xref: /openbmc/qemu/tcg/optimize.c (revision 6e99f574)
1 /*
2  * Optimizations for Tiny Code Generator for QEMU
3  *
4  * Copyright (c) 2010 Samsung Electronics.
5  * Contributed by Kirill Batuzov <batuzovk@ispras.ru>
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 #include "qemu-common.h"
28 #include "exec/cpu-common.h"
29 #include "tcg-op.h"
30 
31 #define CASE_OP_32_64(x)                        \
32         glue(glue(case INDEX_op_, x), _i32):    \
33         glue(glue(case INDEX_op_, x), _i64)
34 
35 struct tcg_temp_info {
36     bool is_const;
37     uint16_t prev_copy;
38     uint16_t next_copy;
39     tcg_target_ulong val;
40     tcg_target_ulong mask;
41 };
42 
43 static struct tcg_temp_info temps[TCG_MAX_TEMPS];
44 static TCGTempSet temps_used;
45 
46 static inline bool temp_is_const(TCGArg arg)
47 {
48     return temps[arg].is_const;
49 }
50 
51 static inline bool temp_is_copy(TCGArg arg)
52 {
53     return temps[arg].next_copy != arg;
54 }
55 
56 /* Reset TEMP's state, possibly removing the temp for the list of copies.  */
57 static void reset_temp(TCGArg temp)
58 {
59     temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy;
60     temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy;
61     temps[temp].next_copy = temp;
62     temps[temp].prev_copy = temp;
63     temps[temp].is_const = false;
64     temps[temp].mask = -1;
65 }
66 
67 /* Reset all temporaries, given that there are NB_TEMPS of them.  */
68 static void reset_all_temps(int nb_temps)
69 {
70     bitmap_zero(temps_used.l, nb_temps);
71 }
72 
73 /* Initialize and activate a temporary.  */
74 static void init_temp_info(TCGArg temp)
75 {
76     if (!test_bit(temp, temps_used.l)) {
77         temps[temp].next_copy = temp;
78         temps[temp].prev_copy = temp;
79         temps[temp].is_const = false;
80         temps[temp].mask = -1;
81         set_bit(temp, temps_used.l);
82     }
83 }
84 
85 static int op_bits(TCGOpcode op)
86 {
87     const TCGOpDef *def = &tcg_op_defs[op];
88     return def->flags & TCG_OPF_64BIT ? 64 : 32;
89 }
90 
91 static TCGOpcode op_to_mov(TCGOpcode op)
92 {
93     switch (op_bits(op)) {
94     case 32:
95         return INDEX_op_mov_i32;
96     case 64:
97         return INDEX_op_mov_i64;
98     default:
99         fprintf(stderr, "op_to_mov: unexpected return value of "
100                 "function op_bits.\n");
101         tcg_abort();
102     }
103 }
104 
105 static TCGOpcode op_to_movi(TCGOpcode op)
106 {
107     switch (op_bits(op)) {
108     case 32:
109         return INDEX_op_movi_i32;
110     case 64:
111         return INDEX_op_movi_i64;
112     default:
113         fprintf(stderr, "op_to_movi: unexpected return value of "
114                 "function op_bits.\n");
115         tcg_abort();
116     }
117 }
118 
119 static TCGArg find_better_copy(TCGContext *s, TCGArg temp)
120 {
121     TCGArg i;
122 
123     /* If this is already a global, we can't do better. */
124     if (temp < s->nb_globals) {
125         return temp;
126     }
127 
128     /* Search for a global first. */
129     for (i = temps[temp].next_copy ; i != temp ; i = temps[i].next_copy) {
130         if (i < s->nb_globals) {
131             return i;
132         }
133     }
134 
135     /* If it is a temp, search for a temp local. */
136     if (!s->temps[temp].temp_local) {
137         for (i = temps[temp].next_copy ; i != temp ; i = temps[i].next_copy) {
138             if (s->temps[i].temp_local) {
139                 return i;
140             }
141         }
142     }
143 
144     /* Failure to find a better representation, return the same temp. */
145     return temp;
146 }
147 
148 static bool temps_are_copies(TCGArg arg1, TCGArg arg2)
149 {
150     TCGArg i;
151 
152     if (arg1 == arg2) {
153         return true;
154     }
155 
156     if (!temp_is_copy(arg1) || !temp_is_copy(arg2)) {
157         return false;
158     }
159 
160     for (i = temps[arg1].next_copy ; i != arg1 ; i = temps[i].next_copy) {
161         if (i == arg2) {
162             return true;
163         }
164     }
165 
166     return false;
167 }
168 
169 static void tcg_opt_gen_movi(TCGContext *s, TCGOp *op, TCGArg *args,
170                              TCGArg dst, TCGArg val)
171 {
172     TCGOpcode new_op = op_to_movi(op->opc);
173     tcg_target_ulong mask;
174 
175     op->opc = new_op;
176 
177     reset_temp(dst);
178     temps[dst].is_const = true;
179     temps[dst].val = val;
180     mask = val;
181     if (TCG_TARGET_REG_BITS > 32 && new_op == INDEX_op_movi_i32) {
182         /* High bits of the destination are now garbage.  */
183         mask |= ~0xffffffffull;
184     }
185     temps[dst].mask = mask;
186 
187     args[0] = dst;
188     args[1] = val;
189 }
190 
191 static void tcg_opt_gen_mov(TCGContext *s, TCGOp *op, TCGArg *args,
192                             TCGArg dst, TCGArg src)
193 {
194     if (temps_are_copies(dst, src)) {
195         tcg_op_remove(s, op);
196         return;
197     }
198 
199     TCGOpcode new_op = op_to_mov(op->opc);
200     tcg_target_ulong mask;
201 
202     op->opc = new_op;
203 
204     reset_temp(dst);
205     mask = temps[src].mask;
206     if (TCG_TARGET_REG_BITS > 32 && new_op == INDEX_op_mov_i32) {
207         /* High bits of the destination are now garbage.  */
208         mask |= ~0xffffffffull;
209     }
210     temps[dst].mask = mask;
211 
212     if (s->temps[src].type == s->temps[dst].type) {
213         temps[dst].next_copy = temps[src].next_copy;
214         temps[dst].prev_copy = src;
215         temps[temps[dst].next_copy].prev_copy = dst;
216         temps[src].next_copy = dst;
217         temps[dst].is_const = temps[src].is_const;
218         temps[dst].val = temps[src].val;
219     }
220 
221     args[0] = dst;
222     args[1] = src;
223 }
224 
225 static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y)
226 {
227     uint64_t l64, h64;
228 
229     switch (op) {
230     CASE_OP_32_64(add):
231         return x + y;
232 
233     CASE_OP_32_64(sub):
234         return x - y;
235 
236     CASE_OP_32_64(mul):
237         return x * y;
238 
239     CASE_OP_32_64(and):
240         return x & y;
241 
242     CASE_OP_32_64(or):
243         return x | y;
244 
245     CASE_OP_32_64(xor):
246         return x ^ y;
247 
248     case INDEX_op_shl_i32:
249         return (uint32_t)x << (y & 31);
250 
251     case INDEX_op_shl_i64:
252         return (uint64_t)x << (y & 63);
253 
254     case INDEX_op_shr_i32:
255         return (uint32_t)x >> (y & 31);
256 
257     case INDEX_op_shr_i64:
258         return (uint64_t)x >> (y & 63);
259 
260     case INDEX_op_sar_i32:
261         return (int32_t)x >> (y & 31);
262 
263     case INDEX_op_sar_i64:
264         return (int64_t)x >> (y & 63);
265 
266     case INDEX_op_rotr_i32:
267         return ror32(x, y & 31);
268 
269     case INDEX_op_rotr_i64:
270         return ror64(x, y & 63);
271 
272     case INDEX_op_rotl_i32:
273         return rol32(x, y & 31);
274 
275     case INDEX_op_rotl_i64:
276         return rol64(x, y & 63);
277 
278     CASE_OP_32_64(not):
279         return ~x;
280 
281     CASE_OP_32_64(neg):
282         return -x;
283 
284     CASE_OP_32_64(andc):
285         return x & ~y;
286 
287     CASE_OP_32_64(orc):
288         return x | ~y;
289 
290     CASE_OP_32_64(eqv):
291         return ~(x ^ y);
292 
293     CASE_OP_32_64(nand):
294         return ~(x & y);
295 
296     CASE_OP_32_64(nor):
297         return ~(x | y);
298 
299     CASE_OP_32_64(ext8s):
300         return (int8_t)x;
301 
302     CASE_OP_32_64(ext16s):
303         return (int16_t)x;
304 
305     CASE_OP_32_64(ext8u):
306         return (uint8_t)x;
307 
308     CASE_OP_32_64(ext16u):
309         return (uint16_t)x;
310 
311     case INDEX_op_ext_i32_i64:
312     case INDEX_op_ext32s_i64:
313         return (int32_t)x;
314 
315     case INDEX_op_extu_i32_i64:
316     case INDEX_op_extrl_i64_i32:
317     case INDEX_op_ext32u_i64:
318         return (uint32_t)x;
319 
320     case INDEX_op_extrh_i64_i32:
321         return (uint64_t)x >> 32;
322 
323     case INDEX_op_muluh_i32:
324         return ((uint64_t)(uint32_t)x * (uint32_t)y) >> 32;
325     case INDEX_op_mulsh_i32:
326         return ((int64_t)(int32_t)x * (int32_t)y) >> 32;
327 
328     case INDEX_op_muluh_i64:
329         mulu64(&l64, &h64, x, y);
330         return h64;
331     case INDEX_op_mulsh_i64:
332         muls64(&l64, &h64, x, y);
333         return h64;
334 
335     case INDEX_op_div_i32:
336         /* Avoid crashing on divide by zero, otherwise undefined.  */
337         return (int32_t)x / ((int32_t)y ? : 1);
338     case INDEX_op_divu_i32:
339         return (uint32_t)x / ((uint32_t)y ? : 1);
340     case INDEX_op_div_i64:
341         return (int64_t)x / ((int64_t)y ? : 1);
342     case INDEX_op_divu_i64:
343         return (uint64_t)x / ((uint64_t)y ? : 1);
344 
345     case INDEX_op_rem_i32:
346         return (int32_t)x % ((int32_t)y ? : 1);
347     case INDEX_op_remu_i32:
348         return (uint32_t)x % ((uint32_t)y ? : 1);
349     case INDEX_op_rem_i64:
350         return (int64_t)x % ((int64_t)y ? : 1);
351     case INDEX_op_remu_i64:
352         return (uint64_t)x % ((uint64_t)y ? : 1);
353 
354     default:
355         fprintf(stderr,
356                 "Unrecognized operation %d in do_constant_folding.\n", op);
357         tcg_abort();
358     }
359 }
360 
361 static TCGArg do_constant_folding(TCGOpcode op, TCGArg x, TCGArg y)
362 {
363     TCGArg res = do_constant_folding_2(op, x, y);
364     if (op_bits(op) == 32) {
365         res = (int32_t)res;
366     }
367     return res;
368 }
369 
370 static bool do_constant_folding_cond_32(uint32_t x, uint32_t y, TCGCond c)
371 {
372     switch (c) {
373     case TCG_COND_EQ:
374         return x == y;
375     case TCG_COND_NE:
376         return x != y;
377     case TCG_COND_LT:
378         return (int32_t)x < (int32_t)y;
379     case TCG_COND_GE:
380         return (int32_t)x >= (int32_t)y;
381     case TCG_COND_LE:
382         return (int32_t)x <= (int32_t)y;
383     case TCG_COND_GT:
384         return (int32_t)x > (int32_t)y;
385     case TCG_COND_LTU:
386         return x < y;
387     case TCG_COND_GEU:
388         return x >= y;
389     case TCG_COND_LEU:
390         return x <= y;
391     case TCG_COND_GTU:
392         return x > y;
393     default:
394         tcg_abort();
395     }
396 }
397 
398 static bool do_constant_folding_cond_64(uint64_t x, uint64_t y, TCGCond c)
399 {
400     switch (c) {
401     case TCG_COND_EQ:
402         return x == y;
403     case TCG_COND_NE:
404         return x != y;
405     case TCG_COND_LT:
406         return (int64_t)x < (int64_t)y;
407     case TCG_COND_GE:
408         return (int64_t)x >= (int64_t)y;
409     case TCG_COND_LE:
410         return (int64_t)x <= (int64_t)y;
411     case TCG_COND_GT:
412         return (int64_t)x > (int64_t)y;
413     case TCG_COND_LTU:
414         return x < y;
415     case TCG_COND_GEU:
416         return x >= y;
417     case TCG_COND_LEU:
418         return x <= y;
419     case TCG_COND_GTU:
420         return x > y;
421     default:
422         tcg_abort();
423     }
424 }
425 
426 static bool do_constant_folding_cond_eq(TCGCond c)
427 {
428     switch (c) {
429     case TCG_COND_GT:
430     case TCG_COND_LTU:
431     case TCG_COND_LT:
432     case TCG_COND_GTU:
433     case TCG_COND_NE:
434         return 0;
435     case TCG_COND_GE:
436     case TCG_COND_GEU:
437     case TCG_COND_LE:
438     case TCG_COND_LEU:
439     case TCG_COND_EQ:
440         return 1;
441     default:
442         tcg_abort();
443     }
444 }
445 
446 /* Return 2 if the condition can't be simplified, and the result
447    of the condition (0 or 1) if it can */
448 static TCGArg do_constant_folding_cond(TCGOpcode op, TCGArg x,
449                                        TCGArg y, TCGCond c)
450 {
451     if (temp_is_const(x) && temp_is_const(y)) {
452         switch (op_bits(op)) {
453         case 32:
454             return do_constant_folding_cond_32(temps[x].val, temps[y].val, c);
455         case 64:
456             return do_constant_folding_cond_64(temps[x].val, temps[y].val, c);
457         default:
458             tcg_abort();
459         }
460     } else if (temps_are_copies(x, y)) {
461         return do_constant_folding_cond_eq(c);
462     } else if (temp_is_const(y) && temps[y].val == 0) {
463         switch (c) {
464         case TCG_COND_LTU:
465             return 0;
466         case TCG_COND_GEU:
467             return 1;
468         default:
469             return 2;
470         }
471     }
472     return 2;
473 }
474 
475 /* Return 2 if the condition can't be simplified, and the result
476    of the condition (0 or 1) if it can */
477 static TCGArg do_constant_folding_cond2(TCGArg *p1, TCGArg *p2, TCGCond c)
478 {
479     TCGArg al = p1[0], ah = p1[1];
480     TCGArg bl = p2[0], bh = p2[1];
481 
482     if (temp_is_const(bl) && temp_is_const(bh)) {
483         uint64_t b = ((uint64_t)temps[bh].val << 32) | (uint32_t)temps[bl].val;
484 
485         if (temp_is_const(al) && temp_is_const(ah)) {
486             uint64_t a;
487             a = ((uint64_t)temps[ah].val << 32) | (uint32_t)temps[al].val;
488             return do_constant_folding_cond_64(a, b, c);
489         }
490         if (b == 0) {
491             switch (c) {
492             case TCG_COND_LTU:
493                 return 0;
494             case TCG_COND_GEU:
495                 return 1;
496             default:
497                 break;
498             }
499         }
500     }
501     if (temps_are_copies(al, bl) && temps_are_copies(ah, bh)) {
502         return do_constant_folding_cond_eq(c);
503     }
504     return 2;
505 }
506 
507 static bool swap_commutative(TCGArg dest, TCGArg *p1, TCGArg *p2)
508 {
509     TCGArg a1 = *p1, a2 = *p2;
510     int sum = 0;
511     sum += temp_is_const(a1);
512     sum -= temp_is_const(a2);
513 
514     /* Prefer the constant in second argument, and then the form
515        op a, a, b, which is better handled on non-RISC hosts. */
516     if (sum > 0 || (sum == 0 && dest == a2)) {
517         *p1 = a2;
518         *p2 = a1;
519         return true;
520     }
521     return false;
522 }
523 
524 static bool swap_commutative2(TCGArg *p1, TCGArg *p2)
525 {
526     int sum = 0;
527     sum += temp_is_const(p1[0]);
528     sum += temp_is_const(p1[1]);
529     sum -= temp_is_const(p2[0]);
530     sum -= temp_is_const(p2[1]);
531     if (sum > 0) {
532         TCGArg t;
533         t = p1[0], p1[0] = p2[0], p2[0] = t;
534         t = p1[1], p1[1] = p2[1], p2[1] = t;
535         return true;
536     }
537     return false;
538 }
539 
540 /* Propagate constants and copies, fold constant expressions. */
541 void tcg_optimize(TCGContext *s)
542 {
543     int oi, oi_next, nb_temps, nb_globals;
544     TCGArg *prev_mb_args = NULL;
545 
546     /* Array VALS has an element for each temp.
547        If this temp holds a constant then its value is kept in VALS' element.
548        If this temp is a copy of other ones then the other copies are
549        available through the doubly linked circular list. */
550 
551     nb_temps = s->nb_temps;
552     nb_globals = s->nb_globals;
553     reset_all_temps(nb_temps);
554 
555     for (oi = s->gen_op_buf[0].next; oi != 0; oi = oi_next) {
556         tcg_target_ulong mask, partmask, affected;
557         int nb_oargs, nb_iargs, i;
558         TCGArg tmp;
559 
560         TCGOp * const op = &s->gen_op_buf[oi];
561         TCGArg * const args = &s->gen_opparam_buf[op->args];
562         TCGOpcode opc = op->opc;
563         const TCGOpDef *def = &tcg_op_defs[opc];
564 
565         oi_next = op->next;
566 
567         /* Count the arguments, and initialize the temps that are
568            going to be used */
569         if (opc == INDEX_op_call) {
570             nb_oargs = op->callo;
571             nb_iargs = op->calli;
572             for (i = 0; i < nb_oargs + nb_iargs; i++) {
573                 tmp = args[i];
574                 if (tmp != TCG_CALL_DUMMY_ARG) {
575                     init_temp_info(tmp);
576                 }
577             }
578         } else {
579             nb_oargs = def->nb_oargs;
580             nb_iargs = def->nb_iargs;
581             for (i = 0; i < nb_oargs + nb_iargs; i++) {
582                 init_temp_info(args[i]);
583             }
584         }
585 
586         /* Do copy propagation */
587         for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) {
588             if (temp_is_copy(args[i])) {
589                 args[i] = find_better_copy(s, args[i]);
590             }
591         }
592 
593         /* For commutative operations make constant second argument */
594         switch (opc) {
595         CASE_OP_32_64(add):
596         CASE_OP_32_64(mul):
597         CASE_OP_32_64(and):
598         CASE_OP_32_64(or):
599         CASE_OP_32_64(xor):
600         CASE_OP_32_64(eqv):
601         CASE_OP_32_64(nand):
602         CASE_OP_32_64(nor):
603         CASE_OP_32_64(muluh):
604         CASE_OP_32_64(mulsh):
605             swap_commutative(args[0], &args[1], &args[2]);
606             break;
607         CASE_OP_32_64(brcond):
608             if (swap_commutative(-1, &args[0], &args[1])) {
609                 args[2] = tcg_swap_cond(args[2]);
610             }
611             break;
612         CASE_OP_32_64(setcond):
613             if (swap_commutative(args[0], &args[1], &args[2])) {
614                 args[3] = tcg_swap_cond(args[3]);
615             }
616             break;
617         CASE_OP_32_64(movcond):
618             if (swap_commutative(-1, &args[1], &args[2])) {
619                 args[5] = tcg_swap_cond(args[5]);
620             }
621             /* For movcond, we canonicalize the "false" input reg to match
622                the destination reg so that the tcg backend can implement
623                a "move if true" operation.  */
624             if (swap_commutative(args[0], &args[4], &args[3])) {
625                 args[5] = tcg_invert_cond(args[5]);
626             }
627             break;
628         CASE_OP_32_64(add2):
629             swap_commutative(args[0], &args[2], &args[4]);
630             swap_commutative(args[1], &args[3], &args[5]);
631             break;
632         CASE_OP_32_64(mulu2):
633         CASE_OP_32_64(muls2):
634             swap_commutative(args[0], &args[2], &args[3]);
635             break;
636         case INDEX_op_brcond2_i32:
637             if (swap_commutative2(&args[0], &args[2])) {
638                 args[4] = tcg_swap_cond(args[4]);
639             }
640             break;
641         case INDEX_op_setcond2_i32:
642             if (swap_commutative2(&args[1], &args[3])) {
643                 args[5] = tcg_swap_cond(args[5]);
644             }
645             break;
646         default:
647             break;
648         }
649 
650         /* Simplify expressions for "shift/rot r, 0, a => movi r, 0",
651            and "sub r, 0, a => neg r, a" case.  */
652         switch (opc) {
653         CASE_OP_32_64(shl):
654         CASE_OP_32_64(shr):
655         CASE_OP_32_64(sar):
656         CASE_OP_32_64(rotl):
657         CASE_OP_32_64(rotr):
658             if (temp_is_const(args[1]) && temps[args[1]].val == 0) {
659                 tcg_opt_gen_movi(s, op, args, args[0], 0);
660                 continue;
661             }
662             break;
663         CASE_OP_32_64(sub):
664             {
665                 TCGOpcode neg_op;
666                 bool have_neg;
667 
668                 if (temp_is_const(args[2])) {
669                     /* Proceed with possible constant folding. */
670                     break;
671                 }
672                 if (opc == INDEX_op_sub_i32) {
673                     neg_op = INDEX_op_neg_i32;
674                     have_neg = TCG_TARGET_HAS_neg_i32;
675                 } else {
676                     neg_op = INDEX_op_neg_i64;
677                     have_neg = TCG_TARGET_HAS_neg_i64;
678                 }
679                 if (!have_neg) {
680                     break;
681                 }
682                 if (temp_is_const(args[1]) && temps[args[1]].val == 0) {
683                     op->opc = neg_op;
684                     reset_temp(args[0]);
685                     args[1] = args[2];
686                     continue;
687                 }
688             }
689             break;
690         CASE_OP_32_64(xor):
691         CASE_OP_32_64(nand):
692             if (!temp_is_const(args[1])
693                 && temp_is_const(args[2]) && temps[args[2]].val == -1) {
694                 i = 1;
695                 goto try_not;
696             }
697             break;
698         CASE_OP_32_64(nor):
699             if (!temp_is_const(args[1])
700                 && temp_is_const(args[2]) && temps[args[2]].val == 0) {
701                 i = 1;
702                 goto try_not;
703             }
704             break;
705         CASE_OP_32_64(andc):
706             if (!temp_is_const(args[2])
707                 && temp_is_const(args[1]) && temps[args[1]].val == -1) {
708                 i = 2;
709                 goto try_not;
710             }
711             break;
712         CASE_OP_32_64(orc):
713         CASE_OP_32_64(eqv):
714             if (!temp_is_const(args[2])
715                 && temp_is_const(args[1]) && temps[args[1]].val == 0) {
716                 i = 2;
717                 goto try_not;
718             }
719             break;
720         try_not:
721             {
722                 TCGOpcode not_op;
723                 bool have_not;
724 
725                 if (def->flags & TCG_OPF_64BIT) {
726                     not_op = INDEX_op_not_i64;
727                     have_not = TCG_TARGET_HAS_not_i64;
728                 } else {
729                     not_op = INDEX_op_not_i32;
730                     have_not = TCG_TARGET_HAS_not_i32;
731                 }
732                 if (!have_not) {
733                     break;
734                 }
735                 op->opc = not_op;
736                 reset_temp(args[0]);
737                 args[1] = args[i];
738                 continue;
739             }
740         default:
741             break;
742         }
743 
744         /* Simplify expression for "op r, a, const => mov r, a" cases */
745         switch (opc) {
746         CASE_OP_32_64(add):
747         CASE_OP_32_64(sub):
748         CASE_OP_32_64(shl):
749         CASE_OP_32_64(shr):
750         CASE_OP_32_64(sar):
751         CASE_OP_32_64(rotl):
752         CASE_OP_32_64(rotr):
753         CASE_OP_32_64(or):
754         CASE_OP_32_64(xor):
755         CASE_OP_32_64(andc):
756             if (!temp_is_const(args[1])
757                 && temp_is_const(args[2]) && temps[args[2]].val == 0) {
758                 tcg_opt_gen_mov(s, op, args, args[0], args[1]);
759                 continue;
760             }
761             break;
762         CASE_OP_32_64(and):
763         CASE_OP_32_64(orc):
764         CASE_OP_32_64(eqv):
765             if (!temp_is_const(args[1])
766                 && temp_is_const(args[2]) && temps[args[2]].val == -1) {
767                 tcg_opt_gen_mov(s, op, args, args[0], args[1]);
768                 continue;
769             }
770             break;
771         default:
772             break;
773         }
774 
775         /* Simplify using known-zero bits. Currently only ops with a single
776            output argument is supported. */
777         mask = -1;
778         affected = -1;
779         switch (opc) {
780         CASE_OP_32_64(ext8s):
781             if ((temps[args[1]].mask & 0x80) != 0) {
782                 break;
783             }
784         CASE_OP_32_64(ext8u):
785             mask = 0xff;
786             goto and_const;
787         CASE_OP_32_64(ext16s):
788             if ((temps[args[1]].mask & 0x8000) != 0) {
789                 break;
790             }
791         CASE_OP_32_64(ext16u):
792             mask = 0xffff;
793             goto and_const;
794         case INDEX_op_ext32s_i64:
795             if ((temps[args[1]].mask & 0x80000000) != 0) {
796                 break;
797             }
798         case INDEX_op_ext32u_i64:
799             mask = 0xffffffffU;
800             goto and_const;
801 
802         CASE_OP_32_64(and):
803             mask = temps[args[2]].mask;
804             if (temp_is_const(args[2])) {
805         and_const:
806                 affected = temps[args[1]].mask & ~mask;
807             }
808             mask = temps[args[1]].mask & mask;
809             break;
810 
811         case INDEX_op_ext_i32_i64:
812             if ((temps[args[1]].mask & 0x80000000) != 0) {
813                 break;
814             }
815         case INDEX_op_extu_i32_i64:
816             /* We do not compute affected as it is a size changing op.  */
817             mask = (uint32_t)temps[args[1]].mask;
818             break;
819 
820         CASE_OP_32_64(andc):
821             /* Known-zeros does not imply known-ones.  Therefore unless
822                args[2] is constant, we can't infer anything from it.  */
823             if (temp_is_const(args[2])) {
824                 mask = ~temps[args[2]].mask;
825                 goto and_const;
826             }
827             /* But we certainly know nothing outside args[1] may be set. */
828             mask = temps[args[1]].mask;
829             break;
830 
831         case INDEX_op_sar_i32:
832             if (temp_is_const(args[2])) {
833                 tmp = temps[args[2]].val & 31;
834                 mask = (int32_t)temps[args[1]].mask >> tmp;
835             }
836             break;
837         case INDEX_op_sar_i64:
838             if (temp_is_const(args[2])) {
839                 tmp = temps[args[2]].val & 63;
840                 mask = (int64_t)temps[args[1]].mask >> tmp;
841             }
842             break;
843 
844         case INDEX_op_shr_i32:
845             if (temp_is_const(args[2])) {
846                 tmp = temps[args[2]].val & 31;
847                 mask = (uint32_t)temps[args[1]].mask >> tmp;
848             }
849             break;
850         case INDEX_op_shr_i64:
851             if (temp_is_const(args[2])) {
852                 tmp = temps[args[2]].val & 63;
853                 mask = (uint64_t)temps[args[1]].mask >> tmp;
854             }
855             break;
856 
857         case INDEX_op_extrl_i64_i32:
858             mask = (uint32_t)temps[args[1]].mask;
859             break;
860         case INDEX_op_extrh_i64_i32:
861             mask = (uint64_t)temps[args[1]].mask >> 32;
862             break;
863 
864         CASE_OP_32_64(shl):
865             if (temp_is_const(args[2])) {
866                 tmp = temps[args[2]].val & (TCG_TARGET_REG_BITS - 1);
867                 mask = temps[args[1]].mask << tmp;
868             }
869             break;
870 
871         CASE_OP_32_64(neg):
872             /* Set to 1 all bits to the left of the rightmost.  */
873             mask = -(temps[args[1]].mask & -temps[args[1]].mask);
874             break;
875 
876         CASE_OP_32_64(deposit):
877             mask = deposit64(temps[args[1]].mask, args[3], args[4],
878                              temps[args[2]].mask);
879             break;
880 
881         CASE_OP_32_64(or):
882         CASE_OP_32_64(xor):
883             mask = temps[args[1]].mask | temps[args[2]].mask;
884             break;
885 
886         CASE_OP_32_64(setcond):
887         case INDEX_op_setcond2_i32:
888             mask = 1;
889             break;
890 
891         CASE_OP_32_64(movcond):
892             mask = temps[args[3]].mask | temps[args[4]].mask;
893             break;
894 
895         CASE_OP_32_64(ld8u):
896             mask = 0xff;
897             break;
898         CASE_OP_32_64(ld16u):
899             mask = 0xffff;
900             break;
901         case INDEX_op_ld32u_i64:
902             mask = 0xffffffffu;
903             break;
904 
905         CASE_OP_32_64(qemu_ld):
906             {
907                 TCGMemOpIdx oi = args[nb_oargs + nb_iargs];
908                 TCGMemOp mop = get_memop(oi);
909                 if (!(mop & MO_SIGN)) {
910                     mask = (2ULL << ((8 << (mop & MO_SIZE)) - 1)) - 1;
911                 }
912             }
913             break;
914 
915         default:
916             break;
917         }
918 
919         /* 32-bit ops generate 32-bit results.  For the result is zero test
920            below, we can ignore high bits, but for further optimizations we
921            need to record that the high bits contain garbage.  */
922         partmask = mask;
923         if (!(def->flags & TCG_OPF_64BIT)) {
924             mask |= ~(tcg_target_ulong)0xffffffffu;
925             partmask &= 0xffffffffu;
926             affected &= 0xffffffffu;
927         }
928 
929         if (partmask == 0) {
930             tcg_debug_assert(nb_oargs == 1);
931             tcg_opt_gen_movi(s, op, args, args[0], 0);
932             continue;
933         }
934         if (affected == 0) {
935             tcg_debug_assert(nb_oargs == 1);
936             tcg_opt_gen_mov(s, op, args, args[0], args[1]);
937             continue;
938         }
939 
940         /* Simplify expression for "op r, a, 0 => movi r, 0" cases */
941         switch (opc) {
942         CASE_OP_32_64(and):
943         CASE_OP_32_64(mul):
944         CASE_OP_32_64(muluh):
945         CASE_OP_32_64(mulsh):
946             if ((temp_is_const(args[2]) && temps[args[2]].val == 0)) {
947                 tcg_opt_gen_movi(s, op, args, args[0], 0);
948                 continue;
949             }
950             break;
951         default:
952             break;
953         }
954 
955         /* Simplify expression for "op r, a, a => mov r, a" cases */
956         switch (opc) {
957         CASE_OP_32_64(or):
958         CASE_OP_32_64(and):
959             if (temps_are_copies(args[1], args[2])) {
960                 tcg_opt_gen_mov(s, op, args, args[0], args[1]);
961                 continue;
962             }
963             break;
964         default:
965             break;
966         }
967 
968         /* Simplify expression for "op r, a, a => movi r, 0" cases */
969         switch (opc) {
970         CASE_OP_32_64(andc):
971         CASE_OP_32_64(sub):
972         CASE_OP_32_64(xor):
973             if (temps_are_copies(args[1], args[2])) {
974                 tcg_opt_gen_movi(s, op, args, args[0], 0);
975                 continue;
976             }
977             break;
978         default:
979             break;
980         }
981 
982         /* Propagate constants through copy operations and do constant
983            folding.  Constants will be substituted to arguments by register
984            allocator where needed and possible.  Also detect copies. */
985         switch (opc) {
986         CASE_OP_32_64(mov):
987             tcg_opt_gen_mov(s, op, args, args[0], args[1]);
988             break;
989         CASE_OP_32_64(movi):
990             tcg_opt_gen_movi(s, op, args, args[0], args[1]);
991             break;
992 
993         CASE_OP_32_64(not):
994         CASE_OP_32_64(neg):
995         CASE_OP_32_64(ext8s):
996         CASE_OP_32_64(ext8u):
997         CASE_OP_32_64(ext16s):
998         CASE_OP_32_64(ext16u):
999         case INDEX_op_ext32s_i64:
1000         case INDEX_op_ext32u_i64:
1001         case INDEX_op_ext_i32_i64:
1002         case INDEX_op_extu_i32_i64:
1003         case INDEX_op_extrl_i64_i32:
1004         case INDEX_op_extrh_i64_i32:
1005             if (temp_is_const(args[1])) {
1006                 tmp = do_constant_folding(opc, temps[args[1]].val, 0);
1007                 tcg_opt_gen_movi(s, op, args, args[0], tmp);
1008                 break;
1009             }
1010             goto do_default;
1011 
1012         CASE_OP_32_64(add):
1013         CASE_OP_32_64(sub):
1014         CASE_OP_32_64(mul):
1015         CASE_OP_32_64(or):
1016         CASE_OP_32_64(and):
1017         CASE_OP_32_64(xor):
1018         CASE_OP_32_64(shl):
1019         CASE_OP_32_64(shr):
1020         CASE_OP_32_64(sar):
1021         CASE_OP_32_64(rotl):
1022         CASE_OP_32_64(rotr):
1023         CASE_OP_32_64(andc):
1024         CASE_OP_32_64(orc):
1025         CASE_OP_32_64(eqv):
1026         CASE_OP_32_64(nand):
1027         CASE_OP_32_64(nor):
1028         CASE_OP_32_64(muluh):
1029         CASE_OP_32_64(mulsh):
1030         CASE_OP_32_64(div):
1031         CASE_OP_32_64(divu):
1032         CASE_OP_32_64(rem):
1033         CASE_OP_32_64(remu):
1034             if (temp_is_const(args[1]) && temp_is_const(args[2])) {
1035                 tmp = do_constant_folding(opc, temps[args[1]].val,
1036                                           temps[args[2]].val);
1037                 tcg_opt_gen_movi(s, op, args, args[0], tmp);
1038                 break;
1039             }
1040             goto do_default;
1041 
1042         CASE_OP_32_64(deposit):
1043             if (temp_is_const(args[1]) && temp_is_const(args[2])) {
1044                 tmp = deposit64(temps[args[1]].val, args[3], args[4],
1045                                 temps[args[2]].val);
1046                 tcg_opt_gen_movi(s, op, args, args[0], tmp);
1047                 break;
1048             }
1049             goto do_default;
1050 
1051         CASE_OP_32_64(setcond):
1052             tmp = do_constant_folding_cond(opc, args[1], args[2], args[3]);
1053             if (tmp != 2) {
1054                 tcg_opt_gen_movi(s, op, args, args[0], tmp);
1055                 break;
1056             }
1057             goto do_default;
1058 
1059         CASE_OP_32_64(brcond):
1060             tmp = do_constant_folding_cond(opc, args[0], args[1], args[2]);
1061             if (tmp != 2) {
1062                 if (tmp) {
1063                     reset_all_temps(nb_temps);
1064                     op->opc = INDEX_op_br;
1065                     args[0] = args[3];
1066                 } else {
1067                     tcg_op_remove(s, op);
1068                 }
1069                 break;
1070             }
1071             goto do_default;
1072 
1073         CASE_OP_32_64(movcond):
1074             tmp = do_constant_folding_cond(opc, args[1], args[2], args[5]);
1075             if (tmp != 2) {
1076                 tcg_opt_gen_mov(s, op, args, args[0], args[4-tmp]);
1077                 break;
1078             }
1079             goto do_default;
1080 
1081         case INDEX_op_add2_i32:
1082         case INDEX_op_sub2_i32:
1083             if (temp_is_const(args[2]) && temp_is_const(args[3])
1084                 && temp_is_const(args[4]) && temp_is_const(args[5])) {
1085                 uint32_t al = temps[args[2]].val;
1086                 uint32_t ah = temps[args[3]].val;
1087                 uint32_t bl = temps[args[4]].val;
1088                 uint32_t bh = temps[args[5]].val;
1089                 uint64_t a = ((uint64_t)ah << 32) | al;
1090                 uint64_t b = ((uint64_t)bh << 32) | bl;
1091                 TCGArg rl, rh;
1092                 TCGOp *op2 = tcg_op_insert_before(s, op, INDEX_op_movi_i32, 2);
1093                 TCGArg *args2 = &s->gen_opparam_buf[op2->args];
1094 
1095                 if (opc == INDEX_op_add2_i32) {
1096                     a += b;
1097                 } else {
1098                     a -= b;
1099                 }
1100 
1101                 rl = args[0];
1102                 rh = args[1];
1103                 tcg_opt_gen_movi(s, op, args, rl, (int32_t)a);
1104                 tcg_opt_gen_movi(s, op2, args2, rh, (int32_t)(a >> 32));
1105 
1106                 /* We've done all we need to do with the movi.  Skip it.  */
1107                 oi_next = op2->next;
1108                 break;
1109             }
1110             goto do_default;
1111 
1112         case INDEX_op_mulu2_i32:
1113             if (temp_is_const(args[2]) && temp_is_const(args[3])) {
1114                 uint32_t a = temps[args[2]].val;
1115                 uint32_t b = temps[args[3]].val;
1116                 uint64_t r = (uint64_t)a * b;
1117                 TCGArg rl, rh;
1118                 TCGOp *op2 = tcg_op_insert_before(s, op, INDEX_op_movi_i32, 2);
1119                 TCGArg *args2 = &s->gen_opparam_buf[op2->args];
1120 
1121                 rl = args[0];
1122                 rh = args[1];
1123                 tcg_opt_gen_movi(s, op, args, rl, (int32_t)r);
1124                 tcg_opt_gen_movi(s, op2, args2, rh, (int32_t)(r >> 32));
1125 
1126                 /* We've done all we need to do with the movi.  Skip it.  */
1127                 oi_next = op2->next;
1128                 break;
1129             }
1130             goto do_default;
1131 
1132         case INDEX_op_brcond2_i32:
1133             tmp = do_constant_folding_cond2(&args[0], &args[2], args[4]);
1134             if (tmp != 2) {
1135                 if (tmp) {
1136             do_brcond_true:
1137                     reset_all_temps(nb_temps);
1138                     op->opc = INDEX_op_br;
1139                     args[0] = args[5];
1140                 } else {
1141             do_brcond_false:
1142                     tcg_op_remove(s, op);
1143                 }
1144             } else if ((args[4] == TCG_COND_LT || args[4] == TCG_COND_GE)
1145                        && temp_is_const(args[2]) && temps[args[2]].val == 0
1146                        && temp_is_const(args[3]) && temps[args[3]].val == 0) {
1147                 /* Simplify LT/GE comparisons vs zero to a single compare
1148                    vs the high word of the input.  */
1149             do_brcond_high:
1150                 reset_all_temps(nb_temps);
1151                 op->opc = INDEX_op_brcond_i32;
1152                 args[0] = args[1];
1153                 args[1] = args[3];
1154                 args[2] = args[4];
1155                 args[3] = args[5];
1156             } else if (args[4] == TCG_COND_EQ) {
1157                 /* Simplify EQ comparisons where one of the pairs
1158                    can be simplified.  */
1159                 tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
1160                                                args[0], args[2], TCG_COND_EQ);
1161                 if (tmp == 0) {
1162                     goto do_brcond_false;
1163                 } else if (tmp == 1) {
1164                     goto do_brcond_high;
1165                 }
1166                 tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
1167                                                args[1], args[3], TCG_COND_EQ);
1168                 if (tmp == 0) {
1169                     goto do_brcond_false;
1170                 } else if (tmp != 1) {
1171                     goto do_default;
1172                 }
1173             do_brcond_low:
1174                 reset_all_temps(nb_temps);
1175                 op->opc = INDEX_op_brcond_i32;
1176                 args[1] = args[2];
1177                 args[2] = args[4];
1178                 args[3] = args[5];
1179             } else if (args[4] == TCG_COND_NE) {
1180                 /* Simplify NE comparisons where one of the pairs
1181                    can be simplified.  */
1182                 tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
1183                                                args[0], args[2], TCG_COND_NE);
1184                 if (tmp == 0) {
1185                     goto do_brcond_high;
1186                 } else if (tmp == 1) {
1187                     goto do_brcond_true;
1188                 }
1189                 tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
1190                                                args[1], args[3], TCG_COND_NE);
1191                 if (tmp == 0) {
1192                     goto do_brcond_low;
1193                 } else if (tmp == 1) {
1194                     goto do_brcond_true;
1195                 }
1196                 goto do_default;
1197             } else {
1198                 goto do_default;
1199             }
1200             break;
1201 
1202         case INDEX_op_setcond2_i32:
1203             tmp = do_constant_folding_cond2(&args[1], &args[3], args[5]);
1204             if (tmp != 2) {
1205             do_setcond_const:
1206                 tcg_opt_gen_movi(s, op, args, args[0], tmp);
1207             } else if ((args[5] == TCG_COND_LT || args[5] == TCG_COND_GE)
1208                        && temp_is_const(args[3]) && temps[args[3]].val == 0
1209                        && temp_is_const(args[4]) && temps[args[4]].val == 0) {
1210                 /* Simplify LT/GE comparisons vs zero to a single compare
1211                    vs the high word of the input.  */
1212             do_setcond_high:
1213                 reset_temp(args[0]);
1214                 temps[args[0]].mask = 1;
1215                 op->opc = INDEX_op_setcond_i32;
1216                 args[1] = args[2];
1217                 args[2] = args[4];
1218                 args[3] = args[5];
1219             } else if (args[5] == TCG_COND_EQ) {
1220                 /* Simplify EQ comparisons where one of the pairs
1221                    can be simplified.  */
1222                 tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
1223                                                args[1], args[3], TCG_COND_EQ);
1224                 if (tmp == 0) {
1225                     goto do_setcond_const;
1226                 } else if (tmp == 1) {
1227                     goto do_setcond_high;
1228                 }
1229                 tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
1230                                                args[2], args[4], TCG_COND_EQ);
1231                 if (tmp == 0) {
1232                     goto do_setcond_high;
1233                 } else if (tmp != 1) {
1234                     goto do_default;
1235                 }
1236             do_setcond_low:
1237                 reset_temp(args[0]);
1238                 temps[args[0]].mask = 1;
1239                 op->opc = INDEX_op_setcond_i32;
1240                 args[2] = args[3];
1241                 args[3] = args[5];
1242             } else if (args[5] == TCG_COND_NE) {
1243                 /* Simplify NE comparisons where one of the pairs
1244                    can be simplified.  */
1245                 tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
1246                                                args[1], args[3], TCG_COND_NE);
1247                 if (tmp == 0) {
1248                     goto do_setcond_high;
1249                 } else if (tmp == 1) {
1250                     goto do_setcond_const;
1251                 }
1252                 tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
1253                                                args[2], args[4], TCG_COND_NE);
1254                 if (tmp == 0) {
1255                     goto do_setcond_low;
1256                 } else if (tmp == 1) {
1257                     goto do_setcond_const;
1258                 }
1259                 goto do_default;
1260             } else {
1261                 goto do_default;
1262             }
1263             break;
1264 
1265         case INDEX_op_call:
1266             if (!(args[nb_oargs + nb_iargs + 1]
1267                   & (TCG_CALL_NO_READ_GLOBALS | TCG_CALL_NO_WRITE_GLOBALS))) {
1268                 for (i = 0; i < nb_globals; i++) {
1269                     if (test_bit(i, temps_used.l)) {
1270                         reset_temp(i);
1271                     }
1272                 }
1273             }
1274             goto do_reset_output;
1275 
1276         default:
1277         do_default:
1278             /* Default case: we know nothing about operation (or were unable
1279                to compute the operation result) so no propagation is done.
1280                We trash everything if the operation is the end of a basic
1281                block, otherwise we only trash the output args.  "mask" is
1282                the non-zero bits mask for the first output arg.  */
1283             if (def->flags & TCG_OPF_BB_END) {
1284                 reset_all_temps(nb_temps);
1285             } else {
1286         do_reset_output:
1287                 for (i = 0; i < nb_oargs; i++) {
1288                     reset_temp(args[i]);
1289                     /* Save the corresponding known-zero bits mask for the
1290                        first output argument (only one supported so far). */
1291                     if (i == 0) {
1292                         temps[args[i]].mask = mask;
1293                     }
1294                 }
1295             }
1296             break;
1297         }
1298 
1299         /* Eliminate duplicate and redundant fence instructions.  */
1300         if (prev_mb_args) {
1301             switch (opc) {
1302             case INDEX_op_mb:
1303                 /* Merge two barriers of the same type into one,
1304                  * or a weaker barrier into a stronger one,
1305                  * or two weaker barriers into a stronger one.
1306                  *   mb X; mb Y => mb X|Y
1307                  *   mb; strl => mb; st
1308                  *   ldaq; mb => ld; mb
1309                  *   ldaq; strl => ld; mb; st
1310                  * Other combinations are also merged into a strong
1311                  * barrier.  This is stricter than specified but for
1312                  * the purposes of TCG is better than not optimizing.
1313                  */
1314                 prev_mb_args[0] |= args[0];
1315                 tcg_op_remove(s, op);
1316                 break;
1317 
1318             default:
1319                 /* Opcodes that end the block stop the optimization.  */
1320                 if ((def->flags & TCG_OPF_BB_END) == 0) {
1321                     break;
1322                 }
1323                 /* fallthru */
1324             case INDEX_op_qemu_ld_i32:
1325             case INDEX_op_qemu_ld_i64:
1326             case INDEX_op_qemu_st_i32:
1327             case INDEX_op_qemu_st_i64:
1328             case INDEX_op_call:
1329                 /* Opcodes that touch guest memory stop the optimization.  */
1330                 prev_mb_args = NULL;
1331                 break;
1332             }
1333         } else if (opc == INDEX_op_mb) {
1334             prev_mb_args = args;
1335         }
1336     }
1337 }
1338