Annotation of gforth/engine/threaded.h, revision 1.25
1.1 anton 1: /* This file defines a number of threading schemes.
2:
1.22 anton 3: Copyright (C) 1995, 1996,1997,1999,2003 Free Software Foundation, Inc.
1.1 anton 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
1.7 anton 19: Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
1.1 anton 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
1.19 pazsan 95: # ifndef DEBUG_DITC
96: # define DEBUG_DITC 0
97: # endif
98: /* define to 1 if you want to check consistency */
1.23 anton 99: # define NEXT_P0 ({cfa1=cfa; cfa=*ip;})
100: # define CFA cfa1
101: # define MORE_VARS Xt cfa1;
1.1 anton 102: # define IP (ip)
1.23 anton 103: # define SET_IP(p) ({ip=(p); cfa=*ip;})
1.1 anton 104: # define NEXT_INST (cfa)
105: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
106: # define DEF_CA Label ca;
1.13 anton 107: # define NEXT_P1 ({\
1.19 pazsan 108: if (DEBUG_DITC && (cfa<=vm_prims+DOESJUMP || cfa>=vm_prims+npriminfos)) \
109: fprintf(stderr,"NEXT encountered prim %p at ip=%p\n", cfa, ip); \
1.21 anton 110: ip++;})
111: # define NEXT_P2 ({ca=**cfa; goto *ca;})
1.13 anton 112: # define EXEC(XT) ({DEF_CA cfa=(XT);\
1.19 pazsan 113: if (DEBUG_DITC && (cfa>vm_prims+DOESJUMP && cfa<vm_prims+npriminfos)) \
1.14 anton 114: fprintf(stderr,"EXEC encountered xt %p at ip=%p, vm_prims=%p, xts=%p\n", cfa, ip, vm_prims, xts); \
1.13 anton 115: ca=**cfa; goto *ca;})
1.1 anton 116:
1.16 anton 117: #elif defined(NO_IP)
118:
119: #define NEXT_P0
1.25 ! anton 120: # define CFA cfa
1.16 anton 121: #define SET_IP(target) assert(0)
122: #define INC_IP(n) ((void)0)
123: #define DEF_CA
124: #define NEXT_P1
125: #define NEXT_P2 ({goto *next_code;})
126: /* set next_code to the return address before performing EXEC */
127: #define EXEC(XT) ({cfa=(XT); goto **cfa;})
128:
129: #else /* !defined(DOUBLY_INDIRECT) && !defined(NO_IP) */
1.1 anton 130:
1.3 anton 131: #if defined(DIRECT_THREADED)
132:
1.17 anton 133: /* This lets the compiler know that cfa is dead before; we place it at
134: "goto *"s that perform direct threaded dispatch (i.e., not EXECUTE
135: etc.), and thus do not reach doers, which would use cfa; the only
136: way to a doer is through EXECUTE etc., which set the cfa
137: themselves.
138:
139: Some of these direct threaded schemes use "cfa" to hold the code
140: address in normal direct threaded code. Of course we cannot use
141: KILLS there.
142:
143: KILLS works by having an empty asm instruction, and claiming to the
144: compiler that it writes to cfa.
145:
146: KILLS is optional. You can write
147:
148: #define KILLS
149:
150: and lose just a little performance.
151: */
152: #define KILLS asm("":"=X"(cfa));
153:
1.20 anton 154: #ifndef THREADING_SCHEME
1.24 anton 155: #define THREADING_SCHEME 7
1.20 anton 156: #endif
157:
1.3 anton 158: #if THREADING_SCHEME==1
159: #warning direct threading scheme 1: autoinc, long latency, cfa live
1.23 anton 160: # define NEXT_P0 ({cfa1=cfa; cfa=*ip++;})
161: # define CFA cfa1
162: # define MORE_VARS Xt cfa1;
1.1 anton 163: # define IP (ip-1)
1.23 anton 164: # define SET_IP(p) ({ip=(p); cfa=*ip++;})
1.1 anton 165: # define NEXT_INST (cfa)
166: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
167: # define DEF_CA
168: # define NEXT_P1
169: # define NEXT_P2 ({goto *cfa;})
1.15 anton 170: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.1 anton 171: #endif
172:
1.3 anton 173: #if THREADING_SCHEME==2
174: #warning direct threading scheme 2: autoinc, long latency, cfa dead
1.1 anton 175: # define NEXT_P0 (ip++)
1.23 anton 176: # define CFA cfa
1.1 anton 177: # define IP (ip-1)
1.3 anton 178: # define SET_IP(p) ({ip=(p); NEXT_P0;})
1.1 anton 179: # define NEXT_INST (*(ip-1))
180: # define INC_IP(const_inc) ({ ip+=(const_inc);})
181: # define DEF_CA
182: # define NEXT_P1
1.17 anton 183: # define NEXT_P2 ({KILLS goto **(ip-1);})
1.15 anton 184: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.1 anton 185: #endif
186:
187:
1.3 anton 188: #if THREADING_SCHEME==3
189: #warning direct threading scheme 3: autoinc, low latency, cfa live
1.1 anton 190: # define NEXT_P0
1.23 anton 191: # define CFA cfa
1.1 anton 192: # define IP (ip)
1.3 anton 193: # define SET_IP(p) ({ip=(p); NEXT_P0;})
1.1 anton 194: # define NEXT_INST (*ip)
195: # define INC_IP(const_inc) ({ip+=(const_inc);})
196: # define DEF_CA
197: # define NEXT_P1 ({cfa=*ip++;})
198: # define NEXT_P2 ({goto *cfa;})
1.15 anton 199: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.1 anton 200: #endif
201:
1.3 anton 202: #if THREADING_SCHEME==4
203: #warning direct threading scheme 4: autoinc, low latency, cfa dead
1.1 anton 204: # define NEXT_P0
1.23 anton 205: # define CFA cfa
1.1 anton 206: # define IP (ip)
1.3 anton 207: # define SET_IP(p) ({ip=(p); NEXT_P0;})
1.1 anton 208: # define NEXT_INST (*ip)
209: # define INC_IP(const_inc) ({ ip+=(const_inc);})
210: # define DEF_CA
211: # define NEXT_P1
1.17 anton 212: # define NEXT_P2 ({KILLS goto **(ip++);})
1.15 anton 213: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.1 anton 214: #endif
215:
1.3 anton 216: #if THREADING_SCHEME==5
217: #warning direct threading scheme 5: long latency, cfa live
1.23 anton 218: # define NEXT_P0 ({cfa1=cfa; cfa=*ip;})
219: # define CFA cfa1
220: # define MORE_VARS Xt cfa1;
1.1 anton 221: # define IP (ip)
1.23 anton 222: # define SET_IP(p) ({ip=(p); cfa=*ip;})
1.1 anton 223: # define NEXT_INST (cfa)
224: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
225: # define DEF_CA
226: # define NEXT_P1 (ip++)
227: # define NEXT_P2 ({goto *cfa;})
1.15 anton 228: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.1 anton 229: #endif
230:
1.3 anton 231: #if THREADING_SCHEME==6
232: #warning direct threading scheme 6: long latency, cfa dead
1.1 anton 233: # define NEXT_P0
1.23 anton 234: # define CFA cfa
1.1 anton 235: # define IP (ip)
1.3 anton 236: # define SET_IP(p) ({ip=(p); NEXT_P0;})
1.1 anton 237: # define NEXT_INST (*ip)
238: # define INC_IP(const_inc) ({ip+=(const_inc);})
239: # define DEF_CA
240: # define NEXT_P1 (ip++)
1.17 anton 241: # define NEXT_P2 ({KILLS goto **(ip-1);})
1.15 anton 242: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.1 anton 243: #endif
244:
245:
1.3 anton 246: #if THREADING_SCHEME==7
247: #warning direct threading scheme 7: low latency, cfa live
1.1 anton 248: # define NEXT_P0
1.23 anton 249: # define CFA cfa
1.1 anton 250: # define IP (ip)
1.3 anton 251: # define SET_IP(p) ({ip=(p); NEXT_P0;})
1.1 anton 252: # define NEXT_INST (*ip)
253: # define INC_IP(const_inc) ({ip+=(const_inc);})
254: # define DEF_CA
255: # define NEXT_P1 ({cfa=*ip++;})
256: # define NEXT_P2 ({goto *cfa;})
1.15 anton 257: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.1 anton 258: #endif
259:
1.3 anton 260: #if THREADING_SCHEME==8
261: #warning direct threading scheme 8: cfa dead, i386 hack
1.1 anton 262: # define NEXT_P0
1.23 anton 263: # define CFA cfa
1.1 anton 264: # define IP (ip)
1.3 anton 265: # define SET_IP(p) ({ip=(p); NEXT_P0;})
1.1 anton 266: # define NEXT_INST (*IP)
267: # define INC_IP(const_inc) ({ ip+=(const_inc);})
268: # define DEF_CA
269: # define NEXT_P1 (ip++)
1.17 anton 270: # define NEXT_P2 ({KILLS goto **(ip-1);})
1.15 anton 271: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.1 anton 272: #endif
273:
1.3 anton 274: #if THREADING_SCHEME==9
275: #warning direct threading scheme 9: Power/PPC hack, long latency
276: /* Power uses a prepare-to-branch instruction, and the latency between
277: this inst and the branch is 5 cycles on a PPC604; so we utilize this
278: to do some prefetching in between */
279: # define NEXT_P0
1.23 anton 280: # define CFA cfa
1.3 anton 281: # define IP ip
282: # define SET_IP(p) ({ip=(p); next_cfa=*ip; NEXT_P0;})
283: # define NEXT_INST (next_cfa)
284: # define INC_IP(const_inc) ({next_cfa=IP[const_inc]; ip+=(const_inc);})
1.8 anton 285: # define DEF_CA
286: # define NEXT_P1 ({cfa=next_cfa; ip++; next_cfa=*ip;})
287: # define NEXT_P2 ({goto *cfa;})
1.15 anton 288: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.3 anton 289: # define MORE_VARS Xt next_cfa;
290: #endif
1.1 anton 291:
1.3 anton 292: #if THREADING_SCHEME==10
293: #warning direct threading scheme 10: plain (no attempt at scheduling)
294: # define NEXT_P0
1.23 anton 295: # define CFA cfa
1.3 anton 296: # define IP (ip)
297: # define SET_IP(p) ({ip=(p); NEXT_P0;})
298: # define NEXT_INST (*ip)
299: # define INC_IP(const_inc) ({ip+=(const_inc);})
300: # define DEF_CA
301: # define NEXT_P1
302: # define NEXT_P2 ({cfa=*ip++; goto *cfa;})
1.15 anton 303: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
1.3 anton 304: #endif
1.1 anton 305:
1.3 anton 306: /* direct threaded */
307: #else
1.1 anton 308: /* indirect THREADED */
1.20 anton 309:
310: #ifndef THREADING_SCHEME
311: #define THREADING_SCHEME 6
312: #endif
1.1 anton 313:
1.3 anton 314: #if THREADING_SCHEME==1
315: #warning indirect threading scheme 1: autoinc, long latency, cisc
1.23 anton 316: # define NEXT_P0 ({cfa1=cfa; cfa=*ip++;})
317: # define CFA cfa1
318: # define MORE_VARS Xt cfa1;
1.1 anton 319: # define IP (ip-1)
1.23 anton 320: # define SET_IP(p) ({ip=(p); cfa=*ip++;})
1.1 anton 321: # define NEXT_INST (cfa)
322: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
323: # define DEF_CA
324: # define NEXT_P1
325: # define NEXT_P2 ({goto **cfa;})
326: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
327: #endif
328:
1.3 anton 329: #if THREADING_SCHEME==2
330: #warning indirect threading scheme 2: autoinc, long latency
1.23 anton 331: # define NEXT_P0 ({cfa1=cfa; cfa=*ip++;})
332: # define CFA cfa1
333: # define MORE_VARS Xt cfa1;
1.1 anton 334: # define IP (ip-1)
1.23 anton 335: # define SET_IP(p) ({ip=(p); cfa=*ip++;})
1.1 anton 336: # define NEXT_INST (cfa)
337: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
338: # define DEF_CA Label ca;
339: # define NEXT_P1 ({ca=*cfa;})
340: # define NEXT_P2 ({goto *ca;})
341: # define EXEC(XT) ({DEF_CA cfa=(XT); ca=*cfa; goto *ca;})
342: #endif
343:
344:
1.3 anton 345: #if THREADING_SCHEME==3
346: #warning indirect threading scheme 3: autoinc, low latency, cisc
1.1 anton 347: # define NEXT_P0
1.23 anton 348: # define CFA cfa
1.1 anton 349: # define IP (ip)
1.3 anton 350: # define SET_IP(p) ({ip=(p); NEXT_P0;})
1.1 anton 351: # define NEXT_INST (*ip)
352: # define INC_IP(const_inc) ({ip+=(const_inc);})
353: # define DEF_CA
354: # define NEXT_P1
355: # define NEXT_P2 ({cfa=*ip++; goto **cfa;})
356: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
357: #endif
358:
1.3 anton 359: #if THREADING_SCHEME==4
360: #warning indirect threading scheme 4: autoinc, low latency
1.23 anton 361: # define NEXT_P0 ({cfa1=cfa; cfa=*ip++;})
362: # define CFA cfa1
363: # define MORE_VARS Xt cfa1;
1.1 anton 364: # define IP (ip-1)
1.23 anton 365: # define SET_IP(p) ({ip=(p); cfa=*ip++;})
1.1 anton 366: # define NEXT_INST (cfa)
367: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
368: # define DEF_CA Label ca;
369: # define NEXT_P1 ({ca=*cfa;})
370: # define NEXT_P2 ({goto *ca;})
371: # define EXEC(XT) ({DEF_CA cfa=(XT); ca=*cfa; goto *ca;})
372: #endif
373:
374:
1.3 anton 375: #if THREADING_SCHEME==5
376: #warning indirect threading scheme 5: long latency, cisc
1.23 anton 377: # define NEXT_P0 ({cfa1=cfa; cfa=*ip;})
378: # define CFA cfa1
379: # define MORE_VARS Xt cfa1;
1.1 anton 380: # define IP (ip)
1.23 anton 381: # define SET_IP(p) ({ip=(p); cfa=*ip;})
1.1 anton 382: # define NEXT_INST (cfa)
383: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
384: # define DEF_CA
385: # define NEXT_P1 (ip++)
386: # define NEXT_P2 ({goto **cfa;})
387: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
388: #endif
389:
1.3 anton 390: #if THREADING_SCHEME==6
391: #warning indirect threading scheme 6: long latency
1.23 anton 392: # define NEXT_P0 ({cfa1=cfa; cfa=*ip;})
393: # define CFA cfa1
394: # define MORE_VARS Xt cfa1;
1.1 anton 395: # define IP (ip)
1.23 anton 396: # define SET_IP(p) ({ip=(p); cfa=*ip;})
1.1 anton 397: # define NEXT_INST (cfa)
398: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
399: # define DEF_CA Label ca;
400: # define NEXT_P1 ({ip++; ca=*cfa;})
401: # define NEXT_P2 ({goto *ca;})
402: # define EXEC(XT) ({DEF_CA cfa=(XT); ca=*cfa; goto *ca;})
403: #endif
404:
1.3 anton 405: #if THREADING_SCHEME==7
406: #warning indirect threading scheme 7: low latency
1.23 anton 407: # define NEXT_P0 ({cfa1=cfa; cfa=*ip;})
408: # define CFA cfa1
409: # define MORE_VARS Xt cfa1;
1.3 anton 410: # define IP (ip)
1.23 anton 411: # define SET_IP(p) ({ip=(p); cfa=*ip;})
1.3 anton 412: # define NEXT_INST (cfa)
413: # define INC_IP(const_inc) ({cfa=IP[const_inc]; ip+=(const_inc);})
414: # define DEF_CA Label ca;
415: # define NEXT_P1 ({ip++; ca=*cfa;})
416: # define NEXT_P2 ({goto *ca;})
417: # define EXEC(XT) ({DEF_CA cfa=(XT); ca=*cfa; goto *ca;})
418: #endif
1.1 anton 419:
1.3 anton 420: #if THREADING_SCHEME==8
421: #warning indirect threading scheme 8: low latency,cisc
1.1 anton 422: # define NEXT_P0
1.23 anton 423: # define CFA cfa
1.1 anton 424: # define IP (ip)
1.3 anton 425: # define SET_IP(p) ({ip=(p); NEXT_P0;})
1.1 anton 426: # define NEXT_INST (*ip)
427: # define INC_IP(const_inc) ({ip+=(const_inc);})
428: # define DEF_CA
429: # define NEXT_P1
430: # define NEXT_P2 ({cfa=*ip++; goto **cfa;})
431: # define EXEC(XT) ({cfa=(XT); goto **cfa;})
432: #endif
433:
1.3 anton 434: /* indirect threaded */
1.1 anton 435: #endif
436:
1.16 anton 437: #endif /* !defined(DOUBLY_INDIRECT) && !defined(NO_IP) */
1.1 anton 438:
439: #define NEXT ({DEF_CA NEXT_P1; NEXT_P2;})
1.10 anton 440: #define IPTOS NEXT_INST
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>