Merge remote-tracking branch 'remotes/stefanha/tags/net-pull-request' into staging
[qemu.git] / tcg / optimize.c
blob37c11103a62b44bd6a83b32d68b043a2b13857cd
1 /*
2 * Optimizations for Tiny Code Generator for QEMU
4 * Copyright (c) 2010 Samsung Electronics.
5 * Contributed by Kirill Batuzov <batuzovk@ispras.ru>
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:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
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.
26 #include "config.h"
28 #include <stdlib.h>
29 #include <stdio.h>
31 #include "qemu-common.h"
32 #include "tcg-op.h"
34 #define CASE_OP_32_64(x) \
35 glue(glue(case INDEX_op_, x), _i32): \
36 glue(glue(case INDEX_op_, x), _i64)
38 typedef enum {
39 TCG_TEMP_UNDEF = 0,
40 TCG_TEMP_CONST,
41 TCG_TEMP_COPY,
42 } tcg_temp_state;
44 struct tcg_temp_info {
45 tcg_temp_state state;
46 uint16_t prev_copy;
47 uint16_t next_copy;
48 tcg_target_ulong val;
49 tcg_target_ulong mask;
52 static struct tcg_temp_info temps[TCG_MAX_TEMPS];
54 /* Reset TEMP's state to TCG_TEMP_UNDEF. If TEMP only had one copy, remove
55 the copy flag from the left temp. */
56 static void reset_temp(TCGArg temp)
58 if (temps[temp].state == TCG_TEMP_COPY) {
59 if (temps[temp].prev_copy == temps[temp].next_copy) {
60 temps[temps[temp].next_copy].state = TCG_TEMP_UNDEF;
61 } else {
62 temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy;
63 temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy;
66 temps[temp].state = TCG_TEMP_UNDEF;
67 temps[temp].mask = -1;
70 static TCGOp *insert_op_before(TCGContext *s, TCGOp *old_op,
71 TCGOpcode opc, int nargs)
73 int oi = s->gen_next_op_idx;
74 int pi = s->gen_next_parm_idx;
75 int prev = old_op->prev;
76 int next = old_op - s->gen_op_buf;
77 TCGOp *new_op;
79 tcg_debug_assert(oi < OPC_BUF_SIZE);
80 tcg_debug_assert(pi + nargs <= OPPARAM_BUF_SIZE);
81 s->gen_next_op_idx = oi + 1;
82 s->gen_next_parm_idx = pi + nargs;
84 new_op = &s->gen_op_buf[oi];
85 *new_op = (TCGOp){
86 .opc = opc,
87 .args = pi,
88 .prev = prev,
89 .next = next
91 if (prev >= 0) {
92 s->gen_op_buf[prev].next = oi;
93 } else {
94 s->gen_first_op_idx = oi;
96 old_op->prev = oi;
98 return new_op;
101 /* Reset all temporaries, given that there are NB_TEMPS of them. */
102 static void reset_all_temps(int nb_temps)
104 int i;
105 for (i = 0; i < nb_temps; i++) {
106 temps[i].state = TCG_TEMP_UNDEF;
107 temps[i].mask = -1;
111 static int op_bits(TCGOpcode op)
113 const TCGOpDef *def = &tcg_op_defs[op];
114 return def->flags & TCG_OPF_64BIT ? 64 : 32;
117 static TCGOpcode op_to_mov(TCGOpcode op)
119 switch (op_bits(op)) {
120 case 32:
121 return INDEX_op_mov_i32;
122 case 64:
123 return INDEX_op_mov_i64;
124 default:
125 fprintf(stderr, "op_to_mov: unexpected return value of "
126 "function op_bits.\n");
127 tcg_abort();
131 static TCGOpcode op_to_movi(TCGOpcode op)
133 switch (op_bits(op)) {
134 case 32:
135 return INDEX_op_movi_i32;
136 case 64:
137 return INDEX_op_movi_i64;
138 default:
139 fprintf(stderr, "op_to_movi: unexpected return value of "
140 "function op_bits.\n");
141 tcg_abort();
145 static TCGArg find_better_copy(TCGContext *s, TCGArg temp)
147 TCGArg i;
149 /* If this is already a global, we can't do better. */
150 if (temp < s->nb_globals) {
151 return temp;
154 /* Search for a global first. */
155 for (i = temps[temp].next_copy ; i != temp ; i = temps[i].next_copy) {
156 if (i < s->nb_globals) {
157 return i;
161 /* If it is a temp, search for a temp local. */
162 if (!s->temps[temp].temp_local) {
163 for (i = temps[temp].next_copy ; i != temp ; i = temps[i].next_copy) {
164 if (s->temps[i].temp_local) {
165 return i;
170 /* Failure to find a better representation, return the same temp. */
171 return temp;
174 static bool temps_are_copies(TCGArg arg1, TCGArg arg2)
176 TCGArg i;
178 if (arg1 == arg2) {
179 return true;
182 if (temps[arg1].state != TCG_TEMP_COPY
183 || temps[arg2].state != TCG_TEMP_COPY) {
184 return false;
187 for (i = temps[arg1].next_copy ; i != arg1 ; i = temps[i].next_copy) {
188 if (i == arg2) {
189 return true;
193 return false;
196 static void tcg_opt_gen_mov(TCGContext *s, TCGOp *op, TCGArg *args,
197 TCGOpcode old_op, TCGArg dst, TCGArg src)
199 TCGOpcode new_op = op_to_mov(old_op);
200 tcg_target_ulong mask;
202 op->opc = new_op;
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;
210 temps[dst].mask = mask;
212 assert(temps[src].state != TCG_TEMP_CONST);
214 if (s->temps[src].type == s->temps[dst].type) {
215 if (temps[src].state != TCG_TEMP_COPY) {
216 temps[src].state = TCG_TEMP_COPY;
217 temps[src].next_copy = src;
218 temps[src].prev_copy = src;
220 temps[dst].state = TCG_TEMP_COPY;
221 temps[dst].next_copy = temps[src].next_copy;
222 temps[dst].prev_copy = src;
223 temps[temps[dst].next_copy].prev_copy = dst;
224 temps[src].next_copy = dst;
227 args[0] = dst;
228 args[1] = src;
231 static void tcg_opt_gen_movi(TCGContext *s, TCGOp *op, TCGArg *args,
232 TCGOpcode old_op, TCGArg dst, TCGArg val)
234 TCGOpcode new_op = op_to_movi(old_op);
235 tcg_target_ulong mask;
237 op->opc = new_op;
239 reset_temp(dst);
240 temps[dst].state = TCG_TEMP_CONST;
241 temps[dst].val = val;
242 mask = val;
243 if (TCG_TARGET_REG_BITS > 32 && new_op == INDEX_op_mov_i32) {
244 /* High bits of the destination are now garbage. */
245 mask |= ~0xffffffffull;
247 temps[dst].mask = mask;
249 args[0] = dst;
250 args[1] = val;
253 static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y)
255 uint64_t l64, h64;
257 switch (op) {
258 CASE_OP_32_64(add):
259 return x + y;
261 CASE_OP_32_64(sub):
262 return x - y;
264 CASE_OP_32_64(mul):
265 return x * y;
267 CASE_OP_32_64(and):
268 return x & y;
270 CASE_OP_32_64(or):
271 return x | y;
273 CASE_OP_32_64(xor):
274 return x ^ y;
276 case INDEX_op_shl_i32:
277 return (uint32_t)x << (y & 31);
279 case INDEX_op_shl_i64:
280 return (uint64_t)x << (y & 63);
282 case INDEX_op_shr_i32:
283 return (uint32_t)x >> (y & 31);
285 case INDEX_op_trunc_shr_i32:
286 case INDEX_op_shr_i64:
287 return (uint64_t)x >> (y & 63);
289 case INDEX_op_sar_i32:
290 return (int32_t)x >> (y & 31);
292 case INDEX_op_sar_i64:
293 return (int64_t)x >> (y & 63);
295 case INDEX_op_rotr_i32:
296 return ror32(x, y & 31);
298 case INDEX_op_rotr_i64:
299 return ror64(x, y & 63);
301 case INDEX_op_rotl_i32:
302 return rol32(x, y & 31);
304 case INDEX_op_rotl_i64:
305 return rol64(x, y & 63);
307 CASE_OP_32_64(not):
308 return ~x;
310 CASE_OP_32_64(neg):
311 return -x;
313 CASE_OP_32_64(andc):
314 return x & ~y;
316 CASE_OP_32_64(orc):
317 return x | ~y;
319 CASE_OP_32_64(eqv):
320 return ~(x ^ y);
322 CASE_OP_32_64(nand):
323 return ~(x & y);
325 CASE_OP_32_64(nor):
326 return ~(x | y);
328 CASE_OP_32_64(ext8s):
329 return (int8_t)x;
331 CASE_OP_32_64(ext16s):
332 return (int16_t)x;
334 CASE_OP_32_64(ext8u):
335 return (uint8_t)x;
337 CASE_OP_32_64(ext16u):
338 return (uint16_t)x;
340 case INDEX_op_ext32s_i64:
341 return (int32_t)x;
343 case INDEX_op_ext32u_i64:
344 return (uint32_t)x;
346 case INDEX_op_muluh_i32:
347 return ((uint64_t)(uint32_t)x * (uint32_t)y) >> 32;
348 case INDEX_op_mulsh_i32:
349 return ((int64_t)(int32_t)x * (int32_t)y) >> 32;
351 case INDEX_op_muluh_i64:
352 mulu64(&l64, &h64, x, y);
353 return h64;
354 case INDEX_op_mulsh_i64:
355 muls64(&l64, &h64, x, y);
356 return h64;
358 case INDEX_op_div_i32:
359 /* Avoid crashing on divide by zero, otherwise undefined. */
360 return (int32_t)x / ((int32_t)y ? : 1);
361 case INDEX_op_divu_i32:
362 return (uint32_t)x / ((uint32_t)y ? : 1);
363 case INDEX_op_div_i64:
364 return (int64_t)x / ((int64_t)y ? : 1);
365 case INDEX_op_divu_i64:
366 return (uint64_t)x / ((uint64_t)y ? : 1);
368 case INDEX_op_rem_i32:
369 return (int32_t)x % ((int32_t)y ? : 1);
370 case INDEX_op_remu_i32:
371 return (uint32_t)x % ((uint32_t)y ? : 1);
372 case INDEX_op_rem_i64:
373 return (int64_t)x % ((int64_t)y ? : 1);
374 case INDEX_op_remu_i64:
375 return (uint64_t)x % ((uint64_t)y ? : 1);
377 default:
378 fprintf(stderr,
379 "Unrecognized operation %d in do_constant_folding.\n", op);
380 tcg_abort();
384 static TCGArg do_constant_folding(TCGOpcode op, TCGArg x, TCGArg y)
386 TCGArg res = do_constant_folding_2(op, x, y);
387 if (op_bits(op) == 32) {
388 res &= 0xffffffff;
390 return res;
393 static bool do_constant_folding_cond_32(uint32_t x, uint32_t y, TCGCond c)
395 switch (c) {
396 case TCG_COND_EQ:
397 return x == y;
398 case TCG_COND_NE:
399 return x != y;
400 case TCG_COND_LT:
401 return (int32_t)x < (int32_t)y;
402 case TCG_COND_GE:
403 return (int32_t)x >= (int32_t)y;
404 case TCG_COND_LE:
405 return (int32_t)x <= (int32_t)y;
406 case TCG_COND_GT:
407 return (int32_t)x > (int32_t)y;
408 case TCG_COND_LTU:
409 return x < y;
410 case TCG_COND_GEU:
411 return x >= y;
412 case TCG_COND_LEU:
413 return x <= y;
414 case TCG_COND_GTU:
415 return x > y;
416 default:
417 tcg_abort();
421 static bool do_constant_folding_cond_64(uint64_t x, uint64_t y, TCGCond c)
423 switch (c) {
424 case TCG_COND_EQ:
425 return x == y;
426 case TCG_COND_NE:
427 return x != y;
428 case TCG_COND_LT:
429 return (int64_t)x < (int64_t)y;
430 case TCG_COND_GE:
431 return (int64_t)x >= (int64_t)y;
432 case TCG_COND_LE:
433 return (int64_t)x <= (int64_t)y;
434 case TCG_COND_GT:
435 return (int64_t)x > (int64_t)y;
436 case TCG_COND_LTU:
437 return x < y;
438 case TCG_COND_GEU:
439 return x >= y;
440 case TCG_COND_LEU:
441 return x <= y;
442 case TCG_COND_GTU:
443 return x > y;
444 default:
445 tcg_abort();
449 static bool do_constant_folding_cond_eq(TCGCond c)
451 switch (c) {
452 case TCG_COND_GT:
453 case TCG_COND_LTU:
454 case TCG_COND_LT:
455 case TCG_COND_GTU:
456 case TCG_COND_NE:
457 return 0;
458 case TCG_COND_GE:
459 case TCG_COND_GEU:
460 case TCG_COND_LE:
461 case TCG_COND_LEU:
462 case TCG_COND_EQ:
463 return 1;
464 default:
465 tcg_abort();
469 /* Return 2 if the condition can't be simplified, and the result
470 of the condition (0 or 1) if it can */
471 static TCGArg do_constant_folding_cond(TCGOpcode op, TCGArg x,
472 TCGArg y, TCGCond c)
474 if (temps[x].state == TCG_TEMP_CONST && temps[y].state == TCG_TEMP_CONST) {
475 switch (op_bits(op)) {
476 case 32:
477 return do_constant_folding_cond_32(temps[x].val, temps[y].val, c);
478 case 64:
479 return do_constant_folding_cond_64(temps[x].val, temps[y].val, c);
480 default:
481 tcg_abort();
483 } else if (temps_are_copies(x, y)) {
484 return do_constant_folding_cond_eq(c);
485 } else if (temps[y].state == TCG_TEMP_CONST && temps[y].val == 0) {
486 switch (c) {
487 case TCG_COND_LTU:
488 return 0;
489 case TCG_COND_GEU:
490 return 1;
491 default:
492 return 2;
494 } else {
495 return 2;
499 /* Return 2 if the condition can't be simplified, and the result
500 of the condition (0 or 1) if it can */
501 static TCGArg do_constant_folding_cond2(TCGArg *p1, TCGArg *p2, TCGCond c)
503 TCGArg al = p1[0], ah = p1[1];
504 TCGArg bl = p2[0], bh = p2[1];
506 if (temps[bl].state == TCG_TEMP_CONST
507 && temps[bh].state == TCG_TEMP_CONST) {
508 uint64_t b = ((uint64_t)temps[bh].val << 32) | (uint32_t)temps[bl].val;
510 if (temps[al].state == TCG_TEMP_CONST
511 && temps[ah].state == TCG_TEMP_CONST) {
512 uint64_t a;
513 a = ((uint64_t)temps[ah].val << 32) | (uint32_t)temps[al].val;
514 return do_constant_folding_cond_64(a, b, c);
516 if (b == 0) {
517 switch (c) {
518 case TCG_COND_LTU:
519 return 0;
520 case TCG_COND_GEU:
521 return 1;
522 default:
523 break;
527 if (temps_are_copies(al, bl) && temps_are_copies(ah, bh)) {
528 return do_constant_folding_cond_eq(c);
530 return 2;
533 static bool swap_commutative(TCGArg dest, TCGArg *p1, TCGArg *p2)
535 TCGArg a1 = *p1, a2 = *p2;
536 int sum = 0;
537 sum += temps[a1].state == TCG_TEMP_CONST;
538 sum -= temps[a2].state == TCG_TEMP_CONST;
540 /* Prefer the constant in second argument, and then the form
541 op a, a, b, which is better handled on non-RISC hosts. */
542 if (sum > 0 || (sum == 0 && dest == a2)) {
543 *p1 = a2;
544 *p2 = a1;
545 return true;
547 return false;
550 static bool swap_commutative2(TCGArg *p1, TCGArg *p2)
552 int sum = 0;
553 sum += temps[p1[0]].state == TCG_TEMP_CONST;
554 sum += temps[p1[1]].state == TCG_TEMP_CONST;
555 sum -= temps[p2[0]].state == TCG_TEMP_CONST;
556 sum -= temps[p2[1]].state == TCG_TEMP_CONST;
557 if (sum > 0) {
558 TCGArg t;
559 t = p1[0], p1[0] = p2[0], p2[0] = t;
560 t = p1[1], p1[1] = p2[1], p2[1] = t;
561 return true;
563 return false;
566 /* Propagate constants and copies, fold constant expressions. */
567 static void tcg_constant_folding(TCGContext *s)
569 int oi, oi_next, nb_temps, nb_globals;
571 /* Array VALS has an element for each temp.
572 If this temp holds a constant then its value is kept in VALS' element.
573 If this temp is a copy of other ones then the other copies are
574 available through the doubly linked circular list. */
576 nb_temps = s->nb_temps;
577 nb_globals = s->nb_globals;
578 reset_all_temps(nb_temps);
580 for (oi = s->gen_first_op_idx; oi >= 0; oi = oi_next) {
581 tcg_target_ulong mask, partmask, affected;
582 int nb_oargs, nb_iargs, i;
583 TCGArg tmp;
585 TCGOp * const op = &s->gen_op_buf[oi];
586 TCGArg * const args = &s->gen_opparam_buf[op->args];
587 TCGOpcode opc = op->opc;
588 const TCGOpDef *def = &tcg_op_defs[opc];
590 oi_next = op->next;
591 if (opc == INDEX_op_call) {
592 nb_oargs = op->callo;
593 nb_iargs = op->calli;
594 } else {
595 nb_oargs = def->nb_oargs;
596 nb_iargs = def->nb_iargs;
599 /* Do copy propagation */
600 for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) {
601 if (temps[args[i]].state == TCG_TEMP_COPY) {
602 args[i] = find_better_copy(s, args[i]);
606 /* For commutative operations make constant second argument */
607 switch (opc) {
608 CASE_OP_32_64(add):
609 CASE_OP_32_64(mul):
610 CASE_OP_32_64(and):
611 CASE_OP_32_64(or):
612 CASE_OP_32_64(xor):
613 CASE_OP_32_64(eqv):
614 CASE_OP_32_64(nand):
615 CASE_OP_32_64(nor):
616 CASE_OP_32_64(muluh):
617 CASE_OP_32_64(mulsh):
618 swap_commutative(args[0], &args[1], &args[2]);
619 break;
620 CASE_OP_32_64(brcond):
621 if (swap_commutative(-1, &args[0], &args[1])) {
622 args[2] = tcg_swap_cond(args[2]);
624 break;
625 CASE_OP_32_64(setcond):
626 if (swap_commutative(args[0], &args[1], &args[2])) {
627 args[3] = tcg_swap_cond(args[3]);
629 break;
630 CASE_OP_32_64(movcond):
631 if (swap_commutative(-1, &args[1], &args[2])) {
632 args[5] = tcg_swap_cond(args[5]);
634 /* For movcond, we canonicalize the "false" input reg to match
635 the destination reg so that the tcg backend can implement
636 a "move if true" operation. */
637 if (swap_commutative(args[0], &args[4], &args[3])) {
638 args[5] = tcg_invert_cond(args[5]);
640 break;
641 CASE_OP_32_64(add2):
642 swap_commutative(args[0], &args[2], &args[4]);
643 swap_commutative(args[1], &args[3], &args[5]);
644 break;
645 CASE_OP_32_64(mulu2):
646 CASE_OP_32_64(muls2):
647 swap_commutative(args[0], &args[2], &args[3]);
648 break;
649 case INDEX_op_brcond2_i32:
650 if (swap_commutative2(&args[0], &args[2])) {
651 args[4] = tcg_swap_cond(args[4]);
653 break;
654 case INDEX_op_setcond2_i32:
655 if (swap_commutative2(&args[1], &args[3])) {
656 args[5] = tcg_swap_cond(args[5]);
658 break;
659 default:
660 break;
663 /* Simplify expressions for "shift/rot r, 0, a => movi r, 0",
664 and "sub r, 0, a => neg r, a" case. */
665 switch (opc) {
666 CASE_OP_32_64(shl):
667 CASE_OP_32_64(shr):
668 CASE_OP_32_64(sar):
669 CASE_OP_32_64(rotl):
670 CASE_OP_32_64(rotr):
671 if (temps[args[1]].state == TCG_TEMP_CONST
672 && temps[args[1]].val == 0) {
673 tcg_opt_gen_movi(s, op, args, opc, args[0], 0);
674 continue;
676 break;
677 CASE_OP_32_64(sub):
679 TCGOpcode neg_op;
680 bool have_neg;
682 if (temps[args[2]].state == TCG_TEMP_CONST) {
683 /* Proceed with possible constant folding. */
684 break;
686 if (opc == INDEX_op_sub_i32) {
687 neg_op = INDEX_op_neg_i32;
688 have_neg = TCG_TARGET_HAS_neg_i32;
689 } else {
690 neg_op = INDEX_op_neg_i64;
691 have_neg = TCG_TARGET_HAS_neg_i64;
693 if (!have_neg) {
694 break;
696 if (temps[args[1]].state == TCG_TEMP_CONST
697 && temps[args[1]].val == 0) {
698 op->opc = neg_op;
699 reset_temp(args[0]);
700 args[1] = args[2];
701 continue;
704 break;
705 CASE_OP_32_64(xor):
706 CASE_OP_32_64(nand):
707 if (temps[args[1]].state != TCG_TEMP_CONST
708 && temps[args[2]].state == TCG_TEMP_CONST
709 && temps[args[2]].val == -1) {
710 i = 1;
711 goto try_not;
713 break;
714 CASE_OP_32_64(nor):
715 if (temps[args[1]].state != TCG_TEMP_CONST
716 && temps[args[2]].state == TCG_TEMP_CONST
717 && temps[args[2]].val == 0) {
718 i = 1;
719 goto try_not;
721 break;
722 CASE_OP_32_64(andc):
723 if (temps[args[2]].state != TCG_TEMP_CONST
724 && temps[args[1]].state == TCG_TEMP_CONST
725 && temps[args[1]].val == -1) {
726 i = 2;
727 goto try_not;
729 break;
730 CASE_OP_32_64(orc):
731 CASE_OP_32_64(eqv):
732 if (temps[args[2]].state != TCG_TEMP_CONST
733 && temps[args[1]].state == TCG_TEMP_CONST
734 && temps[args[1]].val == 0) {
735 i = 2;
736 goto try_not;
738 break;
739 try_not:
741 TCGOpcode not_op;
742 bool have_not;
744 if (def->flags & TCG_OPF_64BIT) {
745 not_op = INDEX_op_not_i64;
746 have_not = TCG_TARGET_HAS_not_i64;
747 } else {
748 not_op = INDEX_op_not_i32;
749 have_not = TCG_TARGET_HAS_not_i32;
751 if (!have_not) {
752 break;
754 op->opc = not_op;
755 reset_temp(args[0]);
756 args[1] = args[i];
757 continue;
759 default:
760 break;
763 /* Simplify expression for "op r, a, const => mov r, a" cases */
764 switch (opc) {
765 CASE_OP_32_64(add):
766 CASE_OP_32_64(sub):
767 CASE_OP_32_64(shl):
768 CASE_OP_32_64(shr):
769 CASE_OP_32_64(sar):
770 CASE_OP_32_64(rotl):
771 CASE_OP_32_64(rotr):
772 CASE_OP_32_64(or):
773 CASE_OP_32_64(xor):
774 CASE_OP_32_64(andc):
775 if (temps[args[1]].state != TCG_TEMP_CONST
776 && temps[args[2]].state == TCG_TEMP_CONST
777 && temps[args[2]].val == 0) {
778 goto do_mov3;
780 break;
781 CASE_OP_32_64(and):
782 CASE_OP_32_64(orc):
783 CASE_OP_32_64(eqv):
784 if (temps[args[1]].state != TCG_TEMP_CONST
785 && temps[args[2]].state == TCG_TEMP_CONST
786 && temps[args[2]].val == -1) {
787 goto do_mov3;
789 break;
790 do_mov3:
791 if (temps_are_copies(args[0], args[1])) {
792 tcg_op_remove(s, op);
793 } else {
794 tcg_opt_gen_mov(s, op, args, opc, args[0], args[1]);
796 continue;
797 default:
798 break;
801 /* Simplify using known-zero bits. Currently only ops with a single
802 output argument is supported. */
803 mask = -1;
804 affected = -1;
805 switch (opc) {
806 CASE_OP_32_64(ext8s):
807 if ((temps[args[1]].mask & 0x80) != 0) {
808 break;
810 CASE_OP_32_64(ext8u):
811 mask = 0xff;
812 goto and_const;
813 CASE_OP_32_64(ext16s):
814 if ((temps[args[1]].mask & 0x8000) != 0) {
815 break;
817 CASE_OP_32_64(ext16u):
818 mask = 0xffff;
819 goto and_const;
820 case INDEX_op_ext32s_i64:
821 if ((temps[args[1]].mask & 0x80000000) != 0) {
822 break;
824 case INDEX_op_ext32u_i64:
825 mask = 0xffffffffU;
826 goto and_const;
828 CASE_OP_32_64(and):
829 mask = temps[args[2]].mask;
830 if (temps[args[2]].state == TCG_TEMP_CONST) {
831 and_const:
832 affected = temps[args[1]].mask & ~mask;
834 mask = temps[args[1]].mask & mask;
835 break;
837 CASE_OP_32_64(andc):
838 /* Known-zeros does not imply known-ones. Therefore unless
839 args[2] is constant, we can't infer anything from it. */
840 if (temps[args[2]].state == TCG_TEMP_CONST) {
841 mask = ~temps[args[2]].mask;
842 goto and_const;
844 /* But we certainly know nothing outside args[1] may be set. */
845 mask = temps[args[1]].mask;
846 break;
848 case INDEX_op_sar_i32:
849 if (temps[args[2]].state == TCG_TEMP_CONST) {
850 tmp = temps[args[2]].val & 31;
851 mask = (int32_t)temps[args[1]].mask >> tmp;
853 break;
854 case INDEX_op_sar_i64:
855 if (temps[args[2]].state == TCG_TEMP_CONST) {
856 tmp = temps[args[2]].val & 63;
857 mask = (int64_t)temps[args[1]].mask >> tmp;
859 break;
861 case INDEX_op_shr_i32:
862 if (temps[args[2]].state == TCG_TEMP_CONST) {
863 tmp = temps[args[2]].val & 31;
864 mask = (uint32_t)temps[args[1]].mask >> tmp;
866 break;
867 case INDEX_op_shr_i64:
868 if (temps[args[2]].state == TCG_TEMP_CONST) {
869 tmp = temps[args[2]].val & 63;
870 mask = (uint64_t)temps[args[1]].mask >> tmp;
872 break;
874 case INDEX_op_trunc_shr_i32:
875 mask = (uint64_t)temps[args[1]].mask >> args[2];
876 break;
878 CASE_OP_32_64(shl):
879 if (temps[args[2]].state == TCG_TEMP_CONST) {
880 tmp = temps[args[2]].val & (TCG_TARGET_REG_BITS - 1);
881 mask = temps[args[1]].mask << tmp;
883 break;
885 CASE_OP_32_64(neg):
886 /* Set to 1 all bits to the left of the rightmost. */
887 mask = -(temps[args[1]].mask & -temps[args[1]].mask);
888 break;
890 CASE_OP_32_64(deposit):
891 mask = deposit64(temps[args[1]].mask, args[3], args[4],
892 temps[args[2]].mask);
893 break;
895 CASE_OP_32_64(or):
896 CASE_OP_32_64(xor):
897 mask = temps[args[1]].mask | temps[args[2]].mask;
898 break;
900 CASE_OP_32_64(setcond):
901 case INDEX_op_setcond2_i32:
902 mask = 1;
903 break;
905 CASE_OP_32_64(movcond):
906 mask = temps[args[3]].mask | temps[args[4]].mask;
907 break;
909 CASE_OP_32_64(ld8u):
910 mask = 0xff;
911 break;
912 CASE_OP_32_64(ld16u):
913 mask = 0xffff;
914 break;
915 case INDEX_op_ld32u_i64:
916 mask = 0xffffffffu;
917 break;
919 CASE_OP_32_64(qemu_ld):
921 TCGMemOp mop = args[nb_oargs + nb_iargs];
922 if (!(mop & MO_SIGN)) {
923 mask = (2ULL << ((8 << (mop & MO_SIZE)) - 1)) - 1;
926 break;
928 default:
929 break;
932 /* 32-bit ops generate 32-bit results. For the result is zero test
933 below, we can ignore high bits, but for further optimizations we
934 need to record that the high bits contain garbage. */
935 partmask = mask;
936 if (!(def->flags & TCG_OPF_64BIT)) {
937 mask |= ~(tcg_target_ulong)0xffffffffu;
938 partmask &= 0xffffffffu;
939 affected &= 0xffffffffu;
942 if (partmask == 0) {
943 assert(nb_oargs == 1);
944 tcg_opt_gen_movi(s, op, args, opc, args[0], 0);
945 continue;
947 if (affected == 0) {
948 assert(nb_oargs == 1);
949 if (temps_are_copies(args[0], args[1])) {
950 tcg_op_remove(s, op);
951 } else if (temps[args[1]].state != TCG_TEMP_CONST) {
952 tcg_opt_gen_mov(s, op, args, opc, args[0], args[1]);
953 } else {
954 tcg_opt_gen_movi(s, op, args, opc,
955 args[0], temps[args[1]].val);
957 continue;
960 /* Simplify expression for "op r, a, 0 => movi r, 0" cases */
961 switch (opc) {
962 CASE_OP_32_64(and):
963 CASE_OP_32_64(mul):
964 CASE_OP_32_64(muluh):
965 CASE_OP_32_64(mulsh):
966 if ((temps[args[2]].state == TCG_TEMP_CONST
967 && temps[args[2]].val == 0)) {
968 tcg_opt_gen_movi(s, op, args, opc, args[0], 0);
969 continue;
971 break;
972 default:
973 break;
976 /* Simplify expression for "op r, a, a => mov r, a" cases */
977 switch (opc) {
978 CASE_OP_32_64(or):
979 CASE_OP_32_64(and):
980 if (temps_are_copies(args[1], args[2])) {
981 if (temps_are_copies(args[0], args[1])) {
982 tcg_op_remove(s, op);
983 } else if (temps[args[1]].state != TCG_TEMP_CONST) {
984 tcg_opt_gen_mov(s, op, args, opc, args[0], args[1]);
985 } else {
986 tcg_opt_gen_movi(s, op, args, opc,
987 args[0], temps[args[1]].val);
989 continue;
991 break;
992 default:
993 break;
996 /* Simplify expression for "op r, a, a => movi r, 0" cases */
997 switch (opc) {
998 CASE_OP_32_64(andc):
999 CASE_OP_32_64(sub):
1000 CASE_OP_32_64(xor):
1001 if (temps_are_copies(args[1], args[2])) {
1002 tcg_opt_gen_movi(s, op, args, opc, args[0], 0);
1003 continue;
1005 break;
1006 default:
1007 break;
1010 /* Propagate constants through copy operations and do constant
1011 folding. Constants will be substituted to arguments by register
1012 allocator where needed and possible. Also detect copies. */
1013 switch (opc) {
1014 CASE_OP_32_64(mov):
1015 if (temps_are_copies(args[0], args[1])) {
1016 tcg_op_remove(s, op);
1017 break;
1019 if (temps[args[1]].state != TCG_TEMP_CONST) {
1020 tcg_opt_gen_mov(s, op, args, opc, args[0], args[1]);
1021 break;
1023 /* Source argument is constant. Rewrite the operation and
1024 let movi case handle it. */
1025 args[1] = temps[args[1]].val;
1026 /* fallthrough */
1027 CASE_OP_32_64(movi):
1028 tcg_opt_gen_movi(s, op, args, opc, args[0], args[1]);
1029 break;
1031 CASE_OP_32_64(not):
1032 CASE_OP_32_64(neg):
1033 CASE_OP_32_64(ext8s):
1034 CASE_OP_32_64(ext8u):
1035 CASE_OP_32_64(ext16s):
1036 CASE_OP_32_64(ext16u):
1037 case INDEX_op_ext32s_i64:
1038 case INDEX_op_ext32u_i64:
1039 if (temps[args[1]].state == TCG_TEMP_CONST) {
1040 tmp = do_constant_folding(opc, temps[args[1]].val, 0);
1041 tcg_opt_gen_movi(s, op, args, opc, args[0], tmp);
1042 break;
1044 goto do_default;
1046 case INDEX_op_trunc_shr_i32:
1047 if (temps[args[1]].state == TCG_TEMP_CONST) {
1048 tmp = do_constant_folding(opc, temps[args[1]].val, args[2]);
1049 tcg_opt_gen_movi(s, op, args, opc, args[0], tmp);
1050 break;
1052 goto do_default;
1054 CASE_OP_32_64(add):
1055 CASE_OP_32_64(sub):
1056 CASE_OP_32_64(mul):
1057 CASE_OP_32_64(or):
1058 CASE_OP_32_64(and):
1059 CASE_OP_32_64(xor):
1060 CASE_OP_32_64(shl):
1061 CASE_OP_32_64(shr):
1062 CASE_OP_32_64(sar):
1063 CASE_OP_32_64(rotl):
1064 CASE_OP_32_64(rotr):
1065 CASE_OP_32_64(andc):
1066 CASE_OP_32_64(orc):
1067 CASE_OP_32_64(eqv):
1068 CASE_OP_32_64(nand):
1069 CASE_OP_32_64(nor):
1070 CASE_OP_32_64(muluh):
1071 CASE_OP_32_64(mulsh):
1072 CASE_OP_32_64(div):
1073 CASE_OP_32_64(divu):
1074 CASE_OP_32_64(rem):
1075 CASE_OP_32_64(remu):
1076 if (temps[args[1]].state == TCG_TEMP_CONST
1077 && temps[args[2]].state == TCG_TEMP_CONST) {
1078 tmp = do_constant_folding(opc, temps[args[1]].val,
1079 temps[args[2]].val);
1080 tcg_opt_gen_movi(s, op, args, opc, args[0], tmp);
1081 break;
1083 goto do_default;
1085 CASE_OP_32_64(deposit):
1086 if (temps[args[1]].state == TCG_TEMP_CONST
1087 && temps[args[2]].state == TCG_TEMP_CONST) {
1088 tmp = deposit64(temps[args[1]].val, args[3], args[4],
1089 temps[args[2]].val);
1090 tcg_opt_gen_movi(s, op, args, opc, args[0], tmp);
1091 break;
1093 goto do_default;
1095 CASE_OP_32_64(setcond):
1096 tmp = do_constant_folding_cond(opc, args[1], args[2], args[3]);
1097 if (tmp != 2) {
1098 tcg_opt_gen_movi(s, op, args, opc, args[0], tmp);
1099 break;
1101 goto do_default;
1103 CASE_OP_32_64(brcond):
1104 tmp = do_constant_folding_cond(opc, args[0], args[1], args[2]);
1105 if (tmp != 2) {
1106 if (tmp) {
1107 reset_all_temps(nb_temps);
1108 op->opc = INDEX_op_br;
1109 args[0] = args[3];
1110 } else {
1111 tcg_op_remove(s, op);
1113 break;
1115 goto do_default;
1117 CASE_OP_32_64(movcond):
1118 tmp = do_constant_folding_cond(opc, args[1], args[2], args[5]);
1119 if (tmp != 2) {
1120 if (temps_are_copies(args[0], args[4-tmp])) {
1121 tcg_op_remove(s, op);
1122 } else if (temps[args[4-tmp]].state == TCG_TEMP_CONST) {
1123 tcg_opt_gen_movi(s, op, args, opc,
1124 args[0], temps[args[4-tmp]].val);
1125 } else {
1126 tcg_opt_gen_mov(s, op, args, opc, args[0], args[4-tmp]);
1128 break;
1130 goto do_default;
1132 case INDEX_op_add2_i32:
1133 case INDEX_op_sub2_i32:
1134 if (temps[args[2]].state == TCG_TEMP_CONST
1135 && temps[args[3]].state == TCG_TEMP_CONST
1136 && temps[args[4]].state == TCG_TEMP_CONST
1137 && temps[args[5]].state == TCG_TEMP_CONST) {
1138 uint32_t al = temps[args[2]].val;
1139 uint32_t ah = temps[args[3]].val;
1140 uint32_t bl = temps[args[4]].val;
1141 uint32_t bh = temps[args[5]].val;
1142 uint64_t a = ((uint64_t)ah << 32) | al;
1143 uint64_t b = ((uint64_t)bh << 32) | bl;
1144 TCGArg rl, rh;
1145 TCGOp *op2 = insert_op_before(s, op, INDEX_op_movi_i32, 2);
1146 TCGArg *args2 = &s->gen_opparam_buf[op2->args];
1148 if (opc == INDEX_op_add2_i32) {
1149 a += b;
1150 } else {
1151 a -= b;
1154 rl = args[0];
1155 rh = args[1];
1156 tcg_opt_gen_movi(s, op, args, opc, rl, (uint32_t)a);
1157 tcg_opt_gen_movi(s, op2, args2, opc, rh, (uint32_t)(a >> 32));
1159 /* We've done all we need to do with the movi. Skip it. */
1160 oi_next = op2->next;
1161 break;
1163 goto do_default;
1165 case INDEX_op_mulu2_i32:
1166 if (temps[args[2]].state == TCG_TEMP_CONST
1167 && temps[args[3]].state == TCG_TEMP_CONST) {
1168 uint32_t a = temps[args[2]].val;
1169 uint32_t b = temps[args[3]].val;
1170 uint64_t r = (uint64_t)a * b;
1171 TCGArg rl, rh;
1172 TCGOp *op2 = insert_op_before(s, op, INDEX_op_movi_i32, 2);
1173 TCGArg *args2 = &s->gen_opparam_buf[op2->args];
1175 rl = args[0];
1176 rh = args[1];
1177 tcg_opt_gen_movi(s, op, args, opc, rl, (uint32_t)r);
1178 tcg_opt_gen_movi(s, op2, args2, opc, rh, (uint32_t)(r >> 32));
1180 /* We've done all we need to do with the movi. Skip it. */
1181 oi_next = op2->next;
1182 break;
1184 goto do_default;
1186 case INDEX_op_brcond2_i32:
1187 tmp = do_constant_folding_cond2(&args[0], &args[2], args[4]);
1188 if (tmp != 2) {
1189 if (tmp) {
1190 do_brcond_true:
1191 reset_all_temps(nb_temps);
1192 op->opc = INDEX_op_br;
1193 args[0] = args[5];
1194 } else {
1195 do_brcond_false:
1196 tcg_op_remove(s, op);
1198 } else if ((args[4] == TCG_COND_LT || args[4] == TCG_COND_GE)
1199 && temps[args[2]].state == TCG_TEMP_CONST
1200 && temps[args[3]].state == TCG_TEMP_CONST
1201 && temps[args[2]].val == 0
1202 && temps[args[3]].val == 0) {
1203 /* Simplify LT/GE comparisons vs zero to a single compare
1204 vs the high word of the input. */
1205 do_brcond_high:
1206 reset_all_temps(nb_temps);
1207 op->opc = INDEX_op_brcond_i32;
1208 args[0] = args[1];
1209 args[1] = args[3];
1210 args[2] = args[4];
1211 args[3] = args[5];
1212 } else if (args[4] == TCG_COND_EQ) {
1213 /* Simplify EQ comparisons where one of the pairs
1214 can be simplified. */
1215 tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
1216 args[0], args[2], TCG_COND_EQ);
1217 if (tmp == 0) {
1218 goto do_brcond_false;
1219 } else if (tmp == 1) {
1220 goto do_brcond_high;
1222 tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
1223 args[1], args[3], TCG_COND_EQ);
1224 if (tmp == 0) {
1225 goto do_brcond_false;
1226 } else if (tmp != 1) {
1227 goto do_default;
1229 do_brcond_low:
1230 reset_all_temps(nb_temps);
1231 op->opc = INDEX_op_brcond_i32;
1232 args[1] = args[2];
1233 args[2] = args[4];
1234 args[3] = args[5];
1235 } else if (args[4] == TCG_COND_NE) {
1236 /* Simplify NE comparisons where one of the pairs
1237 can be simplified. */
1238 tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
1239 args[0], args[2], TCG_COND_NE);
1240 if (tmp == 0) {
1241 goto do_brcond_high;
1242 } else if (tmp == 1) {
1243 goto do_brcond_true;
1245 tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
1246 args[1], args[3], TCG_COND_NE);
1247 if (tmp == 0) {
1248 goto do_brcond_low;
1249 } else if (tmp == 1) {
1250 goto do_brcond_true;
1252 goto do_default;
1253 } else {
1254 goto do_default;
1256 break;
1258 case INDEX_op_setcond2_i32:
1259 tmp = do_constant_folding_cond2(&args[1], &args[3], args[5]);
1260 if (tmp != 2) {
1261 do_setcond_const:
1262 tcg_opt_gen_movi(s, op, args, opc, args[0], tmp);
1263 } else if ((args[5] == TCG_COND_LT || args[5] == TCG_COND_GE)
1264 && temps[args[3]].state == TCG_TEMP_CONST
1265 && temps[args[4]].state == TCG_TEMP_CONST
1266 && temps[args[3]].val == 0
1267 && temps[args[4]].val == 0) {
1268 /* Simplify LT/GE comparisons vs zero to a single compare
1269 vs the high word of the input. */
1270 do_setcond_high:
1271 reset_temp(args[0]);
1272 temps[args[0]].mask = 1;
1273 op->opc = INDEX_op_setcond_i32;
1274 args[1] = args[2];
1275 args[2] = args[4];
1276 args[3] = args[5];
1277 } else if (args[5] == TCG_COND_EQ) {
1278 /* Simplify EQ comparisons where one of the pairs
1279 can be simplified. */
1280 tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
1281 args[1], args[3], TCG_COND_EQ);
1282 if (tmp == 0) {
1283 goto do_setcond_const;
1284 } else if (tmp == 1) {
1285 goto do_setcond_high;
1287 tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
1288 args[2], args[4], TCG_COND_EQ);
1289 if (tmp == 0) {
1290 goto do_setcond_high;
1291 } else if (tmp != 1) {
1292 goto do_default;
1294 do_setcond_low:
1295 reset_temp(args[0]);
1296 temps[args[0]].mask = 1;
1297 op->opc = INDEX_op_setcond_i32;
1298 args[2] = args[3];
1299 args[3] = args[5];
1300 } else if (args[5] == TCG_COND_NE) {
1301 /* Simplify NE comparisons where one of the pairs
1302 can be simplified. */
1303 tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
1304 args[1], args[3], TCG_COND_NE);
1305 if (tmp == 0) {
1306 goto do_setcond_high;
1307 } else if (tmp == 1) {
1308 goto do_setcond_const;
1310 tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
1311 args[2], args[4], TCG_COND_NE);
1312 if (tmp == 0) {
1313 goto do_setcond_low;
1314 } else if (tmp == 1) {
1315 goto do_setcond_const;
1317 goto do_default;
1318 } else {
1319 goto do_default;
1321 break;
1323 case INDEX_op_call:
1324 if (!(args[nb_oargs + nb_iargs + 1]
1325 & (TCG_CALL_NO_READ_GLOBALS | TCG_CALL_NO_WRITE_GLOBALS))) {
1326 for (i = 0; i < nb_globals; i++) {
1327 reset_temp(i);
1330 goto do_reset_output;
1332 default:
1333 do_default:
1334 /* Default case: we know nothing about operation (or were unable
1335 to compute the operation result) so no propagation is done.
1336 We trash everything if the operation is the end of a basic
1337 block, otherwise we only trash the output args. "mask" is
1338 the non-zero bits mask for the first output arg. */
1339 if (def->flags & TCG_OPF_BB_END) {
1340 reset_all_temps(nb_temps);
1341 } else {
1342 do_reset_output:
1343 for (i = 0; i < nb_oargs; i++) {
1344 reset_temp(args[i]);
1345 /* Save the corresponding known-zero bits mask for the
1346 first output argument (only one supported so far). */
1347 if (i == 0) {
1348 temps[args[i]].mask = mask;
1352 break;
1357 void tcg_optimize(TCGContext *s)
1359 tcg_constant_folding(s);