Annotation of gforth/engine/main.c, revision 1.173

1.1       anton       1: /* command line interpretation, image loading etc. for Gforth
                      2: 
                      3: 
1.170     anton       4:   Copyright (C) 1995,1996,1997,1998,2000,2003,2004,2005,2006 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;
1.168     pazsan     78:   va_alist clist = gforth_clist;
1.140     pazsan     79: 
1.161     pazsan     80:   gforth_clist = (va_alist)alist;
1.140     pazsan     81: 
1.161     pazsan     82:   gforth_engine(fcall, sp, rp, fp, lp);
1.140     pazsan     83: 
                     84:   /* restore global variables */
1.161     pazsan     85:   gforth_RP = rp;
                     86:   gforth_SP = sp;
                     87:   gforth_FP = fp;
                     88:   gforth_LP = lp;
1.168     pazsan     89:   gforth_clist = clist;
1.115     pazsan     90: }
                     91: #endif
                     92: 
1.153     pazsan     93: #ifdef HAS_LIBFFI
1.161     pazsan     94: Cell *gforth_RP;
                     95: Address gforth_LP;
1.153     pazsan     96: 
                     97: #include <ffi.h>
                     98: 
1.164     pazsan     99: void ** gforth_clist;
                    100: void * gforth_ritem;
1.153     pazsan    101: 
1.162     pazsan    102: void gforth_callback(ffi_cif * cif, void * resp, void ** args, void * ip)
1.153     pazsan    103: {
1.161     pazsan    104:   Cell *rp = gforth_RP;
                    105:   Cell *sp = gforth_SP;
                    106:   Float *fp = gforth_FP;
                    107:   Address lp = gforth_LP;
1.168     pazsan    108:   void ** clist = gforth_clist;
                    109:   void * ritem = gforth_ritem;
1.153     pazsan    110: 
1.164     pazsan    111:   gforth_clist = args;
                    112:   gforth_ritem = resp;
1.153     pazsan    113: 
1.164     pazsan    114:   gforth_engine((Xt *)ip, sp, rp, fp, lp);
1.153     pazsan    115: 
                    116:   /* restore global variables */
1.161     pazsan    117:   gforth_RP = rp;
                    118:   gforth_SP = sp;
                    119:   gforth_FP = fp;
                    120:   gforth_LP = lp;
1.168     pazsan    121:   gforth_clist = clist;
                    122:   gforth_ritem = ritem;
1.153     pazsan    123: }
                    124: #endif
                    125: 
1.79      anton     126: #ifdef GFORTH_DEBUGGING
                    127: /* define some VM registers as global variables, so they survive exceptions;
                    128:    global register variables are not up to the task (according to the 
                    129:    GNU C manual) */
                    130: Xt *saved_ip;
                    131: Cell *rp;
                    132: #endif
                    133: 
                    134: #ifdef NO_IP
                    135: Label next_code;
                    136: #endif
                    137: 
                    138: #ifdef HAS_FILE
                    139: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
                    140: char* pfileattr[6]={"r","r","r+","r+","w","w"};
                    141: 
                    142: #ifndef O_BINARY
                    143: #define O_BINARY 0
                    144: #endif
                    145: #ifndef O_TEXT
                    146: #define O_TEXT 0
                    147: #endif
                    148: 
                    149: int ufileattr[6]= {
                    150:   O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
                    151:   O_RDWR  |O_BINARY, O_RDWR  |O_BINARY,
                    152:   O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
                    153: #endif
                    154: /* end global vars for engine.c */
                    155: 
1.1       anton     156: #define PRIM_VERSION 1
                    157: /* increment this whenever the primitives change in an incompatible way */
                    158: 
1.14      pazsan    159: #ifndef DEFAULTPATH
1.39      anton     160: #  define DEFAULTPATH "."
1.14      pazsan    161: #endif
                    162: 
1.1       anton     163: #ifdef MSDOS
                    164: jmp_buf throw_jmp_buf;
                    165: #endif
                    166: 
1.56      anton     167: #if defined(DOUBLY_INDIRECT)
                    168: #  define CFA(n)       ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
1.1       anton     169: #else
1.56      anton     170: #  define CFA(n)       ((Cell)(symbols+((n)&~0x4000UL)))
1.1       anton     171: #endif
                    172: 
                    173: #define maxaligned(n)  (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
                    174: 
                    175: static UCell dictsize=0;
                    176: static UCell dsize=0;
                    177: static UCell rsize=0;
                    178: static UCell fsize=0;
                    179: static UCell lsize=0;
                    180: int offset_image=0;
1.4       anton     181: int die_on_signal=0;
1.169     anton     182: int ignore_async_signals=0;
1.13      pazsan    183: #ifndef INCLUDE_IMAGE
1.1       anton     184: static int clear_dictionary=0;
1.24      anton     185: UCell pagesize=1;
1.22      pazsan    186: char *progname;
                    187: #else
                    188: char *progname = "gforth";
                    189: int optind = 1;
1.13      pazsan    190: #endif
1.31      pazsan    191: 
1.167     anton     192: #define CODE_BLOCK_SIZE (512*1024) /* !! overflow handling for -native */
1.48      anton     193: Address code_area=0;
1.73      anton     194: Cell code_area_size = CODE_BLOCK_SIZE;
1.75      anton     195: Address code_here=NULL+CODE_BLOCK_SIZE; /* does for code-area what HERE
                    196:                                           does for the dictionary */
1.100     anton     197: Address start_flush=NULL; /* start of unflushed code */
1.74      anton     198: Cell last_jump=0; /* if the last prim was compiled without jump, this
                    199:                      is it's number, otherwise this contains 0 */
1.48      anton     200: 
1.60      anton     201: static int no_super=0;   /* true if compile_prim should not fuse prims */
1.81      anton     202: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
                    203:                                             dynamically */
1.110     anton     204: static int print_metrics=0; /* if true, print metrics on exit */
1.171     anton     205: static int static_super_number = 0; /* number of ss used if available */
                    206:                                     /* disabled because of tpa */
1.152     anton     207: #define MAX_STATE 9 /* maximum number of states */
1.125     anton     208: static int maxstates = MAX_STATE; /* number of states for stack caching */
1.110     anton     209: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
1.144     pazsan    210: static int diag = 0; /* if true: print diagnostic informations */
1.158     anton     211: static int tpa_noequiv = 0;     /* if true: no state equivalence checking */
                    212: static int tpa_noautomaton = 0; /* if true: no tree parsing automaton */
                    213: static int tpa_trace = 0; /* if true: data for line graph of new states etc. */
1.144     pazsan    214: static int relocs = 0;
                    215: static int nonrelocs = 0;
1.60      anton     216: 
1.30      pazsan    217: #ifdef HAS_DEBUG
1.68      anton     218: int debug=0;
1.144     pazsan    219: # define debugp(x...) if (debug) fprintf(x);
1.31      pazsan    220: #else
                    221: # define perror(x...)
                    222: # define fprintf(x...)
1.144     pazsan    223: # define debugp(x...)
1.30      pazsan    224: #endif
1.31      pazsan    225: 
1.24      anton     226: ImageHeader *gforth_header;
1.43      anton     227: Label *vm_prims;
1.53      anton     228: #ifdef DOUBLY_INDIRECT
                    229: Label *xts; /* same content as vm_prims, but should only be used for xts */
                    230: #endif
1.1       anton     231: 
1.125     anton     232: #ifndef NO_DYNAMIC
                    233: #define MAX_IMMARGS 2
                    234: 
                    235: typedef struct {
                    236:   Label start; /* NULL if not relocatable */
                    237:   Cell length; /* only includes the jump iff superend is true*/
                    238:   Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
                    239:   char superend; /* true if primitive ends superinstruction, i.e.,
                    240:                      unconditional branch, execute, etc. */
                    241:   Cell nimmargs;
                    242:   struct immarg {
                    243:     Cell offset; /* offset of immarg within prim */
                    244:     char rel;    /* true if immarg is relative */
                    245:   } immargs[MAX_IMMARGS];
                    246: } PrimInfo;
                    247: 
                    248: PrimInfo *priminfos;
                    249: PrimInfo **decomp_prims;
                    250: 
1.139     anton     251: const char const* const prim_names[]={
                    252: #include PRIM_NAMES_I
                    253: };
                    254: 
1.148     anton     255: void init_ss_cost(void);
                    256: 
1.125     anton     257: static int is_relocatable(int p)
                    258: {
                    259:   return !no_dynamic && priminfos[p].start != NULL;
                    260: }
                    261: #else /* defined(NO_DYNAMIC) */
                    262: static int is_relocatable(int p)
                    263: {
                    264:   return 0;
                    265: }
                    266: #endif /* defined(NO_DYNAMIC) */
                    267: 
1.30      pazsan    268: #ifdef MEMCMP_AS_SUBROUTINE
                    269: int gforth_memcmp(const char * s1, const char * s2, size_t n)
                    270: {
                    271:   return memcmp(s1, s2, n);
                    272: }
                    273: #endif
                    274: 
1.125     anton     275: static Cell max(Cell a, Cell b)
                    276: {
                    277:   return a>b?a:b;
                    278: }
                    279: 
                    280: static Cell min(Cell a, Cell b)
                    281: {
                    282:   return a<b?a:b;
                    283: }
                    284: 
1.1       anton     285: /* image file format:
1.15      pazsan    286:  *  "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
1.1       anton     287:  *   padding to a multiple of 8
1.84      anton     288:  *   magic: "Gforth3x" means format 0.6,
1.15      pazsan    289:  *              where x is a byte with
                    290:  *              bit 7:   reserved = 0
                    291:  *              bit 6:5: address unit size 2^n octets
                    292:  *              bit 4:3: character size 2^n octets
                    293:  *              bit 2:1: cell size 2^n octets
                    294:  *              bit 0:   endian, big=0, little=1.
                    295:  *  The magic are always 8 octets, no matter what the native AU/character size is
1.1       anton     296:  *  padding to max alignment (no padding necessary on current machines)
1.24      anton     297:  *  ImageHeader structure (see forth.h)
1.1       anton     298:  *  data (size in ImageHeader.image_size)
                    299:  *  tags ((if relocatable, 1 bit/data cell)
                    300:  *
                    301:  * tag==1 means that the corresponding word is an address;
                    302:  * If the word is >=0, the address is within the image;
                    303:  * addresses within the image are given relative to the start of the image.
                    304:  * If the word =-1 (CF_NIL), the address is NIL,
                    305:  * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
                    306:  * If the word =CF(DODOES), it's a DOES> CFA
                    307:  * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
                    308:  *                                     possibly containing a jump to dodoes)
1.51      anton     309:  * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
                    310:  * If the word is <CF(DOESJUMP) and bit 14 is clear, 
                    311:  *                                        it's the threaded code of a primitive
1.85      pazsan    312:  * bits 13..9 of a primitive token state which group the primitive belongs to,
                    313:  * bits 8..0 of a primitive token index into the group
1.1       anton     314:  */
                    315: 
1.115     pazsan    316: Cell groups[32] = {
1.85      pazsan    317:   0,
1.121     anton     318:   0
1.90      anton     319: #undef GROUP
1.115     pazsan    320: #undef GROUPADD
                    321: #define GROUPADD(n) +n
                    322: #define GROUP(x, n) , 0
1.126     anton     323: #include PRIM_GRP_I
1.90      anton     324: #undef GROUP
1.115     pazsan    325: #undef GROUPADD
1.85      pazsan    326: #define GROUP(x, n)
1.115     pazsan    327: #define GROUPADD(n)
1.85      pazsan    328: };
                    329: 
1.161     pazsan    330: static unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
1.125     anton     331:                              int size, Cell base)
                    332:      /* produce a bitmask marking all the branch targets */
                    333: {
1.130     anton     334:   int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
1.125     anton     335:   Cell token;
                    336:   unsigned char bits;
1.130     anton     337:   unsigned char *result=malloc(steps);
                    338: 
                    339:   memset(result, 0, steps);
                    340:   for(k=0; k<steps; k++) {
1.125     anton     341:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
1.130     anton     342:       if(bits & (1U << (RELINFOBITS-1))) {
                    343:        assert(i*sizeof(Cell) < size);
1.125     anton     344:         token=image[i];
                    345:        if (token>=base) { /* relocatable address */
                    346:          UCell bitnum=(token-base)/sizeof(Cell);
1.154     anton     347:          if (bitnum/RELINFOBITS < (UCell)steps)
                    348:            result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
1.125     anton     349:        }
                    350:       }
                    351:     }
                    352:   }
                    353:   return result;
                    354: }
                    355: 
1.162     pazsan    356: void gforth_relocate(Cell *image, const Char *bitstring, 
                    357:                     UCell size, Cell base, Label symbols[])
1.1       anton     358: {
1.130     anton     359:   int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
1.11      pazsan    360:   Cell token;
1.1       anton     361:   char bits;
1.37      anton     362:   Cell max_symbols;
1.46      jwilke    363:   /* 
1.85      pazsan    364:    * A virtual start address that's the real start address minus 
1.46      jwilke    365:    * the one in the image 
                    366:    */
1.45      jwilke    367:   Cell *start = (Cell * ) (((void *) image) - ((void *) base));
1.125     anton     368:   unsigned char *targets = branch_targets(image, bitstring, size, base);
1.1       anton     369: 
1.85      pazsan    370:   /* group index into table */
1.115     pazsan    371:   if(groups[31]==0) {
                    372:     int groupsum=0;
                    373:     for(i=0; i<32; i++) {
                    374:       groupsum += groups[i];
                    375:       groups[i] = groupsum;
                    376:       /* printf("group[%d]=%d\n",i,groupsum); */
                    377:     }
                    378:     i=0;
                    379:   }
1.46      jwilke    380:   
                    381: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
1.37      anton     382:   
1.121     anton     383:   for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
1.37      anton     384:     ;
1.47      anton     385:   max_symbols--;
1.35      pazsan    386: 
1.130     anton     387:   for(k=0; k<steps; k++) {
1.13      pazsan    388:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
1.1       anton     389:       /*      fprintf(stderr,"relocate: image[%d]\n", i);*/
1.130     anton     390:       if(bits & (1U << (RELINFOBITS-1))) {
                    391:        assert(i*sizeof(Cell) < size);
1.35      pazsan    392:        /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
1.45      jwilke    393:         token=image[i];
1.85      pazsan    394:        if(token<0) {
                    395:          int group = (-token & 0x3E00) >> 9;
                    396:          if(group == 0) {
                    397:            switch(token|0x4000) {
1.1       anton     398:            case CF_NIL      : image[i]=0; break;
                    399: #if !defined(DOUBLY_INDIRECT)
                    400:            case CF(DOCOL)   :
                    401:            case CF(DOVAR)   :
                    402:            case CF(DOCON)   :
                    403:            case CF(DOUSER)  : 
                    404:            case CF(DODEFER) : 
1.11      pazsan    405:            case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
1.92      anton     406:            case CF(DOESJUMP): image[i]=0; break;
1.1       anton     407: #endif /* !defined(DOUBLY_INDIRECT) */
                    408:            case CF(DODOES)  :
1.45      jwilke    409:              MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
1.1       anton     410:              break;
1.85      pazsan    411:            default          : /* backward compatibility */
1.56      anton     412: /*           printf("Code field generation image[%x]:=CFA(%x)\n",
1.1       anton     413:                     i, CF(image[i])); */
1.55      anton     414:              if (CF((token | 0x4000))<max_symbols) {
1.56      anton     415:                image[i]=(Cell)CFA(CF(token));
                    416: #ifdef DIRECT_THREADED
1.125     anton     417:                if ((token & 0x4000) == 0) { /* threade code, no CFA */
                    418:                  if (targets[k] & (1U<<(RELINFOBITS-1-j)))
                    419:                    compile_prim1(0);
1.70      anton     420:                  compile_prim1(&image[i]);
1.125     anton     421:                }
1.56      anton     422: #endif
1.55      anton     423:              } else
1.115     pazsan    424:                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     425:            }
1.85      pazsan    426:          } else {
                    427:            int tok = -token & 0x1FF;
                    428:            if (tok < (groups[group+1]-groups[group])) {
                    429: #if defined(DOUBLY_INDIRECT)
                    430:              image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
                    431: #else
                    432:              image[i]=(Cell)CFA((groups[group]+tok));
                    433: #endif
                    434: #ifdef DIRECT_THREADED
1.125     anton     435:              if ((token & 0x4000) == 0) { /* threade code, no CFA */
                    436:                if (targets[k] & (1U<<(RELINFOBITS-1-j)))
                    437:                  compile_prim1(0);
1.85      pazsan    438:                compile_prim1(&image[i]);
1.125     anton     439:              }
1.85      pazsan    440: #endif
                    441:            } else
1.115     pazsan    442:              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    443:          }
                    444:        } else {
1.101     anton     445:           /* if base is > 0: 0 is a null reference so don't adjust*/
1.45      jwilke    446:           if (token>=base) {
                    447:             image[i]+=(Cell)start;
                    448:           }
1.46      jwilke    449:         }
1.1       anton     450:       }
                    451:     }
1.31      pazsan    452:   }
1.125     anton     453:   free(targets);
1.70      anton     454:   finish_code();
1.26      jwilke    455:   ((ImageHeader*)(image))->base = (Address) image;
1.1       anton     456: }
                    457: 
1.162     pazsan    458: #ifndef DOUBLY_INDIRECT
1.161     pazsan    459: static UCell checksum(Label symbols[])
1.1       anton     460: {
                    461:   UCell r=PRIM_VERSION;
                    462:   Cell i;
                    463: 
                    464:   for (i=DOCOL; i<=DOESJUMP; i++) {
                    465:     r ^= (UCell)(symbols[i]);
                    466:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
                    467:   }
                    468: #ifdef DIRECT_THREADED
                    469:   /* we have to consider all the primitives */
                    470:   for (; symbols[i]!=(Label)0; i++) {
                    471:     r ^= (UCell)(symbols[i]);
                    472:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
                    473:   }
                    474: #else
                    475:   /* in indirect threaded code all primitives are accessed through the
                    476:      symbols table, so we just have to put the base address of symbols
                    477:      in the checksum */
                    478:   r ^= (UCell)symbols;
                    479: #endif
                    480:   return r;
                    481: }
1.162     pazsan    482: #endif
1.1       anton     483: 
1.161     pazsan    484: static Address verbose_malloc(Cell size)
1.3       anton     485: {
                    486:   Address r;
                    487:   /* leave a little room (64B) for stack underflows */
                    488:   if ((r = malloc(size+64))==NULL) {
                    489:     perror(progname);
                    490:     exit(1);
                    491:   }
                    492:   r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
1.144     pazsan    493:   debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
1.3       anton     494:   return r;
                    495: }
                    496: 
1.33      anton     497: static Address next_address=0;
1.161     pazsan    498: static void after_alloc(Address r, Cell size)
1.33      anton     499: {
                    500:   if (r != (Address)-1) {
1.144     pazsan    501:     debugp(stderr, "success, address=$%lx\n", (long) r);
1.173   ! anton     502: #if 0
        !           503:     /* not needed now that we protect the stacks with mprotect */
1.33      anton     504:     if (pagesize != 1)
                    505:       next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
1.173   ! anton     506: #endif
1.33      anton     507:   } else {
1.144     pazsan    508:     debugp(stderr, "failed: %s\n", strerror(errno));
1.33      anton     509:   }
                    510: }
                    511: 
1.34      anton     512: #ifndef MAP_FAILED
                    513: #define MAP_FAILED ((Address) -1)
                    514: #endif
                    515: #ifndef MAP_FILE
                    516: # define MAP_FILE 0
                    517: #endif
                    518: #ifndef MAP_PRIVATE
                    519: # define MAP_PRIVATE 0
                    520: #endif
1.91      anton     521: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
                    522: # define MAP_ANON MAP_ANONYMOUS
                    523: #endif
1.34      anton     524: 
                    525: #if defined(HAVE_MMAP)
                    526: static Address alloc_mmap(Cell size)
1.1       anton     527: {
                    528:   Address r;
                    529: 
                    530: #if defined(MAP_ANON)
1.144     pazsan    531:   debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
1.34      anton     532:   r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
1.1       anton     533: #else /* !defined(MAP_ANON) */
1.17      anton     534:   /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
                    535:      apparently defaults) */
1.1       anton     536:   static int dev_zero=-1;
                    537: 
                    538:   if (dev_zero == -1)
                    539:     dev_zero = open("/dev/zero", O_RDONLY);
                    540:   if (dev_zero == -1) {
1.34      anton     541:     r = MAP_FAILED;
1.144     pazsan    542:     debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ", 
1.1       anton     543:              strerror(errno));
                    544:   } else {
1.144     pazsan    545:     debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
1.1       anton     546:     r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE, dev_zero, 0);
                    547:   }
                    548: #endif /* !defined(MAP_ANON) */
1.34      anton     549:   after_alloc(r, size);
                    550:   return r;  
                    551: }
1.172     anton     552: 
                    553: static void page_noaccess(Address a)
                    554: {
                    555:   /* try mprotect first; with munmap the page might be allocated later */
                    556:   debugp(stderr, "try mprotect(%p,%ld,PROT_NONE); ", a, (long)pagesize);
                    557:   if (mprotect(a, pagesize, PROT_NONE)==0) {
                    558:     debugp(stderr, "ok\n");
                    559:     return;
                    560:   }
                    561:   debugp(stderr, "failed: %s\n", strerror(errno));
                    562:   debugp(stderr, "try munmap(%p,%ld); ", a, (long)pagesize);
                    563:   if (munmap(a,pagesize)==0) {
                    564:     debugp(stderr, "ok\n");
                    565:     return;
                    566:   }
                    567:   debugp(stderr, "failed: %s\n", strerror(errno));
                    568: }  
                    569: 
1.173   ! anton     570: static size_t wholepage(size_t n)
1.172     anton     571: {
                    572:   return (n+pagesize-1)&~(pagesize-1);
                    573: }
1.34      anton     574: #endif
                    575: 
1.161     pazsan    576: Address gforth_alloc(Cell size)
1.34      anton     577: {
                    578: #if HAVE_MMAP
                    579:   Address r;
                    580: 
                    581:   r=alloc_mmap(size);
1.117     anton     582:   if (r!=(Address)MAP_FAILED)
1.1       anton     583:     return r;
                    584: #endif /* HAVE_MMAP */
1.3       anton     585:   /* use malloc as fallback */
                    586:   return verbose_malloc(size);
1.1       anton     587: }
                    588: 
1.161     pazsan    589: static Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
1.33      anton     590: {
1.34      anton     591:   Address image = MAP_FAILED;
1.33      anton     592: 
1.56      anton     593: #if defined(HAVE_MMAP)
1.33      anton     594:   if (offset==0) {
1.34      anton     595:     image=alloc_mmap(dictsize);
1.150     anton     596:     if (image != (Address)MAP_FAILED) {
                    597:       Address image1;
                    598:       debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
                    599:       image1 = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE, fileno(file), 0);
                    600:       after_alloc(image1,dictsize);
                    601:       if (image1 == (Address)MAP_FAILED)
                    602:        goto read_image;
                    603:     }
1.33      anton     604:   }
1.56      anton     605: #endif /* defined(HAVE_MMAP) */
1.117     anton     606:   if (image == (Address)MAP_FAILED) {
1.161     pazsan    607:     image = gforth_alloc(dictsize+offset)+offset;
1.149     anton     608:   read_image:
1.33      anton     609:     rewind(file);  /* fseek(imagefile,0L,SEEK_SET); */
1.34      anton     610:     fread(image, 1, imagesize, file);
1.33      anton     611:   }
                    612:   return image;
                    613: }
                    614: 
1.10      pazsan    615: void set_stack_sizes(ImageHeader * header)
                    616: {
                    617:   if (dictsize==0)
                    618:     dictsize = header->dict_size;
                    619:   if (dsize==0)
                    620:     dsize = header->data_stack_size;
                    621:   if (rsize==0)
                    622:     rsize = header->return_stack_size;
                    623:   if (fsize==0)
                    624:     fsize = header->fp_stack_size;
                    625:   if (lsize==0)
                    626:     lsize = header->locals_stack_size;
                    627:   dictsize=maxaligned(dictsize);
                    628:   dsize=maxaligned(dsize);
                    629:   rsize=maxaligned(rsize);
                    630:   lsize=maxaligned(lsize);
                    631:   fsize=maxaligned(fsize);
                    632: }
                    633: 
1.173   ! anton     634: void alloc_stacks(ImageHeader * h)
1.10      pazsan    635: {
1.173   ! anton     636:   h->dict_size=dictsize;
        !           637:   h->data_stack_size=dsize;
        !           638:   h->fp_stack_size=fsize;
        !           639:   h->return_stack_size=rsize;
        !           640:   h->locals_stack_size=lsize;
1.10      pazsan    641: 
1.172     anton     642: #if defined(HAVE_MMAP)
                    643:   if (pagesize > 1) {
1.173   ! anton     644:     size_t p = pagesize;
        !           645:     size_t totalsize =
        !           646:       wholepage(dsize)+wholepage(fsize)+wholepage(rsize)+wholepage(lsize)+5*p;
1.172     anton     647:     Address a = alloc_mmap(totalsize);
                    648:     if (a != (Address)MAP_FAILED) {
1.173   ! anton     649:       page_noaccess(a); a+=p; h->  data_stack_base=a; a+=wholepage(dsize);
        !           650:       page_noaccess(a); a+=p; h->    fp_stack_base=a; a+=wholepage(fsize);
        !           651:       page_noaccess(a); a+=p; h->return_stack_base=a; a+=wholepage(rsize);
        !           652:       page_noaccess(a); a+=p; h->locals_stack_base=a; a+=wholepage(lsize);
1.172     anton     653:       page_noaccess(a);
                    654:       debugp(stderr,"stack addresses: d=%p f=%p r=%p l=%p\n",
1.173   ! anton     655:             h->data_stack_base,
        !           656:             h->fp_stack_base,
        !           657:             h->return_stack_base,
        !           658:             h->locals_stack_base);
1.172     anton     659:       return;
                    660:     }
                    661:   }
                    662: #endif
1.173   ! anton     663:   h->data_stack_base=gforth_alloc(dsize);
        !           664:   h->fp_stack_base=gforth_alloc(fsize);
        !           665:   h->return_stack_base=gforth_alloc(rsize);
        !           666:   h->locals_stack_base=gforth_alloc(lsize);
1.10      pazsan    667: }
                    668: 
1.161     pazsan    669: #warning You can ignore the warnings about clobbered variables in gforth_go
                    670: int gforth_go(Address image, int stack, Cell *entries)
1.11      pazsan    671: {
1.38      anton     672:   volatile ImageHeader *image_header = (ImageHeader *)image;
1.18      anton     673:   Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
1.44      pazsan    674:   Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
1.18      anton     675:   Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
1.44      pazsan    676: #ifdef GFORTH_DEBUGGING
1.38      anton     677:   volatile Cell *orig_rp0=rp0;
1.44      pazsan    678: #endif
1.18      anton     679:   Address lp0=image_header->locals_stack_base + lsize;
                    680:   Xt *ip0=(Xt *)(image_header->boot_entry);
1.13      pazsan    681: #ifdef SYSSIGNALS
1.11      pazsan    682:   int throw_code;
1.13      pazsan    683: #endif
1.11      pazsan    684: 
                    685:   /* ensure that the cached elements (if any) are accessible */
1.151     anton     686: #if !(defined(GFORTH_DEBUGGING) || defined(INDIRECT_THREADED) || defined(DOUBLY_INDIRECT) || defined(VM_PROFILING))
                    687:   sp0 -= 8; /* make stuff below bottom accessible for stack caching */
                    688: #endif
1.41      anton     689:   IF_fpTOS(fp0--);
1.11      pazsan    690:   
                    691:   for(;stack>0;stack--)
1.18      anton     692:     *--sp0=entries[stack-1];
1.11      pazsan    693: 
1.30      pazsan    694: #ifdef SYSSIGNALS
1.11      pazsan    695:   get_winsize();
                    696:    
                    697:   install_signal_handlers(); /* right place? */
                    698:   
                    699:   if ((throw_code=setjmp(throw_jmp_buf))) {
1.152     anton     700:     static Cell signal_data_stack[24];
                    701:     static Cell signal_return_stack[16];
1.11      pazsan    702:     static Float signal_fp_stack[1];
1.13      pazsan    703: 
1.152     anton     704:     signal_data_stack[15]=throw_code;
1.18      anton     705: 
                    706: #ifdef GFORTH_DEBUGGING
1.144     pazsan    707:     debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
1.97      anton     708:              throw_code, saved_ip, rp);
1.38      anton     709:     if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
1.18      anton     710:       /* no rstack overflow or underflow */
                    711:       rp0 = rp;
1.63      anton     712:       *--rp0 = (Cell)saved_ip;
1.18      anton     713:     }
                    714:     else /* I love non-syntactic ifdefs :-) */
1.152     anton     715:       rp0 = signal_return_stack+16;
1.97      anton     716: #else  /* !defined(GFORTH_DEBUGGING) */
1.144     pazsan    717:     debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
1.152     anton     718:       rp0 = signal_return_stack+16;
1.97      anton     719: #endif /* !defined(GFORTH_DEBUGGING) */
1.25      anton     720:     /* fprintf(stderr, "rp=$%x\n",rp0);*/
1.11      pazsan    721:     
1.164     pazsan    722:     return((int)(Cell)gforth_engine(image_header->throw_entry, signal_data_stack+15,
1.18      anton     723:                       rp0, signal_fp_stack, 0));
1.11      pazsan    724:   }
1.13      pazsan    725: #endif
1.11      pazsan    726: 
1.164     pazsan    727:   return((int)(Cell)gforth_engine(ip0,sp0,rp0,fp0,lp0));
1.11      pazsan    728: }
                    729: 
1.30      pazsan    730: #ifndef INCLUDE_IMAGE
1.161     pazsan    731: static void print_sizes(Cell sizebyte)
1.21      anton     732:      /* print size information */
                    733: {
                    734:   static char* endianstring[]= { "   big","little" };
                    735:   
                    736:   fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
                    737:          endianstring[sizebyte & 1],
                    738:          1 << ((sizebyte >> 1) & 3),
                    739:          1 << ((sizebyte >> 3) & 3),
                    740:          1 << ((sizebyte >> 5) & 3));
                    741: }
                    742: 
1.106     anton     743: /* static superinstruction stuff */
                    744: 
1.141     anton     745: struct cost { /* super_info might be a more accurate name */
1.106     anton     746:   char loads;       /* number of stack loads */
                    747:   char stores;      /* number of stack stores */
                    748:   char updates;     /* number of stack pointer updates */
1.123     anton     749:   char branch;     /* is it a branch (SET_IP) */
1.125     anton     750:   unsigned char state_in;    /* state on entry */
                    751:   unsigned char state_out;   /* state on exit */
1.142     anton     752:   unsigned char imm_ops;     /* number of immediate operands */
1.123     anton     753:   short offset;     /* offset into super2 table */
1.125     anton     754:   unsigned char length;      /* number of components */
1.106     anton     755: };
                    756: 
1.121     anton     757: PrimNum super2[] = {
1.126     anton     758: #include SUPER2_I
1.106     anton     759: };
                    760: 
                    761: struct cost super_costs[] = {
1.126     anton     762: #include COSTS_I
1.106     anton     763: };
                    764: 
1.125     anton     765: struct super_state {
                    766:   struct super_state *next;
                    767:   PrimNum super;
                    768: };
                    769: 
1.106     anton     770: #define HASH_SIZE 256
                    771: 
                    772: struct super_table_entry {
                    773:   struct super_table_entry *next;
1.121     anton     774:   PrimNum *start;
1.106     anton     775:   short length;
1.125     anton     776:   struct super_state *ss_list; /* list of supers */
1.106     anton     777: } *super_table[HASH_SIZE];
                    778: int max_super=2;
                    779: 
1.125     anton     780: struct super_state *state_transitions=NULL;
                    781: 
1.161     pazsan    782: static int hash_super(PrimNum *start, int length)
1.106     anton     783: {
                    784:   int i, r;
                    785:   
                    786:   for (i=0, r=0; i<length; i++) {
                    787:     r <<= 1;
                    788:     r += start[i];
                    789:   }
                    790:   return r & (HASH_SIZE-1);
                    791: }
                    792: 
1.161     pazsan    793: static struct super_state **lookup_super(PrimNum *start, int length)
1.106     anton     794: {
                    795:   int hash=hash_super(start,length);
                    796:   struct super_table_entry *p = super_table[hash];
                    797: 
1.125     anton     798:   /* assert(length >= 2); */
1.106     anton     799:   for (; p!=NULL; p = p->next) {
                    800:     if (length == p->length &&
1.121     anton     801:        memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
1.125     anton     802:       return &(p->ss_list);
1.106     anton     803:   }
1.125     anton     804:   return NULL;
1.106     anton     805: }
                    806: 
1.161     pazsan    807: static void prepare_super_table()
1.106     anton     808: {
                    809:   int i;
1.109     anton     810:   int nsupers = 0;
1.106     anton     811: 
                    812:   for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
                    813:     struct cost *c = &super_costs[i];
1.125     anton     814:     if ((c->length < 2 || nsupers < static_super_number) &&
                    815:        c->state_in < maxstates && c->state_out < maxstates) {
                    816:       struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
                    817:       struct super_state *ss = malloc(sizeof(struct super_state));
                    818:       ss->super= i;
                    819:       if (c->offset==N_noop && i != N_noop) {
                    820:        if (is_relocatable(i)) {
                    821:          ss->next = state_transitions;
                    822:          state_transitions = ss;
                    823:        }
                    824:       } else if (ss_listp != NULL) {
                    825:        ss->next = *ss_listp;
                    826:        *ss_listp = ss;
                    827:       } else {
                    828:        int hash = hash_super(super2+c->offset, c->length);
                    829:        struct super_table_entry **p = &super_table[hash];
                    830:        struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
                    831:        ss->next = NULL;
                    832:        e->next = *p;
                    833:        e->start = super2 + c->offset;
                    834:        e->length = c->length;
                    835:        e->ss_list = ss;
                    836:        *p = e;
                    837:       }
1.106     anton     838:       if (c->length > max_super)
                    839:        max_super = c->length;
1.125     anton     840:       if (c->length >= 2)
                    841:        nsupers++;
1.106     anton     842:     }
                    843:   }
1.144     pazsan    844:   debugp(stderr, "Using %d static superinsts\n", nsupers);
1.106     anton     845: }
                    846: 
                    847: /* dynamic replication/superinstruction stuff */
                    848: 
1.69      anton     849: #ifndef NO_DYNAMIC
1.161     pazsan    850: static int compare_priminfo_length(const void *_a, const void *_b)
1.76      anton     851: {
1.90      anton     852:   PrimInfo **a = (PrimInfo **)_a;
                    853:   PrimInfo **b = (PrimInfo **)_b;
1.77      anton     854:   Cell diff = (*a)->length - (*b)->length;
                    855:   if (diff)
                    856:     return diff;
                    857:   else /* break ties by start address; thus the decompiler produces
                    858:           the earliest primitive with the same code (e.g. noop instead
                    859:           of (char) and @ instead of >code-address */
                    860:     return (*b)->start - (*a)->start;
1.76      anton     861: }
1.112     anton     862: #endif /* !defined(NO_DYNAMIC) */
1.76      anton     863: 
1.125     anton     864: static char MAYBE_UNUSED superend[]={
1.126     anton     865: #include PRIM_SUPEREND_I
1.106     anton     866: };
1.107     anton     867: 
                    868: Cell npriminfos=0;
1.76      anton     869: 
1.146     anton     870: Label goto_start;
                    871: Cell goto_len;
                    872: 
1.162     pazsan    873: #ifndef NO_DYNAMIC
1.161     pazsan    874: static int compare_labels(const void *pa, const void *pb)
1.113     anton     875: {
1.114     anton     876:   Label a = *(Label *)pa;
                    877:   Label b = *(Label *)pb;
                    878:   return a-b;
                    879: }
1.162     pazsan    880: #endif
1.113     anton     881: 
1.161     pazsan    882: static Label bsearch_next(Label key, Label *a, UCell n)
1.114     anton     883:      /* a is sorted; return the label >=key that is the closest in a;
                    884:         return NULL if there is no label in a >=key */
                    885: {
                    886:   int mid = (n-1)/2;
                    887:   if (n<1)
                    888:     return NULL;
                    889:   if (n == 1) {
                    890:     if (a[0] < key)
                    891:       return NULL;
                    892:     else
                    893:       return a[0];
                    894:   }
                    895:   if (a[mid] < key)
                    896:     return bsearch_next(key, a+mid+1, n-mid-1);
                    897:   else
                    898:     return bsearch_next(key, a, mid+1);
1.113     anton     899: }
                    900: 
1.161     pazsan    901: static void check_prims(Label symbols1[])
1.47      anton     902: {
                    903:   int i;
1.90      anton     904: #ifndef NO_DYNAMIC
1.146     anton     905:   Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted, *goto_p;
1.119     anton     906:   int nends1j;
1.90      anton     907: #endif
1.47      anton     908: 
1.66      anton     909:   if (debug)
                    910: #ifdef __VERSION__
                    911:     fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
                    912: #else
                    913: #define xstr(s) str(s)
                    914: #define str(s) #s
                    915:   fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n"); 
                    916: #endif
1.121     anton     917:   for (i=0; symbols1[i]!=0; i++)
1.47      anton     918:     ;
1.55      anton     919:   npriminfos = i;
1.70      anton     920:   
                    921: #ifndef NO_DYNAMIC
1.66      anton     922:   if (no_dynamic)
                    923:     return;
1.164     pazsan    924:   symbols2=gforth_engine2(0,0,0,0,0);
1.70      anton     925: #if NO_IP
1.164     pazsan    926:   symbols3=gforth_engine3(0,0,0,0,0);
1.70      anton     927: #else
                    928:   symbols3=symbols1;
                    929: #endif
1.121     anton     930:   ends1 = symbols1+i+1;
1.119     anton     931:   ends1j =   ends1+i;
1.146     anton     932:   goto_p = ends1j+i+1; /* goto_p[0]==before; ...[1]==after;*/
1.121     anton     933:   nends1j = i+1;
1.119     anton     934:   ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
                    935:   memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
                    936:   qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
1.146     anton     937: 
                    938:   /* check whether the "goto *" is relocatable */
                    939:   goto_len = goto_p[1]-goto_p[0];
                    940:   debugp(stderr, "goto * %p %p len=%ld\n",
                    941:         goto_p[0],symbols2[goto_p-symbols1],goto_len);
                    942:   if (memcmp(goto_p[0],symbols2[goto_p-symbols1],goto_len)!=0) { /* unequal */
                    943:     no_dynamic=1;
                    944:     debugp(stderr,"  not relocatable, disabling dynamic code generation\n");
1.148     anton     945:     init_ss_cost();
1.146     anton     946:     return;
                    947:   }
                    948:   goto_start = goto_p[0];
1.113     anton     949:   
1.47      anton     950:   priminfos = calloc(i,sizeof(PrimInfo));
1.121     anton     951:   for (i=0; symbols1[i]!=0; i++) {
1.70      anton     952:     int prim_len = ends1[i]-symbols1[i];
1.47      anton     953:     PrimInfo *pi=&priminfos[i];
1.154     anton     954:     struct cost *sc=&super_costs[i];
1.70      anton     955:     int j=0;
                    956:     char *s1 = (char *)symbols1[i];
                    957:     char *s2 = (char *)symbols2[i];
                    958:     char *s3 = (char *)symbols3[i];
1.119     anton     959:     Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
1.70      anton     960: 
                    961:     pi->start = s1;
1.121     anton     962:     pi->superend = superend[i]|no_super;
1.147     anton     963:     pi->length = prim_len;
1.113     anton     964:     pi->restlength = endlabel - symbols1[i] - pi->length;
1.70      anton     965:     pi->nimmargs = 0;
1.144     pazsan    966:     relocs++;
1.154     anton     967:     debugp(stderr, "%-15s %d-%d %4d %p %p %p len=%3ld rest=%2ld send=%1d",
                    968:           prim_names[i], sc->state_in, sc->state_out,
                    969:           i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength),
                    970:           pi->superend);
1.114     anton     971:     if (endlabel == NULL) {
                    972:       pi->start = NULL; /* not relocatable */
1.122     anton     973:       if (pi->length<0) pi->length=100;
1.144     pazsan    974:       debugp(stderr,"\n   non_reloc: no J label > start found\n");
                    975:       relocs--;
                    976:       nonrelocs++;
1.114     anton     977:       continue;
                    978:     }
                    979:     if (ends1[i] > endlabel && !pi->superend) {
1.113     anton     980:       pi->start = NULL; /* not relocatable */
1.122     anton     981:       pi->length = endlabel-symbols1[i];
1.144     pazsan    982:       debugp(stderr,"\n   non_reloc: there is a J label before the K label (restlength<0)\n");
                    983:       relocs--;
                    984:       nonrelocs++;
1.113     anton     985:       continue;
                    986:     }
1.114     anton     987:     if (ends1[i] < pi->start && !pi->superend) {
1.113     anton     988:       pi->start = NULL; /* not relocatable */
1.122     anton     989:       pi->length = endlabel-symbols1[i];
1.144     pazsan    990:       debugp(stderr,"\n   non_reloc: K label before I label (length<0)\n");
                    991:       relocs--;
                    992:       nonrelocs++;
1.113     anton     993:       continue;
                    994:     }
1.138     anton     995:     assert(pi->length>=0);
1.113     anton     996:     assert(pi->restlength >=0);
1.74      anton     997:     while (j<(pi->length+pi->restlength)) {
1.70      anton     998:       if (s1[j]==s3[j]) {
                    999:        if (s1[j] != s2[j]) {
                   1000:          pi->start = NULL; /* not relocatable */
1.144     pazsan   1001:          debugp(stderr,"\n   non_reloc: engine1!=engine2 offset %3d",j);
1.74      anton    1002:          /* assert(j<prim_len); */
1.144     pazsan   1003:          relocs--;
                   1004:          nonrelocs++;
1.70      anton    1005:          break;
                   1006:        }
                   1007:        j++;
                   1008:       } else {
                   1009:        struct immarg *ia=&pi->immargs[pi->nimmargs];
                   1010: 
                   1011:        pi->nimmargs++;
                   1012:        ia->offset=j;
                   1013:        if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
                   1014:          ia->rel=0;
1.144     pazsan   1015:          debugp(stderr,"\n   absolute immarg: offset %3d",j);
1.70      anton    1016:        } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
                   1017:                   symbols1[DOESJUMP+1]) {
                   1018:          ia->rel=1;
1.144     pazsan   1019:          debugp(stderr,"\n   relative immarg: offset %3d",j);
1.70      anton    1020:        } else {
                   1021:          pi->start = NULL; /* not relocatable */
1.144     pazsan   1022:          debugp(stderr,"\n   non_reloc: engine1!=engine3 offset %3d",j);
1.74      anton    1023:          /* assert(j<prim_len);*/
1.144     pazsan   1024:          relocs--;
                   1025:          nonrelocs++;
1.70      anton    1026:          break;
                   1027:        }
                   1028:        j+=4;
1.47      anton    1029:       }
                   1030:     }
1.144     pazsan   1031:     debugp(stderr,"\n");
1.70      anton    1032:   }
1.76      anton    1033:   decomp_prims = calloc(i,sizeof(PrimInfo *));
                   1034:   for (i=DOESJUMP+1; i<npriminfos; i++)
                   1035:     decomp_prims[i] = &(priminfos[i]);
                   1036:   qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
                   1037:        compare_priminfo_length);
1.70      anton    1038: #endif
                   1039: }
                   1040: 
1.161     pazsan   1041: static void flush_to_here(void)
1.74      anton    1042: {
1.93      anton    1043: #ifndef NO_DYNAMIC
1.100     anton    1044:   if (start_flush)
                   1045:     FLUSH_ICACHE(start_flush, code_here-start_flush);
1.74      anton    1046:   start_flush=code_here;
1.93      anton    1047: #endif
1.74      anton    1048: }
                   1049: 
1.93      anton    1050: #ifndef NO_DYNAMIC
1.161     pazsan   1051: static void append_jump(void)
1.74      anton    1052: {
                   1053:   if (last_jump) {
                   1054:     PrimInfo *pi = &priminfos[last_jump];
                   1055:     
                   1056:     memcpy(code_here, pi->start+pi->length, pi->restlength);
                   1057:     code_here += pi->restlength;
1.147     anton    1058:     memcpy(code_here, goto_start, goto_len);
                   1059:     code_here += goto_len;
1.74      anton    1060:     last_jump=0;
                   1061:   }
                   1062: }
                   1063: 
1.75      anton    1064: /* Gforth remembers all code blocks in this list.  On forgetting (by
                   1065: executing a marker) the code blocks are not freed (because Gforth does
                   1066: not remember how they were allocated; hmm, remembering that might be
                   1067: easier and cleaner).  Instead, code_here etc. are reset to the old
                   1068: value, and the "forgotten" code blocks are reused when they are
                   1069: needed. */
                   1070: 
                   1071: struct code_block_list {
                   1072:   struct code_block_list *next;
                   1073:   Address block;
                   1074:   Cell size;
                   1075: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
                   1076: 
1.161     pazsan   1077: static Address append_prim(Cell p)
1.74      anton    1078: {
                   1079:   PrimInfo *pi = &priminfos[p];
                   1080:   Address old_code_here = code_here;
                   1081: 
1.159     anton    1082:   if (code_area+code_area_size < code_here+pi->length+pi->restlength+goto_len) {
1.75      anton    1083:     struct code_block_list *p;
1.74      anton    1084:     append_jump();
1.93      anton    1085:     flush_to_here();
1.75      anton    1086:     if (*next_code_blockp == NULL) {
1.161     pazsan   1087:       code_here = start_flush = code_area = gforth_alloc(code_area_size);
1.75      anton    1088:       p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
                   1089:       *next_code_blockp = p;
                   1090:       p->next = NULL;
                   1091:       p->block = code_here;
                   1092:       p->size = code_area_size;
                   1093:     } else {
                   1094:       p = *next_code_blockp;
                   1095:       code_here = start_flush = code_area = p->block;
                   1096:     }
1.74      anton    1097:     old_code_here = code_here;
1.75      anton    1098:     next_code_blockp = &(p->next);
1.74      anton    1099:   }
                   1100:   memcpy(code_here, pi->start, pi->length);
                   1101:   code_here += pi->length;
                   1102:   return old_code_here;
                   1103: }
                   1104: #endif
1.75      anton    1105: 
                   1106: int forget_dyncode(Address code)
                   1107: {
                   1108: #ifdef NO_DYNAMIC
                   1109:   return -1;
                   1110: #else
                   1111:   struct code_block_list *p, **pp;
                   1112: 
                   1113:   for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
                   1114:     if (code >= p->block && code < p->block+p->size) {
                   1115:       next_code_blockp = &(p->next);
                   1116:       code_here = start_flush = code;
                   1117:       code_area = p->block;
                   1118:       last_jump = 0;
                   1119:       return -1;
                   1120:     }
                   1121:   }
1.78      anton    1122:   return -no_dynamic;
1.75      anton    1123: #endif /* !defined(NO_DYNAMIC) */
                   1124: }
                   1125: 
1.161     pazsan   1126: static long dyncodesize(void)
1.104     anton    1127: {
                   1128: #ifndef NO_DYNAMIC
1.106     anton    1129:   struct code_block_list *p;
1.104     anton    1130:   long size=0;
                   1131:   for (p=code_block_list; p!=NULL; p=p->next) {
                   1132:     if (code_here >= p->block && code_here < p->block+p->size)
                   1133:       return size + (code_here - p->block);
                   1134:     else
                   1135:       size += p->size;
                   1136:   }
                   1137: #endif /* !defined(NO_DYNAMIC) */
                   1138:   return 0;
                   1139: }
                   1140: 
1.90      anton    1141: Label decompile_code(Label _code)
1.75      anton    1142: {
1.76      anton    1143: #ifdef NO_DYNAMIC
1.90      anton    1144:   return _code;
1.76      anton    1145: #else /* !defined(NO_DYNAMIC) */
                   1146:   Cell i;
1.77      anton    1147:   struct code_block_list *p;
1.90      anton    1148:   Address code=_code;
1.76      anton    1149: 
1.77      anton    1150:   /* first, check if we are in code at all */
                   1151:   for (p = code_block_list;; p = p->next) {
                   1152:     if (p == NULL)
                   1153:       return code;
                   1154:     if (code >= p->block && code < p->block+p->size)
                   1155:       break;
                   1156:   }
1.76      anton    1157:   /* reverse order because NOOP might match other prims */
                   1158:   for (i=npriminfos-1; i>DOESJUMP; i--) {
                   1159:     PrimInfo *pi=decomp_prims[i];
                   1160:     if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1.121     anton    1161:       return vm_prims[super2[super_costs[pi-priminfos].offset]];
1.118     anton    1162:     /* return pi->start;*/
1.76      anton    1163:   }
                   1164:   return code;
                   1165: #endif /* !defined(NO_DYNAMIC) */
1.75      anton    1166: }
1.74      anton    1167: 
1.70      anton    1168: #ifdef NO_IP
                   1169: int nbranchinfos=0;
                   1170: 
                   1171: struct branchinfo {
1.136     anton    1172:   Label **targetpp; /* **(bi->targetpp) is the target */
1.70      anton    1173:   Cell *addressptr; /* store the target here */
                   1174: } branchinfos[100000];
                   1175: 
                   1176: int ndoesexecinfos=0;
                   1177: struct doesexecinfo {
                   1178:   int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1.136     anton    1179:   Label *targetp; /*target for branch (because this is not in threaded code)*/
1.70      anton    1180:   Cell *xt; /* cfa of word whose does-code needs calling */
                   1181: } doesexecinfos[10000];
                   1182: 
1.161     pazsan   1183: static void set_rel_target(Cell *source, Label target)
1.70      anton    1184: {
                   1185:   *source = ((Cell)target)-(((Cell)source)+4);
                   1186: }
                   1187: 
1.161     pazsan   1188: static void register_branchinfo(Label source, Cell *targetpp)
1.70      anton    1189: {
                   1190:   struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1.136     anton    1191:   bi->targetpp = (Label **)targetpp;
1.70      anton    1192:   bi->addressptr = (Cell *)source;
                   1193:   nbranchinfos++;
                   1194: }
                   1195: 
1.161     pazsan   1196: static Address compile_prim1arg(PrimNum p, Cell **argp)
1.70      anton    1197: {
1.133     anton    1198:   Address old_code_here=append_prim(p);
1.70      anton    1199: 
1.74      anton    1200:   assert(vm_prims[p]==priminfos[p].start);
1.133     anton    1201:   *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
                   1202:   return old_code_here;
1.70      anton    1203: }
                   1204: 
1.161     pazsan   1205: static Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1.70      anton    1206: {
1.73      anton    1207:   PrimInfo *pi = &priminfos[N_call2];
1.74      anton    1208:   Address old_code_here = append_prim(N_call2);
1.70      anton    1209: 
1.134     anton    1210:   *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1.136     anton    1211:   register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1.134     anton    1212:   return old_code_here;
1.70      anton    1213: }
                   1214: #endif
                   1215: 
                   1216: void finish_code(void)
                   1217: {
                   1218: #ifdef NO_IP
                   1219:   Cell i;
                   1220: 
                   1221:   compile_prim1(NULL);
                   1222:   for (i=0; i<ndoesexecinfos; i++) {
                   1223:     struct doesexecinfo *dei = &doesexecinfos[i];
1.136     anton    1224:     dei->targetp = (Label *)DOES_CODE1((dei->xt));
                   1225:     branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1.70      anton    1226:   }
                   1227:   ndoesexecinfos = 0;
                   1228:   for (i=0; i<nbranchinfos; i++) {
                   1229:     struct branchinfo *bi=&branchinfos[i];
1.136     anton    1230:     set_rel_target(bi->addressptr, **(bi->targetpp));
1.70      anton    1231:   }
                   1232:   nbranchinfos = 0;
1.128     anton    1233: #else
                   1234:   compile_prim1(NULL);
1.48      anton    1235: #endif
1.93      anton    1236:   flush_to_here();
1.48      anton    1237: }
                   1238: 
1.162     pazsan   1239: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1.128     anton    1240: #ifdef NO_IP
1.161     pazsan   1241: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1.128     anton    1242:      /* compile prim #p dynamically (mod flags etc.) and return start
                   1243:        address of generated code for putting it into the threaded
                   1244:        code. This function is only called if all the associated
                   1245:        inline arguments of p are already in place (at tcp[1] etc.) */
                   1246: {
                   1247:   PrimInfo *pi=&priminfos[p];
                   1248:   Cell *next_code_target=NULL;
1.135     anton    1249:   Address codeaddr;
                   1250:   Address primstart;
1.128     anton    1251:   
                   1252:   assert(p<npriminfos);
                   1253:   if (p==N_execute || p==N_perform || p==N_lit_perform) {
1.134     anton    1254:     codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1.135     anton    1255:     primstart = append_prim(p);
                   1256:     goto other_prim;
                   1257:   } else if (p==N_call) {
1.136     anton    1258:     codeaddr = compile_call2(tcp+1, &next_code_target);
1.128     anton    1259:   } else if (p==N_does_exec) {
                   1260:     struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1.133     anton    1261:     Cell *arg;
                   1262:     codeaddr = compile_prim1arg(N_lit,&arg);
                   1263:     *arg = (Cell)PFA(tcp[1]);
1.128     anton    1264:     /* we cannot determine the callee now (last_start[1] may be a
                   1265:        forward reference), so just register an arbitrary target, and
                   1266:        register in dei that we need to fix this before resolving
                   1267:        branches */
                   1268:     dei->branchinfo = nbranchinfos;
                   1269:     dei->xt = (Cell *)(tcp[1]);
1.134     anton    1270:     compile_call2(0, &next_code_target);
1.128     anton    1271:   } else if (!is_relocatable(p)) {
1.133     anton    1272:     Cell *branch_target;
                   1273:     codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
                   1274:     compile_prim1arg(N_branch,&branch_target);
                   1275:     set_rel_target(branch_target,vm_prims[p]);
1.128     anton    1276:   } else {
                   1277:     unsigned j;
1.135     anton    1278: 
                   1279:     codeaddr = primstart = append_prim(p);
                   1280:   other_prim:
1.128     anton    1281:     for (j=0; j<pi->nimmargs; j++) {
                   1282:       struct immarg *ia = &(pi->immargs[j]);
1.136     anton    1283:       Cell *argp = tcp + pi->nimmargs - j;
                   1284:       Cell argval = *argp; /* !! specific to prims */
1.128     anton    1285:       if (ia->rel) { /* !! assumption: relative refs are branches */
1.136     anton    1286:        register_branchinfo(primstart + ia->offset, argp);
1.128     anton    1287:       } else /* plain argument */
1.135     anton    1288:        *(Cell *)(primstart + ia->offset) = argval;
1.128     anton    1289:     }
                   1290:   }
                   1291:   if (next_code_target!=NULL)
                   1292:     *next_code_target = (Cell)code_here;
1.135     anton    1293:   return (Cell)codeaddr;
1.128     anton    1294: }
                   1295: #else /* !defined(NO_IP) */
1.161     pazsan   1296: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1.128     anton    1297:      /* compile prim #p dynamically (mod flags etc.) and return start
                   1298:         address of generated code for putting it into the threaded code */
1.108     anton    1299: {
1.121     anton    1300:   Cell static_prim = (Cell)vm_prims[p];
1.108     anton    1301: #if defined(NO_DYNAMIC)
                   1302:   return static_prim;
                   1303: #else /* !defined(NO_DYNAMIC) */
                   1304:   Address old_code_here;
                   1305: 
                   1306:   if (no_dynamic)
                   1307:     return static_prim;
1.125     anton    1308:   if (p>=npriminfos || !is_relocatable(p)) {
1.108     anton    1309:     append_jump();
                   1310:     return static_prim;
                   1311:   }
                   1312:   old_code_here = append_prim(p);
1.147     anton    1313:   last_jump = p;
                   1314:   if (priminfos[p].superend)
                   1315:     append_jump();
1.108     anton    1316:   return (Cell)old_code_here;
                   1317: #endif  /* !defined(NO_DYNAMIC) */
                   1318: }
1.128     anton    1319: #endif /* !defined(NO_IP) */
1.162     pazsan   1320: #endif
1.70      anton    1321: 
1.109     anton    1322: #ifndef NO_DYNAMIC
1.161     pazsan   1323: static int cost_codesize(int prim)
1.109     anton    1324: {
1.121     anton    1325:   return priminfos[prim].length;
1.109     anton    1326: }
                   1327: #endif
                   1328: 
1.161     pazsan   1329: static int cost_ls(int prim)
1.109     anton    1330: {
                   1331:   struct cost *c = super_costs+prim;
                   1332: 
                   1333:   return c->loads + c->stores;
                   1334: }
                   1335: 
1.161     pazsan   1336: static int cost_lsu(int prim)
1.109     anton    1337: {
                   1338:   struct cost *c = super_costs+prim;
                   1339: 
                   1340:   return c->loads + c->stores + c->updates;
                   1341: }
                   1342: 
1.161     pazsan   1343: static int cost_nexts(int prim)
1.109     anton    1344: {
                   1345:   return 1;
                   1346: }
                   1347: 
                   1348: typedef int Costfunc(int);
                   1349: Costfunc *ss_cost =  /* cost function for optimize_bb */
                   1350: #ifdef NO_DYNAMIC
                   1351: cost_lsu;
                   1352: #else
                   1353: cost_codesize;
                   1354: #endif
                   1355: 
1.110     anton    1356: struct {
                   1357:   Costfunc *costfunc;
                   1358:   char *metricname;
                   1359:   long sum;
                   1360: } cost_sums[] = {
                   1361: #ifndef NO_DYNAMIC
                   1362:   { cost_codesize, "codesize", 0 },
                   1363: #endif
                   1364:   { cost_ls,       "ls",       0 },
                   1365:   { cost_lsu,      "lsu",      0 },
                   1366:   { cost_nexts,    "nexts",    0 }
                   1367: };
                   1368: 
1.148     anton    1369: #ifndef NO_DYNAMIC
                   1370: void init_ss_cost(void) {
                   1371:   if (no_dynamic && ss_cost == cost_codesize) {
                   1372:     ss_cost = cost_nexts;
                   1373:     cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
                   1374:     debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
                   1375:   }
                   1376: }
                   1377: #endif
                   1378: 
1.106     anton    1379: #define MAX_BB 128 /* maximum number of instructions in BB */
1.125     anton    1380: #define INF_COST 1000000 /* infinite cost */
                   1381: #define CANONICAL_STATE 0
                   1382: 
                   1383: struct waypoint {
                   1384:   int cost;     /* the cost from here to the end */
                   1385:   PrimNum inst; /* the inst used from here to the next waypoint */
                   1386:   char relocatable; /* the last non-transition was relocatable */
                   1387:   char no_transition; /* don't use the next transition (relocatability)
                   1388:                       * or this transition (does not change state) */
                   1389: };
                   1390: 
1.156     anton    1391: struct tpa_state { /* tree parsing automaton (like) state */
1.155     anton    1392:   /* labeling is back-to-front */
                   1393:   struct waypoint *inst;  /* in front of instruction */
                   1394:   struct waypoint *trans; /* in front of instruction and transition */
                   1395: }; 
                   1396: 
1.156     anton    1397: struct tpa_state *termstate = NULL; /* initialized in loader() */
1.155     anton    1398: 
1.158     anton    1399: /* statistics about tree parsing (lazyburg) stuff */
                   1400: long lb_basic_blocks = 0;
                   1401: long lb_labeler_steps = 0;
                   1402: long lb_labeler_automaton = 0;
                   1403: long lb_labeler_dynprog = 0;
                   1404: long lb_newstate_equiv = 0;
                   1405: long lb_newstate_new = 0;
                   1406: long lb_applicable_base_rules = 0;
                   1407: long lb_applicable_chain_rules = 0;
                   1408: 
1.162     pazsan   1409: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1.161     pazsan   1410: static void init_waypoints(struct waypoint ws[])
1.125     anton    1411: {
                   1412:   int k;
                   1413: 
                   1414:   for (k=0; k<maxstates; k++)
                   1415:     ws[k].cost=INF_COST;
                   1416: }
1.106     anton    1417: 
1.161     pazsan   1418: static struct tpa_state *empty_tpa_state()
1.155     anton    1419: {
1.156     anton    1420:   struct tpa_state *s = malloc(sizeof(struct tpa_state));
1.155     anton    1421: 
1.157     anton    1422:   s->inst  = calloc(maxstates,sizeof(struct waypoint));
1.155     anton    1423:   init_waypoints(s->inst);
1.157     anton    1424:   s->trans = calloc(maxstates,sizeof(struct waypoint));
1.155     anton    1425:   /* init_waypoints(s->trans);*/
                   1426:   return s;
                   1427: }
                   1428: 
1.161     pazsan   1429: static void transitions(struct tpa_state *t)
1.107     anton    1430: {
1.125     anton    1431:   int k;
                   1432:   struct super_state *l;
                   1433:   
                   1434:   for (k=0; k<maxstates; k++) {
1.155     anton    1435:     t->trans[k] = t->inst[k];
                   1436:     t->trans[k].no_transition = 1;
1.125     anton    1437:   }
                   1438:   for (l = state_transitions; l != NULL; l = l->next) {
                   1439:     PrimNum s = l->super;
                   1440:     int jcost;
                   1441:     struct cost *c=super_costs+s;
1.155     anton    1442:     struct waypoint *wi=&(t->trans[c->state_in]);
                   1443:     struct waypoint *wo=&(t->inst[c->state_out]);
1.158     anton    1444:     lb_applicable_chain_rules++;
1.125     anton    1445:     if (wo->cost == INF_COST)
                   1446:       continue;
                   1447:     jcost = wo->cost + ss_cost(s);
                   1448:     if (jcost <= wi->cost) {
                   1449:       wi->cost = jcost;
                   1450:       wi->inst = s;
                   1451:       wi->relocatable = wo->relocatable;
                   1452:       wi->no_transition = 0;
                   1453:       /* if (ss_greedy) wi->cost = wo->cost ? */
                   1454:     }
                   1455:   }
                   1456: }
1.107     anton    1457: 
1.161     pazsan   1458: static struct tpa_state *make_termstate()
1.155     anton    1459: {
1.157     anton    1460:   struct tpa_state *s = empty_tpa_state();
1.155     anton    1461: 
                   1462:   s->inst[CANONICAL_STATE].cost = 0;
                   1463:   transitions(s);
                   1464:   return s;
                   1465: }
1.162     pazsan   1466: #endif
1.155     anton    1467: 
1.156     anton    1468: #define TPA_SIZE 16384
                   1469: 
                   1470: struct tpa_entry {
                   1471:   struct tpa_entry *next;
                   1472:   PrimNum inst;
                   1473:   struct tpa_state *state_behind;  /* note: brack-to-front labeling */
                   1474:   struct tpa_state *state_infront; /* note: brack-to-front labeling */
                   1475: } *tpa_table[TPA_SIZE];
                   1476: 
1.162     pazsan   1477: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1.161     pazsan   1478: static Cell hash_tpa(PrimNum p, struct tpa_state *t)
1.156     anton    1479: {
                   1480:   UCell it = (UCell )t;
                   1481:   return (p+it+(it>>14))&(TPA_SIZE-1);
                   1482: }
                   1483: 
1.161     pazsan   1484: static struct tpa_state **lookup_tpa(PrimNum p, struct tpa_state *t2)
1.156     anton    1485: {
                   1486:   int hash=hash_tpa(p, t2);
                   1487:   struct tpa_entry *te = tpa_table[hash];
                   1488: 
1.158     anton    1489:   if (tpa_noautomaton) {
                   1490:     static struct tpa_state *t;
                   1491:     t = NULL;
                   1492:     return &t;
                   1493:   }
1.156     anton    1494:   for (; te!=NULL; te = te->next) {
                   1495:     if (p == te->inst && t2 == te->state_behind)
                   1496:       return &(te->state_infront);
                   1497:   }
                   1498:   te = (struct tpa_entry *)malloc(sizeof(struct tpa_entry));
                   1499:   te->next = tpa_table[hash];
                   1500:   te->inst = p;
                   1501:   te->state_behind = t2;
                   1502:   te->state_infront = NULL;
                   1503:   tpa_table[hash] = te;
                   1504:   return &(te->state_infront);
                   1505: }
                   1506: 
1.161     pazsan   1507: static void tpa_state_normalize(struct tpa_state *t)
1.157     anton    1508: {
                   1509:   /* normalize so cost of canonical state=0; this may result in
                   1510:      negative states for some states */
                   1511:   int d = t->inst[CANONICAL_STATE].cost;
                   1512:   int i;
                   1513: 
                   1514:   for (i=0; i<maxstates; i++) {
                   1515:     if (t->inst[i].cost != INF_COST)
                   1516:       t->inst[i].cost -= d;
                   1517:     if (t->trans[i].cost != INF_COST)
                   1518:       t->trans[i].cost -= d;
                   1519:   }
                   1520: }
                   1521: 
1.161     pazsan   1522: static int tpa_state_equivalent(struct tpa_state *t1, struct tpa_state *t2)
1.157     anton    1523: {
                   1524:   return (memcmp(t1->inst, t2->inst, maxstates*sizeof(struct waypoint)) == 0 &&
                   1525:          memcmp(t1->trans,t2->trans,maxstates*sizeof(struct waypoint)) == 0);
                   1526: }
1.162     pazsan   1527: #endif
1.157     anton    1528: 
                   1529: struct tpa_state_entry {
                   1530:   struct tpa_state_entry *next;
                   1531:   struct tpa_state *state;
                   1532: } *tpa_state_table[TPA_SIZE];
                   1533: 
1.163     pazsan   1534: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1.161     pazsan   1535: static Cell hash_tpa_state(struct tpa_state *t)
1.157     anton    1536: {
                   1537:   int *ti = (int *)(t->inst);
                   1538:   int *tt = (int *)(t->trans);
                   1539:   int r=0;
                   1540:   int i;
                   1541: 
                   1542:   for (i=0; ti+i < (int *)(t->inst+maxstates); i++)
                   1543:     r += ti[i]+tt[i];
                   1544:   return (r+(r>>14)+(r>>22)) & (TPA_SIZE-1);
                   1545: }
                   1546: 
1.161     pazsan   1547: static struct tpa_state *lookup_tpa_state(struct tpa_state *t)
1.157     anton    1548: {
                   1549:   Cell hash = hash_tpa_state(t);
                   1550:   struct tpa_state_entry *te = tpa_state_table[hash];
                   1551:   struct tpa_state_entry *tn;
                   1552: 
1.158     anton    1553:   if (!tpa_noequiv) {
                   1554:     for (; te!=NULL; te = te->next) {
                   1555:       if (tpa_state_equivalent(t, te->state)) {
                   1556:        lb_newstate_equiv++;
                   1557:        free(t->inst);
                   1558:        free(t->trans);
                   1559:        free(t);
                   1560:        return te->state;
                   1561:       }
1.157     anton    1562:     }
1.158     anton    1563:     tn = (struct tpa_state_entry *)malloc(sizeof(struct tpa_state_entry));
                   1564:     tn->next = te;
                   1565:     tn->state = t;
                   1566:     tpa_state_table[hash] = tn;
                   1567:   }
                   1568:   lb_newstate_new++;
                   1569:   if (tpa_trace)
                   1570:     fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
1.157     anton    1571:   return t;
                   1572: }
                   1573: 
1.125     anton    1574: /* use dynamic programming to find the shortest paths within the basic
                   1575:    block origs[0..ninsts-1] and rewrite the instructions pointed to by
                   1576:    instps to use it */
1.161     pazsan   1577: static void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1.125     anton    1578: {
                   1579:   int i,j;
1.156     anton    1580:   struct tpa_state *ts[ninsts+1];
1.125     anton    1581:   int nextdyn, nextstate, no_transition;
                   1582:   
1.158     anton    1583:   lb_basic_blocks++;
1.155     anton    1584:   ts[ninsts] = termstate;
1.107     anton    1585:   for (i=ninsts-1; i>=0; i--) {
1.156     anton    1586:     struct tpa_state **tp = lookup_tpa(origs[i],ts[i+1]);
                   1587:     struct tpa_state *t = *tp;
1.158     anton    1588:     lb_labeler_steps++;
                   1589:     if (t) {
1.156     anton    1590:       ts[i] = t;
1.158     anton    1591:       lb_labeler_automaton++;
                   1592:     }
1.156     anton    1593:     else {
1.158     anton    1594:       lb_labeler_dynprog++;
1.156     anton    1595:       ts[i] = empty_tpa_state();
                   1596:       for (j=1; j<=max_super && i+j<=ninsts; j++) {
                   1597:        struct super_state **superp = lookup_super(origs+i, j);
                   1598:        if (superp!=NULL) {
                   1599:          struct super_state *supers = *superp;
                   1600:          for (; supers!=NULL; supers = supers->next) {
                   1601:            PrimNum s = supers->super;
                   1602:            int jcost;
                   1603:            struct cost *c=super_costs+s;
                   1604:            struct waypoint *wi=&(ts[i]->inst[c->state_in]);
                   1605:            struct waypoint *wo=&(ts[i+j]->trans[c->state_out]);
                   1606:            int no_transition = wo->no_transition;
1.158     anton    1607:            lb_applicable_base_rules++;
1.156     anton    1608:            if (!(is_relocatable(s)) && !wo->relocatable) {
                   1609:              wo=&(ts[i+j]->inst[c->state_out]);
                   1610:              no_transition=1;
                   1611:            }
                   1612:            if (wo->cost == INF_COST) 
                   1613:              continue;
                   1614:            jcost = wo->cost + ss_cost(s);
                   1615:            if (jcost <= wi->cost) {
                   1616:              wi->cost = jcost;
                   1617:              wi->inst = s;
                   1618:              wi->relocatable = is_relocatable(s);
                   1619:              wi->no_transition = no_transition;
                   1620:              /* if (ss_greedy) wi->cost = wo->cost ? */
                   1621:            }
1.125     anton    1622:          }
1.107     anton    1623:        }
                   1624:       }
1.156     anton    1625:       transitions(ts[i]);
1.157     anton    1626:       tpa_state_normalize(ts[i]);
                   1627:       *tp = ts[i] = lookup_tpa_state(ts[i]);
1.158     anton    1628:       if (tpa_trace)
                   1629:        fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1.107     anton    1630:     }
1.125     anton    1631:   }
                   1632:   /* now rewrite the instructions */
                   1633:   nextdyn=0;
                   1634:   nextstate=CANONICAL_STATE;
1.155     anton    1635:   no_transition = ((!ts[0]->trans[nextstate].relocatable) 
                   1636:                   ||ts[0]->trans[nextstate].no_transition);
1.125     anton    1637:   for (i=0; i<ninsts; i++) {
                   1638:     Cell tc=0, tc2;
                   1639:     if (i==nextdyn) {
                   1640:       if (!no_transition) {
                   1641:        /* process trans */
1.155     anton    1642:        PrimNum p = ts[i]->trans[nextstate].inst;
1.125     anton    1643:        struct cost *c = super_costs+p;
1.155     anton    1644:        assert(ts[i]->trans[nextstate].cost != INF_COST);
1.125     anton    1645:        assert(c->state_in==nextstate);
1.128     anton    1646:        tc = compile_prim_dyn(p,NULL);
1.125     anton    1647:        nextstate = c->state_out;
                   1648:       }
                   1649:       {
                   1650:        /* process inst */
1.155     anton    1651:        PrimNum p = ts[i]->inst[nextstate].inst;
1.125     anton    1652:        struct cost *c=super_costs+p;
                   1653:        assert(c->state_in==nextstate);
1.155     anton    1654:        assert(ts[i]->inst[nextstate].cost != INF_COST);
1.125     anton    1655: #if defined(GFORTH_DEBUGGING)
                   1656:        assert(p == origs[i]);
                   1657: #endif
1.128     anton    1658:        tc2 = compile_prim_dyn(p,instps[i]);
1.125     anton    1659:        if (no_transition || !is_relocatable(p))
                   1660:          /* !! actually what we care about is if and where
                   1661:           * compile_prim_dyn() puts NEXTs */
                   1662:          tc=tc2;
1.155     anton    1663:        no_transition = ts[i]->inst[nextstate].no_transition;
1.125     anton    1664:        nextstate = c->state_out;
                   1665:        nextdyn += c->length;
                   1666:       }
                   1667:     } else {
                   1668: #if defined(GFORTH_DEBUGGING)
                   1669:       assert(0);
                   1670: #endif
                   1671:       tc=0;
1.155     anton    1672:       /* tc= (Cell)vm_prims[ts[i]->inst[CANONICAL_STATE].inst]; */
1.125     anton    1673:     }
                   1674:     *(instps[i]) = tc;
                   1675:   }      
                   1676:   if (!no_transition) {
1.155     anton    1677:     PrimNum p = ts[i]->trans[nextstate].inst;
1.125     anton    1678:     struct cost *c = super_costs+p;
                   1679:     assert(c->state_in==nextstate);
1.155     anton    1680:     assert(ts[i]->trans[nextstate].cost != INF_COST);
1.125     anton    1681:     assert(i==nextdyn);
1.128     anton    1682:     (void)compile_prim_dyn(p,NULL);
1.125     anton    1683:     nextstate = c->state_out;
1.107     anton    1684:   }
1.125     anton    1685:   assert(nextstate==CANONICAL_STATE);
1.107     anton    1686: }
1.162     pazsan   1687: #endif
1.107     anton    1688: 
1.105     anton    1689: /* compile *start, possibly rewriting it into a static and/or dynamic
                   1690:    superinstruction */
                   1691: void compile_prim1(Cell *start)
1.70      anton    1692: {
1.108     anton    1693: #if defined(DOUBLY_INDIRECT)
1.125     anton    1694:   Label prim;
                   1695: 
                   1696:   if (start==NULL)
                   1697:     return;
                   1698:   prim = (Label)*start;
1.108     anton    1699:   if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
                   1700:     fprintf(stderr,"compile_prim encountered xt %p\n", prim);
                   1701:     *start=(Cell)prim;
                   1702:     return;
                   1703:   } else {
                   1704:     *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
                   1705:     return;
                   1706:   }
                   1707: #elif defined(INDIRECT_THREADED)
                   1708:   return;
1.112     anton    1709: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1.128     anton    1710:   /* !! does not work, for unknown reasons; but something like this is
                   1711:      probably needed to ensure that we don't call compile_prim_dyn
                   1712:      before the inline arguments are there */
                   1713:   static Cell *instps[MAX_BB];
                   1714:   static PrimNum origs[MAX_BB];
                   1715:   static int ninsts=0;
                   1716:   PrimNum prim_num;
                   1717: 
                   1718:   if (start==NULL || ninsts >= MAX_BB ||
                   1719:       (ninsts>0 && superend[origs[ninsts-1]])) {
                   1720:     /* after bb, or at the start of the next bb */
                   1721:     optimize_rewrite(instps,origs,ninsts);
                   1722:     /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
                   1723:     ninsts=0;
                   1724:     if (start==NULL)
                   1725:       return;
                   1726:   }
                   1727:   prim_num = ((Xt)*start)-vm_prims;
                   1728:   if(prim_num >= npriminfos) {
                   1729:     optimize_rewrite(instps,origs,ninsts);
1.129     anton    1730:     /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1.128     anton    1731:     ninsts=0;
                   1732:     return;
                   1733:   }    
                   1734:   assert(ninsts<MAX_BB);
                   1735:   instps[ninsts] = start;
                   1736:   origs[ninsts] = prim_num;
                   1737:   ninsts++;
1.112     anton    1738: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1.47      anton    1739: }
                   1740: 
1.161     pazsan   1741: Address gforth_loader(FILE *imagefile, char* filename)
1.1       anton    1742: /* returns the address of the image proper (after the preamble) */
                   1743: {
                   1744:   ImageHeader header;
                   1745:   Address image;
                   1746:   Address imp; /* image+preamble */
1.17      anton    1747:   Char magic[8];
                   1748:   char magic7; /* size byte of magic number */
1.1       anton    1749:   Cell preamblesize=0;
1.6       pazsan   1750:   Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1.1       anton    1751:   UCell check_sum;
1.15      pazsan   1752:   Cell ausize = ((RELINFOBITS ==  8) ? 0 :
                   1753:                 (RELINFOBITS == 16) ? 1 :
                   1754:                 (RELINFOBITS == 32) ? 2 : 3);
                   1755:   Cell charsize = ((sizeof(Char) == 1) ? 0 :
                   1756:                   (sizeof(Char) == 2) ? 1 :
                   1757:                   (sizeof(Char) == 4) ? 2 : 3) + ausize;
                   1758:   Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
                   1759:                   (sizeof(Cell) == 2) ? 1 :
                   1760:                   (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1.21      anton    1761:   Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
                   1762: #ifdef WORDS_BIGENDIAN
                   1763:        0
                   1764: #else
                   1765:        1
                   1766: #endif
                   1767:     ;
1.1       anton    1768: 
1.164     pazsan   1769:   vm_prims = gforth_engine(0,0,0,0,0);
1.47      anton    1770:   check_prims(vm_prims);
1.106     anton    1771:   prepare_super_table();
1.1       anton    1772: #ifndef DOUBLY_INDIRECT
1.59      anton    1773: #ifdef PRINT_SUPER_LENGTHS
                   1774:   print_super_lengths();
                   1775: #endif
1.43      anton    1776:   check_sum = checksum(vm_prims);
1.1       anton    1777: #else /* defined(DOUBLY_INDIRECT) */
1.43      anton    1778:   check_sum = (UCell)vm_prims;
1.1       anton    1779: #endif /* defined(DOUBLY_INDIRECT) */
1.155     anton    1780: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
                   1781:   termstate = make_termstate();
                   1782: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1.10      pazsan   1783:   
                   1784:   do {
                   1785:     if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1.84      anton    1786:       fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1.10      pazsan   1787:              progname, filename);
                   1788:       exit(1);
1.1       anton    1789:     }
1.10      pazsan   1790:     preamblesize+=8;
1.84      anton    1791:   } while(memcmp(magic,"Gforth3",7));
1.17      anton    1792:   magic7 = magic[7];
1.1       anton    1793:   if (debug) {
1.17      anton    1794:     magic[7]='\0';
1.21      anton    1795:     fprintf(stderr,"Magic found: %s ", magic);
                   1796:     print_sizes(magic7);
1.1       anton    1797:   }
                   1798: 
1.21      anton    1799:   if (magic7 != sizebyte)
                   1800:     {
                   1801:       fprintf(stderr,"This image is:         ");
                   1802:       print_sizes(magic7);
                   1803:       fprintf(stderr,"whereas the machine is ");
                   1804:       print_sizes(sizebyte);
1.1       anton    1805:       exit(-2);
                   1806:     };
                   1807: 
                   1808:   fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1.10      pazsan   1809: 
                   1810:   set_stack_sizes(&header);
1.1       anton    1811:   
                   1812: #if HAVE_GETPAGESIZE
                   1813:   pagesize=getpagesize(); /* Linux/GNU libc offers this */
                   1814: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
                   1815:   pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
                   1816: #elif PAGESIZE
                   1817:   pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
                   1818: #endif
1.144     pazsan   1819:   debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1.1       anton    1820: 
1.34      anton    1821:   image = dict_alloc_read(imagefile, preamblesize+header.image_size,
                   1822:                          preamblesize+dictsize, data_offset);
1.33      anton    1823:   imp=image+preamblesize;
1.57      anton    1824:   alloc_stacks((ImageHeader *)imp);
1.1       anton    1825:   if (clear_dictionary)
1.33      anton    1826:     memset(imp+header.image_size, 0, dictsize-header.image_size);
1.90      anton    1827:   if(header.base==0 || header.base  == (Address)0x100) {
1.1       anton    1828:     Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1.162     pazsan   1829:     Char reloc_bits[reloc_size];
1.33      anton    1830:     fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1.10      pazsan   1831:     fread(reloc_bits, 1, reloc_size, imagefile);
1.161     pazsan   1832:     gforth_relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1.1       anton    1833: #if 0
                   1834:     { /* let's see what the relocator did */
                   1835:       FILE *snapshot=fopen("snapshot.fi","wb");
                   1836:       fwrite(image,1,imagesize,snapshot);
                   1837:       fclose(snapshot);
                   1838:     }
                   1839: #endif
1.46      jwilke   1840:   }
                   1841:   else if(header.base!=imp) {
                   1842:     fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
                   1843:            progname, (unsigned long)header.base, (unsigned long)imp);
                   1844:     exit(1);
1.1       anton    1845:   }
                   1846:   if (header.checksum==0)
                   1847:     ((ImageHeader *)imp)->checksum=check_sum;
                   1848:   else if (header.checksum != check_sum) {
                   1849:     fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
                   1850:            progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
                   1851:     exit(1);
                   1852:   }
1.53      anton    1853: #ifdef DOUBLY_INDIRECT
                   1854:   ((ImageHeader *)imp)->xt_base = xts;
                   1855: #endif
1.1       anton    1856:   fclose(imagefile);
                   1857: 
1.56      anton    1858:   /* unnecessary, except maybe for CODE words */
                   1859:   /* FLUSH_ICACHE(imp, header.image_size);*/
1.1       anton    1860: 
                   1861:   return imp;
                   1862: }
                   1863: 
1.72      anton    1864: /* pointer to last '/' or '\' in file, 0 if there is none. */
1.161     pazsan   1865: static char *onlypath(char *filename)
1.10      pazsan   1866: {
1.72      anton    1867:   return strrchr(filename, DIRSEP);
1.1       anton    1868: }
                   1869: 
1.161     pazsan   1870: static FILE *openimage(char *fullfilename)
1.10      pazsan   1871: {
                   1872:   FILE *image_file;
1.162     pazsan   1873:   char * expfilename = tilde_cstr((Char *)fullfilename, strlen(fullfilename), 1);
1.10      pazsan   1874: 
1.28      anton    1875:   image_file=fopen(expfilename,"rb");
1.1       anton    1876:   if (image_file!=NULL && debug)
1.28      anton    1877:     fprintf(stderr, "Opened image file: %s\n", expfilename);
1.10      pazsan   1878:   return image_file;
1.1       anton    1879: }
                   1880: 
1.28      anton    1881: /* try to open image file concat(path[0:len],imagename) */
1.161     pazsan   1882: static FILE *checkimage(char *path, int len, char *imagename)
1.10      pazsan   1883: {
                   1884:   int dirlen=len;
1.162     pazsan   1885:   char fullfilename[dirlen+strlen((char *)imagename)+2];
1.10      pazsan   1886: 
1.1       anton    1887:   memcpy(fullfilename, path, dirlen);
1.71      pazsan   1888:   if (fullfilename[dirlen-1]!=DIRSEP)
                   1889:     fullfilename[dirlen++]=DIRSEP;
1.1       anton    1890:   strcpy(fullfilename+dirlen,imagename);
1.10      pazsan   1891:   return openimage(fullfilename);
1.1       anton    1892: }
                   1893: 
1.161     pazsan   1894: static FILE * open_image_file(char * imagename, char * path)
1.1       anton    1895: {
1.10      pazsan   1896:   FILE * image_file=NULL;
1.28      anton    1897:   char *origpath=path;
1.10      pazsan   1898:   
1.71      pazsan   1899:   if(strchr(imagename, DIRSEP)==NULL) {
1.10      pazsan   1900:     /* first check the directory where the exe file is in !! 01may97jaw */
                   1901:     if (onlypath(progname))
1.72      anton    1902:       image_file=checkimage(progname, onlypath(progname)-progname, imagename);
1.10      pazsan   1903:     if (!image_file)
                   1904:       do {
                   1905:        char *pend=strchr(path, PATHSEP);
                   1906:        if (pend==NULL)
                   1907:          pend=path+strlen(path);
                   1908:        if (strlen(path)==0) break;
                   1909:        image_file=checkimage(path, pend-path, imagename);
                   1910:        path=pend+(*pend==PATHSEP);
                   1911:       } while (image_file==NULL);
                   1912:   } else {
                   1913:     image_file=openimage(imagename);
                   1914:   }
1.1       anton    1915: 
1.10      pazsan   1916:   if (!image_file) {
                   1917:     fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
1.28      anton    1918:            progname, imagename, origpath);
1.10      pazsan   1919:     exit(1);
1.7       anton    1920:   }
                   1921: 
1.10      pazsan   1922:   return image_file;
                   1923: }
1.11      pazsan   1924: #endif
                   1925: 
                   1926: #ifdef HAS_OS
1.161     pazsan   1927: static UCell convsize(char *s, UCell elemsize)
1.11      pazsan   1928: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
                   1929:    of bytes.  the letter at the end indicates the unit, where e stands
                   1930:    for the element size. default is e */
                   1931: {
                   1932:   char *endp;
                   1933:   UCell n,m;
                   1934: 
                   1935:   m = elemsize;
                   1936:   n = strtoul(s,&endp,0);
                   1937:   if (endp!=NULL) {
                   1938:     if (strcmp(endp,"b")==0)
                   1939:       m=1;
                   1940:     else if (strcmp(endp,"k")==0)
                   1941:       m=1024;
                   1942:     else if (strcmp(endp,"M")==0)
                   1943:       m=1024*1024;
                   1944:     else if (strcmp(endp,"G")==0)
                   1945:       m=1024*1024*1024;
                   1946:     else if (strcmp(endp,"T")==0) {
                   1947: #if (SIZEOF_CHAR_P > 4)
1.24      anton    1948:       m=1024L*1024*1024*1024;
1.11      pazsan   1949: #else
                   1950:       fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
                   1951:       exit(1);
                   1952: #endif
                   1953:     } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
                   1954:       fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
                   1955:       exit(1);
                   1956:     }
                   1957:   }
                   1958:   return n*m;
                   1959: }
1.10      pazsan   1960: 
1.109     anton    1961: enum {
                   1962:   ss_number = 256,
1.125     anton    1963:   ss_states,
1.109     anton    1964:   ss_min_codesize,
                   1965:   ss_min_ls,
                   1966:   ss_min_lsu,
                   1967:   ss_min_nexts,
                   1968: };
                   1969: 
1.10      pazsan   1970: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
                   1971: {
                   1972:   int c;
                   1973: 
1.1       anton    1974:   opterr=0;
                   1975:   while (1) {
                   1976:     int option_index=0;
                   1977:     static struct option opts[] = {
1.29      anton    1978:       {"appl-image", required_argument, NULL, 'a'},
1.1       anton    1979:       {"image-file", required_argument, NULL, 'i'},
                   1980:       {"dictionary-size", required_argument, NULL, 'm'},
                   1981:       {"data-stack-size", required_argument, NULL, 'd'},
                   1982:       {"return-stack-size", required_argument, NULL, 'r'},
                   1983:       {"fp-stack-size", required_argument, NULL, 'f'},
                   1984:       {"locals-stack-size", required_argument, NULL, 'l'},
                   1985:       {"path", required_argument, NULL, 'p'},
                   1986:       {"version", no_argument, NULL, 'v'},
                   1987:       {"help", no_argument, NULL, 'h'},
                   1988:       /* put something != 0 into offset_image */
                   1989:       {"offset-image", no_argument, &offset_image, 1},
                   1990:       {"no-offset-im", no_argument, &offset_image, 0},
                   1991:       {"clear-dictionary", no_argument, &clear_dictionary, 1},
1.4       anton    1992:       {"die-on-signal", no_argument, &die_on_signal, 1},
1.169     anton    1993:       {"ignore-async-signals", no_argument, &ignore_async_signals, 1},
1.1       anton    1994:       {"debug", no_argument, &debug, 1},
1.144     pazsan   1995:       {"diag", no_argument, &diag, 1},
1.60      anton    1996:       {"no-super", no_argument, &no_super, 1},
                   1997:       {"no-dynamic", no_argument, &no_dynamic, 1},
1.66      anton    1998:       {"dynamic", no_argument, &no_dynamic, 0},
1.110     anton    1999:       {"print-metrics", no_argument, &print_metrics, 1},
1.109     anton    2000:       {"ss-number", required_argument, NULL, ss_number},
1.125     anton    2001:       {"ss-states", required_argument, NULL, ss_states},
1.109     anton    2002: #ifndef NO_DYNAMIC
                   2003:       {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
                   2004: #endif
                   2005:       {"ss-min-ls",       no_argument, NULL, ss_min_ls},
                   2006:       {"ss-min-lsu",      no_argument, NULL, ss_min_lsu},
                   2007:       {"ss-min-nexts",    no_argument, NULL, ss_min_nexts},
1.110     anton    2008:       {"ss-greedy",       no_argument, &ss_greedy, 1},
1.158     anton    2009:       {"tpa-noequiv",     no_argument, &tpa_noequiv, 1},
                   2010:       {"tpa-noautomaton", no_argument, &tpa_noautomaton, 1},
                   2011:       {"tpa-trace",      no_argument, &tpa_trace, 1},
1.1       anton    2012:       {0,0,0,0}
                   2013:       /* no-init-file, no-rc? */
                   2014:     };
                   2015:     
1.36      pazsan   2016:     c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
1.1       anton    2017:     
                   2018:     switch (c) {
1.29      anton    2019:     case EOF: return;
                   2020:     case '?': optind--; return;
                   2021:     case 'a': *imagename = optarg; return;
1.10      pazsan   2022:     case 'i': *imagename = optarg; break;
1.1       anton    2023:     case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
                   2024:     case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
                   2025:     case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
                   2026:     case 'f': fsize = convsize(optarg,sizeof(Float)); break;
                   2027:     case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
1.10      pazsan   2028:     case 'p': *path = optarg; break;
1.36      pazsan   2029:     case 'o': offset_image = 1; break;
                   2030:     case 'n': offset_image = 0; break;
                   2031:     case 'c': clear_dictionary = 1; break;
                   2032:     case 's': die_on_signal = 1; break;
                   2033:     case 'x': debug = 1; break;
1.83      anton    2034:     case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
1.109     anton    2035:     case ss_number: static_super_number = atoi(optarg); break;
1.125     anton    2036:     case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
1.109     anton    2037: #ifndef NO_DYNAMIC
                   2038:     case ss_min_codesize: ss_cost = cost_codesize; break;
                   2039: #endif
                   2040:     case ss_min_ls:       ss_cost = cost_ls;       break;
                   2041:     case ss_min_lsu:      ss_cost = cost_lsu;      break;
                   2042:     case ss_min_nexts:    ss_cost = cost_nexts;    break;
1.1       anton    2043:     case 'h': 
1.29      anton    2044:       fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
1.1       anton    2045: Engine Options:\n\
1.29      anton    2046:   --appl-image FILE                equivalent to '--image-file=FILE --'\n\
1.10      pazsan   2047:   --clear-dictionary               Initialize the dictionary with 0 bytes\n\
                   2048:   -d SIZE, --data-stack-size=SIZE   Specify data stack size\n\
                   2049:   --debug                          Print debugging information during startup\n\
1.144     pazsan   2050:   --diag                           Print diagnostic information during startup\n\
1.169     anton    2051:   --die-on-signal                  exit instead of THROWing some signals\n\
1.66      anton    2052:   --dynamic                        use dynamic native code\n\
1.10      pazsan   2053:   -f SIZE, --fp-stack-size=SIZE            Specify floating point stack size\n\
                   2054:   -h, --help                       Print this message and exit\n\
1.169     anton    2055:   --ignore-async-signals            ignore instead of THROWing async. signals\n\
1.10      pazsan   2056:   -i FILE, --image-file=FILE       Use image FILE instead of `gforth.fi'\n\
                   2057:   -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
                   2058:   -m SIZE, --dictionary-size=SIZE   Specify Forth dictionary size\n\
1.60      anton    2059:   --no-dynamic                     Use only statically compiled primitives\n\
1.10      pazsan   2060:   --no-offset-im                   Load image at normal position\n\
1.60      anton    2061:   --no-super                        No dynamically formed superinstructions\n\
1.10      pazsan   2062:   --offset-image                   Load image at a different position\n\
                   2063:   -p PATH, --path=PATH             Search path for finding image and sources\n\
1.110     anton    2064:   --print-metrics                  Print some code generation metrics on exit\n\
1.10      pazsan   2065:   -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
1.111     anton    2066:   --ss-greedy                       greedy, not optimal superinst selection\n\
                   2067:   --ss-min-codesize                 select superinsts for smallest native code\n\
                   2068:   --ss-min-ls                       minimize loads and stores\n\
                   2069:   --ss-min-lsu                      minimize loads, stores, and pointer updates\n\
                   2070:   --ss-min-nexts                    minimize the number of static superinsts\n\
                   2071:   --ss-number=N                     use N static superinsts (default max)\n\
1.125     anton    2072:   --ss-states=N                     N states for stack caching (default max)\n\
1.158     anton    2073:   --tpa-noequiv                     automaton without state equivalence\n\
                   2074:   --tpa-noautomaton                 dynamic programming only\n\
                   2075:   --tpa-trace                       report new states etc.\n\
1.66      anton    2076:   -v, --version                            Print engine version and exit\n\
1.1       anton    2077: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
1.10      pazsan   2078:   `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
                   2079:              argv[0]);
                   2080:       optind--;
                   2081:       return;
1.1       anton    2082:     }
                   2083:   }
1.10      pazsan   2084: }
1.11      pazsan   2085: #endif
1.10      pazsan   2086: 
1.161     pazsan   2087: static void print_diag()
1.144     pazsan   2088: {
                   2089: 
1.153     pazsan   2090: #if !defined(HAVE_GETRUSAGE) || (!defined(HAS_FFCALL) && !defined(HAS_LIBFFI))
1.145     pazsan   2091:   fprintf(stderr, "*** missing functionality ***\n"
1.144     pazsan   2092: #ifndef HAVE_GETRUSAGE
                   2093:          "    no getrusage -> CPUTIME broken\n"
                   2094: #endif
1.153     pazsan   2095: #if !defined(HAS_FFCALL) && !defined(HAS_LIBFFI)
1.144     pazsan   2096:          "    no ffcall -> only old-style foreign function calls (no fflib.fs)\n"
                   2097: #endif
                   2098:          );
                   2099: #endif
                   2100:   if((relocs < nonrelocs) ||
                   2101: #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)
                   2102:      1
                   2103: #else
                   2104:      0
                   2105: #endif
                   2106:      )
                   2107:     debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
1.165     pazsan   2108:     fprintf(stderr, "*** %sperformance problems ***\n%s",
                   2109: #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) || !defined(FORCE_REG) || defined(BUGGY_LONG_LONG)
                   2110:            "",
                   2111: #else
                   2112:            "no ",
                   2113: #endif
1.144     pazsan   2114: #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)
                   2115:            "    double-cell integer type buggy ->\n        "
                   2116: #ifdef BUGGY_LL_CMP
                   2117:            "CMP, "
                   2118: #endif
                   2119: #ifdef BUGGY_LL_MUL
                   2120:            "MUL, "
                   2121: #endif
                   2122: #ifdef BUGGY_LL_DIV
                   2123:            "DIV, "
                   2124: #endif
                   2125: #ifdef BUGGY_LL_ADD
                   2126:            "ADD, "
                   2127: #endif
                   2128: #ifdef BUGGY_LL_SHIFT
                   2129:            "SHIFT, "
                   2130: #endif
                   2131: #ifdef BUGGY_LL_D2F
                   2132:            "D2F, "
                   2133: #endif
                   2134: #ifdef BUGGY_LL_F2D
                   2135:            "F2D, "
                   2136: #endif
                   2137:            "\b\b slow\n"
1.145     pazsan   2138: #endif
                   2139: #ifndef FORCE_REG
                   2140:            "    automatic register allocation: performance degradation possible\n"
                   2141: #endif
                   2142: #if !defined(FORCE_REG) || defined(BUGGY_LONG_LONG)
                   2143:            "*** Suggested remedy: try ./configure"
                   2144: #ifndef FORCE_REG
                   2145:            " --enable-force-reg"
                   2146: #endif
                   2147: #ifdef BUGGY_LONG_LONG
                   2148:            " --enable-force-ll"
                   2149: #endif
                   2150:            "\n"
1.166     pazsan   2151: #else
                   2152:            ""
1.144     pazsan   2153: #endif
                   2154:            ,
                   2155:            (relocs < nonrelocs) ? "    gcc PR 15242 -> no dynamic code generation (use gcc-2.95 instead)\n" : "");
                   2156: }
                   2157: 
1.10      pazsan   2158: #ifdef INCLUDE_IMAGE
                   2159: extern Cell image[];
                   2160: extern const char reloc_bits[];
                   2161: #endif
1.67      pazsan   2162: 
1.10      pazsan   2163: int main(int argc, char **argv, char **env)
                   2164: {
1.30      pazsan   2165: #ifdef HAS_OS
1.10      pazsan   2166:   char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
1.30      pazsan   2167: #else
                   2168:   char *path = DEFAULTPATH;
                   2169: #endif
1.13      pazsan   2170: #ifndef INCLUDE_IMAGE
1.10      pazsan   2171:   char *imagename="gforth.fi";
                   2172:   FILE *image_file;
                   2173:   Address image;
                   2174: #endif
                   2175:   int retvalue;
                   2176:          
1.56      anton    2177: #if defined(i386) && defined(ALIGNMENT_CHECK)
1.10      pazsan   2178:   /* turn on alignment checks on the 486.
                   2179:    * on the 386 this should have no effect. */
                   2180:   __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
                   2181:   /* this is unusable with Linux' libc.4.6.27, because this library is
                   2182:      not alignment-clean; we would have to replace some library
                   2183:      functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
                   2184:      the stack FP-aligned. */
                   2185: #endif
                   2186: 
                   2187:   /* buffering of the user output device */
1.11      pazsan   2188: #ifdef _IONBF
1.10      pazsan   2189:   if (isatty(fileno(stdout))) {
                   2190:     fflush(stdout);
                   2191:     setvbuf(stdout,NULL,_IONBF,0);
1.1       anton    2192:   }
1.11      pazsan   2193: #endif
1.1       anton    2194: 
1.10      pazsan   2195:   progname = argv[0];
                   2196: 
1.11      pazsan   2197: #ifdef HAS_OS
1.10      pazsan   2198:   gforth_args(argc, argv, &path, &imagename);
1.109     anton    2199: #ifndef NO_DYNAMIC
1.148     anton    2200:   init_ss_cost();
1.109     anton    2201: #endif /* !defined(NO_DYNAMIC) */
                   2202: #endif /* defined(HAS_OS) */
1.10      pazsan   2203: 
                   2204: #ifdef INCLUDE_IMAGE
                   2205:   set_stack_sizes((ImageHeader *)image);
1.22      pazsan   2206:   if(((ImageHeader *)image)->base != image)
1.161     pazsan   2207:     gforth_relocate(image, reloc_bits, ((ImageHeader *)image)->image_size,
1.164     pazsan   2208:                    (Label*)gforth_engine(0, 0, 0, 0, 0));
1.10      pazsan   2209:   alloc_stacks((ImageHeader *)image);
                   2210: #else
                   2211:   image_file = open_image_file(imagename, path);
1.161     pazsan   2212:   image = gforth_loader(image_file, imagename);
1.10      pazsan   2213: #endif
1.24      anton    2214:   gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
1.1       anton    2215: 
1.144     pazsan   2216:   if (diag)
                   2217:     print_diag();
1.1       anton    2218:   {
1.10      pazsan   2219:     char path2[strlen(path)+1];
1.1       anton    2220:     char *p1, *p2;
                   2221:     Cell environ[]= {
                   2222:       (Cell)argc-(optind-1),
                   2223:       (Cell)(argv+(optind-1)),
1.10      pazsan   2224:       (Cell)strlen(path),
1.1       anton    2225:       (Cell)path2};
                   2226:     argv[optind-1] = progname;
                   2227:     /*
                   2228:        for (i=0; i<environ[0]; i++)
                   2229:        printf("%s\n", ((char **)(environ[1]))[i]);
                   2230:        */
                   2231:     /* make path OS-independent by replacing path separators with NUL */
1.10      pazsan   2232:     for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
1.1       anton    2233:       if (*p1==PATHSEP)
                   2234:        *p2 = '\0';
                   2235:       else
                   2236:        *p2 = *p1;
                   2237:     *p2='\0';
1.161     pazsan   2238:     retvalue = gforth_go(image, 4, environ);
1.102     anton    2239: #ifdef SIGPIPE
                   2240:     bsd_signal(SIGPIPE, SIG_IGN);
                   2241: #endif
1.42      anton    2242: #ifdef VM_PROFILING
                   2243:     vm_print_profile(stderr);
                   2244: #endif
1.1       anton    2245:     deprep_terminal();
1.104     anton    2246:   }
1.110     anton    2247:   if (print_metrics) {
                   2248:     int i;
                   2249:     fprintf(stderr, "code size = %8ld\n", dyncodesize());
                   2250:     for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
                   2251:       fprintf(stderr, "metric %8s: %8ld\n",
                   2252:              cost_sums[i].metricname, cost_sums[i].sum);
1.158     anton    2253:     fprintf(stderr,"lb_basic_blocks = %ld\n", lb_basic_blocks);
                   2254:     fprintf(stderr,"lb_labeler_steps = %ld\n", lb_labeler_steps);
                   2255:     fprintf(stderr,"lb_labeler_automaton = %ld\n", lb_labeler_automaton);
                   2256:     fprintf(stderr,"lb_labeler_dynprog = %ld\n", lb_labeler_dynprog);
                   2257:     fprintf(stderr,"lb_newstate_equiv = %ld\n", lb_newstate_equiv);
                   2258:     fprintf(stderr,"lb_newstate_new = %ld\n", lb_newstate_new);
                   2259:     fprintf(stderr,"lb_applicable_base_rules = %ld\n", lb_applicable_base_rules);
                   2260:     fprintf(stderr,"lb_applicable_chain_rules = %ld\n", lb_applicable_chain_rules);
                   2261:   }
                   2262:   if (tpa_trace) {
                   2263:     fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
                   2264:     fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1.1       anton    2265:   }
1.13      pazsan   2266:   return retvalue;
1.1       anton    2267: }

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