1: /* This file defines a number of threading schemes.
2:
3: Copyright (C) 1995, 1996,1997,1999,2003,2004 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 GCC_PR15242_WORKAROUND
95: #define DO_GOTO goto before_goto
96: #else
97: #define DO_GOTO goto *real_ca
98: #endif
99: #ifndef GOTO_ALIGN
100: #define GOTO_ALIGN
101: #endif
102:
103: #define GOTO(target) do {(real_ca=(target));} while(0)
104: #define NEXT_P2 do {NEXT_P1_5; DO_GOTO;} while(0)
105: #define EXEC(XT) do { real_ca=EXEC1(XT); DO_GOTO;} while (0)
106: #define VM_JUMP(target) do {GOTO(target); DO_GOTO;} while (0)
107: #define NEXT do {DEF_CA NEXT_P1; NEXT_P2;} while(0)
108: #define FIRST_NEXT_P2 NEXT_P1_5; GOTO_ALIGN; \
109: before_goto: goto *real_ca; after_goto:
110: #define FIRST_NEXT do {DEF_CA NEXT_P1; FIRST_NEXT_P2;} while(0)
111: #define IPTOS NEXT_INST
112:
113:
114: #ifdef DOUBLY_INDIRECT
115: # ifndef DEBUG_DITC
116: # define DEBUG_DITC 0
117: # endif
118: /* define to 1 if you want to check consistency */
119: # define NEXT_P0 do {cfa1=cfa; cfa=*ip;} while(0)
120: # define CFA cfa1
121: # define MORE_VARS Xt cfa1;
122: # define IP (ip)
123: # define SET_IP(p) do {ip=(p); cfa=*ip;} while(0)
124: # define NEXT_INST (cfa)
125: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
126: # define DEF_CA Label ca;
127: # define NEXT_P1 do {\
128: if (DEBUG_DITC && (cfa<=vm_prims+DOESJUMP || cfa>=vm_prims+npriminfos)) \
129: fprintf(stderr,"NEXT encountered prim %p at ip=%p\n", cfa, ip); \
130: ip++;} while(0)
131: # define NEXT_P1_5 do {ca=**cfa; GOTO(ca);} while(0)
132: # define EXEC1(XT) ({DEF_CA cfa=(XT);\
133: if (DEBUG_DITC && (cfa>vm_prims+DOESJUMP && cfa<vm_prims+npriminfos)) \
134: fprintf(stderr,"EXEC encountered xt %p at ip=%p, vm_prims=%p, xts=%p\n", cfa, ip, vm_prims, xts); \
135: ca=**cfa; ca;})
136:
137: #elif defined(NO_IP)
138:
139: #define NEXT_P0
140: # define CFA cfa
141: #define SET_IP(target) assert(0)
142: #define INC_IP(n) ((void)0)
143: #define DEF_CA
144: #define NEXT_P1
145: #define NEXT_P1_5 do {goto *next_code;} while(0)
146: /* set next_code to the return address before performing EXEC */
147: /* original: */
148: /* #define EXEC1(XT) do {cfa=(XT); goto **cfa;} while(0) */
149: /* fake, to make syntax check work */
150: #define EXEC1(XT) ({cfa=(XT); *cfa;})
151:
152: #else /* !defined(DOUBLY_INDIRECT) && !defined(NO_IP) */
153:
154: #if defined(DIRECT_THREADED)
155:
156: /* This lets the compiler know that cfa is dead before; we place it at
157: "goto *"s that perform direct threaded dispatch (i.e., not EXECUTE
158: etc.), and thus do not reach doers, which would use cfa; the only
159: way to a doer is through EXECUTE etc., which set the cfa
160: themselves.
161:
162: Some of these direct threaded schemes use "cfa" to hold the code
163: address in normal direct threaded code. Of course we cannot use
164: KILLS there.
165:
166: KILLS works by having an empty asm instruction, and claiming to the
167: compiler that it writes to cfa.
168:
169: KILLS is optional. You can write
170:
171: #define KILLS
172:
173: and lose just a little performance.
174: */
175: #define KILLS asm("":"=X"(cfa));
176:
177: #ifndef THREADING_SCHEME
178: #define THREADING_SCHEME 7
179: #endif
180:
181: #if THREADING_SCHEME==1
182: #warning direct threading scheme 1: autoinc, long latency, cfa live
183: # define NEXT_P0 do {cfa1=cfa; cfa=*ip++;} while(0)
184: # define CFA cfa1
185: # define MORE_VARS Xt cfa1;
186: # define IP (ip-1)
187: # define SET_IP(p) do {ip=(p); cfa=*ip++;} while(0)
188: # define NEXT_INST (cfa)
189: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
190: # define DEF_CA
191: # define NEXT_P1
192: # define NEXT_P1_5 do {GOTO(cfa);} while(0)
193: # define EXEC1(XT) ({cfa=(XT); *cfa;})
194: #endif
195:
196: #if THREADING_SCHEME==2
197: #warning direct threading scheme 2: autoinc, long latency, cfa dead
198: # define NEXT_P0 (ip++)
199: # define CFA cfa
200: # define IP (ip-1)
201: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
202: # define NEXT_INST (*(ip-1))
203: # define INC_IP(const_inc) do { ip+=(const_inc);} while(0)
204: # define DEF_CA
205: # define NEXT_P1
206: # define NEXT_P1_5 do {KILLS GOTO(*(ip-1));} while(0)
207: # define EXEC1(XT) ({cfa=(XT); *cfa;})
208: #endif
209:
210:
211: #if THREADING_SCHEME==3
212: #warning direct threading scheme 3: autoinc, low latency, cfa live
213: # define NEXT_P0
214: # define CFA cfa
215: # define IP (ip)
216: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
217: # define NEXT_INST (*ip)
218: # define INC_IP(const_inc) do {ip+=(const_inc);} while(0)
219: # define DEF_CA
220: # define NEXT_P1 do {cfa=*ip++;} while(0)
221: # define NEXT_P1_5 do {GOTO(cfa);} while(0)
222: # define EXEC1(XT) ({cfa=(XT); *cfa;})
223: #endif
224:
225: #if THREADING_SCHEME==4
226: #warning direct threading scheme 4: autoinc, low latency, cfa dead
227: # define NEXT_P0
228: # define CFA cfa
229: # define IP (ip)
230: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
231: # define NEXT_INST (*ip)
232: # define INC_IP(const_inc) do { ip+=(const_inc);} while(0)
233: # define DEF_CA
234: # define NEXT_P1
235: # define NEXT_P1_5 do {KILLS GOTO(*(ip++));} while(0)
236: # define EXEC1(XT) ({cfa=(XT); *cfa;})
237: #endif
238:
239: #if THREADING_SCHEME==5
240: #warning direct threading scheme 5: long latency, cfa live
241: # define NEXT_P0 do {cfa1=cfa; cfa=*ip;} while(0)
242: # define CFA cfa1
243: # define MORE_VARS Xt cfa1;
244: # define IP (ip)
245: # define SET_IP(p) do {ip=(p); cfa=*ip;} while(0)
246: # define NEXT_INST (cfa)
247: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
248: # define DEF_CA
249: # define NEXT_P1 (ip++)
250: # define NEXT_P1_5 do {GOTO(cfa);} while(0)
251: # define EXEC1(XT) ({cfa=(XT); *cfa;})
252: #endif
253:
254: #if THREADING_SCHEME==6
255: #warning direct threading scheme 6: long latency, cfa dead
256: # define NEXT_P0
257: # define CFA cfa
258: # define IP (ip)
259: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
260: # define NEXT_INST (*ip)
261: # define INC_IP(const_inc) do {ip+=(const_inc);} while(0)
262: # define DEF_CA
263: # define NEXT_P1 (ip++)
264: # define NEXT_P1_5 do {KILLS GOTO(*(ip-1));} while(0)
265: # define EXEC1(XT) ({cfa=(XT); *cfa;})
266: #endif
267:
268:
269: #if THREADING_SCHEME==7
270: #warning direct threading scheme 7: low latency, cfa live
271: # define NEXT_P0
272: # define CFA cfa
273: # define IP (ip)
274: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
275: # define NEXT_INST (*ip)
276: # define INC_IP(const_inc) do {ip+=(const_inc);} while(0)
277: # define DEF_CA
278: # define NEXT_P1 do {cfa=*ip++;} while(0)
279: # define NEXT_P1_5 do {GOTO(cfa);} while(0)
280: # define EXEC1(XT) ({cfa=(XT); *cfa;})
281: #endif
282:
283: #if THREADING_SCHEME==8
284: #warning direct threading scheme 8: cfa dead, i386 hack
285: # define NEXT_P0
286: # define CFA cfa
287: # define IP (ip)
288: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
289: # define NEXT_INST (*IP)
290: # define INC_IP(const_inc) do { ip+=(const_inc);} while(0)
291: # define DEF_CA
292: # define NEXT_P1 (ip++)
293: # define NEXT_P1_5 do {KILLS GOTO(*(ip-1));} while(0)
294: # define EXEC1(XT) ({cfa=(XT); *cfa;})
295: #endif
296:
297: #if THREADING_SCHEME==9
298: #warning direct threading scheme 9: Power/PPC hack, long latency
299: /* Power uses a prepare-to-branch instruction, and the latency between
300: this inst and the branch is 5 cycles on a PPC604; so we utilize this
301: to do some prefetching in between */
302: # define NEXT_P0
303: # define CFA cfa
304: # define IP ip
305: # define SET_IP(p) do {ip=(p); next_cfa=*ip; NEXT_P0;} while(0)
306: # define NEXT_INST (next_cfa)
307: # define INC_IP(const_inc) do {next_cfa=IP[const_inc]; ip+=(const_inc);} while(0)
308: # define DEF_CA
309: # define NEXT_P1 do {cfa=next_cfa; ip++; next_cfa=*ip;} while(0)
310: # define NEXT_P1_5 do {GOTO(cfa);} while(0)
311: # define EXEC1(XT) ({cfa=(XT); *cfa;})
312: # define MORE_VARS Xt next_cfa;
313: #endif
314:
315: #if THREADING_SCHEME==10
316: #warning direct threading scheme 10: plain (no attempt at scheduling)
317: # define NEXT_P0
318: # define CFA cfa
319: # define IP (ip)
320: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
321: # define NEXT_INST (*ip)
322: # define INC_IP(const_inc) do {ip+=(const_inc);} while(0)
323: # define DEF_CA
324: # define NEXT_P1
325: # define NEXT_P1_5 do {cfa=*ip++; GOTO(cfa);} while(0)
326: # define EXEC1(XT) ({cfa=(XT); *cfa;})
327: #endif
328:
329: /* direct threaded */
330: #else
331: /* indirect THREADED */
332:
333: #ifndef THREADING_SCHEME
334: #define THREADING_SCHEME 6
335: #endif
336:
337: #if THREADING_SCHEME==1
338: #warning indirect threading scheme 1: autoinc, long latency, cisc
339: # define NEXT_P0 do {cfa1=cfa; cfa=*ip++;} while(0)
340: # define CFA cfa1
341: # define MORE_VARS Xt cfa1;
342: # define IP (ip-1)
343: # define SET_IP(p) do {ip=(p); cfa=*ip++;} while(0)
344: # define NEXT_INST (cfa)
345: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
346: # define DEF_CA
347: # define NEXT_P1
348: # define NEXT_P1_5 do {GOTO(*cfa);} while(0)
349: # define EXEC1(XT) ({cfa=(XT); *cfa;})
350: #endif
351:
352: #if THREADING_SCHEME==2
353: #warning indirect threading scheme 2: autoinc, long latency
354: # define NEXT_P0 do {cfa1=cfa; cfa=*ip++;} while(0)
355: # define CFA cfa1
356: # define MORE_VARS Xt cfa1;
357: # define IP (ip-1)
358: # define SET_IP(p) do {ip=(p); cfa=*ip++;} while(0)
359: # define NEXT_INST (cfa)
360: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
361: # define DEF_CA Label ca;
362: # define NEXT_P1 do {ca=*cfa;} while(0)
363: # define NEXT_P1_5 do {GOTO(ca);} while(0)
364: # define EXEC1(XT) ({DEF_CA cfa=(XT); ca=*cfa; ca;})
365: #endif
366:
367:
368: #if THREADING_SCHEME==3
369: #warning indirect threading scheme 3: autoinc, low latency, cisc
370: # define NEXT_P0
371: # define CFA cfa
372: # define IP (ip)
373: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
374: # define NEXT_INST (*ip)
375: # define INC_IP(const_inc) do {ip+=(const_inc);} while(0)
376: # define DEF_CA
377: # define NEXT_P1
378: # define NEXT_P1_5 do {cfa=*ip++; GOTO(*cfa);} while(0)
379: # define EXEC1(XT) ({cfa=(XT); *cfa;})
380: #endif
381:
382: #if THREADING_SCHEME==4
383: #warning indirect threading scheme 4: autoinc, low latency
384: # define NEXT_P0 do {cfa1=cfa; cfa=*ip++;} while(0)
385: # define CFA cfa1
386: # define MORE_VARS Xt cfa1;
387: # define IP (ip-1)
388: # define SET_IP(p) do {ip=(p); cfa=*ip++;} while(0)
389: # define NEXT_INST (cfa)
390: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
391: # define DEF_CA Label ca;
392: # define NEXT_P1 do {ca=*cfa;} while(0)
393: # define NEXT_P1_5 do {GOTO(ca);} while(0)
394: # define EXEC1(XT) ({DEF_CA cfa=(XT); ca=*cfa; ca;})
395: #endif
396:
397:
398: #if THREADING_SCHEME==5
399: #warning indirect threading scheme 5: long latency, cisc
400: # define NEXT_P0 do {cfa1=cfa; cfa=*ip;} while(0)
401: # define CFA cfa1
402: # define MORE_VARS Xt cfa1;
403: # define IP (ip)
404: # define SET_IP(p) do {ip=(p); cfa=*ip;} while(0)
405: # define NEXT_INST (cfa)
406: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
407: # define DEF_CA
408: # define NEXT_P1 (ip++)
409: # define NEXT_P1_5 do {GOTO(*cfa);} while(0)
410: # define EXEC1(XT) ({cfa=(XT); *cfa;})
411: #endif
412:
413: #if THREADING_SCHEME==6
414: #warning indirect threading scheme 6: long latency
415: # define NEXT_P0 do {cfa1=cfa; cfa=*ip;} while(0)
416: # define CFA cfa1
417: # define MORE_VARS Xt cfa1;
418: # define IP (ip)
419: # define SET_IP(p) do {ip=(p); cfa=*ip;} while(0)
420: # define NEXT_INST (cfa)
421: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
422: # define DEF_CA Label ca;
423: # define NEXT_P1 do {ip++; ca=*cfa;} while(0)
424: # define NEXT_P1_5 do {GOTO(ca);} while(0)
425: # define EXEC1(XT) ({DEF_CA cfa=(XT); ca=*cfa; ca;})
426: #endif
427:
428: #if THREADING_SCHEME==7
429: #warning indirect threading scheme 7: low latency
430: # define NEXT_P0 do {cfa1=cfa; cfa=*ip;} while(0)
431: # define CFA cfa1
432: # define MORE_VARS Xt cfa1;
433: # define IP (ip)
434: # define SET_IP(p) do {ip=(p); cfa=*ip;} while(0)
435: # define NEXT_INST (cfa)
436: # define INC_IP(const_inc) do {cfa=IP[const_inc]; ip+=(const_inc);} while(0)
437: # define DEF_CA Label ca;
438: # define NEXT_P1 do {ip++; ca=*cfa;} while(0)
439: # define NEXT_P1_5 do {GOTO(ca);} while(0)
440: # define EXEC1(XT) ({DEF_CA cfa=(XT); ca=*cfa; ca;})
441: #endif
442:
443: #if THREADING_SCHEME==8
444: #warning indirect threading scheme 8: low latency,cisc
445: # define NEXT_P0
446: # define CFA cfa
447: # define IP (ip)
448: # define SET_IP(p) do {ip=(p); NEXT_P0;} while(0)
449: # define NEXT_INST (*ip)
450: # define INC_IP(const_inc) do {ip+=(const_inc);} while(0)
451: # define DEF_CA
452: # define NEXT_P1
453: # define NEXT_P1_5 do {cfa=*ip++; GOTO(*cfa);} while(0)
454: # define EXEC1(XT) ({cfa=(XT); *cfa;})
455: #endif
456:
457: /* indirect threaded */
458: #endif
459:
460: #endif /* !defined(DOUBLY_INDIRECT) && !defined(NO_IP) */
461:
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>