cells + ! - current-depth i cells + @ - - q prim-stacks-out i cells + ! + 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 ; + +: compute-stack-max-back-depths ( stack -- ) + stack-number @ { stack# } + current-depth stack# th @ dup + dup stack# num-combined @ s-c-max-back-depth ! + -1 num-combined @ 1- -do ( max-depth current-depth ) + combined-prims i th @ { p } + p prim-stacks-out stack# th @ + + dup >r max r> + over stack# i s-c-max-back-depth ! + p prim-stacks-in stack# th @ - + stack# inst-stream-adjustment - + 1 -loop + assert( dup stack# inst-stream-adjustment negate = ) + assert( over max-depth stack# th @ = ) + 2drop ; + +: compute-max-back-depths ( -- ) + \ compute max-back-depths. + \ assumes that current-depths is correct for the end of the combination + ['] compute-stack-max-back-depths map-stacks ; + : process-combined ( -- ) - prim compute-effects ; + 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 + compute-max-back-depths + 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 + in-part @ >r in-part off + combined ['] output-c-tail-no-stores prim-context + r> in-part ! ; + +: part-stack-pointer-updates ( -- ) + max-stacks 0 +do + i part-num @ 1+ s-c-max-depth @ dup + i num-combined @ s-c-max-depth @ = \ final depth + swap i part-num @ s-c-max-depth @ <> \ just reached now + part-num @ 0= \ first part + or and if + stacks i th @ stack-pointer-update + endif + loop ; + +: 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 + combined ['] part-stack-pointer-updates prim-context + 1 part-num +! + 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 in-part on + current-depth max-stacks cells erase + 0 part-num ! + ['] output-part map-combined + 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 \ now in parts + \ print-debug-args + \ stack-pointer-updates now in parts + output-parts + output-c-tail2-no-stores + ." }" cr + cr ; + +: output-forth-combined ( -- ) +; + + +\ 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 superinstruction 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 ; + + +\ 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 @@ -816,16 +1314,16 @@ 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 compare if rdrop 1 line +! EXIT 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! @@ -865,18 +1363,20 @@ 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 ** )) @@ -915,9 +1415,15 @@ 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 @@ -931,24 +1437,30 @@ 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 )) @@ -957,6 +1469,18 @@ 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 ! @@ -964,14 +1488,14 @@ warnings @ [IF] checksyncline primitives2something ; -: process-file ( addr u xt -- ) - output ! +: process-file ( addr u xt-simple x-combined -- ) + output-combined ! output ! save-mem 2dup filename 2! slurp-file warnings @ if ." ------------ CUT HERE -------------" cr endif primfilter ; -: process ( xt -- ) - bl word count rot - process-file ; +\ : process ( xt -- ) +\ bl word count rot +\ process-file ;