Annotation of gforth/engine/main.c, revision 1.161

1.1       anton       1: /* command line interpretation, image loading etc. for Gforth
                      2: 
                      3: 
1.160     anton       4:   Copyright (C) 1995,1996,1997,1998,2000,2003,2004,2005 Free Software Foundation, Inc.
1.1       anton       5: 
                      6:   This file is part of Gforth.
                      7: 
                      8:   Gforth is free software; you can redistribute it and/or
                      9:   modify it under the terms of the GNU General Public License
                     10:   as published by the Free Software Foundation; either version 2
                     11:   of the License, or (at your option) any later version.
                     12: 
                     13:   This program is distributed in the hope that it will be useful,
                     14:   but WITHOUT ANY WARRANTY; without even the implied warranty of
                     15:   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
                     16:   GNU General Public License for more details.
                     17: 
                     18:   You should have received a copy of the GNU General Public License
                     19:   along with this program; if not, write to the Free Software
1.40      anton      20:   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
1.1       anton      21: */
                     22: 
                     23: #include "config.h"
1.82      anton      24: #include "forth.h"
1.1       anton      25: #include <errno.h>
                     26: #include <ctype.h>
                     27: #include <stdio.h>
1.2       pazsan     28: #include <unistd.h>
1.1       anton      29: #include <string.h>
                     30: #include <math.h>
                     31: #include <sys/types.h>
1.32      pazsan     32: #ifndef STANDALONE
1.1       anton      33: #include <sys/stat.h>
1.32      pazsan     34: #endif
1.1       anton      35: #include <fcntl.h>
                     36: #include <assert.h>
                     37: #include <stdlib.h>
1.102     anton      38: #include <signal.h>
1.11      pazsan     39: #ifndef STANDALONE
1.1       anton      40: #if HAVE_SYS_MMAN_H
                     41: #include <sys/mman.h>
                     42: #endif
1.11      pazsan     43: #endif
1.1       anton      44: #include "io.h"
                     45: #include "getopt.h"
1.11      pazsan     46: #ifdef STANDALONE
                     47: #include <systypes.h>
                     48: #endif
1.1       anton      49: 
1.121     anton      50: typedef enum prim_num {
1.119     anton      51: /* definitions of N_execute etc. */
1.126     anton      52: #include PRIM_NUM_I
1.119     anton      53:   N_START_SUPER
1.121     anton      54: } PrimNum;
1.119     anton      55: 
1.79      anton      56: /* global variables for engine.c 
                     57:    We put them here because engine.c is compiled several times in
                     58:    different ways for the same engine. */
1.161   ! pazsan     59: Cell *gforth_SP;
        !            60: Float *gforth_FP;
        !            61: Address gforth_UP=NULL;
1.79      anton      62: 
1.115     pazsan     63: #ifdef HAS_FFCALL
1.161   ! pazsan     64: Cell *gforth_RP;
        !            65: Address gforth_LP;
1.115     pazsan     66: 
                     67: #include <callback.h>
                     68: 
1.161   ! pazsan     69: va_alist gforth_clist;
1.115     pazsan     70: 
1.161   ! pazsan     71: void gforth_callback(Xt* fcall, void * alist)
1.115     pazsan     72: {
1.140     pazsan     73:   /* save global valiables */
1.161   ! pazsan     74:   Cell *rp = gforth_RP;
        !            75:   Cell *sp = gforth_SP;
        !            76:   Float *fp = gforth_FP;
        !            77:   Address lp = gforth_LP;
1.140     pazsan     78: 
1.161   ! pazsan     79:   gforth_clist = (va_alist)alist;
1.140     pazsan     80: 
1.161   ! pazsan     81:   gforth_engine(fcall, sp, rp, fp, lp);
1.140     pazsan     82: 
                     83:   /* restore global variables */
1.161   ! pazsan     84:   gforth_RP = rp;
        !            85:   gforth_SP = sp;
        !            86:   gforth_FP = fp;
        !            87:   gforth_LP = lp;
1.115     pazsan     88: }
                     89: #endif
                     90: 
1.153     pazsan     91: #ifdef HAS_LIBFFI
1.161   ! pazsan     92: Cell *gforth_RP;
        !            93: Address gforth_LP;
1.153     pazsan     94: 
                     95: #include <ffi.h>
                     96: 
                     97: void ** clist;
                     98: void * ritem;
                     99: 
1.161   ! pazsan    100: void gforth_callback(ffi_cif * cif, void * resp, void ** args, Xt * ip)
1.153     pazsan    101: {
1.161   ! pazsan    102:   Cell *rp = gforth_RP;
        !           103:   Cell *sp = gforth_SP;
        !           104:   Float *fp = gforth_FP;
        !           105:   Address lp = gforth_LP;
1.153     pazsan    106: 
                    107:   clist = args;
                    108:   ritem = resp;
                    109: 
                    110:   engine(ip, sp, rp, fp, lp);
                    111: 
                    112:   /* restore global variables */
1.161   ! pazsan    113:   gforth_RP = rp;
        !           114:   gforth_SP = sp;
        !           115:   gforth_FP = fp;
        !           116:   gforth_LP = lp;
1.153     pazsan    117: }
                    118: #endif
                    119: 
1.79      anton     120: #ifdef GFORTH_DEBUGGING
                    121: /* define some VM registers as global variables, so they survive exceptions;
                    122:    global register variables are not up to the task (according to the 
                    123:    GNU C manual) */
                    124: Xt *saved_ip;
                    125: Cell *rp;
                    126: #endif
                    127: 
                    128: #ifdef NO_IP
                    129: Label next_code;
                    130: #endif
                    131: 
                    132: #ifdef HAS_FILE
                    133: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
                    134: char* pfileattr[6]={"r","r","r+","r+","w","w"};
                    135: 
                    136: #ifndef O_BINARY
                    137: #define O_BINARY 0
                    138: #endif
                    139: #ifndef O_TEXT
                    140: #define O_TEXT 0
                    141: #endif
                    142: 
                    143: int ufileattr[6]= {
                    144:   O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
                    145:   O_RDWR  |O_BINARY, O_RDWR  |O_BINARY,
                    146:   O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
                    147: #endif
                    148: /* end global vars for engine.c */
                    149: 
1.1       anton     150: #define PRIM_VERSION 1
                    151: /* increment this whenever the primitives change in an incompatible way */
                    152: 
1.14      pazsan    153: #ifndef DEFAULTPATH
1.39      anton     154: #  define DEFAULTPATH "."
1.14      pazsan    155: #endif
                    156: 
1.1       anton     157: #ifdef MSDOS
                    158: jmp_buf throw_jmp_buf;
                    159: #endif
                    160: 
1.56      anton     161: #if defined(DOUBLY_INDIRECT)
                    162: #  define CFA(n)       ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
1.1       anton     163: #else
1.56      anton     164: #  define CFA(n)       ((Cell)(symbols+((n)&~0x4000UL)))
1.1       anton     165: #endif
                    166: 
                    167: #define maxaligned(n)  (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
                    168: 
                    169: static UCell dictsize=0;
                    170: static UCell dsize=0;
                    171: static UCell rsize=0;
                    172: static UCell fsize=0;
                    173: static UCell lsize=0;
                    174: int offset_image=0;
1.4       anton     175: int die_on_signal=0;
1.13      pazsan    176: #ifndef INCLUDE_IMAGE
1.1       anton     177: static int clear_dictionary=0;
1.24      anton     178: UCell pagesize=1;
1.22      pazsan    179: char *progname;
                    180: #else
                    181: char *progname = "gforth";
                    182: int optind = 1;
1.13      pazsan    183: #endif
1.31      pazsan    184: 
1.131     anton     185: #define CODE_BLOCK_SIZE (4096*1024) /* !! overflow handling for -native */
1.48      anton     186: Address code_area=0;
1.73      anton     187: Cell code_area_size = CODE_BLOCK_SIZE;
1.75      anton     188: Address code_here=NULL+CODE_BLOCK_SIZE; /* does for code-area what HERE
                    189:                                           does for the dictionary */
1.100     anton     190: Address start_flush=NULL; /* start of unflushed code */
1.74      anton     191: Cell last_jump=0; /* if the last prim was compiled without jump, this
                    192:                      is it's number, otherwise this contains 0 */
1.48      anton     193: 
1.60      anton     194: static int no_super=0;   /* true if compile_prim should not fuse prims */
1.81      anton     195: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
                    196:                                             dynamically */
1.110     anton     197: static int print_metrics=0; /* if true, print metrics on exit */
1.157     anton     198: static int static_super_number = 0; /*10000000;*/ /* number of ss used if available */
1.152     anton     199: #define MAX_STATE 9 /* maximum number of states */
1.125     anton     200: static int maxstates = MAX_STATE; /* number of states for stack caching */
1.110     anton     201: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
1.144     pazsan    202: static int diag = 0; /* if true: print diagnostic informations */
1.158     anton     203: static int tpa_noequiv = 0;     /* if true: no state equivalence checking */
                    204: static int tpa_noautomaton = 0; /* if true: no tree parsing automaton */
                    205: static int tpa_trace = 0; /* if true: data for line graph of new states etc. */
1.144     pazsan    206: static int relocs = 0;
                    207: static int nonrelocs = 0;
1.60      anton     208: 
1.30      pazsan    209: #ifdef HAS_DEBUG
1.68      anton     210: int debug=0;
1.144     pazsan    211: # define debugp(x...) if (debug) fprintf(x);
1.31      pazsan    212: #else
                    213: # define perror(x...)
                    214: # define fprintf(x...)
1.144     pazsan    215: # define debugp(x...)
1.30      pazsan    216: #endif
1.31      pazsan    217: 
1.24      anton     218: ImageHeader *gforth_header;
1.43      anton     219: Label *vm_prims;
1.53      anton     220: #ifdef DOUBLY_INDIRECT
                    221: Label *xts; /* same content as vm_prims, but should only be used for xts */
                    222: #endif
1.1       anton     223: 
1.125     anton     224: #ifndef NO_DYNAMIC
                    225: #define MAX_IMMARGS 2
                    226: 
                    227: typedef struct {
                    228:   Label start; /* NULL if not relocatable */
                    229:   Cell length; /* only includes the jump iff superend is true*/
                    230:   Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
                    231:   char superend; /* true if primitive ends superinstruction, i.e.,
                    232:                      unconditional branch, execute, etc. */
                    233:   Cell nimmargs;
                    234:   struct immarg {
                    235:     Cell offset; /* offset of immarg within prim */
                    236:     char rel;    /* true if immarg is relative */
                    237:   } immargs[MAX_IMMARGS];
                    238: } PrimInfo;
                    239: 
                    240: PrimInfo *priminfos;
                    241: PrimInfo **decomp_prims;
                    242: 
1.139     anton     243: const char const* const prim_names[]={
                    244: #include PRIM_NAMES_I
                    245: };
                    246: 
1.148     anton     247: void init_ss_cost(void);
                    248: 
1.125     anton     249: static int is_relocatable(int p)
                    250: {
                    251:   return !no_dynamic && priminfos[p].start != NULL;
                    252: }
                    253: #else /* defined(NO_DYNAMIC) */
                    254: static int is_relocatable(int p)
                    255: {
                    256:   return 0;
                    257: }
                    258: #endif /* defined(NO_DYNAMIC) */
                    259: 
1.30      pazsan    260: #ifdef MEMCMP_AS_SUBROUTINE
                    261: int gforth_memcmp(const char * s1, const char * s2, size_t n)
                    262: {
                    263:   return memcmp(s1, s2, n);
                    264: }
                    265: #endif
                    266: 
1.125     anton     267: static Cell max(Cell a, Cell b)
                    268: {
                    269:   return a>b?a:b;
                    270: }
                    271: 
                    272: static Cell min(Cell a, Cell b)
                    273: {
                    274:   return a<b?a:b;
                    275: }
                    276: 
1.1       anton     277: /* image file format:
1.15      pazsan    278:  *  "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
1.1       anton     279:  *   padding to a multiple of 8
1.84      anton     280:  *   magic: "Gforth3x" means format 0.6,
1.15      pazsan    281:  *              where x is a byte with
                    282:  *              bit 7:   reserved = 0
                    283:  *              bit 6:5: address unit size 2^n octets
                    284:  *              bit 4:3: character size 2^n octets
                    285:  *              bit 2:1: cell size 2^n octets
                    286:  *              bit 0:   endian, big=0, little=1.
                    287:  *  The magic are always 8 octets, no matter what the native AU/character size is
1.1       anton     288:  *  padding to max alignment (no padding necessary on current machines)
1.24      anton     289:  *  ImageHeader structure (see forth.h)
1.1       anton     290:  *  data (size in ImageHeader.image_size)
                    291:  *  tags ((if relocatable, 1 bit/data cell)
                    292:  *
                    293:  * tag==1 means that the corresponding word is an address;
                    294:  * If the word is >=0, the address is within the image;
                    295:  * addresses within the image are given relative to the start of the image.
                    296:  * If the word =-1 (CF_NIL), the address is NIL,
                    297:  * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
                    298:  * If the word =CF(DODOES), it's a DOES> CFA
                    299:  * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
                    300:  *                                     possibly containing a jump to dodoes)
1.51      anton     301:  * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
                    302:  * If the word is <CF(DOESJUMP) and bit 14 is clear, 
                    303:  *                                        it's the threaded code of a primitive
1.85      pazsan    304:  * bits 13..9 of a primitive token state which group the primitive belongs to,
                    305:  * bits 8..0 of a primitive token index into the group
1.1       anton     306:  */
                    307: 
1.115     pazsan    308: Cell groups[32] = {
1.85      pazsan    309:   0,
1.121     anton     310:   0
1.90      anton     311: #undef GROUP
1.115     pazsan    312: #undef GROUPADD
                    313: #define GROUPADD(n) +n
                    314: #define GROUP(x, n) , 0
1.126     anton     315: #include PRIM_GRP_I
1.90      anton     316: #undef GROUP
1.115     pazsan    317: #undef GROUPADD
1.85      pazsan    318: #define GROUP(x, n)
1.115     pazsan    319: #define GROUPADD(n)
1.85      pazsan    320: };
                    321: 
1.161   ! pazsan    322: static unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
1.125     anton     323:                              int size, Cell base)
                    324:      /* produce a bitmask marking all the branch targets */
                    325: {
1.130     anton     326:   int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
1.125     anton     327:   Cell token;
                    328:   unsigned char bits;
1.130     anton     329:   unsigned char *result=malloc(steps);
                    330: 
                    331:   memset(result, 0, steps);
                    332:   for(k=0; k<steps; k++) {
1.125     anton     333:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
1.130     anton     334:       if(bits & (1U << (RELINFOBITS-1))) {
                    335:        assert(i*sizeof(Cell) < size);
1.125     anton     336:         token=image[i];
                    337:        if (token>=base) { /* relocatable address */
                    338:          UCell bitnum=(token-base)/sizeof(Cell);
1.154     anton     339:          if (bitnum/RELINFOBITS < (UCell)steps)
                    340:            result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
1.125     anton     341:        }
                    342:       }
                    343:     }
                    344:   }
                    345:   return result;
                    346: }
                    347: 
1.161   ! pazsan    348: void gforth_relocate(Cell *image, const unsigned char *bitstring, 
        !           349:                     int size, Cell base, Label symbols[])
1.1       anton     350: {
1.130     anton     351:   int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
1.11      pazsan    352:   Cell token;
1.1       anton     353:   char bits;
1.37      anton     354:   Cell max_symbols;
1.46      jwilke    355:   /* 
1.85      pazsan    356:    * A virtual start address that's the real start address minus 
1.46      jwilke    357:    * the one in the image 
                    358:    */
1.45      jwilke    359:   Cell *start = (Cell * ) (((void *) image) - ((void *) base));
1.125     anton     360:   unsigned char *targets = branch_targets(image, bitstring, size, base);
1.1       anton     361: 
1.85      pazsan    362:   /* group index into table */
1.115     pazsan    363:   if(groups[31]==0) {
                    364:     int groupsum=0;
                    365:     for(i=0; i<32; i++) {
                    366:       groupsum += groups[i];
                    367:       groups[i] = groupsum;
                    368:       /* printf("group[%d]=%d\n",i,groupsum); */
                    369:     }
                    370:     i=0;
                    371:   }
1.46      jwilke    372:   
                    373: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
1.37      anton     374:   
1.121     anton     375:   for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
1.37      anton     376:     ;
1.47      anton     377:   max_symbols--;
1.35      pazsan    378: 
1.130     anton     379:   for(k=0; k<steps; k++) {
1.13      pazsan    380:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
1.1       anton     381:       /*      fprintf(stderr,"relocate: image[%d]\n", i);*/
1.130     anton     382:       if(bits & (1U << (RELINFOBITS-1))) {
                    383:        assert(i*sizeof(Cell) < size);
1.35      pazsan    384:        /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
1.45      jwilke    385:         token=image[i];
1.85      pazsan    386:        if(token<0) {
                    387:          int group = (-token & 0x3E00) >> 9;
                    388:          if(group == 0) {
                    389:            switch(token|0x4000) {
1.1       anton     390:            case CF_NIL      : image[i]=0; break;
                    391: #if !defined(DOUBLY_INDIRECT)
                    392:            case CF(DOCOL)   :
                    393:            case CF(DOVAR)   :
                    394:            case CF(DOCON)   :
                    395:            case CF(DOUSER)  : 
                    396:            case CF(DODEFER) : 
1.11      pazsan    397:            case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
1.92      anton     398:            case CF(DOESJUMP): image[i]=0; break;
1.1       anton     399: #endif /* !defined(DOUBLY_INDIRECT) */
                    400:            case CF(DODOES)  :
1.45      jwilke    401:              MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
1.1       anton     402:              break;
1.85      pazsan    403:            default          : /* backward compatibility */
1.56      anton     404: /*           printf("Code field generation image[%x]:=CFA(%x)\n",
1.1       anton     405:                     i, CF(image[i])); */
1.55      anton     406:              if (CF((token | 0x4000))<max_symbols) {
1.56      anton     407:                image[i]=(Cell)CFA(CF(token));
                    408: #ifdef DIRECT_THREADED
1.125     anton     409:                if ((token & 0x4000) == 0) { /* threade code, no CFA */
                    410:                  if (targets[k] & (1U<<(RELINFOBITS-1-j)))
                    411:                    compile_prim1(0);
1.70      anton     412:                  compile_prim1(&image[i]);
1.125     anton     413:                }
1.56      anton     414: #endif
1.55      anton     415:              } else
1.115     pazsan    416:                fprintf(stderr,"Primitive %ld used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n",(long)CF(token),(long)&image[i], i, PACKAGE_VERSION);
1.1       anton     417:            }
1.85      pazsan    418:          } else {
                    419:            int tok = -token & 0x1FF;
                    420:            if (tok < (groups[group+1]-groups[group])) {
                    421: #if defined(DOUBLY_INDIRECT)
                    422:              image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
                    423: #else
                    424:              image[i]=(Cell)CFA((groups[group]+tok));
                    425: #endif
                    426: #ifdef DIRECT_THREADED
1.125     anton     427:              if ((token & 0x4000) == 0) { /* threade code, no CFA */
                    428:                if (targets[k] & (1U<<(RELINFOBITS-1-j)))
                    429:                  compile_prim1(0);
1.85      pazsan    430:                compile_prim1(&image[i]);
1.125     anton     431:              }
1.85      pazsan    432: #endif
                    433:            } else
1.115     pazsan    434:              fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
1.85      pazsan    435:          }
                    436:        } else {
1.101     anton     437:           /* if base is > 0: 0 is a null reference so don't adjust*/
1.45      jwilke    438:           if (token>=base) {
                    439:             image[i]+=(Cell)start;
                    440:           }
1.46      jwilke    441:         }
1.1       anton     442:       }
                    443:     }
1.31      pazsan    444:   }
1.125     anton     445:   free(targets);
1.70      anton     446:   finish_code();
1.26      jwilke    447:   ((ImageHeader*)(image))->base = (Address) image;
1.1       anton     448: }
                    449: 
1.161   ! pazsan    450: static UCell checksum(Label symbols[])
1.1       anton     451: {
                    452:   UCell r=PRIM_VERSION;
                    453:   Cell i;
                    454: 
                    455:   for (i=DOCOL; i<=DOESJUMP; i++) {
                    456:     r ^= (UCell)(symbols[i]);
                    457:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
                    458:   }
                    459: #ifdef DIRECT_THREADED
                    460:   /* we have to consider all the primitives */
                    461:   for (; symbols[i]!=(Label)0; i++) {
                    462:     r ^= (UCell)(symbols[i]);
                    463:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
                    464:   }
                    465: #else
                    466:   /* in indirect threaded code all primitives are accessed through the
                    467:      symbols table, so we just have to put the base address of symbols
                    468:      in the checksum */
                    469:   r ^= (UCell)symbols;
                    470: #endif
                    471:   return r;
                    472: }
                    473: 
1.161   ! pazsan    474: static Address verbose_malloc(Cell size)
1.3       anton     475: {
                    476:   Address r;
                    477:   /* leave a little room (64B) for stack underflows */
                    478:   if ((r = malloc(size+64))==NULL) {
                    479:     perror(progname);
                    480:     exit(1);
                    481:   }
                    482:   r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
1.144     pazsan    483:   debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
1.3       anton     484:   return r;
                    485: }
                    486: 
1.33      anton     487: static Address next_address=0;
1.161   ! pazsan    488: static void after_alloc(Address r, Cell size)
1.33      anton     489: {
                    490:   if (r != (Address)-1) {
1.144     pazsan    491:     debugp(stderr, "success, address=$%lx\n", (long) r);
1.33      anton     492:     if (pagesize != 1)
                    493:       next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
                    494:   } else {
1.144     pazsan    495:     debugp(stderr, "failed: %s\n", strerror(errno));
1.33      anton     496:   }
                    497: }
                    498: 
1.34      anton     499: #ifndef MAP_FAILED
                    500: #define MAP_FAILED ((Address) -1)
                    501: #endif
                    502: #ifndef MAP_FILE
                    503: # define MAP_FILE 0
                    504: #endif
                    505: #ifndef MAP_PRIVATE
                    506: # define MAP_PRIVATE 0
                    507: #endif
1.91      anton     508: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
                    509: # define MAP_ANON MAP_ANONYMOUS
                    510: #endif
1.34      anton     511: 
                    512: #if defined(HAVE_MMAP)
                    513: static Address alloc_mmap(Cell size)
1.1       anton     514: {
                    515:   Address r;
                    516: 
                    517: #if defined(MAP_ANON)
1.144     pazsan    518:   debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
1.34      anton     519:   r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
1.1       anton     520: #else /* !defined(MAP_ANON) */
1.17      anton     521:   /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
                    522:      apparently defaults) */
1.1       anton     523:   static int dev_zero=-1;
                    524: 
                    525:   if (dev_zero == -1)
                    526:     dev_zero = open("/dev/zero", O_RDONLY);
                    527:   if (dev_zero == -1) {
1.34      anton     528:     r = MAP_FAILED;
1.144     pazsan    529:     debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ", 
1.1       anton     530:              strerror(errno));
                    531:   } else {
1.144     pazsan    532:     debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
1.1       anton     533:     r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE, dev_zero, 0);
                    534:   }
                    535: #endif /* !defined(MAP_ANON) */
1.34      anton     536:   after_alloc(r, size);
                    537:   return r;  
                    538: }
                    539: #endif
                    540: 
1.161   ! pazsan    541: Address gforth_alloc(Cell size)
1.34      anton     542: {
                    543: #if HAVE_MMAP
                    544:   Address r;
                    545: 
                    546:   r=alloc_mmap(size);
1.117     anton     547:   if (r!=(Address)MAP_FAILED)
1.1       anton     548:     return r;
                    549: #endif /* HAVE_MMAP */
1.3       anton     550:   /* use malloc as fallback */
                    551:   return verbose_malloc(size);
1.1       anton     552: }
                    553: 
1.161   ! pazsan    554: static Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
1.33      anton     555: {
1.34      anton     556:   Address image = MAP_FAILED;
1.33      anton     557: 
1.56      anton     558: #if defined(HAVE_MMAP)
1.33      anton     559:   if (offset==0) {
1.34      anton     560:     image=alloc_mmap(dictsize);
1.150     anton     561:     if (image != (Address)MAP_FAILED) {
                    562:       Address image1;
                    563:       debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
                    564:       image1 = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE, fileno(file), 0);
                    565:       after_alloc(image1,dictsize);
                    566:       if (image1 == (Address)MAP_FAILED)
                    567:        goto read_image;
                    568:     }
1.33      anton     569:   }
1.56      anton     570: #endif /* defined(HAVE_MMAP) */
1.117     anton     571:   if (image == (Address)MAP_FAILED) {
1.161   ! pazsan    572:     image = gforth_alloc(dictsize+offset)+offset;
1.149     anton     573:   read_image:
1.33      anton     574:     rewind(file);  /* fseek(imagefile,0L,SEEK_SET); */
1.34      anton     575:     fread(image, 1, imagesize, file);
1.33      anton     576:   }
                    577:   return image;
                    578: }
                    579: 
1.10      pazsan    580: void set_stack_sizes(ImageHeader * header)
                    581: {
                    582:   if (dictsize==0)
                    583:     dictsize = header->dict_size;
                    584:   if (dsize==0)
                    585:     dsize = header->data_stack_size;
                    586:   if (rsize==0)
                    587:     rsize = header->return_stack_size;
                    588:   if (fsize==0)
                    589:     fsize = header->fp_stack_size;
                    590:   if (lsize==0)
                    591:     lsize = header->locals_stack_size;
                    592:   dictsize=maxaligned(dictsize);
                    593:   dsize=maxaligned(dsize);
                    594:   rsize=maxaligned(rsize);
                    595:   lsize=maxaligned(lsize);
                    596:   fsize=maxaligned(fsize);
                    597: }
                    598: 
                    599: void alloc_stacks(ImageHeader * header)
                    600: {
                    601:   header->dict_size=dictsize;
                    602:   header->data_stack_size=dsize;
                    603:   header->fp_stack_size=fsize;
                    604:   header->return_stack_size=rsize;
                    605:   header->locals_stack_size=lsize;
                    606: 
1.161   ! pazsan    607:   header->data_stack_base=gforth_alloc(dsize);
        !           608:   header->fp_stack_base=gforth_alloc(fsize);
        !           609:   header->return_stack_base=gforth_alloc(rsize);
        !           610:   header->locals_stack_base=gforth_alloc(lsize);
1.10      pazsan    611: }
                    612: 
1.161   ! pazsan    613: #warning You can ignore the warnings about clobbered variables in gforth_go
        !           614: int gforth_go(Address image, int stack, Cell *entries)
1.11      pazsan    615: {
1.38      anton     616:   volatile ImageHeader *image_header = (ImageHeader *)image;
1.18      anton     617:   Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
1.44      pazsan    618:   Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
1.18      anton     619:   Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
1.44      pazsan    620: #ifdef GFORTH_DEBUGGING
1.38      anton     621:   volatile Cell *orig_rp0=rp0;
1.44      pazsan    622: #endif
1.18      anton     623:   Address lp0=image_header->locals_stack_base + lsize;
                    624:   Xt *ip0=(Xt *)(image_header->boot_entry);
1.13      pazsan    625: #ifdef SYSSIGNALS
1.11      pazsan    626:   int throw_code;
1.13      pazsan    627: #endif
1.11      pazsan    628: 
                    629:   /* ensure that the cached elements (if any) are accessible */
1.151     anton     630: #if !(defined(GFORTH_DEBUGGING) || defined(INDIRECT_THREADED) || defined(DOUBLY_INDIRECT) || defined(VM_PROFILING))
                    631:   sp0 -= 8; /* make stuff below bottom accessible for stack caching */
                    632: #endif
1.41      anton     633:   IF_fpTOS(fp0--);
1.11      pazsan    634:   
                    635:   for(;stack>0;stack--)
1.18      anton     636:     *--sp0=entries[stack-1];
1.11      pazsan    637: 
1.30      pazsan    638: #ifdef SYSSIGNALS
1.11      pazsan    639:   get_winsize();
                    640:    
                    641:   install_signal_handlers(); /* right place? */
                    642:   
                    643:   if ((throw_code=setjmp(throw_jmp_buf))) {
1.152     anton     644:     static Cell signal_data_stack[24];
                    645:     static Cell signal_return_stack[16];
1.11      pazsan    646:     static Float signal_fp_stack[1];
1.13      pazsan    647: 
1.152     anton     648:     signal_data_stack[15]=throw_code;
1.18      anton     649: 
                    650: #ifdef GFORTH_DEBUGGING
1.144     pazsan    651:     debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
1.97      anton     652:              throw_code, saved_ip, rp);
1.38      anton     653:     if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
1.18      anton     654:       /* no rstack overflow or underflow */
                    655:       rp0 = rp;
1.63      anton     656:       *--rp0 = (Cell)saved_ip;
1.18      anton     657:     }
                    658:     else /* I love non-syntactic ifdefs :-) */
1.152     anton     659:       rp0 = signal_return_stack+16;
1.97      anton     660: #else  /* !defined(GFORTH_DEBUGGING) */
1.144     pazsan    661:     debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
1.152     anton     662:       rp0 = signal_return_stack+16;
1.97      anton     663: #endif /* !defined(GFORTH_DEBUGGING) */
1.25      anton     664:     /* fprintf(stderr, "rp=$%x\n",rp0);*/
1.11      pazsan    665:     
1.152     anton     666:     return((int)(Cell)engine(image_header->throw_entry, signal_data_stack+15,
1.18      anton     667:                       rp0, signal_fp_stack, 0));
1.11      pazsan    668:   }
1.13      pazsan    669: #endif
1.11      pazsan    670: 
1.33      anton     671:   return((int)(Cell)engine(ip0,sp0,rp0,fp0,lp0));
1.11      pazsan    672: }
                    673: 
1.30      pazsan    674: #ifndef INCLUDE_IMAGE
1.161   ! pazsan    675: static void print_sizes(Cell sizebyte)
1.21      anton     676:      /* print size information */
                    677: {
                    678:   static char* endianstring[]= { "   big","little" };
                    679:   
                    680:   fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
                    681:          endianstring[sizebyte & 1],
                    682:          1 << ((sizebyte >> 1) & 3),
                    683:          1 << ((sizebyte >> 3) & 3),
                    684:          1 << ((sizebyte >> 5) & 3));
                    685: }
                    686: 
1.106     anton     687: /* static superinstruction stuff */
                    688: 
1.141     anton     689: struct cost { /* super_info might be a more accurate name */
1.106     anton     690:   char loads;       /* number of stack loads */
                    691:   char stores;      /* number of stack stores */
                    692:   char updates;     /* number of stack pointer updates */
1.123     anton     693:   char branch;     /* is it a branch (SET_IP) */
1.125     anton     694:   unsigned char state_in;    /* state on entry */
                    695:   unsigned char state_out;   /* state on exit */
1.142     anton     696:   unsigned char imm_ops;     /* number of immediate operands */
1.123     anton     697:   short offset;     /* offset into super2 table */
1.125     anton     698:   unsigned char length;      /* number of components */
1.106     anton     699: };
                    700: 
1.121     anton     701: PrimNum super2[] = {
1.126     anton     702: #include SUPER2_I
1.106     anton     703: };
                    704: 
                    705: struct cost super_costs[] = {
1.126     anton     706: #include COSTS_I
1.106     anton     707: };
                    708: 
1.125     anton     709: struct super_state {
                    710:   struct super_state *next;
                    711:   PrimNum super;
                    712: };
                    713: 
1.106     anton     714: #define HASH_SIZE 256
                    715: 
                    716: struct super_table_entry {
                    717:   struct super_table_entry *next;
1.121     anton     718:   PrimNum *start;
1.106     anton     719:   short length;
1.125     anton     720:   struct super_state *ss_list; /* list of supers */
1.106     anton     721: } *super_table[HASH_SIZE];
                    722: int max_super=2;
                    723: 
1.125     anton     724: struct super_state *state_transitions=NULL;
                    725: 
1.161   ! pazsan    726: static int hash_super(PrimNum *start, int length)
1.106     anton     727: {
                    728:   int i, r;
                    729:   
                    730:   for (i=0, r=0; i<length; i++) {
                    731:     r <<= 1;
                    732:     r += start[i];
                    733:   }
                    734:   return r & (HASH_SIZE-1);
                    735: }
                    736: 
1.161   ! pazsan    737: static struct super_state **lookup_super(PrimNum *start, int length)
1.106     anton     738: {
                    739:   int hash=hash_super(start,length);
                    740:   struct super_table_entry *p = super_table[hash];
                    741: 
1.125     anton     742:   /* assert(length >= 2); */
1.106     anton     743:   for (; p!=NULL; p = p->next) {
                    744:     if (length == p->length &&
1.121     anton     745:        memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
1.125     anton     746:       return &(p->ss_list);
1.106     anton     747:   }
1.125     anton     748:   return NULL;
1.106     anton     749: }
                    750: 
1.161   ! pazsan    751: static void prepare_super_table()
1.106     anton     752: {
                    753:   int i;
1.109     anton     754:   int nsupers = 0;
1.106     anton     755: 
                    756:   for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
                    757:     struct cost *c = &super_costs[i];
1.125     anton     758:     if ((c->length < 2 || nsupers < static_super_number) &&
                    759:        c->state_in < maxstates && c->state_out < maxstates) {
                    760:       struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
                    761:       struct super_state *ss = malloc(sizeof(struct super_state));
                    762:       ss->super= i;
                    763:       if (c->offset==N_noop && i != N_noop) {
                    764:        if (is_relocatable(i)) {
                    765:          ss->next = state_transitions;
                    766:          state_transitions = ss;
                    767:        }
                    768:       } else if (ss_listp != NULL) {
                    769:        ss->next = *ss_listp;
                    770:        *ss_listp = ss;
                    771:       } else {
                    772:        int hash = hash_super(super2+c->offset, c->length);
                    773:        struct super_table_entry **p = &super_table[hash];
                    774:        struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
                    775:        ss->next = NULL;
                    776:        e->next = *p;
                    777:        e->start = super2 + c->offset;
                    778:        e->length = c->length;
                    779:        e->ss_list = ss;
                    780:        *p = e;
                    781:       }
1.106     anton     782:       if (c->length > max_super)
                    783:        max_super = c->length;
1.125     anton     784:       if (c->length >= 2)
                    785:        nsupers++;
1.106     anton     786:     }
                    787:   }
1.144     pazsan    788:   debugp(stderr, "Using %d static superinsts\n", nsupers);
1.106     anton     789: }
                    790: 
                    791: /* dynamic replication/superinstruction stuff */
                    792: 
1.69      anton     793: #ifndef NO_DYNAMIC
1.161   ! pazsan    794: static int compare_priminfo_length(const void *_a, const void *_b)
1.76      anton     795: {
1.90      anton     796:   PrimInfo **a = (PrimInfo **)_a;
                    797:   PrimInfo **b = (PrimInfo **)_b;
1.77      anton     798:   Cell diff = (*a)->length - (*b)->length;
                    799:   if (diff)
                    800:     return diff;
                    801:   else /* break ties by start address; thus the decompiler produces
                    802:           the earliest primitive with the same code (e.g. noop instead
                    803:           of (char) and @ instead of >code-address */
                    804:     return (*b)->start - (*a)->start;
1.76      anton     805: }
1.112     anton     806: #endif /* !defined(NO_DYNAMIC) */
1.76      anton     807: 
1.125     anton     808: static char MAYBE_UNUSED superend[]={
1.126     anton     809: #include PRIM_SUPEREND_I
1.106     anton     810: };
1.107     anton     811: 
                    812: Cell npriminfos=0;
1.76      anton     813: 
1.146     anton     814: Label goto_start;
                    815: Cell goto_len;
                    816: 
1.161   ! pazsan    817: static int compare_labels(const void *pa, const void *pb)
1.113     anton     818: {
1.114     anton     819:   Label a = *(Label *)pa;
                    820:   Label b = *(Label *)pb;
                    821:   return a-b;
                    822: }
1.113     anton     823: 
1.161   ! pazsan    824: static Label bsearch_next(Label key, Label *a, UCell n)
1.114     anton     825:      /* a is sorted; return the label >=key that is the closest in a;
                    826:         return NULL if there is no label in a >=key */
                    827: {
                    828:   int mid = (n-1)/2;
                    829:   if (n<1)
                    830:     return NULL;
                    831:   if (n == 1) {
                    832:     if (a[0] < key)
                    833:       return NULL;
                    834:     else
                    835:       return a[0];
                    836:   }
                    837:   if (a[mid] < key)
                    838:     return bsearch_next(key, a+mid+1, n-mid-1);
                    839:   else
                    840:     return bsearch_next(key, a, mid+1);
1.113     anton     841: }
                    842: 
1.161   ! pazsan    843: static void check_prims(Label symbols1[])
1.47      anton     844: {
                    845:   int i;
1.90      anton     846: #ifndef NO_DYNAMIC
1.146     anton     847:   Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted, *goto_p;
1.119     anton     848:   int nends1j;
1.90      anton     849: #endif
1.47      anton     850: 
1.66      anton     851:   if (debug)
                    852: #ifdef __VERSION__
                    853:     fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
                    854: #else
                    855: #define xstr(s) str(s)
                    856: #define str(s) #s
                    857:   fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n"); 
                    858: #endif
1.121     anton     859:   for (i=0; symbols1[i]!=0; i++)
1.47      anton     860:     ;
1.55      anton     861:   npriminfos = i;
1.70      anton     862:   
                    863: #ifndef NO_DYNAMIC
1.66      anton     864:   if (no_dynamic)
                    865:     return;
1.55      anton     866:   symbols2=engine2(0,0,0,0,0);
1.70      anton     867: #if NO_IP
                    868:   symbols3=engine3(0,0,0,0,0);
                    869: #else
                    870:   symbols3=symbols1;
                    871: #endif
1.121     anton     872:   ends1 = symbols1+i+1;
1.119     anton     873:   ends1j =   ends1+i;
1.146     anton     874:   goto_p = ends1j+i+1; /* goto_p[0]==before; ...[1]==after;*/
1.121     anton     875:   nends1j = i+1;
1.119     anton     876:   ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
                    877:   memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
                    878:   qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
1.146     anton     879: 
                    880:   /* check whether the "goto *" is relocatable */
                    881:   goto_len = goto_p[1]-goto_p[0];
                    882:   debugp(stderr, "goto * %p %p len=%ld\n",
                    883:         goto_p[0],symbols2[goto_p-symbols1],goto_len);
                    884:   if (memcmp(goto_p[0],symbols2[goto_p-symbols1],goto_len)!=0) { /* unequal */
                    885:     no_dynamic=1;
                    886:     debugp(stderr,"  not relocatable, disabling dynamic code generation\n");
1.148     anton     887:     init_ss_cost();
1.146     anton     888:     return;
                    889:   }
                    890:   goto_start = goto_p[0];
1.113     anton     891:   
1.47      anton     892:   priminfos = calloc(i,sizeof(PrimInfo));
1.121     anton     893:   for (i=0; symbols1[i]!=0; i++) {
1.70      anton     894:     int prim_len = ends1[i]-symbols1[i];
1.47      anton     895:     PrimInfo *pi=&priminfos[i];
1.154     anton     896:     struct cost *sc=&super_costs[i];
1.70      anton     897:     int j=0;
                    898:     char *s1 = (char *)symbols1[i];
                    899:     char *s2 = (char *)symbols2[i];
                    900:     char *s3 = (char *)symbols3[i];
1.119     anton     901:     Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
1.70      anton     902: 
                    903:     pi->start = s1;
1.121     anton     904:     pi->superend = superend[i]|no_super;
1.147     anton     905:     pi->length = prim_len;
1.113     anton     906:     pi->restlength = endlabel - symbols1[i] - pi->length;
1.70      anton     907:     pi->nimmargs = 0;
1.144     pazsan    908:     relocs++;
1.154     anton     909:     debugp(stderr, "%-15s %d-%d %4d %p %p %p len=%3ld rest=%2ld send=%1d",
                    910:           prim_names[i], sc->state_in, sc->state_out,
                    911:           i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength),
                    912:           pi->superend);
1.114     anton     913:     if (endlabel == NULL) {
                    914:       pi->start = NULL; /* not relocatable */
1.122     anton     915:       if (pi->length<0) pi->length=100;
1.144     pazsan    916:       debugp(stderr,"\n   non_reloc: no J label > start found\n");
                    917:       relocs--;
                    918:       nonrelocs++;
1.114     anton     919:       continue;
                    920:     }
                    921:     if (ends1[i] > endlabel && !pi->superend) {
1.113     anton     922:       pi->start = NULL; /* not relocatable */
1.122     anton     923:       pi->length = endlabel-symbols1[i];
1.144     pazsan    924:       debugp(stderr,"\n   non_reloc: there is a J label before the K label (restlength<0)\n");
                    925:       relocs--;
                    926:       nonrelocs++;
1.113     anton     927:       continue;
                    928:     }
1.114     anton     929:     if (ends1[i] < pi->start && !pi->superend) {
1.113     anton     930:       pi->start = NULL; /* not relocatable */
1.122     anton     931:       pi->length = endlabel-symbols1[i];
1.144     pazsan    932:       debugp(stderr,"\n   non_reloc: K label before I label (length<0)\n");
                    933:       relocs--;
                    934:       nonrelocs++;
1.113     anton     935:       continue;
                    936:     }
1.138     anton     937:     assert(pi->length>=0);
1.113     anton     938:     assert(pi->restlength >=0);
1.74      anton     939:     while (j<(pi->length+pi->restlength)) {
1.70      anton     940:       if (s1[j]==s3[j]) {
                    941:        if (s1[j] != s2[j]) {
                    942:          pi->start = NULL; /* not relocatable */
1.144     pazsan    943:          debugp(stderr,"\n   non_reloc: engine1!=engine2 offset %3d",j);
1.74      anton     944:          /* assert(j<prim_len); */
1.144     pazsan    945:          relocs--;
                    946:          nonrelocs++;
1.70      anton     947:          break;
                    948:        }
                    949:        j++;
                    950:       } else {
                    951:        struct immarg *ia=&pi->immargs[pi->nimmargs];
                    952: 
                    953:        pi->nimmargs++;
                    954:        ia->offset=j;
                    955:        if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
                    956:          ia->rel=0;
1.144     pazsan    957:          debugp(stderr,"\n   absolute immarg: offset %3d",j);
1.70      anton     958:        } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
                    959:                   symbols1[DOESJUMP+1]) {
                    960:          ia->rel=1;
1.144     pazsan    961:          debugp(stderr,"\n   relative immarg: offset %3d",j);
1.70      anton     962:        } else {
                    963:          pi->start = NULL; /* not relocatable */
1.144     pazsan    964:          debugp(stderr,"\n   non_reloc: engine1!=engine3 offset %3d",j);
1.74      anton     965:          /* assert(j<prim_len);*/
1.144     pazsan    966:          relocs--;
                    967:          nonrelocs++;
1.70      anton     968:          break;
                    969:        }
                    970:        j+=4;
1.47      anton     971:       }
                    972:     }
1.144     pazsan    973:     debugp(stderr,"\n");
1.70      anton     974:   }
1.76      anton     975:   decomp_prims = calloc(i,sizeof(PrimInfo *));
                    976:   for (i=DOESJUMP+1; i<npriminfos; i++)
                    977:     decomp_prims[i] = &(priminfos[i]);
                    978:   qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
                    979:        compare_priminfo_length);
1.70      anton     980: #endif
                    981: }
                    982: 
1.161   ! pazsan    983: static void flush_to_here(void)
1.74      anton     984: {
1.93      anton     985: #ifndef NO_DYNAMIC
1.100     anton     986:   if (start_flush)
                    987:     FLUSH_ICACHE(start_flush, code_here-start_flush);
1.74      anton     988:   start_flush=code_here;
1.93      anton     989: #endif
1.74      anton     990: }
                    991: 
1.93      anton     992: #ifndef NO_DYNAMIC
1.161   ! pazsan    993: static void append_jump(void)
1.74      anton     994: {
                    995:   if (last_jump) {
                    996:     PrimInfo *pi = &priminfos[last_jump];
                    997:     
                    998:     memcpy(code_here, pi->start+pi->length, pi->restlength);
                    999:     code_here += pi->restlength;
1.147     anton    1000:     memcpy(code_here, goto_start, goto_len);
                   1001:     code_here += goto_len;
1.74      anton    1002:     last_jump=0;
                   1003:   }
                   1004: }
                   1005: 
1.75      anton    1006: /* Gforth remembers all code blocks in this list.  On forgetting (by
                   1007: executing a marker) the code blocks are not freed (because Gforth does
                   1008: not remember how they were allocated; hmm, remembering that might be
                   1009: easier and cleaner).  Instead, code_here etc. are reset to the old
                   1010: value, and the "forgotten" code blocks are reused when they are
                   1011: needed. */
                   1012: 
                   1013: struct code_block_list {
                   1014:   struct code_block_list *next;
                   1015:   Address block;
                   1016:   Cell size;
                   1017: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
                   1018: 
1.161   ! pazsan   1019: static Address append_prim(Cell p)
1.74      anton    1020: {
                   1021:   PrimInfo *pi = &priminfos[p];
                   1022:   Address old_code_here = code_here;
                   1023: 
1.159     anton    1024:   if (code_area+code_area_size < code_here+pi->length+pi->restlength+goto_len) {
1.75      anton    1025:     struct code_block_list *p;
1.74      anton    1026:     append_jump();
1.93      anton    1027:     flush_to_here();
1.75      anton    1028:     if (*next_code_blockp == NULL) {
1.161   ! pazsan   1029:       code_here = start_flush = code_area = gforth_alloc(code_area_size);
1.75      anton    1030:       p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
                   1031:       *next_code_blockp = p;
                   1032:       p->next = NULL;
                   1033:       p->block = code_here;
                   1034:       p->size = code_area_size;
                   1035:     } else {
                   1036:       p = *next_code_blockp;
                   1037:       code_here = start_flush = code_area = p->block;
                   1038:     }
1.74      anton    1039:     old_code_here = code_here;
1.75      anton    1040:     next_code_blockp = &(p->next);
1.74      anton    1041:   }
                   1042:   memcpy(code_here, pi->start, pi->length);
                   1043:   code_here += pi->length;
                   1044:   return old_code_here;
                   1045: }
                   1046: #endif
1.75      anton    1047: 
                   1048: int forget_dyncode(Address code)
                   1049: {
                   1050: #ifdef NO_DYNAMIC
                   1051:   return -1;
                   1052: #else
                   1053:   struct code_block_list *p, **pp;
                   1054: 
                   1055:   for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
                   1056:     if (code >= p->block && code < p->block+p->size) {
                   1057:       next_code_blockp = &(p->next);
                   1058:       code_here = start_flush = code;
                   1059:       code_area = p->block;
                   1060:       last_jump = 0;
                   1061:       return -1;
                   1062:     }
                   1063:   }
1.78      anton    1064:   return -no_dynamic;
1.75      anton    1065: #endif /* !defined(NO_DYNAMIC) */
                   1066: }
                   1067: 
1.161   ! pazsan   1068: static long dyncodesize(void)
1.104     anton    1069: {
                   1070: #ifndef NO_DYNAMIC
1.106     anton    1071:   struct code_block_list *p;
1.104     anton    1072:   long size=0;
                   1073:   for (p=code_block_list; p!=NULL; p=p->next) {
                   1074:     if (code_here >= p->block && code_here < p->block+p->size)
                   1075:       return size + (code_here - p->block);
                   1076:     else
                   1077:       size += p->size;
                   1078:   }
                   1079: #endif /* !defined(NO_DYNAMIC) */
                   1080:   return 0;
                   1081: }
                   1082: 
1.90      anton    1083: Label decompile_code(Label _code)
1.75      anton    1084: {
1.76      anton    1085: #ifdef NO_DYNAMIC
1.90      anton    1086:   return _code;
1.76      anton    1087: #else /* !defined(NO_DYNAMIC) */
                   1088:   Cell i;
1.77      anton    1089:   struct code_block_list *p;
1.90      anton    1090:   Address code=_code;
1.76      anton    1091: 
1.77      anton    1092:   /* first, check if we are in code at all */
                   1093:   for (p = code_block_list;; p = p->next) {
                   1094:     if (p == NULL)
                   1095:       return code;
                   1096:     if (code >= p->block && code < p->block+p->size)
                   1097:       break;
                   1098:   }
1.76      anton    1099:   /* reverse order because NOOP might match other prims */
                   1100:   for (i=npriminfos-1; i>DOESJUMP; i--) {
                   1101:     PrimInfo *pi=decomp_prims[i];
                   1102:     if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1.121     anton    1103:       return vm_prims[super2[super_costs[pi-priminfos].offset]];
1.118     anton    1104:     /* return pi->start;*/
1.76      anton    1105:   }
                   1106:   return code;
                   1107: #endif /* !defined(NO_DYNAMIC) */
1.75      anton    1108: }
1.74      anton    1109: 
1.70      anton    1110: #ifdef NO_IP
                   1111: int nbranchinfos=0;
                   1112: 
                   1113: struct branchinfo {
1.136     anton    1114:   Label **targetpp; /* **(bi->targetpp) is the target */
1.70      anton    1115:   Cell *addressptr; /* store the target here */
                   1116: } branchinfos[100000];
                   1117: 
                   1118: int ndoesexecinfos=0;
                   1119: struct doesexecinfo {
                   1120:   int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1.136     anton    1121:   Label *targetp; /*target for branch (because this is not in threaded code)*/
1.70      anton    1122:   Cell *xt; /* cfa of word whose does-code needs calling */
                   1123: } doesexecinfos[10000];
                   1124: 
1.161   ! pazsan   1125: static void set_rel_target(Cell *source, Label target)
1.70      anton    1126: {
                   1127:   *source = ((Cell)target)-(((Cell)source)+4);
                   1128: }
                   1129: 
1.161   ! pazsan   1130: static void register_branchinfo(Label source, Cell *targetpp)
1.70      anton    1131: {
                   1132:   struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1.136     anton    1133:   bi->targetpp = (Label **)targetpp;
1.70      anton    1134:   bi->addressptr = (Cell *)source;
                   1135:   nbranchinfos++;
                   1136: }
                   1137: 
1.161   ! pazsan   1138: static Address compile_prim1arg(PrimNum p, Cell **argp)
1.70      anton    1139: {
1.133     anton    1140:   Address old_code_here=append_prim(p);
1.70      anton    1141: 
1.74      anton    1142:   assert(vm_prims[p]==priminfos[p].start);
1.133     anton    1143:   *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
                   1144:   return old_code_here;
1.70      anton    1145: }
                   1146: 
1.161   ! pazsan   1147: static Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1.70      anton    1148: {
1.73      anton    1149:   PrimInfo *pi = &priminfos[N_call2];
1.74      anton    1150:   Address old_code_here = append_prim(N_call2);
1.70      anton    1151: 
1.134     anton    1152:   *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1.136     anton    1153:   register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1.134     anton    1154:   return old_code_here;
1.70      anton    1155: }
                   1156: #endif
                   1157: 
                   1158: void finish_code(void)
                   1159: {
                   1160: #ifdef NO_IP
                   1161:   Cell i;
                   1162: 
                   1163:   compile_prim1(NULL);
                   1164:   for (i=0; i<ndoesexecinfos; i++) {
                   1165:     struct doesexecinfo *dei = &doesexecinfos[i];
1.136     anton    1166:     dei->targetp = (Label *)DOES_CODE1((dei->xt));
                   1167:     branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1.70      anton    1168:   }
                   1169:   ndoesexecinfos = 0;
                   1170:   for (i=0; i<nbranchinfos; i++) {
                   1171:     struct branchinfo *bi=&branchinfos[i];
1.136     anton    1172:     set_rel_target(bi->addressptr, **(bi->targetpp));
1.70      anton    1173:   }
                   1174:   nbranchinfos = 0;
1.128     anton    1175: #else
                   1176:   compile_prim1(NULL);
1.48      anton    1177: #endif
1.93      anton    1178:   flush_to_here();
1.48      anton    1179: }
                   1180: 
1.128     anton    1181: #ifdef NO_IP
1.161   ! pazsan   1182: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1.128     anton    1183:      /* compile prim #p dynamically (mod flags etc.) and return start
                   1184:        address of generated code for putting it into the threaded
                   1185:        code. This function is only called if all the associated
                   1186:        inline arguments of p are already in place (at tcp[1] etc.) */
                   1187: {
                   1188:   PrimInfo *pi=&priminfos[p];
                   1189:   Cell *next_code_target=NULL;
1.135     anton    1190:   Address codeaddr;
                   1191:   Address primstart;
1.128     anton    1192:   
                   1193:   assert(p<npriminfos);
                   1194:   if (p==N_execute || p==N_perform || p==N_lit_perform) {
1.134     anton    1195:     codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1.135     anton    1196:     primstart = append_prim(p);
                   1197:     goto other_prim;
                   1198:   } else if (p==N_call) {
1.136     anton    1199:     codeaddr = compile_call2(tcp+1, &next_code_target);
1.128     anton    1200:   } else if (p==N_does_exec) {
                   1201:     struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1.133     anton    1202:     Cell *arg;
                   1203:     codeaddr = compile_prim1arg(N_lit,&arg);
                   1204:     *arg = (Cell)PFA(tcp[1]);
1.128     anton    1205:     /* we cannot determine the callee now (last_start[1] may be a
                   1206:        forward reference), so just register an arbitrary target, and
                   1207:        register in dei that we need to fix this before resolving
                   1208:        branches */
                   1209:     dei->branchinfo = nbranchinfos;
                   1210:     dei->xt = (Cell *)(tcp[1]);
1.134     anton    1211:     compile_call2(0, &next_code_target);
1.128     anton    1212:   } else if (!is_relocatable(p)) {
1.133     anton    1213:     Cell *branch_target;
                   1214:     codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
                   1215:     compile_prim1arg(N_branch,&branch_target);
                   1216:     set_rel_target(branch_target,vm_prims[p]);
1.128     anton    1217:   } else {
                   1218:     unsigned j;
1.135     anton    1219: 
                   1220:     codeaddr = primstart = append_prim(p);
                   1221:   other_prim:
1.128     anton    1222:     for (j=0; j<pi->nimmargs; j++) {
                   1223:       struct immarg *ia = &(pi->immargs[j]);
1.136     anton    1224:       Cell *argp = tcp + pi->nimmargs - j;
                   1225:       Cell argval = *argp; /* !! specific to prims */
1.128     anton    1226:       if (ia->rel) { /* !! assumption: relative refs are branches */
1.136     anton    1227:        register_branchinfo(primstart + ia->offset, argp);
1.128     anton    1228:       } else /* plain argument */
1.135     anton    1229:        *(Cell *)(primstart + ia->offset) = argval;
1.128     anton    1230:     }
                   1231:   }
                   1232:   if (next_code_target!=NULL)
                   1233:     *next_code_target = (Cell)code_here;
1.135     anton    1234:   return (Cell)codeaddr;
1.128     anton    1235: }
                   1236: #else /* !defined(NO_IP) */
1.161   ! pazsan   1237: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1.128     anton    1238:      /* compile prim #p dynamically (mod flags etc.) and return start
                   1239:         address of generated code for putting it into the threaded code */
1.108     anton    1240: {
1.121     anton    1241:   Cell static_prim = (Cell)vm_prims[p];
1.108     anton    1242: #if defined(NO_DYNAMIC)
                   1243:   return static_prim;
                   1244: #else /* !defined(NO_DYNAMIC) */
                   1245:   Address old_code_here;
                   1246: 
                   1247:   if (no_dynamic)
                   1248:     return static_prim;
1.125     anton    1249:   if (p>=npriminfos || !is_relocatable(p)) {
1.108     anton    1250:     append_jump();
                   1251:     return static_prim;
                   1252:   }
                   1253:   old_code_here = append_prim(p);
1.147     anton    1254:   last_jump = p;
                   1255:   if (priminfos[p].superend)
                   1256:     append_jump();
1.108     anton    1257:   return (Cell)old_code_here;
                   1258: #endif  /* !defined(NO_DYNAMIC) */
                   1259: }
1.128     anton    1260: #endif /* !defined(NO_IP) */
1.70      anton    1261: 
1.109     anton    1262: #ifndef NO_DYNAMIC
1.161   ! pazsan   1263: static int cost_codesize(int prim)
1.109     anton    1264: {
1.121     anton    1265:   return priminfos[prim].length;
1.109     anton    1266: }
                   1267: #endif
                   1268: 
1.161   ! pazsan   1269: static int cost_ls(int prim)
1.109     anton    1270: {
                   1271:   struct cost *c = super_costs+prim;
                   1272: 
                   1273:   return c->loads + c->stores;
                   1274: }
                   1275: 
1.161   ! pazsan   1276: static int cost_lsu(int prim)
1.109     anton    1277: {
                   1278:   struct cost *c = super_costs+prim;
                   1279: 
                   1280:   return c->loads + c->stores + c->updates;
                   1281: }
                   1282: 
1.161   ! pazsan   1283: static int cost_nexts(int prim)
1.109     anton    1284: {
                   1285:   return 1;
                   1286: }
                   1287: 
                   1288: typedef int Costfunc(int);
                   1289: Costfunc *ss_cost =  /* cost function for optimize_bb */
                   1290: #ifdef NO_DYNAMIC
                   1291: cost_lsu;
                   1292: #else
                   1293: cost_codesize;
                   1294: #endif
                   1295: 
1.110     anton    1296: struct {
                   1297:   Costfunc *costfunc;
                   1298:   char *metricname;
                   1299:   long sum;
                   1300: } cost_sums[] = {
                   1301: #ifndef NO_DYNAMIC
                   1302:   { cost_codesize, "codesize", 0 },
                   1303: #endif
                   1304:   { cost_ls,       "ls",       0 },
                   1305:   { cost_lsu,      "lsu",      0 },
                   1306:   { cost_nexts,    "nexts",    0 }
                   1307: };
                   1308: 
1.148     anton    1309: #ifndef NO_DYNAMIC
                   1310: void init_ss_cost(void) {
                   1311:   if (no_dynamic && ss_cost == cost_codesize) {
                   1312:     ss_cost = cost_nexts;
                   1313:     cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
                   1314:     debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
                   1315:   }
                   1316: }
                   1317: #endif
                   1318: 
1.106     anton    1319: #define MAX_BB 128 /* maximum number of instructions in BB */
1.125     anton    1320: #define INF_COST 1000000 /* infinite cost */
                   1321: #define CANONICAL_STATE 0
                   1322: 
                   1323: struct waypoint {
                   1324:   int cost;     /* the cost from here to the end */
                   1325:   PrimNum inst; /* the inst used from here to the next waypoint */
                   1326:   char relocatable; /* the last non-transition was relocatable */
                   1327:   char no_transition; /* don't use the next transition (relocatability)
                   1328:                       * or this transition (does not change state) */
                   1329: };
                   1330: 
1.156     anton    1331: struct tpa_state { /* tree parsing automaton (like) state */
1.155     anton    1332:   /* labeling is back-to-front */
                   1333:   struct waypoint *inst;  /* in front of instruction */
                   1334:   struct waypoint *trans; /* in front of instruction and transition */
                   1335: }; 
                   1336: 
1.156     anton    1337: struct tpa_state *termstate = NULL; /* initialized in loader() */
1.155     anton    1338: 
1.158     anton    1339: /* statistics about tree parsing (lazyburg) stuff */
                   1340: long lb_basic_blocks = 0;
                   1341: long lb_labeler_steps = 0;
                   1342: long lb_labeler_automaton = 0;
                   1343: long lb_labeler_dynprog = 0;
                   1344: long lb_newstate_equiv = 0;
                   1345: long lb_newstate_new = 0;
                   1346: long lb_applicable_base_rules = 0;
                   1347: long lb_applicable_chain_rules = 0;
                   1348: 
1.161   ! pazsan   1349: static void init_waypoints(struct waypoint ws[])
1.125     anton    1350: {
                   1351:   int k;
                   1352: 
                   1353:   for (k=0; k<maxstates; k++)
                   1354:     ws[k].cost=INF_COST;
                   1355: }
1.106     anton    1356: 
1.161   ! pazsan   1357: static struct tpa_state *empty_tpa_state()
1.155     anton    1358: {
1.156     anton    1359:   struct tpa_state *s = malloc(sizeof(struct tpa_state));
1.155     anton    1360: 
1.157     anton    1361:   s->inst  = calloc(maxstates,sizeof(struct waypoint));
1.155     anton    1362:   init_waypoints(s->inst);
1.157     anton    1363:   s->trans = calloc(maxstates,sizeof(struct waypoint));
1.155     anton    1364:   /* init_waypoints(s->trans);*/
                   1365:   return s;
                   1366: }
                   1367: 
1.161   ! pazsan   1368: static void transitions(struct tpa_state *t)
1.107     anton    1369: {
1.125     anton    1370:   int k;
                   1371:   struct super_state *l;
                   1372:   
                   1373:   for (k=0; k<maxstates; k++) {
1.155     anton    1374:     t->trans[k] = t->inst[k];
                   1375:     t->trans[k].no_transition = 1;
1.125     anton    1376:   }
                   1377:   for (l = state_transitions; l != NULL; l = l->next) {
                   1378:     PrimNum s = l->super;
                   1379:     int jcost;
                   1380:     struct cost *c=super_costs+s;
1.155     anton    1381:     struct waypoint *wi=&(t->trans[c->state_in]);
                   1382:     struct waypoint *wo=&(t->inst[c->state_out]);
1.158     anton    1383:     lb_applicable_chain_rules++;
1.125     anton    1384:     if (wo->cost == INF_COST)
                   1385:       continue;
                   1386:     jcost = wo->cost + ss_cost(s);
                   1387:     if (jcost <= wi->cost) {
                   1388:       wi->cost = jcost;
                   1389:       wi->inst = s;
                   1390:       wi->relocatable = wo->relocatable;
                   1391:       wi->no_transition = 0;
                   1392:       /* if (ss_greedy) wi->cost = wo->cost ? */
                   1393:     }
                   1394:   }
                   1395: }
1.107     anton    1396: 
1.161   ! pazsan   1397: static struct tpa_state *make_termstate()
1.155     anton    1398: {
1.157     anton    1399:   struct tpa_state *s = empty_tpa_state();
1.155     anton    1400: 
                   1401:   s->inst[CANONICAL_STATE].cost = 0;
                   1402:   transitions(s);
                   1403:   return s;
                   1404: }
                   1405: 
1.156     anton    1406: #define TPA_SIZE 16384
                   1407: 
                   1408: struct tpa_entry {
                   1409:   struct tpa_entry *next;
                   1410:   PrimNum inst;
                   1411:   struct tpa_state *state_behind;  /* note: brack-to-front labeling */
                   1412:   struct tpa_state *state_infront; /* note: brack-to-front labeling */
                   1413: } *tpa_table[TPA_SIZE];
                   1414: 
1.161   ! pazsan   1415: static Cell hash_tpa(PrimNum p, struct tpa_state *t)
1.156     anton    1416: {
                   1417:   UCell it = (UCell )t;
                   1418:   return (p+it+(it>>14))&(TPA_SIZE-1);
                   1419: }
                   1420: 
1.161   ! pazsan   1421: static struct tpa_state **lookup_tpa(PrimNum p, struct tpa_state *t2)
1.156     anton    1422: {
                   1423:   int hash=hash_tpa(p, t2);
                   1424:   struct tpa_entry *te = tpa_table[hash];
                   1425: 
1.158     anton    1426:   if (tpa_noautomaton) {
                   1427:     static struct tpa_state *t;
                   1428:     t = NULL;
                   1429:     return &t;
                   1430:   }
1.156     anton    1431:   for (; te!=NULL; te = te->next) {
                   1432:     if (p == te->inst && t2 == te->state_behind)
                   1433:       return &(te->state_infront);
                   1434:   }
                   1435:   te = (struct tpa_entry *)malloc(sizeof(struct tpa_entry));
                   1436:   te->next = tpa_table[hash];
                   1437:   te->inst = p;
                   1438:   te->state_behind = t2;
                   1439:   te->state_infront = NULL;
                   1440:   tpa_table[hash] = te;
                   1441:   return &(te->state_infront);
                   1442: }
                   1443: 
1.161   ! pazsan   1444: static void tpa_state_normalize(struct tpa_state *t)
1.157     anton    1445: {
                   1446:   /* normalize so cost of canonical state=0; this may result in
                   1447:      negative states for some states */
                   1448:   int d = t->inst[CANONICAL_STATE].cost;
                   1449:   int i;
                   1450: 
                   1451:   for (i=0; i<maxstates; i++) {
                   1452:     if (t->inst[i].cost != INF_COST)
                   1453:       t->inst[i].cost -= d;
                   1454:     if (t->trans[i].cost != INF_COST)
                   1455:       t->trans[i].cost -= d;
                   1456:   }
                   1457: }
                   1458: 
1.161   ! pazsan   1459: static int tpa_state_equivalent(struct tpa_state *t1, struct tpa_state *t2)
1.157     anton    1460: {
                   1461:   return (memcmp(t1->inst, t2->inst, maxstates*sizeof(struct waypoint)) == 0 &&
                   1462:          memcmp(t1->trans,t2->trans,maxstates*sizeof(struct waypoint)) == 0);
                   1463: }
                   1464: 
                   1465: struct tpa_state_entry {
                   1466:   struct tpa_state_entry *next;
                   1467:   struct tpa_state *state;
                   1468: } *tpa_state_table[TPA_SIZE];
                   1469: 
1.161   ! pazsan   1470: static Cell hash_tpa_state(struct tpa_state *t)
1.157     anton    1471: {
                   1472:   int *ti = (int *)(t->inst);
                   1473:   int *tt = (int *)(t->trans);
                   1474:   int r=0;
                   1475:   int i;
                   1476: 
                   1477:   for (i=0; ti+i < (int *)(t->inst+maxstates); i++)
                   1478:     r += ti[i]+tt[i];
                   1479:   return (r+(r>>14)+(r>>22)) & (TPA_SIZE-1);
                   1480: }
                   1481: 
1.161   ! pazsan   1482: static struct tpa_state *lookup_tpa_state(struct tpa_state *t)
1.157     anton    1483: {
                   1484:   Cell hash = hash_tpa_state(t);
                   1485:   struct tpa_state_entry *te = tpa_state_table[hash];
                   1486:   struct tpa_state_entry *tn;
                   1487: 
1.158     anton    1488:   if (!tpa_noequiv) {
                   1489:     for (; te!=NULL; te = te->next) {
                   1490:       if (tpa_state_equivalent(t, te->state)) {
                   1491:        lb_newstate_equiv++;
                   1492:        free(t->inst);
                   1493:        free(t->trans);
                   1494:        free(t);
                   1495:        return te->state;
                   1496:       }
1.157     anton    1497:     }
1.158     anton    1498:     tn = (struct tpa_state_entry *)malloc(sizeof(struct tpa_state_entry));
                   1499:     tn->next = te;
                   1500:     tn->state = t;
                   1501:     tpa_state_table[hash] = tn;
                   1502:   }
                   1503:   lb_newstate_new++;
                   1504:   if (tpa_trace)
                   1505:     fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
1.157     anton    1506:   return t;
                   1507: }
                   1508: 
1.125     anton    1509: /* use dynamic programming to find the shortest paths within the basic
                   1510:    block origs[0..ninsts-1] and rewrite the instructions pointed to by
                   1511:    instps to use it */
1.161   ! pazsan   1512: static void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1.125     anton    1513: {
                   1514:   int i,j;
1.156     anton    1515:   struct tpa_state *ts[ninsts+1];
1.125     anton    1516:   int nextdyn, nextstate, no_transition;
                   1517:   
1.158     anton    1518:   lb_basic_blocks++;
1.155     anton    1519:   ts[ninsts] = termstate;
1.107     anton    1520:   for (i=ninsts-1; i>=0; i--) {
1.156     anton    1521:     struct tpa_state **tp = lookup_tpa(origs[i],ts[i+1]);
                   1522:     struct tpa_state *t = *tp;
1.158     anton    1523:     lb_labeler_steps++;
                   1524:     if (t) {
1.156     anton    1525:       ts[i] = t;
1.158     anton    1526:       lb_labeler_automaton++;
                   1527:     }
1.156     anton    1528:     else {
1.158     anton    1529:       lb_labeler_dynprog++;
1.156     anton    1530:       ts[i] = empty_tpa_state();
                   1531:       for (j=1; j<=max_super && i+j<=ninsts; j++) {
                   1532:        struct super_state **superp = lookup_super(origs+i, j);
                   1533:        if (superp!=NULL) {
                   1534:          struct super_state *supers = *superp;
                   1535:          for (; supers!=NULL; supers = supers->next) {
                   1536:            PrimNum s = supers->super;
                   1537:            int jcost;
                   1538:            struct cost *c=super_costs+s;
                   1539:            struct waypoint *wi=&(ts[i]->inst[c->state_in]);
                   1540:            struct waypoint *wo=&(ts[i+j]->trans[c->state_out]);
                   1541:            int no_transition = wo->no_transition;
1.158     anton    1542:            lb_applicable_base_rules++;
1.156     anton    1543:            if (!(is_relocatable(s)) && !wo->relocatable) {
                   1544:              wo=&(ts[i+j]->inst[c->state_out]);
                   1545:              no_transition=1;
                   1546:            }
                   1547:            if (wo->cost == INF_COST) 
                   1548:              continue;
                   1549:            jcost = wo->cost + ss_cost(s);
                   1550:            if (jcost <= wi->cost) {
                   1551:              wi->cost = jcost;
                   1552:              wi->inst = s;
                   1553:              wi->relocatable = is_relocatable(s);
                   1554:              wi->no_transition = no_transition;
                   1555:              /* if (ss_greedy) wi->cost = wo->cost ? */
                   1556:            }
1.125     anton    1557:          }
1.107     anton    1558:        }
                   1559:       }
1.156     anton    1560:       transitions(ts[i]);
1.157     anton    1561:       tpa_state_normalize(ts[i]);
                   1562:       *tp = ts[i] = lookup_tpa_state(ts[i]);
1.158     anton    1563:       if (tpa_trace)
                   1564:        fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1.107     anton    1565:     }
1.125     anton    1566:   }
                   1567:   /* now rewrite the instructions */
                   1568:   nextdyn=0;
                   1569:   nextstate=CANONICAL_STATE;
1.155     anton    1570:   no_transition = ((!ts[0]->trans[nextstate].relocatable) 
                   1571:                   ||ts[0]->trans[nextstate].no_transition);
1.125     anton    1572:   for (i=0; i<ninsts; i++) {
                   1573:     Cell tc=0, tc2;
                   1574:     if (i==nextdyn) {
                   1575:       if (!no_transition) {
                   1576:        /* process trans */
1.155     anton    1577:        PrimNum p = ts[i]->trans[nextstate].inst;
1.125     anton    1578:        struct cost *c = super_costs+p;
1.155     anton    1579:        assert(ts[i]->trans[nextstate].cost != INF_COST);
1.125     anton    1580:        assert(c->state_in==nextstate);
1.128     anton    1581:        tc = compile_prim_dyn(p,NULL);
1.125     anton    1582:        nextstate = c->state_out;
                   1583:       }
                   1584:       {
                   1585:        /* process inst */
1.155     anton    1586:        PrimNum p = ts[i]->inst[nextstate].inst;
1.125     anton    1587:        struct cost *c=super_costs+p;
                   1588:        assert(c->state_in==nextstate);
1.155     anton    1589:        assert(ts[i]->inst[nextstate].cost != INF_COST);
1.125     anton    1590: #if defined(GFORTH_DEBUGGING)
                   1591:        assert(p == origs[i]);
                   1592: #endif
1.128     anton    1593:        tc2 = compile_prim_dyn(p,instps[i]);
1.125     anton    1594:        if (no_transition || !is_relocatable(p))
                   1595:          /* !! actually what we care about is if and where
                   1596:           * compile_prim_dyn() puts NEXTs */
                   1597:          tc=tc2;
1.155     anton    1598:        no_transition = ts[i]->inst[nextstate].no_transition;
1.125     anton    1599:        nextstate = c->state_out;
                   1600:        nextdyn += c->length;
                   1601:       }
                   1602:     } else {
                   1603: #if defined(GFORTH_DEBUGGING)
                   1604:       assert(0);
                   1605: #endif
                   1606:       tc=0;
1.155     anton    1607:       /* tc= (Cell)vm_prims[ts[i]->inst[CANONICAL_STATE].inst]; */
1.125     anton    1608:     }
                   1609:     *(instps[i]) = tc;
                   1610:   }      
                   1611:   if (!no_transition) {
1.155     anton    1612:     PrimNum p = ts[i]->trans[nextstate].inst;
1.125     anton    1613:     struct cost *c = super_costs+p;
                   1614:     assert(c->state_in==nextstate);
1.155     anton    1615:     assert(ts[i]->trans[nextstate].cost != INF_COST);
1.125     anton    1616:     assert(i==nextdyn);
1.128     anton    1617:     (void)compile_prim_dyn(p,NULL);
1.125     anton    1618:     nextstate = c->state_out;
1.107     anton    1619:   }
1.125     anton    1620:   assert(nextstate==CANONICAL_STATE);
1.107     anton    1621: }
                   1622: 
1.105     anton    1623: /* compile *start, possibly rewriting it into a static and/or dynamic
                   1624:    superinstruction */
                   1625: void compile_prim1(Cell *start)
1.70      anton    1626: {
1.108     anton    1627: #if defined(DOUBLY_INDIRECT)
1.125     anton    1628:   Label prim;
                   1629: 
                   1630:   if (start==NULL)
                   1631:     return;
                   1632:   prim = (Label)*start;
1.108     anton    1633:   if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
                   1634:     fprintf(stderr,"compile_prim encountered xt %p\n", prim);
                   1635:     *start=(Cell)prim;
                   1636:     return;
                   1637:   } else {
                   1638:     *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
                   1639:     return;
                   1640:   }
                   1641: #elif defined(INDIRECT_THREADED)
                   1642:   return;
1.112     anton    1643: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1.128     anton    1644:   /* !! does not work, for unknown reasons; but something like this is
                   1645:      probably needed to ensure that we don't call compile_prim_dyn
                   1646:      before the inline arguments are there */
                   1647:   static Cell *instps[MAX_BB];
                   1648:   static PrimNum origs[MAX_BB];
                   1649:   static int ninsts=0;
                   1650:   PrimNum prim_num;
                   1651: 
                   1652:   if (start==NULL || ninsts >= MAX_BB ||
                   1653:       (ninsts>0 && superend[origs[ninsts-1]])) {
                   1654:     /* after bb, or at the start of the next bb */
                   1655:     optimize_rewrite(instps,origs,ninsts);
                   1656:     /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
                   1657:     ninsts=0;
                   1658:     if (start==NULL)
                   1659:       return;
                   1660:   }
                   1661:   prim_num = ((Xt)*start)-vm_prims;
                   1662:   if(prim_num >= npriminfos) {
                   1663:     optimize_rewrite(instps,origs,ninsts);
1.129     anton    1664:     /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1.128     anton    1665:     ninsts=0;
                   1666:     return;
                   1667:   }    
                   1668:   assert(ninsts<MAX_BB);
                   1669:   instps[ninsts] = start;
                   1670:   origs[ninsts] = prim_num;
                   1671:   ninsts++;
1.112     anton    1672: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1.47      anton    1673: }
                   1674: 
1.161   ! pazsan   1675: Address gforth_loader(FILE *imagefile, char* filename)
1.1       anton    1676: /* returns the address of the image proper (after the preamble) */
                   1677: {
                   1678:   ImageHeader header;
                   1679:   Address image;
                   1680:   Address imp; /* image+preamble */
1.17      anton    1681:   Char magic[8];
                   1682:   char magic7; /* size byte of magic number */
1.1       anton    1683:   Cell preamblesize=0;
1.6       pazsan   1684:   Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1.1       anton    1685:   UCell check_sum;
1.15      pazsan   1686:   Cell ausize = ((RELINFOBITS ==  8) ? 0 :
                   1687:                 (RELINFOBITS == 16) ? 1 :
                   1688:                 (RELINFOBITS == 32) ? 2 : 3);
                   1689:   Cell charsize = ((sizeof(Char) == 1) ? 0 :
                   1690:                   (sizeof(Char) == 2) ? 1 :
                   1691:                   (sizeof(Char) == 4) ? 2 : 3) + ausize;
                   1692:   Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
                   1693:                   (sizeof(Cell) == 2) ? 1 :
                   1694:                   (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1.21      anton    1695:   Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
                   1696: #ifdef WORDS_BIGENDIAN
                   1697:        0
                   1698: #else
                   1699:        1
                   1700: #endif
                   1701:     ;
1.1       anton    1702: 
1.43      anton    1703:   vm_prims = engine(0,0,0,0,0);
1.47      anton    1704:   check_prims(vm_prims);
1.106     anton    1705:   prepare_super_table();
1.1       anton    1706: #ifndef DOUBLY_INDIRECT
1.59      anton    1707: #ifdef PRINT_SUPER_LENGTHS
                   1708:   print_super_lengths();
                   1709: #endif
1.43      anton    1710:   check_sum = checksum(vm_prims);
1.1       anton    1711: #else /* defined(DOUBLY_INDIRECT) */
1.43      anton    1712:   check_sum = (UCell)vm_prims;
1.1       anton    1713: #endif /* defined(DOUBLY_INDIRECT) */
1.155     anton    1714: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
                   1715:   termstate = make_termstate();
                   1716: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1.10      pazsan   1717:   
                   1718:   do {
                   1719:     if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1.84      anton    1720:       fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1.10      pazsan   1721:              progname, filename);
                   1722:       exit(1);
1.1       anton    1723:     }
1.10      pazsan   1724:     preamblesize+=8;
1.84      anton    1725:   } while(memcmp(magic,"Gforth3",7));
1.17      anton    1726:   magic7 = magic[7];
1.1       anton    1727:   if (debug) {
1.17      anton    1728:     magic[7]='\0';
1.21      anton    1729:     fprintf(stderr,"Magic found: %s ", magic);
                   1730:     print_sizes(magic7);
1.1       anton    1731:   }
                   1732: 
1.21      anton    1733:   if (magic7 != sizebyte)
                   1734:     {
                   1735:       fprintf(stderr,"This image is:         ");
                   1736:       print_sizes(magic7);
                   1737:       fprintf(stderr,"whereas the machine is ");
                   1738:       print_sizes(sizebyte);
1.1       anton    1739:       exit(-2);
                   1740:     };
                   1741: 
                   1742:   fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1.10      pazsan   1743: 
                   1744:   set_stack_sizes(&header);
1.1       anton    1745:   
                   1746: #if HAVE_GETPAGESIZE
                   1747:   pagesize=getpagesize(); /* Linux/GNU libc offers this */
                   1748: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
                   1749:   pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
                   1750: #elif PAGESIZE
                   1751:   pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
                   1752: #endif
1.144     pazsan   1753:   debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1.1       anton    1754: 
1.34      anton    1755:   image = dict_alloc_read(imagefile, preamblesize+header.image_size,
                   1756:                          preamblesize+dictsize, data_offset);
1.33      anton    1757:   imp=image+preamblesize;
1.57      anton    1758:   alloc_stacks((ImageHeader *)imp);
1.1       anton    1759:   if (clear_dictionary)
1.33      anton    1760:     memset(imp+header.image_size, 0, dictsize-header.image_size);
1.90      anton    1761:   if(header.base==0 || header.base  == (Address)0x100) {
1.1       anton    1762:     Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
                   1763:     char reloc_bits[reloc_size];
1.33      anton    1764:     fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1.10      pazsan   1765:     fread(reloc_bits, 1, reloc_size, imagefile);
1.161   ! pazsan   1766:     gforth_relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1.1       anton    1767: #if 0
                   1768:     { /* let's see what the relocator did */
                   1769:       FILE *snapshot=fopen("snapshot.fi","wb");
                   1770:       fwrite(image,1,imagesize,snapshot);
                   1771:       fclose(snapshot);
                   1772:     }
                   1773: #endif
1.46      jwilke   1774:   }
                   1775:   else if(header.base!=imp) {
                   1776:     fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
                   1777:            progname, (unsigned long)header.base, (unsigned long)imp);
                   1778:     exit(1);
1.1       anton    1779:   }
                   1780:   if (header.checksum==0)
                   1781:     ((ImageHeader *)imp)->checksum=check_sum;
                   1782:   else if (header.checksum != check_sum) {
                   1783:     fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
                   1784:            progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
                   1785:     exit(1);
                   1786:   }
1.53      anton    1787: #ifdef DOUBLY_INDIRECT
                   1788:   ((ImageHeader *)imp)->xt_base = xts;
                   1789: #endif
1.1       anton    1790:   fclose(imagefile);
                   1791: 
1.56      anton    1792:   /* unnecessary, except maybe for CODE words */
                   1793:   /* FLUSH_ICACHE(imp, header.image_size);*/
1.1       anton    1794: 
                   1795:   return imp;
                   1796: }
                   1797: 
1.72      anton    1798: /* pointer to last '/' or '\' in file, 0 if there is none. */
1.161   ! pazsan   1799: static char *onlypath(char *filename)
1.10      pazsan   1800: {
1.72      anton    1801:   return strrchr(filename, DIRSEP);
1.1       anton    1802: }
                   1803: 
1.161   ! pazsan   1804: static FILE *openimage(char *fullfilename)
1.10      pazsan   1805: {
                   1806:   FILE *image_file;
1.28      anton    1807:   char * expfilename = tilde_cstr(fullfilename, strlen(fullfilename), 1);
1.10      pazsan   1808: 
1.28      anton    1809:   image_file=fopen(expfilename,"rb");
1.1       anton    1810:   if (image_file!=NULL && debug)
1.28      anton    1811:     fprintf(stderr, "Opened image file: %s\n", expfilename);
1.10      pazsan   1812:   return image_file;
1.1       anton    1813: }
                   1814: 
1.28      anton    1815: /* try to open image file concat(path[0:len],imagename) */
1.161   ! pazsan   1816: static FILE *checkimage(char *path, int len, char *imagename)
1.10      pazsan   1817: {
                   1818:   int dirlen=len;
1.1       anton    1819:   char fullfilename[dirlen+strlen(imagename)+2];
1.10      pazsan   1820: 
1.1       anton    1821:   memcpy(fullfilename, path, dirlen);
1.71      pazsan   1822:   if (fullfilename[dirlen-1]!=DIRSEP)
                   1823:     fullfilename[dirlen++]=DIRSEP;
1.1       anton    1824:   strcpy(fullfilename+dirlen,imagename);
1.10      pazsan   1825:   return openimage(fullfilename);
1.1       anton    1826: }
                   1827: 
1.161   ! pazsan   1828: static FILE * open_image_file(char * imagename, char * path)
1.1       anton    1829: {
1.10      pazsan   1830:   FILE * image_file=NULL;
1.28      anton    1831:   char *origpath=path;
1.10      pazsan   1832:   
1.71      pazsan   1833:   if(strchr(imagename, DIRSEP)==NULL) {
1.10      pazsan   1834:     /* first check the directory where the exe file is in !! 01may97jaw */
                   1835:     if (onlypath(progname))
1.72      anton    1836:       image_file=checkimage(progname, onlypath(progname)-progname, imagename);
1.10      pazsan   1837:     if (!image_file)
                   1838:       do {
                   1839:        char *pend=strchr(path, PATHSEP);
                   1840:        if (pend==NULL)
                   1841:          pend=path+strlen(path);
                   1842:        if (strlen(path)==0) break;
                   1843:        image_file=checkimage(path, pend-path, imagename);
                   1844:        path=pend+(*pend==PATHSEP);
                   1845:       } while (image_file==NULL);
                   1846:   } else {
                   1847:     image_file=openimage(imagename);
                   1848:   }
1.1       anton    1849: 
1.10      pazsan   1850:   if (!image_file) {
                   1851:     fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
1.28      anton    1852:            progname, imagename, origpath);
1.10      pazsan   1853:     exit(1);
1.7       anton    1854:   }
                   1855: 
1.10      pazsan   1856:   return image_file;
                   1857: }
1.11      pazsan   1858: #endif
                   1859: 
                   1860: #ifdef HAS_OS
1.161   ! pazsan   1861: static UCell convsize(char *s, UCell elemsize)
1.11      pazsan   1862: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
                   1863:    of bytes.  the letter at the end indicates the unit, where e stands
                   1864:    for the element size. default is e */
                   1865: {
                   1866:   char *endp;
                   1867:   UCell n,m;
                   1868: 
                   1869:   m = elemsize;
                   1870:   n = strtoul(s,&endp,0);
                   1871:   if (endp!=NULL) {
                   1872:     if (strcmp(endp,"b")==0)
                   1873:       m=1;
                   1874:     else if (strcmp(endp,"k")==0)
                   1875:       m=1024;
                   1876:     else if (strcmp(endp,"M")==0)
                   1877:       m=1024*1024;
                   1878:     else if (strcmp(endp,"G")==0)
                   1879:       m=1024*1024*1024;
                   1880:     else if (strcmp(endp,"T")==0) {
                   1881: #if (SIZEOF_CHAR_P > 4)
1.24      anton    1882:       m=1024L*1024*1024*1024;
1.11      pazsan   1883: #else
                   1884:       fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
                   1885:       exit(1);
                   1886: #endif
                   1887:     } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
                   1888:       fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
                   1889:       exit(1);
                   1890:     }
                   1891:   }
                   1892:   return n*m;
                   1893: }
1.10      pazsan   1894: 
1.109     anton    1895: enum {
                   1896:   ss_number = 256,
1.125     anton    1897:   ss_states,
1.109     anton    1898:   ss_min_codesize,
                   1899:   ss_min_ls,
                   1900:   ss_min_lsu,
                   1901:   ss_min_nexts,
                   1902: };
                   1903: 
1.10      pazsan   1904: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
                   1905: {
                   1906:   int c;
                   1907: 
1.1       anton    1908:   opterr=0;
                   1909:   while (1) {
                   1910:     int option_index=0;
                   1911:     static struct option opts[] = {
1.29      anton    1912:       {"appl-image", required_argument, NULL, 'a'},
1.1       anton    1913:       {"image-file", required_argument, NULL, 'i'},
                   1914:       {"dictionary-size", required_argument, NULL, 'm'},
                   1915:       {"data-stack-size", required_argument, NULL, 'd'},
                   1916:       {"return-stack-size", required_argument, NULL, 'r'},
                   1917:       {"fp-stack-size", required_argument, NULL, 'f'},
                   1918:       {"locals-stack-size", required_argument, NULL, 'l'},
                   1919:       {"path", required_argument, NULL, 'p'},
                   1920:       {"version", no_argument, NULL, 'v'},
                   1921:       {"help", no_argument, NULL, 'h'},
                   1922:       /* put something != 0 into offset_image */
                   1923:       {"offset-image", no_argument, &offset_image, 1},
                   1924:       {"no-offset-im", no_argument, &offset_image, 0},
                   1925:       {"clear-dictionary", no_argument, &clear_dictionary, 1},
1.4       anton    1926:       {"die-on-signal", no_argument, &die_on_signal, 1},
1.1       anton    1927:       {"debug", no_argument, &debug, 1},
1.144     pazsan   1928:       {"diag", no_argument, &diag, 1},
1.60      anton    1929:       {"no-super", no_argument, &no_super, 1},
                   1930:       {"no-dynamic", no_argument, &no_dynamic, 1},
1.66      anton    1931:       {"dynamic", no_argument, &no_dynamic, 0},
1.110     anton    1932:       {"print-metrics", no_argument, &print_metrics, 1},
1.109     anton    1933:       {"ss-number", required_argument, NULL, ss_number},
1.125     anton    1934:       {"ss-states", required_argument, NULL, ss_states},
1.109     anton    1935: #ifndef NO_DYNAMIC
                   1936:       {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
                   1937: #endif
                   1938:       {"ss-min-ls",       no_argument, NULL, ss_min_ls},
                   1939:       {"ss-min-lsu",      no_argument, NULL, ss_min_lsu},
                   1940:       {"ss-min-nexts",    no_argument, NULL, ss_min_nexts},
1.110     anton    1941:       {"ss-greedy",       no_argument, &ss_greedy, 1},
1.158     anton    1942:       {"tpa-noequiv",     no_argument, &tpa_noequiv, 1},
                   1943:       {"tpa-noautomaton", no_argument, &tpa_noautomaton, 1},
                   1944:       {"tpa-trace",      no_argument, &tpa_trace, 1},
1.1       anton    1945:       {0,0,0,0}
                   1946:       /* no-init-file, no-rc? */
                   1947:     };
                   1948:     
1.36      pazsan   1949:     c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
1.1       anton    1950:     
                   1951:     switch (c) {
1.29      anton    1952:     case EOF: return;
                   1953:     case '?': optind--; return;
                   1954:     case 'a': *imagename = optarg; return;
1.10      pazsan   1955:     case 'i': *imagename = optarg; break;
1.1       anton    1956:     case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
                   1957:     case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
                   1958:     case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
                   1959:     case 'f': fsize = convsize(optarg,sizeof(Float)); break;
                   1960:     case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
1.10      pazsan   1961:     case 'p': *path = optarg; break;
1.36      pazsan   1962:     case 'o': offset_image = 1; break;
                   1963:     case 'n': offset_image = 0; break;
                   1964:     case 'c': clear_dictionary = 1; break;
                   1965:     case 's': die_on_signal = 1; break;
                   1966:     case 'x': debug = 1; break;
1.83      anton    1967:     case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
1.109     anton    1968:     case ss_number: static_super_number = atoi(optarg); break;
1.125     anton    1969:     case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
1.109     anton    1970: #ifndef NO_DYNAMIC
                   1971:     case ss_min_codesize: ss_cost = cost_codesize; break;
                   1972: #endif
                   1973:     case ss_min_ls:       ss_cost = cost_ls;       break;
                   1974:     case ss_min_lsu:      ss_cost = cost_lsu;      break;
                   1975:     case ss_min_nexts:    ss_cost = cost_nexts;    break;
1.1       anton    1976:     case 'h': 
1.29      anton    1977:       fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
1.1       anton    1978: Engine Options:\n\
1.29      anton    1979:   --appl-image FILE                equivalent to '--image-file=FILE --'\n\
1.10      pazsan   1980:   --clear-dictionary               Initialize the dictionary with 0 bytes\n\
                   1981:   -d SIZE, --data-stack-size=SIZE   Specify data stack size\n\
                   1982:   --debug                          Print debugging information during startup\n\
1.144     pazsan   1983:   --diag                           Print diagnostic information during startup\n\
1.10      pazsan   1984:   --die-on-signal                  exit instead of CATCHing some signals\n\
1.66      anton    1985:   --dynamic                        use dynamic native code\n\
1.10      pazsan   1986:   -f SIZE, --fp-stack-size=SIZE            Specify floating point stack size\n\
                   1987:   -h, --help                       Print this message and exit\n\
                   1988:   -i FILE, --image-file=FILE       Use image FILE instead of `gforth.fi'\n\
                   1989:   -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
                   1990:   -m SIZE, --dictionary-size=SIZE   Specify Forth dictionary size\n\
1.60      anton    1991:   --no-dynamic                     Use only statically compiled primitives\n\
1.10      pazsan   1992:   --no-offset-im                   Load image at normal position\n\
1.60      anton    1993:   --no-super                        No dynamically formed superinstructions\n\
1.10      pazsan   1994:   --offset-image                   Load image at a different position\n\
                   1995:   -p PATH, --path=PATH             Search path for finding image and sources\n\
1.110     anton    1996:   --print-metrics                  Print some code generation metrics on exit\n\
1.10      pazsan   1997:   -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
1.111     anton    1998:   --ss-greedy                       greedy, not optimal superinst selection\n\
                   1999:   --ss-min-codesize                 select superinsts for smallest native code\n\
                   2000:   --ss-min-ls                       minimize loads and stores\n\
                   2001:   --ss-min-lsu                      minimize loads, stores, and pointer updates\n\
                   2002:   --ss-min-nexts                    minimize the number of static superinsts\n\
                   2003:   --ss-number=N                     use N static superinsts (default max)\n\
1.125     anton    2004:   --ss-states=N                     N states for stack caching (default max)\n\
1.158     anton    2005:   --tpa-noequiv                     automaton without state equivalence\n\
                   2006:   --tpa-noautomaton                 dynamic programming only\n\
                   2007:   --tpa-trace                       report new states etc.\n\
1.66      anton    2008:   -v, --version                            Print engine version and exit\n\
1.1       anton    2009: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
1.10      pazsan   2010:   `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
                   2011:              argv[0]);
                   2012:       optind--;
                   2013:       return;
1.1       anton    2014:     }
                   2015:   }
1.10      pazsan   2016: }
1.11      pazsan   2017: #endif
1.10      pazsan   2018: 
1.161   ! pazsan   2019: static void print_diag()
1.144     pazsan   2020: {
                   2021: 
1.153     pazsan   2022: #if !defined(HAVE_GETRUSAGE) || (!defined(HAS_FFCALL) && !defined(HAS_LIBFFI))
1.145     pazsan   2023:   fprintf(stderr, "*** missing functionality ***\n"
1.144     pazsan   2024: #ifndef HAVE_GETRUSAGE
                   2025:          "    no getrusage -> CPUTIME broken\n"
                   2026: #endif
1.153     pazsan   2027: #if !defined(HAS_FFCALL) && !defined(HAS_LIBFFI)
1.144     pazsan   2028:          "    no ffcall -> only old-style foreign function calls (no fflib.fs)\n"
                   2029: #endif
                   2030:          );
                   2031: #endif
                   2032:   if((relocs < nonrelocs) ||
                   2033: #if defined(BUGGY_LL_CMP) || defined(BUGGY_LL_MUL) || defined(BUGGY_LL_DIV) || defined(BUGGY_LL_ADD) || defined(BUGGY_LL_SHIFT) || defined(BUGGY_LL_D2F) || defined(BUGGY_LL_F2D)
                   2034:      1
                   2035: #else
                   2036:      0
                   2037: #endif
                   2038:      )
                   2039:     debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
1.145     pazsan   2040:     fprintf(stderr, "*** performance problems ***\n%s"
1.144     pazsan   2041: #if defined(BUGGY_LL_CMP) || defined(BUGGY_LL_MUL) || defined(BUGGY_LL_DIV) || defined(BUGGY_LL_ADD) || defined(BUGGY_LL_SHIFT) || defined(BUGGY_LL_D2F) || defined(BUGGY_LL_F2D)
                   2042:            "    double-cell integer type buggy ->\n        "
                   2043: #ifdef BUGGY_LL_CMP
                   2044:            "CMP, "
                   2045: #endif
                   2046: #ifdef BUGGY_LL_MUL
                   2047:            "MUL, "
                   2048: #endif
                   2049: #ifdef BUGGY_LL_DIV
                   2050:            "DIV, "
                   2051: #endif
                   2052: #ifdef BUGGY_LL_ADD
                   2053:            "ADD, "
                   2054: #endif
                   2055: #ifdef BUGGY_LL_SHIFT
                   2056:            "SHIFT, "
                   2057: #endif
                   2058: #ifdef BUGGY_LL_D2F
                   2059:            "D2F, "
                   2060: #endif
                   2061: #ifdef BUGGY_LL_F2D
                   2062:            "F2D, "
                   2063: #endif
                   2064:            "\b\b slow\n"
1.145     pazsan   2065: #endif
                   2066: #ifndef FORCE_REG
                   2067:            "    automatic register allocation: performance degradation possible\n"
                   2068: #endif
                   2069: #if !defined(FORCE_REG) || defined(BUGGY_LONG_LONG)
                   2070:            "*** Suggested remedy: try ./configure"
                   2071: #ifndef FORCE_REG
                   2072:            " --enable-force-reg"
                   2073: #endif
                   2074: #ifdef BUGGY_LONG_LONG
                   2075:            " --enable-force-ll"
                   2076: #endif
                   2077:            "\n"
1.144     pazsan   2078: #endif
                   2079:            ,
                   2080:            (relocs < nonrelocs) ? "    gcc PR 15242 -> no dynamic code generation (use gcc-2.95 instead)\n" : "");
                   2081: }
                   2082: 
1.10      pazsan   2083: #ifdef INCLUDE_IMAGE
                   2084: extern Cell image[];
                   2085: extern const char reloc_bits[];
                   2086: #endif
1.67      pazsan   2087: 
1.10      pazsan   2088: int main(int argc, char **argv, char **env)
                   2089: {
1.30      pazsan   2090: #ifdef HAS_OS
1.10      pazsan   2091:   char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
1.30      pazsan   2092: #else
                   2093:   char *path = DEFAULTPATH;
                   2094: #endif
1.13      pazsan   2095: #ifndef INCLUDE_IMAGE
1.10      pazsan   2096:   char *imagename="gforth.fi";
                   2097:   FILE *image_file;
                   2098:   Address image;
                   2099: #endif
                   2100:   int retvalue;
                   2101:          
1.56      anton    2102: #if defined(i386) && defined(ALIGNMENT_CHECK)
1.10      pazsan   2103:   /* turn on alignment checks on the 486.
                   2104:    * on the 386 this should have no effect. */
                   2105:   __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
                   2106:   /* this is unusable with Linux' libc.4.6.27, because this library is
                   2107:      not alignment-clean; we would have to replace some library
                   2108:      functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
                   2109:      the stack FP-aligned. */
                   2110: #endif
                   2111: 
                   2112:   /* buffering of the user output device */
1.11      pazsan   2113: #ifdef _IONBF
1.10      pazsan   2114:   if (isatty(fileno(stdout))) {
                   2115:     fflush(stdout);
                   2116:     setvbuf(stdout,NULL,_IONBF,0);
1.1       anton    2117:   }
1.11      pazsan   2118: #endif
1.1       anton    2119: 
1.10      pazsan   2120:   progname = argv[0];
                   2121: 
1.11      pazsan   2122: #ifdef HAS_OS
1.10      pazsan   2123:   gforth_args(argc, argv, &path, &imagename);
1.109     anton    2124: #ifndef NO_DYNAMIC
1.148     anton    2125:   init_ss_cost();
1.109     anton    2126: #endif /* !defined(NO_DYNAMIC) */
                   2127: #endif /* defined(HAS_OS) */
1.10      pazsan   2128: 
                   2129: #ifdef INCLUDE_IMAGE
                   2130:   set_stack_sizes((ImageHeader *)image);
1.22      pazsan   2131:   if(((ImageHeader *)image)->base != image)
1.161   ! pazsan   2132:     gforth_relocate(image, reloc_bits, ((ImageHeader *)image)->image_size,
        !          2133:                    (Label*)engine(0, 0, 0, 0, 0));
1.10      pazsan   2134:   alloc_stacks((ImageHeader *)image);
                   2135: #else
                   2136:   image_file = open_image_file(imagename, path);
1.161   ! pazsan   2137:   image = gforth_loader(image_file, imagename);
1.10      pazsan   2138: #endif
1.24      anton    2139:   gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
1.1       anton    2140: 
1.144     pazsan   2141:   if (diag)
                   2142:     print_diag();
1.1       anton    2143:   {
1.10      pazsan   2144:     char path2[strlen(path)+1];
1.1       anton    2145:     char *p1, *p2;
                   2146:     Cell environ[]= {
                   2147:       (Cell)argc-(optind-1),
                   2148:       (Cell)(argv+(optind-1)),
1.10      pazsan   2149:       (Cell)strlen(path),
1.1       anton    2150:       (Cell)path2};
                   2151:     argv[optind-1] = progname;
                   2152:     /*
                   2153:        for (i=0; i<environ[0]; i++)
                   2154:        printf("%s\n", ((char **)(environ[1]))[i]);
                   2155:        */
                   2156:     /* make path OS-independent by replacing path separators with NUL */
1.10      pazsan   2157:     for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
1.1       anton    2158:       if (*p1==PATHSEP)
                   2159:        *p2 = '\0';
                   2160:       else
                   2161:        *p2 = *p1;
                   2162:     *p2='\0';
1.161   ! pazsan   2163:     retvalue = gforth_go(image, 4, environ);
1.102     anton    2164: #ifdef SIGPIPE
                   2165:     bsd_signal(SIGPIPE, SIG_IGN);
                   2166: #endif
1.42      anton    2167: #ifdef VM_PROFILING
                   2168:     vm_print_profile(stderr);
                   2169: #endif
1.1       anton    2170:     deprep_terminal();
1.104     anton    2171:   }
1.110     anton    2172:   if (print_metrics) {
                   2173:     int i;
                   2174:     fprintf(stderr, "code size = %8ld\n", dyncodesize());
                   2175:     for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
                   2176:       fprintf(stderr, "metric %8s: %8ld\n",
                   2177:              cost_sums[i].metricname, cost_sums[i].sum);
1.158     anton    2178:     fprintf(stderr,"lb_basic_blocks = %ld\n", lb_basic_blocks);
                   2179:     fprintf(stderr,"lb_labeler_steps = %ld\n", lb_labeler_steps);
                   2180:     fprintf(stderr,"lb_labeler_automaton = %ld\n", lb_labeler_automaton);
                   2181:     fprintf(stderr,"lb_labeler_dynprog = %ld\n", lb_labeler_dynprog);
                   2182:     fprintf(stderr,"lb_newstate_equiv = %ld\n", lb_newstate_equiv);
                   2183:     fprintf(stderr,"lb_newstate_new = %ld\n", lb_newstate_new);
                   2184:     fprintf(stderr,"lb_applicable_base_rules = %ld\n", lb_applicable_base_rules);
                   2185:     fprintf(stderr,"lb_applicable_chain_rules = %ld\n", lb_applicable_chain_rules);
                   2186:   }
                   2187:   if (tpa_trace) {
                   2188:     fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
                   2189:     fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1.1       anton    2190:   }
1.13      pazsan   2191:   return retvalue;
1.1       anton    2192: }

FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>