tcg/optimize.c

   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 "config.h"
  27
  28 #include <stdlib.h>
  29 #include <stdio.h>
  30
  31 #include "qemu-common.h"
  32 #include "tcg-op.h"
  33
  34 #define CASE_OP_32_64(x)                        \
  35         glue(glue(case INDEX_op_, x), _i32):    \
  36         glue(glue(case INDEX_op_, x), _i64)
  37
  38 typedef enum {
  39     TCG_TEMP_UNDEF = 0,
  40     TCG_TEMP_CONST,
  41     TCG_TEMP_COPY,
  42     TCG_TEMP_HAS_COPY,
  43     TCG_TEMP_ANY
  44 } tcg_temp_state;
  45
  46 struct tcg_temp_info {
  47     tcg_temp_state state;
  48     uint16_t prev_copy;
  49     uint16_t next_copy;
  50     tcg_target_ulong val;
  51 };
  52
  53 static struct tcg_temp_info temps[TCG_MAX_TEMPS];
  54
  55 /* Reset TEMP's state to TCG_TEMP_ANY.  If TEMP was a representative of some
  56    class of equivalent temp's, a new representative should be chosen in this
  57    class. */
  58 static void reset_temp(TCGArg temp, int nb_temps, int nb_globals)
  59 {
  60     int i;
  61     TCGArg new_base = (TCGArg)-1;
  62     if (temps[temp].state == TCG_TEMP_HAS_COPY) {
  63         for (i = temps[temp].next_copy; i != temp; i = temps[i].next_copy) {
  64             if (i >= nb_globals) {
  65                 temps[i].state = TCG_TEMP_HAS_COPY;
  66                 new_base = i;
  67                 break;
  68             }
  69         }
  70         for (i = temps[temp].next_copy; i != temp; i = temps[i].next_copy) {
  71             if (new_base == (TCGArg)-1) {
  72                 temps[i].state = TCG_TEMP_ANY;
  73             } else {
  74                 temps[i].val = new_base;
  75             }
  76         }
  77         temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy;
  78         temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy;
  79     } else if (temps[temp].state == TCG_TEMP_COPY) {
  80         temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy;
  81         temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy;
  82         new_base = temps[temp].val;
  83     }
  84     temps[temp].state = TCG_TEMP_ANY;
  85     if (new_base != (TCGArg)-1 && temps[new_base].next_copy == new_base) {
  86         temps[new_base].state = TCG_TEMP_ANY;
  87     }
  88 }
  89
  90 static int op_bits(TCGOpcode op)
  91 {
  92     const TCGOpDef *def = &tcg_op_defs[op];
  93     return def->flags & TCG_OPF_64BIT ? 64 : 32;
  94 }
  95
  96 static TCGOpcode op_to_movi(TCGOpcode op)
  97 {
  98     switch (op_bits(op)) {
  99     case 32:
 100         return INDEX_op_movi_i32;
 101     case 64:
 102         return INDEX_op_movi_i64;
 103     default:
 104         fprintf(stderr, "op_to_movi: unexpected return value of "
 105                 "function op_bits.\n");
 106         tcg_abort();
 107     }
 108 }
 109
 110 static void tcg_opt_gen_mov(TCGContext *s, TCGArg *gen_args, TCGArg dst,
 111                             TCGArg src, int nb_temps, int nb_globals)
 112 {
 113         reset_temp(dst, nb_temps, nb_globals);
 114         assert(temps[src].state != TCG_TEMP_COPY);
 115         /* Don't try to copy if one of temps is a global or either one
 116            is local and another is register */
 117         if (src >= nb_globals && dst >= nb_globals &&
 118             tcg_arg_is_local(s, src) == tcg_arg_is_local(s, dst)) {
 119             assert(temps[src].state != TCG_TEMP_CONST);
 120             if (temps[src].state != TCG_TEMP_HAS_COPY) {
 121                 temps[src].state = TCG_TEMP_HAS_COPY;
 122                 temps[src].next_copy = src;
 123                 temps[src].prev_copy = src;
 124             }
 125             temps[dst].state = TCG_TEMP_COPY;
 126             temps[dst].val = src;
 127             temps[dst].next_copy = temps[src].next_copy;
 128             temps[dst].prev_copy = src;
 129             temps[temps[dst].next_copy].prev_copy = dst;
 130             temps[src].next_copy = dst;
 131         }
 132         gen_args[0] = dst;
 133         gen_args[1] = src;
 134 }
 135
 136 static void tcg_opt_gen_movi(TCGArg *gen_args, TCGArg dst, TCGArg val,
 137                              int nb_temps, int nb_globals)
 138 {
 139         reset_temp(dst, nb_temps, nb_globals);
 140         temps[dst].state = TCG_TEMP_CONST;
 141         temps[dst].val = val;
 142         gen_args[0] = dst;
 143         gen_args[1] = val;
 144 }
 145
 146 static TCGOpcode op_to_mov(TCGOpcode op)
 147 {
 148     switch (op_bits(op)) {
 149     case 32:
 150         return INDEX_op_mov_i32;
 151     case 64:
 152         return INDEX_op_mov_i64;
 153     default:
 154         fprintf(stderr, "op_to_mov: unexpected return value of "
 155                 "function op_bits.\n");
 156         tcg_abort();
 157     }
 158 }
 159
 160 static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y)
 161 {
 162     switch (op) {
 163     CASE_OP_32_64(add):
 164         return x + y;
 165
 166     CASE_OP_32_64(sub):
 167         return x - y;
 168
 169     CASE_OP_32_64(mul):
 170         return x * y;
 171
 172     CASE_OP_32_64(and):
 173         return x & y;
 174
 175     CASE_OP_32_64(or):
 176         return x | y;
 177
 178     CASE_OP_32_64(xor):
 179         return x ^ y;
 180
 181     case INDEX_op_shl_i32:
 182         return (uint32_t)x << (uint32_t)y;
 183
 184     case INDEX_op_shl_i64:
 185         return (uint64_t)x << (uint64_t)y;
 186
 187     case INDEX_op_shr_i32:
 188         return (uint32_t)x >> (uint32_t)y;
 189
 190     case INDEX_op_shr_i64:
 191         return (uint64_t)x >> (uint64_t)y;
 192
 193     case INDEX_op_sar_i32:
 194         return (int32_t)x >> (int32_t)y;
 195
 196     case INDEX_op_sar_i64:
 197         return (int64_t)x >> (int64_t)y;
 198
 199     case INDEX_op_rotr_i32:
 200         x = ((uint32_t)x << (32 - y)) | ((uint32_t)x >> y);
 201         return x;
 202
 203     case INDEX_op_rotr_i64:
 204         x = ((uint64_t)x << (64 - y)) | ((uint64_t)x >> y);
 205         return x;
 206
 207     case INDEX_op_rotl_i32:
 208         x = ((uint32_t)x << y) | ((uint32_t)x >> (32 - y));
 209         return x;
 210
 211     case INDEX_op_rotl_i64:
 212         x = ((uint64_t)x << y) | ((uint64_t)x >> (64 - y));
 213         return x;
 214
 215     CASE_OP_32_64(not):
 216         return ~x;
 217
 218     CASE_OP_32_64(neg):
 219         return -x;
 220
 221     CASE_OP_32_64(andc):
 222         return x & ~y;
 223
 224     CASE_OP_32_64(orc):
 225         return x | ~y;
 226
 227     CASE_OP_32_64(eqv):
 228         return ~(x ^ y);
 229
 230     CASE_OP_32_64(nand):
 231         return ~(x & y);
 232
 233     CASE_OP_32_64(nor):
 234         return ~(x | y);
 235
 236     CASE_OP_32_64(ext8s):
 237         return (int8_t)x;
 238
 239     CASE_OP_32_64(ext16s):
 240         return (int16_t)x;
 241
 242     CASE_OP_32_64(ext8u):
 243         return (uint8_t)x;
 244
 245     CASE_OP_32_64(ext16u):
 246         return (uint16_t)x;
 247
 248     case INDEX_op_ext32s_i64:
 249         return (int32_t)x;
 250
 251     case INDEX_op_ext32u_i64:
 252         return (uint32_t)x;
 253
 254     default:
 255         fprintf(stderr,
 256                 "Unrecognized operation %d in do_constant_folding.\n", op);
 257         tcg_abort();
 258     }
 259 }
 260
 261 static TCGArg do_constant_folding(TCGOpcode op, TCGArg x, TCGArg y)
 262 {
 263     TCGArg res = do_constant_folding_2(op, x, y);
 264     if (op_bits(op) == 32) {
 265         res &= 0xffffffff;
 266     }
 267     return res;
 268 }
 269
 270 /* Propagate constants and copies, fold constant expressions. */
 271 static TCGArg *tcg_constant_folding(TCGContext *s, uint16_t *tcg_opc_ptr,
 272                                     TCGArg *args, TCGOpDef *tcg_op_defs)
 273 {
 274     int i, nb_ops, op_index, nb_temps, nb_globals, nb_call_args;
 275     TCGOpcode op;
 276     const TCGOpDef *def;
 277     TCGArg *gen_args;
 278     TCGArg tmp;
 279     /* Array VALS has an element for each temp.
 280        If this temp holds a constant then its value is kept in VALS' element.
 281        If this temp is a copy of other ones then this equivalence class'
 282        representative is kept in VALS' element.
 283        If this temp is neither copy nor constant then corresponding VALS'
 284        element is unused. */
 285
 286     nb_temps = s->nb_temps;
 287     nb_globals = s->nb_globals;
 288     memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info));
 289
 290     nb_ops = tcg_opc_ptr - gen_opc_buf;
 291     gen_args = args;
 292     for (op_index = 0; op_index < nb_ops; op_index++) {
 293         op = gen_opc_buf[op_index];
 294         def = &tcg_op_defs[op];
 295         /* Do copy propagation */
 296         if (!(def->flags & (TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS))) {
 297             assert(op != INDEX_op_call);
 298             for (i = def->nb_oargs; i < def->nb_oargs + def->nb_iargs; i++) {
 299                 if (temps[args[i]].state == TCG_TEMP_COPY) {
 300                     args[i] = temps[args[i]].val;
 301                 }
 302             }
 303         }
 304
 305         /* For commutative operations make constant second argument */
 306         switch (op) {
 307         CASE_OP_32_64(add):
 308         CASE_OP_32_64(mul):
 309         CASE_OP_32_64(and):
 310         CASE_OP_32_64(or):
 311         CASE_OP_32_64(xor):
 312         CASE_OP_32_64(eqv):
 313         CASE_OP_32_64(nand):
 314         CASE_OP_32_64(nor):
 315             if (temps[args[1]].state == TCG_TEMP_CONST) {
 316                 tmp = args[1];
 317                 args[1] = args[2];
 318                 args[2] = tmp;
 319             }
 320             break;
 321         default:
 322             break;
 323         }
 324
 325         /* Simplify expressions for "shift/rot r, 0, a => movi r, 0" */
 326         switch (op) {
 327         CASE_OP_32_64(shl):
 328         CASE_OP_32_64(shr):
 329         CASE_OP_32_64(sar):
 330         CASE_OP_32_64(rotl):
 331         CASE_OP_32_64(rotr):
 332             if (temps[args[1]].state == TCG_TEMP_CONST
 333                 && temps[args[1]].val == 0) {
 334                 gen_opc_buf[op_index] = op_to_movi(op);
 335                 tcg_opt_gen_movi(gen_args, args[0], 0, nb_temps, nb_globals);
 336                 args += 3;
 337                 gen_args += 2;
 338                 continue;
 339             }
 340             break;
 341         default:
 342             break;
 343         }
 344
 345         /* Simplify expression for "op r, a, 0 => mov r, a" cases */
 346         switch (op) {
 347         CASE_OP_32_64(add):
 348         CASE_OP_32_64(sub):
 349         CASE_OP_32_64(shl):
 350         CASE_OP_32_64(shr):
 351         CASE_OP_32_64(sar):
 352         CASE_OP_32_64(rotl):
 353         CASE_OP_32_64(rotr):
 354         CASE_OP_32_64(or):
 355         CASE_OP_32_64(xor):
 356             if (temps[args[1]].state == TCG_TEMP_CONST) {
 357                 /* Proceed with possible constant folding. */
 358                 break;
 359             }
 360             if (temps[args[2]].state == TCG_TEMP_CONST
 361                 && temps[args[2]].val == 0) {
 362                 if ((temps[args[0]].state == TCG_TEMP_COPY
 363                     && temps[args[0]].val == args[1])
 364                     || args[0] == args[1]) {
 365                     args += 3;
 366                     gen_opc_buf[op_index] = INDEX_op_nop;
 367                 } else {
 368                     gen_opc_buf[op_index] = op_to_mov(op);
 369                     tcg_opt_gen_mov(s, gen_args, args[0], args[1],
 370                                     nb_temps, nb_globals);
 371                     gen_args += 2;
 372                     args += 3;
 373                 }
 374                 continue;
 375             }
 376             break;
 377         default:
 378             break;
 379         }
 380
 381         /* Simplify expression for "op r, a, 0 => movi r, 0" cases */
 382         switch (op) {
 383         CASE_OP_32_64(and):
 384         CASE_OP_32_64(mul):
 385             if ((temps[args[2]].state == TCG_TEMP_CONST
 386                 && temps[args[2]].val == 0)) {
 387                 gen_opc_buf[op_index] = op_to_movi(op);
 388                 tcg_opt_gen_movi(gen_args, args[0], 0, nb_temps, nb_globals);
 389                 args += 3;
 390                 gen_args += 2;
 391                 continue;
 392             }
 393             break;
 394         default:
 395             break;
 396         }
 397
 398         /* Simplify expression for "op r, a, a => mov r, a" cases */
 399         switch (op) {
 400         CASE_OP_32_64(or):
 401         CASE_OP_32_64(and):
 402             if (args[1] == args[2]) {
 403                 if (args[1] == args[0]) {
 404                     args += 3;
 405                     gen_opc_buf[op_index] = INDEX_op_nop;
 406                 } else {
 407                     gen_opc_buf[op_index] = op_to_mov(op);
 408                     tcg_opt_gen_mov(s, gen_args, args[0], args[1], nb_temps,
 409                                     nb_globals);
 410                     gen_args += 2;
 411                     args += 3;
 412                 }
 413                 continue;
 414             }
 415             break;
 416         default:
 417             break;
 418         }
 419
 420         /* Propagate constants through copy operations and do constant
 421            folding.  Constants will be substituted to arguments by register
 422            allocator where needed and possible.  Also detect copies. */
 423         switch (op) {
 424         CASE_OP_32_64(mov):
 425             if ((temps[args[1]].state == TCG_TEMP_COPY
 426                 && temps[args[1]].val == args[0])
 427                 || args[0] == args[1]) {
 428                 args += 2;
 429                 gen_opc_buf[op_index] = INDEX_op_nop;
 430                 break;
 431             }
 432             if (temps[args[1]].state != TCG_TEMP_CONST) {
 433                 tcg_opt_gen_mov(s, gen_args, args[0], args[1],
 434                                 nb_temps, nb_globals);
 435                 gen_args += 2;
 436                 args += 2;
 437                 break;
 438             }
 439             /* Source argument is constant.  Rewrite the operation and
 440                let movi case handle it. */
 441             op = op_to_movi(op);
 442             gen_opc_buf[op_index] = op;
 443             args[1] = temps[args[1]].val;
 444             /* fallthrough */
 445         CASE_OP_32_64(movi):
 446             tcg_opt_gen_movi(gen_args, args[0], args[1], nb_temps, nb_globals);
 447             gen_args += 2;
 448             args += 2;
 449             break;
 450         CASE_OP_32_64(not):
 451         CASE_OP_32_64(neg):
 452         CASE_OP_32_64(ext8s):
 453         CASE_OP_32_64(ext8u):
 454         CASE_OP_32_64(ext16s):
 455         CASE_OP_32_64(ext16u):
 456         case INDEX_op_ext32s_i64:
 457         case INDEX_op_ext32u_i64:
 458             if (temps[args[1]].state == TCG_TEMP_CONST) {
 459                 gen_opc_buf[op_index] = op_to_movi(op);
 460                 tmp = do_constant_folding(op, temps[args[1]].val, 0);
 461                 tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals);
 462                 gen_args += 2;
 463                 args += 2;
 464                 break;
 465             } else {
 466                 reset_temp(args[0], nb_temps, nb_globals);
 467                 gen_args[0] = args[0];
 468                 gen_args[1] = args[1];
 469                 gen_args += 2;
 470                 args += 2;
 471                 break;
 472             }
 473         CASE_OP_32_64(add):
 474         CASE_OP_32_64(sub):
 475         CASE_OP_32_64(mul):
 476         CASE_OP_32_64(or):
 477         CASE_OP_32_64(and):
 478         CASE_OP_32_64(xor):
 479         CASE_OP_32_64(shl):
 480         CASE_OP_32_64(shr):
 481         CASE_OP_32_64(sar):
 482         CASE_OP_32_64(rotl):
 483         CASE_OP_32_64(rotr):
 484         CASE_OP_32_64(andc):
 485         CASE_OP_32_64(orc):
 486         CASE_OP_32_64(eqv):
 487         CASE_OP_32_64(nand):
 488         CASE_OP_32_64(nor):
 489             if (temps[args[1]].state == TCG_TEMP_CONST
 490                 && temps[args[2]].state == TCG_TEMP_CONST) {
 491                 gen_opc_buf[op_index] = op_to_movi(op);
 492                 tmp = do_constant_folding(op, temps[args[1]].val,
 493                                           temps[args[2]].val);
 494                 tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals);
 495                 gen_args += 2;
 496                 args += 3;
 497                 break;
 498             } else {
 499                 reset_temp(args[0], nb_temps, nb_globals);
 500                 gen_args[0] = args[0];
 501                 gen_args[1] = args[1];
 502                 gen_args[2] = args[2];
 503                 gen_args += 3;
 504                 args += 3;
 505                 break;
 506             }
 507         case INDEX_op_call:
 508             nb_call_args = (args[0] >> 16) + (args[0] & 0xffff);
 509             if (!(args[nb_call_args + 1] & (TCG_CALL_CONST | TCG_CALL_PURE))) {
 510                 for (i = 0; i < nb_globals; i++) {
 511                     reset_temp(i, nb_temps, nb_globals);
 512                 }
 513             }
 514             for (i = 0; i < (args[0] >> 16); i++) {
 515                 reset_temp(args[i + 1], nb_temps, nb_globals);
 516             }
 517             i = nb_call_args + 3;
 518             while (i) {
 519                 *gen_args = *args;
 520                 args++;
 521                 gen_args++;
 522                 i--;
 523             }
 524             break;
 525         case INDEX_op_set_label:
 526         case INDEX_op_jmp:
 527         case INDEX_op_br:
 528         CASE_OP_32_64(brcond):
 529             memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info));
 530             for (i = 0; i < def->nb_args; i++) {
 531                 *gen_args = *args;
 532                 args++;
 533                 gen_args++;
 534             }
 535             break;
 536         default:
 537             /* Default case: we do know nothing about operation so no
 538                propagation is done.  We only trash output args.  */
 539             for (i = 0; i < def->nb_oargs; i++) {
 540                 reset_temp(args[i], nb_temps, nb_globals);
 541             }
 542             for (i = 0; i < def->nb_args; i++) {
 543                 gen_args[i] = args[i];
 544             }
 545             args += def->nb_args;
 546             gen_args += def->nb_args;
 547             break;
 548         }
 549     }
 550
 551     return gen_args;
 552 }
 553
 554 TCGArg *tcg_optimize(TCGContext *s, uint16_t *tcg_opc_ptr,
 555         TCGArg *args, TCGOpDef *tcg_op_defs)
 556 {
 557     TCGArg *res;
 558     res = tcg_constant_folding(s, tcg_opc_ptr, args, tcg_op_defs);
 559     return res;
 560 }