Commit | Line | Data |
---|---|---|
c906108c SS |
1 | # Generate the main loop of the simulator. |
2 | # Copyright (C) 1996, 1997, 1998, 1999 Free Software Foundation, Inc. | |
3 | # Contributed by Cygnus Support. | |
4 | # | |
5 | # This file is part of the GNU simulators. | |
6 | # | |
7 | # This program is free software; you can redistribute it and/or modify | |
8 | # it under the terms of the GNU General Public License as published by | |
9 | # the Free Software Foundation; either version 2, or (at your option) | |
10 | # 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 along | |
18 | # with this program; if not, write to the Free Software Foundation, Inc., | |
19 | # 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
20 | # | |
21 | # This file creates two files: eng.hin and mloop.cin. | |
22 | # eng.hin defines a few macros that specify what kind of engine was selected | |
23 | # based on the arguments to this script. | |
24 | # mloop.cin contains the engine. | |
25 | # | |
26 | # ??? Rename mloop.c to eng.c? | |
27 | # ??? Rename mainloop.in to engine.in? | |
28 | # ??? Add options to specify output file names? | |
29 | # ??? Rename this file to genengine.sh? | |
30 | # | |
31 | # Syntax: genmloop.sh [options] | |
32 | # | |
33 | # Options: | |
34 | # | |
35 | # -mono | -multi | |
36 | # - specify single cpu or multiple cpus (number specifyable at runtime), | |
37 | # maximum number is a configuration parameter | |
38 | # - -multi wip | |
39 | # | |
40 | # -fast: include support for fast execution in addition to full featured mode | |
41 | # | |
42 | # Full featured mode is for tracing, profiling, etc. and is always | |
43 | # provided. Fast mode contains no frills, except speed. | |
44 | # A target need only provide a "full" version of one of | |
45 | # simple,scache,pbb. If the target wants it can also provide a fast | |
46 | # version of same. It can't provide more than this. | |
47 | # ??? Later add ability to have another set of full/fast semantics | |
48 | # for use in with-devices/with-smp situations (pbb can be inappropriate | |
49 | # here). | |
50 | # | |
51 | # -full-switch: same as -fast but for full featured version of -switch | |
52 | # Only needed if -fast present. | |
53 | # | |
54 | # -simple: simple execution engine (the default) | |
55 | # | |
56 | # This engine fetches and executes one instruction at a time. | |
57 | # Field extraction is done in the semantic routines. | |
58 | # | |
59 | # ??? There are two possible flavours of -simple. One that extracts | |
60 | # fields in the semantic routine (which is what is implemented here), | |
61 | # and one that stores the extracted fields in ARGBUF before calling the | |
62 | # semantic routine. The latter is essentially the -scache case with a | |
63 | # cache size of one (and the scache lookup code removed). There are no | |
64 | # current uses of this and it's not clear when doing this would be a win. | |
65 | # More complicated ISA's that want to use -simple may find this a win. | |
66 | # Should this ever be desirable, implement a new engine style here and | |
67 | # call it -extract (or some such). It's believed that the CGEN-generated | |
68 | # code for the -scache case would be usable here, so no new code | |
69 | # generation option would be needed for CGEN. | |
70 | # | |
71 | # -scache: use the scache to speed things up (not always a win) | |
72 | # | |
73 | # This engine caches the extracted instruction before executing it. | |
74 | # When executing instructions they are first looked up in the scache. | |
75 | # | |
76 | # -pbb: same as -scache but extract a (pseudo-) basic block at a time | |
77 | # | |
78 | # This engine is basically identical to the scache version except that | |
79 | # extraction is done a pseudo-basic-block at a time and the address of | |
80 | # the scache entry of a branch target is recorded as well. | |
81 | # Additional speedups are then possible by defering Ctrl-C checking | |
82 | # to the end of basic blocks and by threading the insns together. | |
83 | # We call them pseudo-basic-block's instead of just basic-blocks because | |
84 | # they're not necessarily basic-blocks, though normally are. | |
85 | # | |
86 | # -parallel-read: support parallel execution with read-before-exec support. | |
87 | # -parallel-write: support parallel execution with write-after-exec support. | |
88 | # | |
89 | # One of these options is specified in addition to -simple, -scache, | |
90 | # -pbb. Note that while the code can determine if the cpu supports | |
91 | # parallel execution with HAVE_PARALLEL_INSNS [and thus this option is | |
92 | # technically unnecessary], having this option cuts down on the clutter | |
93 | # in the result. | |
94 | # | |
95 | # -switch file: specify file containing semantics implemented as a switch() | |
96 | # | |
97 | # -cpu <cpu-family> | |
98 | # | |
99 | # Specify the cpu family name. | |
100 | # | |
101 | # -infile <input-file> | |
102 | # | |
103 | # Specify the mainloop.in input file. | |
104 | # | |
105 | # Only one of -scache/-pbb may be selected. | |
106 | # -simple is the default. | |
107 | # | |
108 | #### | |
109 | # | |
110 | # TODO | |
111 | # - build mainloop.in from .cpu file | |
112 | ||
113 | type=mono | |
114 | #scache= | |
115 | #fast= | |
116 | #full_switch= | |
117 | #pbb= | |
118 | parallel=no | |
119 | switch= | |
120 | cpu="unknown" | |
121 | infile="" | |
122 | ||
123 | while test $# -gt 0 | |
124 | do | |
125 | case $1 in | |
126 | -mono) type=mono ;; | |
127 | -multi) type=multi ;; | |
128 | -no-fast) ;; | |
129 | -fast) fast=yes ;; | |
130 | -full-switch) full_switch=yes ;; | |
131 | -simple) ;; | |
132 | -scache) scache=yes ;; | |
133 | -pbb) pbb=yes ;; | |
134 | -no-parallel) ;; | |
135 | -parallel-read) parallel=read ;; | |
136 | -parallel-write) parallel=write ;; | |
137 | -switch) shift ; switch=$1 ;; | |
138 | -cpu) shift ; cpu=$1 ;; | |
139 | -infile) shift ; infile=$1 ;; | |
140 | *) echo "unknown option: $1" >&2 ; exit 1 ;; | |
141 | esac | |
142 | shift | |
143 | done | |
144 | ||
145 | # Argument validation. | |
146 | ||
147 | if [ x$scache = xyes -a x$pbb = xyes ] ; then | |
148 | echo "only one of -scache and -pbb may be selected" >&2 | |
149 | exit 1 | |
150 | fi | |
151 | ||
152 | if [ "x$cpu" = xunknown ] ; then | |
153 | echo "cpu family not specified" >&2 | |
154 | exit 1 | |
155 | fi | |
156 | ||
157 | if [ "x$infile" = x ] ; then | |
158 | echo "mainloop.in not specified" >&2 | |
159 | exit 1 | |
160 | fi | |
161 | ||
162 | lowercase='abcdefghijklmnopqrstuvwxyz' | |
163 | uppercase='ABCDEFGHIJKLMNOPQRSTUVWXYZ' | |
164 | CPU=`echo ${cpu} | tr "${lowercase}" "${uppercase}"` | |
165 | ||
166 | ########################################################################## | |
167 | ||
168 | rm -f eng.hin | |
169 | exec 1>eng.hin | |
170 | ||
171 | echo "/* engine configuration for ${cpu} */" | |
172 | echo "" | |
173 | ||
174 | echo "/* WITH_FAST: non-zero if a fast version of the engine is available" | |
175 | echo " in addition to the full-featured version. */" | |
176 | if [ x$fast = xyes ] ; then | |
177 | echo "#define WITH_FAST 1" | |
178 | else | |
179 | echo "#define WITH_FAST 0" | |
180 | fi | |
181 | ||
182 | echo "" | |
183 | echo "/* WITH_SCACHE_PBB_${CPU}: non-zero if the pbb engine was selected. */" | |
184 | if [ x$pbb = xyes ] ; then | |
185 | echo "#define WITH_SCACHE_PBB_${CPU} 1" | |
186 | else | |
187 | echo "#define WITH_SCACHE_PBB_${CPU} 0" | |
188 | fi | |
189 | ||
190 | echo "" | |
191 | echo "/* HAVE_PARALLEL_INSNS: non-zero if cpu can parallelly execute > 1 insn. */" | |
192 | if [ x$parallel != xno ] ; then | |
193 | echo "#define HAVE_PARALLEL_INSNS 1" | |
194 | if [ x$parallel = xread ] ; then | |
195 | echo "/* Parallel execution is supported by read-before-exec. */" | |
196 | echo "#define WITH_PARALLEL_READ 1" | |
197 | echo "#define WITH_PARALLEL_WRITE 0" | |
198 | else | |
199 | echo "/* Parallel execution is supported by write-after-exec. */" | |
200 | echo "#define WITH_PARALLEL_READ 0" | |
201 | echo "#define WITH_PARALLEL_WRITE 1" | |
202 | fi | |
203 | else | |
204 | echo "#define HAVE_PARALLEL_INSNS 0" | |
205 | echo "#define WITH_PARALLEL_READ 0" | |
206 | echo "#define WITH_PARALLEL_WRITE 0" | |
207 | fi | |
208 | ||
209 | if [ "x$switch" != x ] ; then | |
210 | echo "" | |
211 | echo "/* WITH_SEM_SWITCH_FULL: non-zero if full-featured engine is" | |
212 | echo " implemented as a switch(). */" | |
213 | if [ x$fast != xyes -o x$full_switch = xyes ] ; then | |
214 | echo "#define WITH_SEM_SWITCH_FULL 1" | |
215 | else | |
216 | echo "#define WITH_SEM_SWITCH_FULL 0" | |
217 | fi | |
218 | echo "" | |
219 | echo "/* WITH_SEM_SWITCH_FAST: non-zero if fast engine is" | |
220 | echo " implemented as a switch(). */" | |
221 | if [ x$fast = xyes ] ; then | |
222 | echo "#define WITH_SEM_SWITCH_FAST 1" | |
223 | else | |
224 | echo "#define WITH_SEM_SWITCH_FAST 0" | |
225 | fi | |
226 | fi | |
227 | ||
228 | # Decls of functions we define. | |
229 | ||
230 | echo "" | |
231 | echo "/* Functions defined in the generated mainloop.c file" | |
232 | echo " (which doesn't necessarily have that file name). */" | |
233 | echo "" | |
234 | echo "extern ENGINE_FN ${cpu}_engine_run_full;" | |
235 | echo "extern ENGINE_FN ${cpu}_engine_run_fast;" | |
236 | ||
237 | if [ x$pbb = xyes ] ; then | |
238 | echo "" | |
239 | echo "extern SEM_PC ${cpu}_pbb_begin (SIM_CPU *, int);" | |
240 | echo "extern SEM_PC ${cpu}_pbb_chain (SIM_CPU *, SEM_ARG);" | |
241 | echo "extern SEM_PC ${cpu}_pbb_cti_chain (SIM_CPU *, SEM_ARG, SEM_PC *, PCADDR);" | |
242 | echo "extern void ${cpu}_pbb_before (SIM_CPU *, SCACHE *);" | |
243 | echo "extern void ${cpu}_pbb_after (SIM_CPU *, SCACHE *);" | |
244 | fi | |
245 | ||
246 | ########################################################################## | |
247 | ||
248 | rm -f tmp-mloop.cin mloop.cin | |
249 | exec 1>tmp-mloop.cin | |
250 | ||
251 | # We use @cpu@ instead of ${cpu} because we still need to run sed to handle | |
252 | # transformation of @cpu@ for mainloop.in, so there's no need to use ${cpu} | |
253 | # here. | |
254 | ||
255 | cat << EOF | |
256 | /* This file is generated by the genmloop script. DO NOT EDIT! */ | |
257 | ||
258 | /* Enable switch() support in cgen headers. */ | |
259 | #define SEM_IN_SWITCH | |
260 | ||
261 | #define WANT_CPU @cpu@ | |
262 | #define WANT_CPU_@CPU@ | |
263 | ||
264 | #include "sim-main.h" | |
265 | #include "bfd.h" | |
266 | #include "cgen-mem.h" | |
267 | #include "cgen-ops.h" | |
268 | #include "sim-assert.h" | |
269 | ||
270 | /* Fill in the administrative ARGBUF fields required by all insns, | |
271 | virtual and real. */ | |
272 | ||
273 | static INLINE void | |
274 | @cpu@_fill_argbuf (const SIM_CPU *cpu, ARGBUF *abuf, const IDESC *idesc, | |
275 | PCADDR pc, int fast_p) | |
276 | { | |
277 | #if WITH_SCACHE | |
278 | SEM_SET_CODE (abuf, idesc, fast_p); | |
279 | ARGBUF_ADDR (abuf) = pc; | |
280 | #endif | |
281 | ARGBUF_IDESC (abuf) = idesc; | |
282 | } | |
283 | ||
284 | /* Fill in tracing/profiling fields of an ARGBUF. */ | |
285 | ||
286 | static INLINE void | |
287 | @cpu@_fill_argbuf_tp (const SIM_CPU *cpu, ARGBUF *abuf, | |
288 | int trace_p, int profile_p) | |
289 | { | |
290 | ARGBUF_TRACE_P (abuf) = trace_p; | |
291 | ARGBUF_PROFILE_P (abuf) = profile_p; | |
292 | } | |
293 | ||
294 | #if WITH_SCACHE_PBB | |
295 | ||
296 | /* Emit the "x-before" handler. | |
297 | x-before is emitted before each insn (serial or parallel). | |
298 | This is as opposed to x-after which is only emitted at the end of a group | |
299 | of parallel insns. */ | |
300 | ||
301 | static INLINE void | |
302 | @cpu@_emit_before (SIM_CPU *current_cpu, SCACHE *sc, PCADDR pc, int first_p) | |
303 | { | |
304 | ARGBUF *abuf = &sc[0].argbuf; | |
305 | const IDESC *id = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEFORE]; | |
306 | ||
307 | abuf->fields.before.first_p = first_p; | |
308 | @cpu@_fill_argbuf (current_cpu, abuf, id, pc, 0); | |
309 | /* no need to set trace_p,profile_p */ | |
310 | } | |
311 | ||
312 | /* Emit the "x-after" handler. | |
313 | x-after is emitted after a serial insn or at the end of a group of | |
314 | parallel insns. */ | |
315 | ||
316 | static INLINE void | |
317 | @cpu@_emit_after (SIM_CPU *current_cpu, SCACHE *sc, PCADDR pc) | |
318 | { | |
319 | ARGBUF *abuf = &sc[0].argbuf; | |
320 | const IDESC *id = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_AFTER]; | |
321 | ||
322 | @cpu@_fill_argbuf (current_cpu, abuf, id, pc, 0); | |
323 | /* no need to set trace_p,profile_p */ | |
324 | } | |
325 | ||
326 | #endif /* WITH_SCACHE_PBB */ | |
327 | ||
328 | EOF | |
329 | ||
330 | ${SHELL} $infile support | |
331 | ||
332 | ########################################################################## | |
333 | ||
334 | # Simple engine: fetch an instruction, execute the instruction. | |
335 | # | |
336 | # Instruction fields are not extracted into ARGBUF, they are extracted in | |
337 | # the semantic routines themselves. However, there is still a need to pass | |
338 | # and return misc. information to the semantic routines so we still use ARGBUF. | |
339 | # [One could certainly implement things differently and remove ARGBUF. | |
340 | # It's not clear this is necessarily always a win.] | |
341 | # ??? The use of the SCACHE struct is for consistency with the with-scache | |
342 | # case though it might be a source of confusion. | |
343 | ||
344 | if [ x$scache != xyes -a x$pbb != xyes ] ; then | |
345 | ||
346 | cat << EOF | |
347 | ||
348 | #define FAST_P 0 | |
349 | ||
350 | void | |
351 | @cpu@_engine_run_full (SIM_CPU *current_cpu) | |
352 | { | |
353 | #define FAST_P 0 | |
354 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
355 | /* ??? Use of SCACHE is a bit of a hack as we don't actually use the scache. | |
356 | We do however use ARGBUF so for consistency with the other engine flavours | |
357 | the SCACHE type is used. */ | |
358 | SCACHE cache[MAX_LIW_INSNS]; | |
359 | SCACHE *sc = &cache[0]; | |
360 | ||
361 | EOF | |
362 | ||
363 | if [ x$parallel != xno ] ; then | |
364 | cat << EOF | |
365 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
366 | PAREXEC *par_exec; | |
367 | ||
368 | EOF | |
369 | fi | |
370 | ||
371 | # Any initialization code before looping starts. | |
372 | # Note that this code may declare some locals. | |
373 | ${SHELL} $infile init | |
374 | ||
375 | if [ x$parallel != xno ] ; then | |
376 | cat << EOF | |
377 | ||
378 | #if defined (HAVE_PARALLEL_EXEC) && defined (__GNUC__) | |
379 | { | |
380 | if (! CPU_IDESC_READ_INIT_P (current_cpu)) | |
381 | { | |
382 | /* ??? Later maybe paste read.c in when building mainloop.c. */ | |
383 | #define DEFINE_LABELS | |
384 | #include "readx.c" | |
385 | CPU_IDESC_READ_INIT_P (current_cpu) = 1; | |
386 | } | |
387 | } | |
388 | #endif | |
389 | ||
390 | EOF | |
391 | fi | |
392 | ||
393 | cat << EOF | |
394 | ||
395 | #if WITH_SEM_SWITCH_FULL && defined (__GNUC__) | |
396 | { | |
397 | if (! CPU_IDESC_SEM_INIT_P (current_cpu)) | |
398 | { | |
399 | /* ??? Later maybe paste sem-switch.c in when building mainloop.c. */ | |
400 | #define DEFINE_LABELS | |
401 | #include "$switch" | |
402 | CPU_IDESC_SEM_INIT_P (current_cpu) = 1; | |
403 | } | |
404 | } | |
405 | #endif | |
406 | ||
407 | do | |
408 | { | |
409 | /* begin full-exec-simple */ | |
410 | EOF | |
411 | ||
412 | ${SHELL} $infile full-exec-simple | |
413 | ||
414 | cat << EOF | |
415 | /* end full-exec-simple */ | |
416 | ||
417 | ++ CPU_INSN_COUNT (current_cpu); | |
418 | } | |
419 | while (0 /*CPU_RUNNING_P (current_cpu)*/); | |
420 | } | |
421 | ||
422 | #undef FAST_P | |
423 | ||
424 | EOF | |
425 | ||
426 | #################################### | |
427 | ||
428 | # Simple engine: fast version. | |
429 | # ??? A somewhat dubious effort, but for completeness' sake. | |
430 | ||
431 | if [ x$fast = xyes ] ; then | |
432 | ||
433 | cat << EOF | |
434 | ||
435 | #define FAST_P 1 | |
436 | ||
437 | FIXME: "fast simple version unimplemented, delete -fast arg to genmloop.sh." | |
438 | ||
439 | #undef FAST_P | |
440 | ||
441 | EOF | |
442 | ||
443 | fi # -fast | |
444 | ||
445 | fi # simple engine | |
446 | ||
447 | ########################################################################## | |
448 | ||
449 | # Scache engine: lookup insn in scache, fetch if missing, then execute it. | |
450 | ||
451 | if [ x$scache = xyes ] ; then | |
452 | ||
453 | cat << EOF | |
454 | ||
455 | static INLINE SCACHE * | |
456 | @cpu@_scache_lookup (SIM_CPU *current_cpu, PCADDR vpc, SCACHE *scache, | |
457 | unsigned int hash_mask, int FAST_P) | |
458 | { | |
459 | /* First step: look up current insn in hash table. */ | |
460 | SCACHE *sc = scache + SCACHE_HASH_PC (vpc, hash_mask); | |
461 | ||
462 | /* If the entry isn't the one we want (cache miss), | |
463 | fetch and decode the instruction. */ | |
464 | if (sc->argbuf.addr != vpc) | |
465 | { | |
c906108c SS |
466 | if (FAST_P) |
467 | PROFILE_COUNT_SCACHE_MISS (current_cpu); | |
468 | ||
469 | /* begin extract-scache */ | |
470 | EOF | |
471 | ||
472 | ${SHELL} $infile extract-scache | |
473 | ||
474 | cat << EOF | |
475 | /* end extract-scache */ | |
476 | } | |
477 | else if (FAST_P) | |
478 | { | |
479 | PROFILE_COUNT_SCACHE_HIT (current_cpu); | |
480 | /* Make core access statistics come out right. | |
481 | The size is a guess, but it's currently not used either. */ | |
482 | PROFILE_COUNT_CORE (current_cpu, vpc, 2, exec_map); | |
483 | } | |
484 | ||
485 | return sc; | |
486 | } | |
487 | ||
488 | #define FAST_P 0 | |
489 | ||
490 | void | |
491 | @cpu@_engine_run_full (SIM_CPU *current_cpu) | |
492 | { | |
493 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
494 | SCACHE *scache = CPU_SCACHE_CACHE (current_cpu); | |
495 | unsigned int hash_mask = CPU_SCACHE_HASH_MASK (current_cpu); | |
496 | SEM_PC vpc; | |
497 | ||
498 | EOF | |
499 | ||
500 | if [ x$parallel != xno ] ; then | |
501 | cat << EOF | |
502 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
503 | PAREXEC *par_exec; | |
504 | ||
505 | EOF | |
506 | fi | |
507 | ||
508 | # Any initialization code before looping starts. | |
509 | # Note that this code may declare some locals. | |
510 | ${SHELL} $infile init | |
511 | ||
512 | if [ x$parallel != xno ] ; then | |
513 | cat << EOF | |
514 | ||
515 | #if defined (HAVE_PARALLEL_EXEC) && defined (__GNUC__) | |
516 | { | |
517 | if (! CPU_IDESC_READ_INIT_P (current_cpu)) | |
518 | { | |
519 | /* ??? Later maybe paste read.c in when building mainloop.c. */ | |
520 | #define DEFINE_LABELS | |
521 | #include "readx.c" | |
522 | CPU_IDESC_READ_INIT_P (current_cpu) = 1; | |
523 | } | |
524 | } | |
525 | #endif | |
526 | ||
527 | EOF | |
528 | fi | |
529 | ||
530 | cat << EOF | |
531 | ||
532 | vpc = GET_H_PC (); | |
533 | ||
534 | do | |
535 | { | |
536 | SCACHE *sc; | |
537 | ||
538 | sc = @cpu@_scache_lookup (current_cpu, vpc, scache, hash_mask, FAST_P); | |
539 | ||
540 | /* begin full-exec-scache */ | |
541 | EOF | |
542 | ||
543 | ${SHELL} $infile full-exec-scache | |
544 | ||
545 | cat << EOF | |
546 | /* end full-exec-scache */ | |
547 | ||
548 | SET_H_PC (vpc); | |
549 | ||
550 | ++ CPU_INSN_COUNT (current_cpu); | |
551 | } | |
552 | while (0 /*CPU_RUNNING_P (current_cpu)*/); | |
553 | } | |
554 | ||
555 | #undef FAST_P | |
556 | ||
557 | EOF | |
558 | ||
559 | #################################### | |
560 | ||
561 | # Scache engine: fast version. | |
562 | ||
563 | if [ x$fast = xyes ] ; then | |
564 | ||
565 | cat << EOF | |
566 | ||
567 | #define FAST_P 1 | |
568 | ||
569 | void | |
570 | @cpu@_engine_run_fast (SIM_CPU *current_cpu) | |
571 | { | |
572 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
573 | SCACHE *scache = CPU_SCACHE_CACHE (current_cpu); | |
574 | unsigned int hash_mask = CPU_SCACHE_HASH_MASK (current_cpu); | |
575 | SEM_PC vpc; | |
576 | ||
577 | EOF | |
578 | ||
579 | if [ x$parallel != xno ] ; then | |
580 | cat << EOF | |
581 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
582 | PAREXEC *par_exec; | |
583 | ||
584 | EOF | |
585 | fi | |
586 | ||
587 | # Any initialization code before looping starts. | |
588 | # Note that this code may declare some locals. | |
589 | ${SHELL} $infile init | |
590 | ||
591 | if [ x$parallel != xno ] ; then | |
592 | cat << EOF | |
593 | ||
594 | #if defined (HAVE_PARALLEL_EXEC) && defined (__GNUC__) | |
595 | { | |
596 | if (! CPU_IDESC_READ_INIT_P (current_cpu)) | |
597 | { | |
598 | /* ??? Later maybe paste read.c in when building mainloop.c. */ | |
599 | #define DEFINE_LABELS | |
600 | #include "readx.c" | |
601 | CPU_IDESC_READ_INIT_P (current_cpu) = 1; | |
602 | } | |
603 | } | |
604 | #endif | |
605 | ||
606 | EOF | |
607 | fi # parallel != no | |
608 | ||
609 | cat << EOF | |
610 | ||
611 | #if WITH_SEM_SWITCH_FAST && defined (__GNUC__) | |
612 | { | |
613 | if (! CPU_IDESC_SEM_INIT_P (current_cpu)) | |
614 | { | |
615 | /* ??? Later maybe paste sem-switch.c in when building mainloop.c. */ | |
616 | #define DEFINE_LABELS | |
617 | #include "$switch" | |
618 | CPU_IDESC_SEM_INIT_P (current_cpu) = 1; | |
619 | } | |
620 | } | |
621 | #endif | |
622 | ||
623 | vpc = GET_H_PC (); | |
624 | ||
625 | do | |
626 | { | |
627 | SCACHE *sc; | |
628 | ||
629 | sc = @cpu@_scache_lookup (current_cpu, vpc, scache, hash_mask, FAST_P); | |
630 | ||
631 | /* begin fast-exec-scache */ | |
632 | EOF | |
633 | ||
634 | ${SHELL} $infile fast-exec-scache | |
635 | ||
636 | cat << EOF | |
637 | /* end fast-exec-scache */ | |
638 | ||
639 | SET_H_PC (vpc); | |
640 | ||
641 | ++ CPU_INSN_COUNT (current_cpu); | |
642 | } | |
643 | while (0 /*CPU_RUNNING_P (current_cpu)*/); | |
644 | } | |
645 | ||
646 | #undef FAST_P | |
647 | ||
648 | EOF | |
649 | ||
650 | fi # -fast | |
651 | ||
652 | fi # -scache | |
653 | ||
654 | ########################################################################## | |
655 | ||
656 | # Compilation engine: lookup insn in scache, extract a pbb | |
657 | # (pseudo-basic-block) if missing, then execute the pbb. | |
658 | # A "pbb" is a sequence of insns up to the next cti insn or until | |
659 | # some prespecified maximum. | |
660 | # CTI: control transfer instruction. | |
661 | ||
662 | if [ x$pbb = xyes ] ; then | |
663 | ||
664 | cat << EOF | |
665 | ||
666 | /* Record address of cti terminating a pbb. */ | |
667 | #define SET_CTI_VPC(sc) do { _cti_sc = (sc); } while (0) | |
668 | /* Record number of [real] insns in pbb. */ | |
669 | #define SET_INSN_COUNT(n) do { _insn_count = (n); } while (0) | |
670 | ||
671 | /* Fetch and extract a pseudo-basic-block. | |
672 | FAST_P is non-zero if no tracing/profiling/etc. is wanted. */ | |
673 | ||
674 | INLINE SEM_PC | |
675 | @cpu@_pbb_begin (SIM_CPU *current_cpu, int FAST_P) | |
676 | { | |
677 | SEM_PC new_vpc; | |
678 | PCADDR pc; | |
679 | SCACHE *sc; | |
680 | int max_insns = CPU_SCACHE_MAX_CHAIN_LENGTH (current_cpu); | |
681 | ||
682 | pc = GET_H_PC (); | |
683 | ||
684 | new_vpc = scache_lookup_or_alloc (current_cpu, pc, max_insns, &sc); | |
685 | if (! new_vpc) | |
686 | { | |
687 | /* Leading '_' to avoid collision with mainloop.in. */ | |
688 | int _insn_count = 0; | |
689 | SCACHE *orig_sc = sc; | |
690 | SCACHE *_cti_sc = NULL; | |
691 | int slice_insns = CPU_MAX_SLICE_INSNS (current_cpu); | |
692 | ||
693 | /* First figure out how many instructions to compile. | |
694 | MAX_INSNS is the size of the allocated buffer, which includes space | |
695 | for before/after handlers if they're being used. | |
696 | SLICE_INSNS is the maxinum number of real insns that can be | |
697 | executed. Zero means "as many as we want". */ | |
698 | /* ??? max_insns is serving two incompatible roles. | |
699 | 1) Number of slots available in scache buffer. | |
700 | 2) Number of real insns to execute. | |
701 | They're incompatible because there are virtual insns emitted too | |
702 | (chain,cti-chain,before,after handlers). */ | |
703 | ||
704 | if (slice_insns == 1) | |
705 | { | |
706 | /* No need to worry about extra slots required for virtual insns | |
707 | and parallel exec support because MAX_CHAIN_LENGTH is | |
708 | guaranteed to be big enough to execute at least 1 insn! */ | |
709 | max_insns = 1; | |
710 | } | |
711 | else | |
712 | { | |
713 | /* Allow enough slop so that while compiling insns, if max_insns > 0 | |
714 | then there's guaranteed to be enough space to emit one real insn. | |
715 | MAX_CHAIN_LENGTH is typically much longer than | |
716 | the normal number of insns between cti's anyway. */ | |
717 | max_insns -= (1 /* one for the trailing chain insn */ | |
718 | + (FAST_P | |
719 | ? 0 | |
720 | : (1 + MAX_PARALLEL_INSNS) /* before+after */) | |
721 | + (MAX_PARALLEL_INSNS > 1 | |
722 | ? (MAX_PARALLEL_INSNS * 2) | |
723 | : 0)); | |
724 | ||
725 | /* Account for before/after handlers. */ | |
726 | if (! FAST_P) | |
727 | slice_insns *= 3; | |
728 | ||
729 | if (slice_insns > 0 | |
730 | && slice_insns < max_insns) | |
731 | max_insns = slice_insns; | |
732 | } | |
733 | ||
734 | new_vpc = sc; | |
735 | ||
736 | /* SC,PC must be updated to point passed the last entry used. | |
737 | SET_CTI_VPC must be called if pbb is terminated by a cti. | |
738 | SET_INSN_COUNT must be called to record number of real insns in | |
739 | pbb [could be computed by us of course, extra cpu but perhaps | |
740 | negligible enough]. */ | |
741 | ||
742 | /* begin extract-pbb */ | |
743 | EOF | |
744 | ||
745 | ${SHELL} $infile extract-pbb | |
746 | ||
747 | cat << EOF | |
748 | /* end extract-pbb */ | |
749 | ||
750 | /* The last one is a pseudo-insn to link to the next chain. | |
751 | It is also used to record the insn count for this chain. */ | |
752 | { | |
753 | const IDESC *id; | |
754 | ||
755 | /* Was pbb terminated by a cti? */ | |
756 | if (_cti_sc) | |
757 | { | |
758 | id = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_CTI_CHAIN]; | |
759 | } | |
760 | else | |
761 | { | |
762 | id = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_CHAIN]; | |
763 | } | |
764 | SEM_SET_CODE (&sc->argbuf, id, FAST_P); | |
765 | sc->argbuf.idesc = id; | |
766 | sc->argbuf.addr = pc; | |
767 | sc->argbuf.fields.chain.insn_count = _insn_count; | |
768 | sc->argbuf.fields.chain.next = 0; | |
769 | ++sc; | |
770 | } | |
771 | ||
085dd6e6 JM |
772 | /* Update the pointer to the next free entry, may not have used as |
773 | many entries as was asked for. */ | |
c906108c SS |
774 | CPU_SCACHE_NEXT_FREE (current_cpu) = sc; |
775 | /* Record length of chain if profiling. | |
776 | This includes virtual insns since they count against | |
777 | max_insns too. */ | |
778 | if (! FAST_P) | |
779 | PROFILE_COUNT_SCACHE_CHAIN_LENGTH (current_cpu, sc - orig_sc); | |
780 | } | |
781 | ||
782 | return new_vpc; | |
783 | } | |
784 | ||
785 | /* Chain to the next block from a non-cti terminated previous block. */ | |
786 | ||
787 | INLINE SEM_PC | |
788 | @cpu@_pbb_chain (SIM_CPU *current_cpu, SEM_ARG sem_arg) | |
789 | { | |
790 | ARGBUF *abuf = SEM_ARGBUF (sem_arg); | |
791 | ||
792 | PBB_UPDATE_INSN_COUNT (current_cpu, sem_arg); | |
793 | ||
794 | SET_H_PC (abuf->addr); | |
795 | ||
796 | /* If not running forever, exit back to main loop. */ | |
797 | if (CPU_MAX_SLICE_INSNS (current_cpu) != 0 | |
798 | /* Also exit back to main loop if there's an event. | |
799 | Note that if CPU_MAX_SLICE_INSNS != 1, events won't get processed | |
800 | at the "right" time, but then that was what was asked for. | |
801 | There is no silver bullet for simulator engines. | |
802 | ??? Clearly this needs a cleaner interface. | |
803 | At present it's just so Ctrl-C works. */ | |
804 | || STATE_EVENTS (CPU_STATE (current_cpu))->work_pending) | |
805 | CPU_RUNNING_P (current_cpu) = 0; | |
806 | ||
807 | /* If chained to next block, go straight to it. */ | |
808 | if (abuf->fields.chain.next) | |
809 | return abuf->fields.chain.next; | |
810 | /* See if next block has already been compiled. */ | |
811 | abuf->fields.chain.next = scache_lookup (current_cpu, abuf->addr); | |
812 | if (abuf->fields.chain.next) | |
813 | return abuf->fields.chain.next; | |
814 | /* Nope, so next insn is a virtual insn to invoke the compiler | |
815 | (begin a pbb). */ | |
816 | return CPU_SCACHE_PBB_BEGIN (current_cpu); | |
817 | } | |
818 | ||
819 | /* Chain to the next block from a cti terminated previous block. | |
820 | NEW_VPC_PTR is one of SEM_BRANCH_UNTAKEN, SEM_BRANCH_UNCACHEABLE, or | |
821 | a pointer to a location containing the SEM_PC of the branch's address. | |
822 | NEW_PC is the target's branch address, and is only valid if | |
823 | NEW_VPC_PTR != SEM_BRANCH_UNTAKEN. */ | |
824 | ||
825 | INLINE SEM_PC | |
826 | @cpu@_pbb_cti_chain (SIM_CPU *current_cpu, SEM_ARG sem_arg, | |
827 | SEM_PC *new_vpc_ptr, PCADDR new_pc) | |
828 | { | |
c906108c SS |
829 | PBB_UPDATE_INSN_COUNT (current_cpu, sem_arg); |
830 | ||
831 | /* If not running forever, exit back to main loop. */ | |
832 | if (CPU_MAX_SLICE_INSNS (current_cpu) != 0 | |
833 | /* Also exit back to main loop if there's an event. | |
834 | Note that if CPU_MAX_SLICE_INSNS != 1, events won't get processed | |
835 | at the "right" time, but then that was what was asked for. | |
836 | There is no silver bullet for simulator engines. | |
837 | ??? Clearly this needs a cleaner interface. | |
838 | At present it's just so Ctrl-C works. */ | |
839 | || STATE_EVENTS (CPU_STATE (current_cpu))->work_pending) | |
840 | CPU_RUNNING_P (current_cpu) = 0; | |
841 | ||
842 | /* Restart compiler if we branched to an uncacheable address | |
843 | (e.g. "j reg"). */ | |
844 | if (new_vpc_ptr == SEM_BRANCH_UNCACHEABLE) | |
845 | { | |
846 | SET_H_PC (new_pc); | |
847 | return CPU_SCACHE_PBB_BEGIN (current_cpu); | |
848 | } | |
849 | ||
850 | /* If branch wasn't taken, update the pc and set BR_ADDR_PTR to our | |
851 | next chain ptr. */ | |
852 | if (new_vpc_ptr == SEM_BRANCH_UNTAKEN) | |
853 | { | |
085dd6e6 | 854 | ARGBUF *abuf = SEM_ARGBUF (sem_arg); |
c906108c SS |
855 | SET_H_PC (abuf->addr); |
856 | new_vpc_ptr = &abuf->fields.chain.next; | |
857 | } | |
858 | else | |
859 | { | |
860 | SET_H_PC (new_pc); | |
861 | } | |
862 | ||
863 | /* If chained to next block, go straight to it. */ | |
864 | if (*new_vpc_ptr) | |
865 | return *new_vpc_ptr; | |
866 | /* See if next block has already been compiled. */ | |
867 | *new_vpc_ptr = scache_lookup (current_cpu, GET_H_PC ()); | |
868 | if (*new_vpc_ptr) | |
869 | return *new_vpc_ptr; | |
870 | /* Nope, so next insn is a virtual insn to invoke the compiler | |
871 | (begin a pbb). */ | |
872 | return CPU_SCACHE_PBB_BEGIN (current_cpu); | |
873 | } | |
874 | ||
875 | /* x-before handler. | |
876 | This is called before each insn. */ | |
877 | ||
878 | void | |
879 | @cpu@_pbb_before (SIM_CPU *current_cpu, SCACHE *sc) | |
880 | { | |
881 | SEM_ARG sem_arg = sc; | |
882 | const ARGBUF *abuf = SEM_ARGBUF (sem_arg); | |
883 | int first_p = abuf->fields.before.first_p; | |
884 | const ARGBUF *cur_abuf = SEM_ARGBUF (sc + 1); | |
885 | const IDESC *cur_idesc = cur_abuf->idesc; | |
886 | PCADDR pc = cur_abuf->addr; | |
887 | ||
888 | if (ARGBUF_PROFILE_P (cur_abuf)) | |
889 | PROFILE_COUNT_INSN (current_cpu, pc, cur_idesc->num); | |
890 | ||
891 | /* If this isn't the first insn, finish up the previous one. */ | |
892 | ||
893 | if (! first_p) | |
894 | { | |
895 | if (PROFILE_MODEL_P (current_cpu)) | |
896 | { | |
897 | const SEM_ARG prev_sem_arg = sc - 1; | |
898 | const ARGBUF *prev_abuf = SEM_ARGBUF (prev_sem_arg); | |
899 | const IDESC *prev_idesc = prev_abuf->idesc; | |
900 | int cycles; | |
901 | ||
902 | /* ??? May want to measure all insns if doing insn tracing. */ | |
903 | if (ARGBUF_PROFILE_P (prev_abuf)) | |
904 | { | |
905 | cycles = (*prev_idesc->timing->model_fn) (current_cpu, prev_sem_arg); | |
906 | @cpu@_model_insn_after (current_cpu, 0 /*last_p*/, cycles); | |
907 | } | |
908 | } | |
909 | ||
910 | TRACE_INSN_FINI (current_cpu, cur_abuf, 0 /*last_p*/); | |
911 | } | |
912 | ||
913 | /* FIXME: Later make cover macros: PROFILE_INSN_{INIT,FINI}. */ | |
914 | if (PROFILE_MODEL_P (current_cpu) | |
915 | && ARGBUF_PROFILE_P (cur_abuf)) | |
916 | @cpu@_model_insn_before (current_cpu, first_p); | |
917 | ||
918 | TRACE_INSN_INIT (current_cpu, cur_abuf, first_p); | |
919 | TRACE_INSN (current_cpu, cur_idesc->idata, cur_abuf, pc); | |
920 | } | |
921 | ||
922 | /* x-after handler. | |
923 | This is called after a serial insn or at the end of a group of parallel | |
924 | insns. */ | |
925 | ||
926 | void | |
927 | @cpu@_pbb_after (SIM_CPU *current_cpu, SCACHE *sc) | |
928 | { | |
929 | SEM_ARG sem_arg = sc; | |
930 | const ARGBUF *abuf = SEM_ARGBUF (sem_arg); | |
931 | const SEM_ARG prev_sem_arg = sc - 1; | |
932 | const ARGBUF *prev_abuf = SEM_ARGBUF (prev_sem_arg); | |
933 | ||
934 | /* ??? May want to measure all insns if doing insn tracing. */ | |
935 | if (PROFILE_MODEL_P (current_cpu) | |
936 | && ARGBUF_PROFILE_P (prev_abuf)) | |
937 | { | |
938 | const IDESC *prev_idesc = prev_abuf->idesc; | |
939 | int cycles; | |
940 | ||
941 | cycles = (*prev_idesc->timing->model_fn) (current_cpu, prev_sem_arg); | |
942 | @cpu@_model_insn_after (current_cpu, 1 /*last_p*/, cycles); | |
943 | } | |
944 | TRACE_INSN_FINI (current_cpu, prev_abuf, 1 /*last_p*/); | |
945 | } | |
946 | ||
947 | #define FAST_P 0 | |
948 | ||
949 | void | |
950 | @cpu@_engine_run_full (SIM_CPU *current_cpu) | |
951 | { | |
952 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
953 | SCACHE *scache = CPU_SCACHE_CACHE (current_cpu); | |
954 | /* virtual program counter */ | |
955 | SEM_PC vpc; | |
956 | #if WITH_SEM_SWITCH_FULL | |
957 | /* For communication between cti's and cti-chain. */ | |
958 | PCADDR pbb_br_npc; | |
959 | SEM_PC *pbb_br_npc_ptr; | |
960 | #endif | |
961 | ||
962 | EOF | |
963 | ||
964 | if [ x$parallel != xno ] ; then | |
965 | cat << EOF | |
966 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
967 | PAREXEC *par_exec = &pbufs[0]; | |
968 | ||
969 | EOF | |
970 | fi | |
971 | ||
972 | # Any initialization code before looping starts. | |
973 | # Note that this code may declare some locals. | |
974 | ${SHELL} $infile init | |
975 | ||
976 | cat << EOF | |
977 | ||
978 | if (! CPU_IDESC_SEM_INIT_P (current_cpu)) | |
979 | { | |
980 | /* ??? 'twould be nice to move this up a level and only call it once. | |
981 | On the other hand, in the "let's go fast" case the test is only done | |
982 | once per pbb (since we only return to the main loop at the end of | |
983 | a pbb). And in the "let's run until we're done" case we don't return | |
984 | until the program exits. */ | |
985 | ||
986 | #if WITH_SEM_SWITCH_FULL && defined (__GNUC__) | |
987 | /* ??? Later maybe paste sem-switch.c in when building mainloop.c. */ | |
988 | #define DEFINE_LABELS | |
989 | #include "$switch" | |
990 | #endif | |
991 | ||
992 | /* Initialize the "begin (compile) a pbb" virtual insn. */ | |
993 | vpc = CPU_SCACHE_PBB_BEGIN (current_cpu); | |
994 | SEM_SET_FULL_CODE (SEM_ARGBUF (vpc), | |
995 | & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEGIN]); | |
996 | vpc->argbuf.idesc = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEGIN]; | |
997 | ||
998 | CPU_IDESC_SEM_INIT_P (current_cpu) = 1; | |
999 | } | |
1000 | ||
1001 | CPU_RUNNING_P (current_cpu) = 1; | |
1002 | /* ??? In the case where we're returning to the main loop after every | |
1003 | pbb we don't want to call pbb_begin each time (which hashes on the pc | |
1004 | and does a table lookup). A way to speed this up is to save vpc | |
1005 | between calls. */ | |
1006 | vpc = @cpu@_pbb_begin (current_cpu, FAST_P); | |
1007 | ||
1008 | do | |
1009 | { | |
1010 | /* begin full-exec-pbb */ | |
1011 | EOF | |
1012 | ||
1013 | ${SHELL} $infile full-exec-pbb | |
1014 | ||
1015 | cat << EOF | |
1016 | /* end full-exec-pbb */ | |
1017 | } | |
1018 | while (CPU_RUNNING_P (current_cpu)); | |
1019 | } | |
1020 | ||
1021 | #undef FAST_P | |
1022 | ||
1023 | EOF | |
1024 | ||
1025 | #################################### | |
1026 | ||
1027 | # Compile engine: fast version. | |
1028 | ||
1029 | if [ x$fast = xyes ] ; then | |
1030 | ||
1031 | cat << EOF | |
1032 | ||
1033 | #define FAST_P 1 | |
1034 | ||
1035 | void | |
1036 | @cpu@_engine_run_fast (SIM_CPU *current_cpu) | |
1037 | { | |
1038 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
1039 | SCACHE *scache = CPU_SCACHE_CACHE (current_cpu); | |
1040 | /* virtual program counter */ | |
1041 | SEM_PC vpc; | |
1042 | #if WITH_SEM_SWITCH_FAST | |
1043 | /* For communication between cti's and cti-chain. */ | |
1044 | PCADDR pbb_br_npc; | |
1045 | SEM_PC *pbb_br_npc_ptr; | |
1046 | #endif | |
1047 | ||
1048 | EOF | |
1049 | ||
1050 | if [ x$parallel != xno ] ; then | |
1051 | cat << EOF | |
1052 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
1053 | PAREXEC *par_exec = &pbufs[0]; | |
1054 | ||
1055 | EOF | |
1056 | fi | |
1057 | ||
1058 | # Any initialization code before looping starts. | |
1059 | # Note that this code may declare some locals. | |
1060 | ${SHELL} $infile init | |
1061 | ||
1062 | cat << EOF | |
1063 | ||
1064 | if (! CPU_IDESC_SEM_INIT_P (current_cpu)) | |
1065 | { | |
1066 | /* ??? 'twould be nice to move this up a level and only call it once. | |
1067 | On the other hand, in the "let's go fast" case the test is only done | |
1068 | once per pbb (since we only return to the main loop at the end of | |
1069 | a pbb). And in the "let's run until we're done" case we don't return | |
1070 | until the program exits. */ | |
1071 | ||
1072 | #if WITH_SEM_SWITCH_FAST && defined (__GNUC__) | |
1073 | /* ??? Later maybe paste sem-switch.c in when building mainloop.c. */ | |
1074 | #define DEFINE_LABELS | |
1075 | #include "$switch" | |
1076 | #endif | |
1077 | ||
1078 | /* Initialize the "begin (compile) a pbb" virtual insn. */ | |
1079 | vpc = CPU_SCACHE_PBB_BEGIN (current_cpu); | |
1080 | SEM_SET_FAST_CODE (SEM_ARGBUF (vpc), | |
1081 | & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEGIN]); | |
1082 | vpc->argbuf.idesc = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEGIN]; | |
1083 | ||
1084 | CPU_IDESC_SEM_INIT_P (current_cpu) = 1; | |
1085 | } | |
1086 | ||
1087 | CPU_RUNNING_P (current_cpu) = 1; | |
1088 | /* ??? In the case where we're returning to the main loop after every | |
1089 | pbb we don't want to call pbb_begin each time (which hashes on the pc | |
1090 | and does a table lookup). A way to speed this up is to save vpc | |
1091 | between calls. */ | |
1092 | vpc = @cpu@_pbb_begin (current_cpu, FAST_P); | |
1093 | ||
1094 | do | |
1095 | { | |
1096 | /* begin fast-exec-pbb */ | |
1097 | EOF | |
1098 | ||
1099 | ${SHELL} $infile fast-exec-pbb | |
1100 | ||
1101 | cat << EOF | |
1102 | /* end fast-exec-pbb */ | |
1103 | } | |
1104 | while (CPU_RUNNING_P (current_cpu)); | |
1105 | } | |
1106 | ||
1107 | #undef FAST_P | |
1108 | ||
1109 | EOF | |
1110 | fi # -fast | |
1111 | ||
1112 | fi # -pbb | |
1113 | ||
1114 | # Process @cpu@,@CPU@ appearing in mainloop.in. | |
1115 | sed -e "s/@cpu@/$cpu/g" -e "s/@CPU@/$CPU/g" < tmp-mloop.cin > mloop.cin | |
1116 | rc=$? | |
1117 | rm -f tmp-mloop.cin | |
1118 | ||
1119 | exit $rc |