gcc/config/arm/aarch-common.c

   1 /* Dependency checks for instruction scheduling, shared between ARM and
   2    AARCH64.
   3
   4    Copyright (C) 1991-2017 Free Software Foundation, Inc.
   5    Contributed by ARM Ltd.
   6
   7    This file is part of GCC.
   8
   9    GCC is free software; you can redistribute it and/or modify it
  10    under the terms of the GNU General Public License as published
  11    by the Free Software Foundation; either version 3, or (at your
  12    option) any later version.
  13
  14    GCC is distributed in the hope that it will be useful, but WITHOUT
  15    ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  16    or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
  17    License for more details.
  18
  19    You should have received a copy of the GNU General Public License
  20    along with GCC; see the file COPYING3.  If not see
  21    <http://www.gnu.org/licenses/>.  */
  22
  23
  24 #include "config.h"
  25 #include "system.h"
  26 #include "coretypes.h"
  27 #include "tm.h"
  28 #include "rtl.h"
  29 #include "rtl-iter.h"
  30
  31 /* In ARMv8-A there's a general expectation that AESE/AESMC
  32    and AESD/AESIMC sequences of the form:
  33
  34    AESE Vn, _
  35    AESMC Vn, Vn
  36
  37    will issue both instructions in a single cycle on super-scalar
  38    implementations.  This function identifies such pairs.  */
  39
  40 int
  41 aarch_crypto_can_dual_issue (rtx_insn *producer_insn, rtx_insn *consumer_insn)
  42 {
  43   rtx producer_set, consumer_set;
  44   rtx producer_src, consumer_src;
  45
  46   producer_set = single_set (producer_insn);
  47   consumer_set = single_set (consumer_insn);
  48
  49   producer_src = producer_set ? SET_SRC (producer_set) : NULL;
  50   consumer_src = consumer_set ? SET_SRC (consumer_set) : NULL;
  51
  52   if (producer_src && consumer_src
  53       && GET_CODE (producer_src) == UNSPEC && GET_CODE (consumer_src) == UNSPEC
  54       && ((XINT (producer_src, 1) == UNSPEC_AESE
  55            && XINT (consumer_src, 1) == UNSPEC_AESMC)
  56           || (XINT (producer_src, 1) == UNSPEC_AESD
  57               && XINT (consumer_src, 1) == UNSPEC_AESIMC)))
  58   {
  59     unsigned int regno = REGNO (SET_DEST (producer_set));
  60
  61     /* Before reload the registers are virtual, so the destination of
  62        consumer_set doesn't need to match.  */
  63
  64     return (REGNO (SET_DEST (consumer_set)) == regno || !reload_completed)
  65             && REGNO (XVECEXP (consumer_src, 0, 0)) == regno;
  66   }
  67
  68   return 0;
  69 }
  70
  71 /* Return TRUE if X is either an arithmetic shift left, or
  72    is a multiplication by a power of two.  */
  73 bool
  74 arm_rtx_shift_left_p (rtx x)
  75 {
  76   enum rtx_code code = GET_CODE (x);
  77
  78   if (code == MULT && CONST_INT_P (XEXP (x, 1))
  79       && exact_log2 (INTVAL (XEXP (x, 1))) > 0)
  80     return true;
  81
  82   if (code == ASHIFT)
  83     return true;
  84
  85   return false;
  86 }
  87
  88 static rtx_code shift_rtx_codes[] =
  89   { ASHIFT, ROTATE, ASHIFTRT, LSHIFTRT,
  90     ROTATERT, ZERO_EXTEND, SIGN_EXTEND };
  91
  92 /* Traverse PATTERN looking for a sub-rtx with RTX_CODE CODE.
  93    If FIND_ANY_SHIFT then we are interested in anything which can
  94    reasonably be described as a SHIFT RTX.  */
  95 static rtx
  96 arm_find_sub_rtx_with_code (rtx pattern, rtx_code code, bool find_any_shift)
  97 {
  98   subrtx_var_iterator::array_type array;
  99   FOR_EACH_SUBRTX_VAR (iter, array, pattern, NONCONST)
 100     {
 101       rtx x = *iter;
 102       if (find_any_shift)
 103         {
 104           /* Left shifts might have been canonicalized to a MULT of some
 105              power of two.  Make sure we catch them.  */
 106           if (arm_rtx_shift_left_p (x))
 107             return x;
 108           else
 109             for (unsigned int i = 0; i < ARRAY_SIZE (shift_rtx_codes); i++)
 110               if (GET_CODE (x) == shift_rtx_codes[i])
 111                 return x;
 112         }
 113
 114       if (GET_CODE (x) == code)
 115         return x;
 116     }
 117   return NULL_RTX;
 118 }
 119
 120 /* Traverse PATTERN looking for any sub-rtx which looks like a shift.  */
 121 static rtx
 122 arm_find_shift_sub_rtx (rtx pattern)
 123 {
 124   return arm_find_sub_rtx_with_code (pattern, ASHIFT, true);
 125 }
 126
 127 /* PRODUCER and CONSUMER are two potentially dependant RTX.  PRODUCER
 128    (possibly) contains a SET which will provide a result we can access
 129    using the SET_DEST macro.  We will place the RTX which would be
 130    written by PRODUCER in SET_SOURCE.
 131    Similarly, CONSUMER (possibly) contains a SET which has an operand
 132    we can access using SET_SRC.  We place this operand in
 133    SET_DESTINATION.
 134
 135    Return nonzero if we found the SET RTX we expected.  */
 136 static int
 137 arm_get_set_operands (rtx producer, rtx consumer,
 138                       rtx *set_source, rtx *set_destination)
 139 {
 140   rtx set_producer = arm_find_sub_rtx_with_code (PATTERN (producer),
 141                                                  SET, false);
 142   rtx set_consumer = arm_find_sub_rtx_with_code (PATTERN (consumer),
 143                                                  SET, false);
 144
 145   if (set_producer && set_consumer)
 146     {
 147       *set_source = SET_DEST (set_producer);
 148       *set_destination = SET_SRC (set_consumer);
 149       return 1;
 150     }
 151   return 0;
 152 }
 153
 154 bool
 155 aarch_rev16_shright_mask_imm_p (rtx val, machine_mode mode)
 156 {
 157   return CONST_INT_P (val)
 158          && INTVAL (val)
 159             == trunc_int_for_mode (HOST_WIDE_INT_C (0xff00ff00ff00ff),
 160                                    mode);
 161 }
 162
 163 bool
 164 aarch_rev16_shleft_mask_imm_p (rtx val, machine_mode mode)
 165 {
 166   return CONST_INT_P (val)
 167          && INTVAL (val)
 168             == trunc_int_for_mode (HOST_WIDE_INT_C (0xff00ff00ff00ff00),
 169                                    mode);
 170 }
 171
 172
 173 static bool
 174 aarch_rev16_p_1 (rtx lhs, rtx rhs, machine_mode mode)
 175 {
 176   if (GET_CODE (lhs) == AND
 177          && GET_CODE (XEXP (lhs, 0)) == ASHIFT
 178             && CONST_INT_P (XEXP (XEXP (lhs, 0), 1))
 179             && INTVAL (XEXP (XEXP (lhs, 0), 1)) == 8
 180             && REG_P (XEXP (XEXP (lhs, 0), 0))
 181          && CONST_INT_P (XEXP (lhs, 1))
 182       && GET_CODE (rhs) == AND
 183          && GET_CODE (XEXP (rhs, 0)) == LSHIFTRT
 184             && REG_P (XEXP (XEXP (rhs, 0), 0))
 185             && CONST_INT_P (XEXP (XEXP (rhs, 0), 1))
 186             && INTVAL (XEXP (XEXP (rhs, 0), 1)) == 8
 187          && CONST_INT_P (XEXP (rhs, 1))
 188       && REGNO (XEXP (XEXP (rhs, 0), 0)) == REGNO (XEXP (XEXP (lhs, 0), 0)))
 189
 190     {
 191       rtx lhs_mask = XEXP (lhs, 1);
 192       rtx rhs_mask = XEXP (rhs, 1);
 193
 194       return aarch_rev16_shright_mask_imm_p (rhs_mask, mode)
 195              && aarch_rev16_shleft_mask_imm_p (lhs_mask, mode);
 196     }
 197
 198   return false;
 199 }
 200
 201 /* Recognise a sequence of bitwise operations corresponding to a rev16 operation.
 202    These will be of the form:
 203      ((x >> 8) & 0x00ff00ff)
 204    | ((x << 8) & 0xff00ff00)
 205    for SImode and with similar but wider bitmasks for DImode.
 206    The two sub-expressions of the IOR can appear on either side so check both
 207    permutations with the help of aarch_rev16_p_1 above.  */
 208
 209 bool
 210 aarch_rev16_p (rtx x)
 211 {
 212   rtx left_sub_rtx, right_sub_rtx;
 213   bool is_rev = false;
 214
 215   if (GET_CODE (x) != IOR)
 216     return false;
 217
 218   left_sub_rtx = XEXP (x, 0);
 219   right_sub_rtx = XEXP (x, 1);
 220
 221   /* There are no canonicalisation rules for the position of the two shifts
 222      involved in a rev, so try both permutations.  */
 223   is_rev = aarch_rev16_p_1 (left_sub_rtx, right_sub_rtx, GET_MODE (x));
 224
 225   if (!is_rev)
 226     is_rev = aarch_rev16_p_1 (right_sub_rtx, left_sub_rtx, GET_MODE (x));
 227
 228   return is_rev;
 229 }
 230
 231 /* Return nonzero if the CONSUMER instruction (a load) does need
 232    PRODUCER's value to calculate the address.  */
 233 int
 234 arm_early_load_addr_dep (rtx producer, rtx consumer)
 235 {
 236   rtx value, addr;
 237
 238   if (!arm_get_set_operands (producer, consumer, &value, &addr))
 239     return 0;
 240
 241   return reg_overlap_mentioned_p (value, addr);
 242 }
 243
 244 /* Return nonzero if the CONSUMER instruction (a load) does need
 245    a Pmode PRODUCER's value to calculate the address.  */
 246
 247 int
 248 arm_early_load_addr_dep_ptr (rtx producer, rtx consumer)
 249 {
 250   rtx value = arm_find_sub_rtx_with_code (PATTERN (producer), SET, false);
 251   rtx addr = arm_find_sub_rtx_with_code (PATTERN (consumer), SET, false);
 252
 253   if (!value || !addr || !MEM_P (SET_SRC (value)))
 254     return 0;
 255
 256   value = SET_DEST (value);
 257   addr = SET_SRC (addr);
 258
 259   return GET_MODE (value) == Pmode && reg_overlap_mentioned_p (value, addr);
 260 }
 261
 262 /* Return nonzero if the CONSUMER instruction (an ALU op) does not
 263    have an early register shift value or amount dependency on the
 264    result of PRODUCER.  */
 265 int
 266 arm_no_early_alu_shift_dep (rtx producer, rtx consumer)
 267 {
 268   rtx value, op;
 269   rtx early_op;
 270
 271   if (!arm_get_set_operands (producer, consumer, &value, &op))
 272     return 0;
 273
 274   if ((early_op = arm_find_shift_sub_rtx (op)))
 275     return !reg_overlap_mentioned_p (value, early_op);
 276
 277   return 0;
 278 }
 279
 280 /* Return nonzero if the CONSUMER instruction (an ALU op) does not
 281    have an early register shift value dependency on the result of
 282    PRODUCER.  */
 283 int
 284 arm_no_early_alu_shift_value_dep (rtx producer, rtx consumer)
 285 {
 286   rtx value, op;
 287   rtx early_op;
 288
 289   if (!arm_get_set_operands (producer, consumer, &value, &op))
 290     return 0;
 291
 292   if ((early_op = arm_find_shift_sub_rtx (op)))
 293     /* We want to check the value being shifted.  */
 294     if (!reg_overlap_mentioned_p (value, XEXP (early_op, 0)))
 295       return 1;
 296
 297   return 0;
 298 }
 299
 300 /* Return nonzero if the CONSUMER (a mul or mac op) does not
 301    have an early register mult dependency on the result of
 302    PRODUCER.  */
 303 int
 304 arm_no_early_mul_dep (rtx producer, rtx consumer)
 305 {
 306   rtx value, op;
 307
 308   if (!arm_get_set_operands (producer, consumer, &value, &op))
 309     return 0;
 310
 311   if (GET_CODE (op) == PLUS || GET_CODE (op) == MINUS)
 312     {
 313       if (GET_CODE (XEXP (op, 0)) == MULT)
 314         return !reg_overlap_mentioned_p (value, XEXP (op, 0));
 315       else
 316         return !reg_overlap_mentioned_p (value, XEXP (op, 1));
 317     }
 318
 319   return 0;
 320 }
 321
 322 /* Return nonzero if the CONSUMER instruction (a store) does not need
 323    PRODUCER's value to calculate the address.  */
 324
 325 int
 326 arm_no_early_store_addr_dep (rtx producer, rtx consumer)
 327 {
 328   rtx value = arm_find_sub_rtx_with_code (PATTERN (producer), SET, false);
 329   rtx addr = arm_find_sub_rtx_with_code (PATTERN (consumer), SET, false);
 330
 331   if (value)
 332     value = SET_DEST (value);
 333
 334   if (addr)
 335     addr = SET_DEST (addr);
 336
 337   if (!value || !addr)
 338     return 0;
 339
 340   return !reg_overlap_mentioned_p (value, addr);
 341 }
 342
 343 /* Return nonzero if the CONSUMER instruction (a store) does need
 344    PRODUCER's value to calculate the address.  */
 345
 346 int
 347 arm_early_store_addr_dep (rtx producer, rtx consumer)
 348 {
 349   return !arm_no_early_store_addr_dep (producer, consumer);
 350 }
 351
 352 /* Return nonzero if the CONSUMER instruction (a store) does need
 353    a Pmode PRODUCER's value to calculate the address.  */
 354
 355 int
 356 arm_early_store_addr_dep_ptr (rtx producer, rtx consumer)
 357 {
 358   rtx value = arm_find_sub_rtx_with_code (PATTERN (producer), SET, false);
 359   rtx addr = arm_find_sub_rtx_with_code (PATTERN (consumer), SET, false);
 360
 361   if (!value || !addr || !MEM_P (SET_SRC (value)))
 362     return 0;
 363
 364   value = SET_DEST (value);
 365   addr = SET_DEST (addr);
 366
 367   return GET_MODE (value) == Pmode && reg_overlap_mentioned_p (value, addr);
 368 }
 369
 370 /* Return non-zero iff the consumer (a multiply-accumulate or a
 371    multiple-subtract instruction) has an accumulator dependency on the
 372    result of the producer and no other dependency on that result.  It
 373    does not check if the producer is multiply-accumulate instruction.  */
 374 int
 375 arm_mac_accumulator_is_result (rtx producer, rtx consumer)
 376 {
 377   rtx result;
 378   rtx op0, op1, acc;
 379
 380   producer = PATTERN (producer);
 381   consumer = PATTERN (consumer);
 382
 383   if (GET_CODE (producer) == COND_EXEC)
 384     producer = COND_EXEC_CODE (producer);
 385   if (GET_CODE (consumer) == COND_EXEC)
 386     consumer = COND_EXEC_CODE (consumer);
 387
 388   if (GET_CODE (producer) != SET)
 389     return 0;
 390
 391   result = XEXP (producer, 0);
 392
 393   if (GET_CODE (consumer) != SET)
 394     return 0;
 395
 396   /* Check that the consumer is of the form
 397      (set (...) (plus (mult ...) (...)))
 398      or
 399      (set (...) (minus (...) (mult ...))).  */
 400   if (GET_CODE (XEXP (consumer, 1)) == PLUS)
 401     {
 402       if (GET_CODE (XEXP (XEXP (consumer, 1), 0)) != MULT)
 403         return 0;
 404
 405       op0 = XEXP (XEXP (XEXP (consumer, 1), 0), 0);
 406       op1 = XEXP (XEXP (XEXP (consumer, 1), 0), 1);
 407       acc = XEXP (XEXP (consumer, 1), 1);
 408     }
 409   else if (GET_CODE (XEXP (consumer, 1)) == MINUS)
 410     {
 411       if (GET_CODE (XEXP (XEXP (consumer, 1), 1)) != MULT)
 412         return 0;
 413
 414       op0 = XEXP (XEXP (XEXP (consumer, 1), 1), 0);
 415       op1 = XEXP (XEXP (XEXP (consumer, 1), 1), 1);
 416       acc = XEXP (XEXP (consumer, 1), 0);
 417     }
 418   else
 419     return 0;
 420
 421   return (reg_overlap_mentioned_p (result, acc)
 422           && !reg_overlap_mentioned_p (result, op0)
 423           && !reg_overlap_mentioned_p (result, op1));
 424 }
 425
 426 /* Return non-zero if the destination of PRODUCER feeds the accumulator
 427    operand of an MLA-like operation.  */
 428
 429 int
 430 aarch_accumulator_forwarding (rtx_insn *producer, rtx_insn *consumer)
 431 {
 432   rtx producer_set = single_set (producer);
 433   rtx consumer_set = single_set (consumer);
 434
 435   /* We are looking for a SET feeding a SET.  */
 436   if (!producer_set || !consumer_set)
 437     return 0;
 438
 439   rtx dest = SET_DEST (producer_set);
 440   rtx mla = SET_SRC (consumer_set);
 441
 442   /* We're looking for a register SET.  */
 443   if (!REG_P (dest))
 444     return 0;
 445
 446   rtx accumulator;
 447
 448   /* Strip a zero_extend.  */
 449   if (GET_CODE (mla) == ZERO_EXTEND)
 450     mla = XEXP (mla, 0);
 451
 452   switch (GET_CODE (mla))
 453     {
 454     case PLUS:
 455       /* Possibly an MADD.  */
 456       if (GET_CODE (XEXP (mla, 0)) == MULT)
 457         accumulator = XEXP (mla, 1);
 458       else
 459         return 0;
 460       break;
 461     case MINUS:
 462       /* Possibly an MSUB.  */
 463       if (GET_CODE (XEXP (mla, 1)) == MULT)
 464         accumulator = XEXP (mla, 0);
 465       else
 466         return 0;
 467       break;
 468     case FMA:
 469         {
 470           /* Possibly an FMADD/FMSUB/FNMADD/FNMSUB.  */
 471           if (REG_P (XEXP (mla, 1))
 472               && REG_P (XEXP (mla, 2))
 473               && (REG_P (XEXP (mla, 0))
 474                   || GET_CODE (XEXP (mla, 0)) == NEG))
 475
 476             {
 477               /* FMADD/FMSUB.  */
 478               accumulator = XEXP (mla, 2);
 479             }
 480           else if (REG_P (XEXP (mla, 1))
 481                    && GET_CODE (XEXP (mla, 2)) == NEG
 482                    && (REG_P (XEXP (mla, 0))
 483                        || GET_CODE (XEXP (mla, 0)) == NEG))
 484             {
 485               /* FNMADD/FNMSUB.  */
 486               accumulator = XEXP (XEXP (mla, 2), 0);
 487             }
 488           else
 489             return 0;
 490           break;
 491         }
 492       default:
 493         /* Not an MLA-like operation.  */
 494         return 0;
 495     }
 496
 497   if (GET_CODE (accumulator) == SUBREG)
 498     accumulator = SUBREG_REG (accumulator);
 499
 500   if (!REG_P (accumulator))
 501     return 0;
 502
 503   return (REGNO (dest) == REGNO (accumulator));
 504 }
 505
 506 /* Return non-zero if the consumer (a multiply-accumulate instruction)
 507    has an accumulator dependency on the result of the producer (a
 508    multiplication instruction) and no other dependency on that result.  */
 509 int
 510 arm_mac_accumulator_is_mul_result (rtx producer, rtx consumer)
 511 {
 512   rtx mul = PATTERN (producer);
 513   rtx mac = PATTERN (consumer);
 514   rtx mul_result;
 515   rtx mac_op0, mac_op1, mac_acc;
 516
 517   if (GET_CODE (mul) == COND_EXEC)
 518     mul = COND_EXEC_CODE (mul);
 519   if (GET_CODE (mac) == COND_EXEC)
 520     mac = COND_EXEC_CODE (mac);
 521
 522   /* Check that mul is of the form (set (...) (mult ...))
 523      and mla is of the form (set (...) (plus (mult ...) (...))).  */
 524   if ((GET_CODE (mul) != SET || GET_CODE (XEXP (mul, 1)) != MULT)
 525       || (GET_CODE (mac) != SET || GET_CODE (XEXP (mac, 1)) != PLUS
 526           || GET_CODE (XEXP (XEXP (mac, 1), 0)) != MULT))
 527     return 0;
 528
 529   mul_result = XEXP (mul, 0);
 530   mac_op0 = XEXP (XEXP (XEXP (mac, 1), 0), 0);
 531   mac_op1 = XEXP (XEXP (XEXP (mac, 1), 0), 1);
 532   mac_acc = XEXP (XEXP (mac, 1), 1);
 533
 534   return (reg_overlap_mentioned_p (mul_result, mac_acc)
 535           && !reg_overlap_mentioned_p (mul_result, mac_op0)
 536           && !reg_overlap_mentioned_p (mul_result, mac_op1));
 537 }