File:  [gforth] / gforth / engine / threaded.h
Revision 1.15: download - view: text, annotated - select for diffs
Sun Jan 20 19:04:11 2002 UTC (18 years, 1 month ago) by anton
Branches: MAIN
CVS tags: HEAD
hybrid direct/indirect threading (replaces direct threading)

    1: /* This file defines a number of threading schemes.
    2: 
    3:   Copyright (C) 1995, 1996,1997,1999 Free Software Foundation, Inc.
    4: 
    5:   This file is part of Gforth.
    6: 
    7:   Gforth is free software; you can redistribute it and/or
    8:   modify it under the terms of the GNU General Public License
    9:   as published by the Free Software Foundation; either version 2
   10:   of the License, or (at your option) any later version.
   11: 
   12:   This program is distributed in the hope that it will be useful,
   13:   but WITHOUT ANY WARRANTY; without even the implied warranty of
   14:   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   15:   GNU General Public License for more details.
   16: 
   17:   You should have received a copy of the GNU General Public License
   18:   along with this program; if not, write to the Free Software
   19:   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
   20: 
   21: 
   22:   This files defines macros for threading. Many sets of macros are
   23:   defined. Functionally they have only one difference: Some implement
   24:   direct threading, some indirect threading. The other differences are
   25:   just variations to help GCC generate faster code for various
   26:   machines.
   27: 
   28:   (Well, to tell the truth, there actually is another functional
   29:   difference in some pathological cases: e.g., a '!' stores into the
   30:   cell where the next executed word comes from; or, the next word
   31:   executed comes from the top-of-stack. These differences are one of
   32:   the reasons why GCC cannot produce the right variation by itself. We
   33:   chose disallowing such practices and using the added implementation
   34:   freedom to achieve a significant speedup, because these practices
   35:   are not common in Forth (I have never heard of or seen anyone using
   36:   them), and it is easy to circumvent problems: A control flow change
   37:   will flush any prefetched words; you may want to do a "0
   38:   drop" before that to write back the top-of-stack cache.)
   39: 
   40:   These macro sets are used in the following ways: After translation
   41:   to C a typical primitive looks like
   42: 
   43:   ...
   44:   {
   45:   DEF_CA
   46:   other declarations
   47:   NEXT_P0;
   48:   main part of the primitive
   49:   NEXT_P1;
   50:   store results to stack
   51:   NEXT_P2;
   52:   }
   53: 
   54:   DEF_CA and all the NEXT_P* together must implement NEXT; In the main
   55:   part the instruction pointer can be read with IP, changed with
   56:   INC_IP(const_inc), and the cell right behind the presently executing
   57:   word (i.e. the value of *IP) is accessed with NEXT_INST.
   58: 
   59:   If a primitive does not fall through the main part, it has to do the
   60:   rest by itself. If it changes ip, it has to redo NEXT_P0 (perhaps we
   61:   should define a macro SET_IP).
   62: 
   63:   Some primitives (execute, dodefer) do not end with NEXT, but with
   64:   EXEC(.). If NEXT_P0 has been called earlier, it has to perform
   65:   "ip=IP;" to ensure that ip has the right value (NEXT_P0 may change
   66:   it).
   67: 
   68:   Finally, there is NEXT1_P1 and NEXT1_P2, which are parts of EXEC
   69:   (EXEC(XT) could be defined as "cfa=XT; NEXT1_P1; NEXT1_P2;" (is this
   70:   true?)) and are used for making docol faster.
   71: 
   72:   We can define the ways in which these macros are used with a regular
   73:   expression:
   74: 
   75:   For a primitive
   76: 
   77:   DEF_CA NEXT_P0 ( IP | INC_IP | NEXT_INST | ip=...; NEXT_P0 ) * ( NEXT_P1 NEXT_P2 | EXEC(...) )
   78: 
   79:   For a run-time routine, e.g., docol:
   80:   PFA1(cfa) ( NEXT_P0 NEXT | cfa=...; NEXT1_P1; NEXT1_P2 | EXEC(...) )
   81: 
   82:   This comment does not yet describe all the dependences that the
   83:   macros have to satisfy.
   84: 
   85:   To organize the former ifdef chaos, each path is separated
   86:   This gives a quite impressive number of paths, but you clearly
   87:   find things that go together.
   88: 
   89:   It should be possible to organize the whole thing in a way that
   90:   contains less redundancy and allows a simpler description.
   91: 
   92: */
   93: 
   94: #ifdef DOUBLY_INDIRECT
   95: #  define NEXT_P0	({cfa=*ip;})
   96: #  define IP		(ip)
   97: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
   98: #  define NEXT_INST	(cfa)
   99: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  100: #  define DEF_CA	Label ca;
  101: #  define NEXT_P1	({\
  102:   if (cfa<=vm_prims+DOESJUMP || cfa>=vm_prims+npriminfos) \
  103:     fprintf(stderr,"NEXT encountered prim %p at ip=%p\n", cfa, ip); \
  104:   ip++; ca=**cfa;})
  105: #  define NEXT_P2	({goto *ca;})
  106: #  define EXEC(XT)	({DEF_CA cfa=(XT);\
  107:   if (cfa>vm_prims+DOESJUMP && cfa<vm_prims+npriminfos) \
  108:     fprintf(stderr,"EXEC encountered xt %p at ip=%p, vm_prims=%p, xts=%p\n", cfa, ip, vm_prims, xts); \
  109:  ca=**cfa; goto *ca;})
  110: 
  111: #else /* !defined(DOUBLY_INDIRECT) */
  112: 
  113: #if defined(DIRECT_THREADED)
  114: 
  115: #if THREADING_SCHEME==1
  116: #warning direct threading scheme 1: autoinc, long latency, cfa live
  117: #  define NEXT_P0	({cfa=*ip++;})
  118: #  define IP		(ip-1)
  119: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  120: #  define NEXT_INST	(cfa)
  121: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  122: #  define DEF_CA
  123: #  define NEXT_P1
  124: #  define NEXT_P2	({goto *cfa;})
  125: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  126: #endif
  127: 
  128: #if THREADING_SCHEME==2
  129: #warning direct threading scheme 2: autoinc, long latency, cfa dead
  130: #  define NEXT_P0	(ip++)
  131: #  define IP		(ip-1)
  132: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  133: #  define NEXT_INST	(*(ip-1))
  134: #  define INC_IP(const_inc)	({ ip+=(const_inc);})
  135: #  define DEF_CA
  136: #  define NEXT_P1
  137: #  define NEXT_P2	({goto **(ip-1);})
  138: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  139: #endif
  140: 
  141: 
  142: #if THREADING_SCHEME==3
  143: #warning direct threading scheme 3: autoinc, low latency, cfa live
  144: #  define NEXT_P0
  145: #  define IP		(ip)
  146: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  147: #  define NEXT_INST	(*ip)
  148: #  define INC_IP(const_inc)	({ip+=(const_inc);})
  149: #  define DEF_CA
  150: #  define NEXT_P1	({cfa=*ip++;})
  151: #  define NEXT_P2	({goto *cfa;})
  152: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  153: #endif
  154: 
  155: #if THREADING_SCHEME==4
  156: #warning direct threading scheme 4: autoinc, low latency, cfa dead
  157: #  define NEXT_P0
  158: #  define IP		(ip)
  159: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  160: #  define NEXT_INST	(*ip)
  161: #  define INC_IP(const_inc)	({ ip+=(const_inc);})
  162: #  define DEF_CA
  163: #  define NEXT_P1
  164: #  define NEXT_P2	({goto **(ip++);})
  165: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  166: #endif
  167: 
  168: #if THREADING_SCHEME==5
  169: #warning direct threading scheme 5: long latency, cfa live
  170: #  define NEXT_P0	({cfa=*ip;})
  171: #  define IP		(ip)
  172: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  173: #  define NEXT_INST	(cfa)
  174: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  175: #  define DEF_CA
  176: #  define NEXT_P1	(ip++)
  177: #  define NEXT_P2	({goto *cfa;})
  178: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  179: #endif
  180: 
  181: #if THREADING_SCHEME==6
  182: #warning direct threading scheme 6: long latency, cfa dead
  183: #  define NEXT_P0
  184: #  define IP		(ip)
  185: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  186: #  define NEXT_INST	(*ip)
  187: #  define INC_IP(const_inc)	({ip+=(const_inc);})
  188: #  define DEF_CA
  189: #  define NEXT_P1	(ip++)
  190: #  define NEXT_P2	({goto **(ip-1);})
  191: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  192: #endif
  193: 
  194: 
  195: #if THREADING_SCHEME==7
  196: #warning direct threading scheme 7: low latency, cfa live
  197: #  define NEXT_P0
  198: #  define IP		(ip)
  199: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  200: #  define NEXT_INST	(*ip)
  201: #  define INC_IP(const_inc)	({ip+=(const_inc);})
  202: #  define DEF_CA
  203: #  define NEXT_P1	({cfa=*ip++;})
  204: #  define NEXT_P2	({goto *cfa;})
  205: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  206: #endif
  207: 
  208: #if THREADING_SCHEME==8
  209: #warning direct threading scheme 8: cfa dead, i386 hack
  210: #  define NEXT_P0
  211: #  define IP		(ip)
  212: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  213: #  define NEXT_INST	(*IP)
  214: #  define INC_IP(const_inc)	({ ip+=(const_inc);})
  215: #  define DEF_CA
  216: #  define NEXT_P1	(ip++)
  217: #  define NEXT_P2	({goto **(ip-1);})
  218: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  219: #endif
  220: 
  221: #if THREADING_SCHEME==9
  222: #warning direct threading scheme 9: Power/PPC hack, long latency
  223: /* Power uses a prepare-to-branch instruction, and the latency between
  224:    this inst and the branch is 5 cycles on a PPC604; so we utilize this
  225:    to do some prefetching in between */
  226: #  define NEXT_P0
  227: #  define IP		ip
  228: #  define SET_IP(p)	({ip=(p); next_cfa=*ip; NEXT_P0;})
  229: #  define NEXT_INST	(next_cfa)
  230: #  define INC_IP(const_inc)	({next_cfa=IP[const_inc]; ip+=(const_inc);})
  231: #  define DEF_CA	
  232: #  define NEXT_P1	({cfa=next_cfa; ip++; next_cfa=*ip;})
  233: #  define NEXT_P2	({goto *cfa;})
  234: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  235: #  define MORE_VARS	Xt next_cfa;
  236: #endif
  237: 
  238: #if THREADING_SCHEME==10
  239: #warning direct threading scheme 10: plain (no attempt at scheduling)
  240: #  define NEXT_P0
  241: #  define IP		(ip)
  242: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  243: #  define NEXT_INST	(*ip)
  244: #  define INC_IP(const_inc)	({ip+=(const_inc);})
  245: #  define DEF_CA
  246: #  define NEXT_P1
  247: #  define NEXT_P2	({cfa=*ip++; goto *cfa;})
  248: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  249: #endif
  250: 
  251: /* direct threaded */
  252: #else
  253: /* indirect THREADED  */
  254: 
  255: #if THREADING_SCHEME==1
  256: #warning indirect threading scheme 1: autoinc, long latency, cisc
  257: #  define NEXT_P0	({cfa=*ip++;})
  258: #  define IP		(ip-1)
  259: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  260: #  define NEXT_INST	(cfa)
  261: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  262: #  define DEF_CA
  263: #  define NEXT_P1
  264: #  define NEXT_P2	({goto **cfa;})
  265: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  266: #endif
  267: 
  268: #if THREADING_SCHEME==2
  269: #warning indirect threading scheme 2: autoinc, long latency
  270: #  define NEXT_P0	({cfa=*ip++;})
  271: #  define IP		(ip-1)
  272: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  273: #  define NEXT_INST	(cfa)
  274: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  275: #  define DEF_CA	Label ca;
  276: #  define NEXT_P1	({ca=*cfa;})
  277: #  define NEXT_P2	({goto *ca;})
  278: #  define EXEC(XT)	({DEF_CA cfa=(XT); ca=*cfa; goto *ca;})
  279: #endif
  280: 
  281: 
  282: #if THREADING_SCHEME==3
  283: #warning indirect threading scheme 3: autoinc, low latency, cisc
  284: #  define NEXT_P0
  285: #  define IP		(ip)
  286: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  287: #  define NEXT_INST	(*ip)
  288: #  define INC_IP(const_inc)	({ip+=(const_inc);})
  289: #  define DEF_CA
  290: #  define NEXT_P1
  291: #  define NEXT_P2	({cfa=*ip++; goto **cfa;})
  292: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  293: #endif
  294: 
  295: #if THREADING_SCHEME==4
  296: #warning indirect threading scheme 4: autoinc, low latency
  297: #  define NEXT_P0	({cfa=*ip++;})
  298: #  define IP		(ip-1)
  299: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  300: #  define NEXT_INST	(cfa)
  301: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  302: #  define DEF_CA	Label ca;
  303: #  define NEXT_P1	({ca=*cfa;})
  304: #  define NEXT_P2	({goto *ca;})
  305: #  define EXEC(XT)	({DEF_CA cfa=(XT); ca=*cfa; goto *ca;})
  306: #endif
  307: 
  308: 
  309: #if THREADING_SCHEME==5
  310: #warning indirect threading scheme 5: long latency, cisc
  311: #  define NEXT_P0	({cfa=*ip;})
  312: #  define IP		(ip)
  313: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  314: #  define NEXT_INST	(cfa)
  315: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  316: #  define DEF_CA
  317: #  define NEXT_P1	(ip++)
  318: #  define NEXT_P2	({goto **cfa;})
  319: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  320: #endif
  321: 
  322: #if THREADING_SCHEME==6
  323: #warning indirect threading scheme 6: long latency
  324: #  define NEXT_P0	({cfa=*ip;})
  325: #  define IP		(ip)
  326: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  327: #  define NEXT_INST	(cfa)
  328: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  329: #  define DEF_CA	Label ca;
  330: #  define NEXT_P1	({ip++; ca=*cfa;})
  331: #  define NEXT_P2	({goto *ca;})
  332: #  define EXEC(XT)	({DEF_CA cfa=(XT); ca=*cfa; goto *ca;})
  333: #endif
  334: 
  335: #if THREADING_SCHEME==7
  336: #warning indirect threading scheme 7: low latency
  337: #  define NEXT_P0	({cfa=*ip;})
  338: #  define IP		(ip)
  339: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  340: #  define NEXT_INST	(cfa)
  341: #  define INC_IP(const_inc)	({cfa=IP[const_inc]; ip+=(const_inc);})
  342: #  define DEF_CA	Label ca;
  343: #  define NEXT_P1	({ip++; ca=*cfa;})
  344: #  define NEXT_P2	({goto *ca;})
  345: #  define EXEC(XT)	({DEF_CA cfa=(XT); ca=*cfa; goto *ca;})
  346: #endif
  347: 
  348: #if THREADING_SCHEME==8
  349: #warning indirect threading scheme 8: low latency,cisc
  350: #  define NEXT_P0
  351: #  define IP		(ip)
  352: #  define SET_IP(p)	({ip=(p); NEXT_P0;})
  353: #  define NEXT_INST	(*ip)
  354: #  define INC_IP(const_inc)	({ip+=(const_inc);})
  355: #  define DEF_CA
  356: #  define NEXT_P1
  357: #  define NEXT_P2	({cfa=*ip++; goto **cfa;})
  358: #  define EXEC(XT)	({cfa=(XT); goto **cfa;})
  359: #endif
  360: 
  361: /* indirect threaded */
  362: #endif
  363: 
  364: #endif /* !defined(DOUBLY_INDIRECT) */
  365: 
  366: #define NEXT ({DEF_CA NEXT_P1; NEXT_P2;})
  367: #define IPTOS NEXT_INST

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