target-arm/iwmmxt_helper.c

   1 /*
   2  * iwMMXt micro operations for XScale.
   3  *
   4  * Copyright (c) 2007 OpenedHand, Ltd.
   5  * Written by Andrzej Zaborowski <andrew@openedhand.com>
   6  * Copyright (c) 2008 CodeSourcery
   7  *
   8  * This library is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2 of the License, or (at your option) any later version.
  12  *
  13  * This library is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  20  */
  21
  22 #include <stdlib.h>
  23 #include <stdio.h>
  24
  25 #include "cpu.h"
  26 #include "exec/exec-all.h"
  27 #include "exec/helper-proto.h"
  28
  29 /* iwMMXt macros extracted from GNU gdb.  */
  30
  31 /* Set the SIMD wCASF flags for 8, 16, 32 or 64-bit operations.  */
  32 #define SIMD8_SET( v, n, b)     ((v != 0) << ((((b) + 1) * 4) + (n)))
  33 #define SIMD16_SET(v, n, h)     ((v != 0) << ((((h) + 1) * 8) + (n)))
  34 #define SIMD32_SET(v, n, w)     ((v != 0) << ((((w) + 1) * 16) + (n)))
  35 #define SIMD64_SET(v, n)        ((v != 0) << (32 + (n)))
  36 /* Flags to pass as "n" above.  */
  37 #define SIMD_NBIT       -1
  38 #define SIMD_ZBIT       -2
  39 #define SIMD_CBIT       -3
  40 #define SIMD_VBIT       -4
  41 /* Various status bit macros.  */
  42 #define NBIT8(x)        ((x) & 0x80)
  43 #define NBIT16(x)       ((x) & 0x8000)
  44 #define NBIT32(x)       ((x) & 0x80000000)
  45 #define NBIT64(x)       ((x) & 0x8000000000000000ULL)
  46 #define ZBIT8(x)        (((x) & 0xff) == 0)
  47 #define ZBIT16(x)       (((x) & 0xffff) == 0)
  48 #define ZBIT32(x)       (((x) & 0xffffffff) == 0)
  49 #define ZBIT64(x)       (x == 0)
  50 /* Sign extension macros.  */
  51 #define EXTEND8H(a)     ((uint16_t) (int8_t) (a))
  52 #define EXTEND8(a)      ((uint32_t) (int8_t) (a))
  53 #define EXTEND16(a)     ((uint32_t) (int16_t) (a))
  54 #define EXTEND16S(a)    ((int32_t) (int16_t) (a))
  55 #define EXTEND32(a)     ((uint64_t) (int32_t) (a))
  56
  57 uint64_t HELPER(iwmmxt_maddsq)(uint64_t a, uint64_t b)
  58 {
  59     a = ((
  60             EXTEND16S((a >> 0) & 0xffff) * EXTEND16S((b >> 0) & 0xffff) +
  61             EXTEND16S((a >> 16) & 0xffff) * EXTEND16S((b >> 16) & 0xffff)
  62         ) & 0xffffffff) | ((uint64_t) (
  63             EXTEND16S((a >> 32) & 0xffff) * EXTEND16S((b >> 32) & 0xffff) +
  64             EXTEND16S((a >> 48) & 0xffff) * EXTEND16S((b >> 48) & 0xffff)
  65         ) << 32);
  66     return a;
  67 }
  68
  69 uint64_t HELPER(iwmmxt_madduq)(uint64_t a, uint64_t b)
  70 {
  71     a = ((
  72             ((a >> 0) & 0xffff) * ((b >> 0) & 0xffff) +
  73             ((a >> 16) & 0xffff) * ((b >> 16) & 0xffff)
  74         ) & 0xffffffff) | ((
  75             ((a >> 32) & 0xffff) * ((b >> 32) & 0xffff) +
  76             ((a >> 48) & 0xffff) * ((b >> 48) & 0xffff)
  77         ) << 32);
  78     return a;
  79 }
  80
  81 uint64_t HELPER(iwmmxt_sadb)(uint64_t a, uint64_t b)
  82 {
  83 #define abs(x) (((x) >= 0) ? x : -x)
  84 #define SADB(SHR) abs((int) ((a >> SHR) & 0xff) - (int) ((b >> SHR) & 0xff))
  85     return
  86         SADB(0) + SADB(8) + SADB(16) + SADB(24) +
  87         SADB(32) + SADB(40) + SADB(48) + SADB(56);
  88 #undef SADB
  89 }
  90
  91 uint64_t HELPER(iwmmxt_sadw)(uint64_t a, uint64_t b)
  92 {
  93 #define SADW(SHR) \
  94     abs((int) ((a >> SHR) & 0xffff) - (int) ((b >> SHR) & 0xffff))
  95     return SADW(0) + SADW(16) + SADW(32) + SADW(48);
  96 #undef SADW
  97 }
  98
  99 uint64_t HELPER(iwmmxt_mulslw)(uint64_t a, uint64_t b)
 100 {
 101 #define MULS(SHR) ((uint64_t) ((( \
 102         EXTEND16S((a >> SHR) & 0xffff) * EXTEND16S((b >> SHR) & 0xffff) \
 103     ) >> 0) & 0xffff) << SHR)
 104     return MULS(0) | MULS(16) | MULS(32) | MULS(48);
 105 #undef MULS
 106 }
 107
 108 uint64_t HELPER(iwmmxt_mulshw)(uint64_t a, uint64_t b)
 109 {
 110 #define MULS(SHR) ((uint64_t) ((( \
 111         EXTEND16S((a >> SHR) & 0xffff) * EXTEND16S((b >> SHR) & 0xffff) \
 112     ) >> 16) & 0xffff) << SHR)
 113     return MULS(0) | MULS(16) | MULS(32) | MULS(48);
 114 #undef MULS
 115 }
 116
 117 uint64_t HELPER(iwmmxt_mululw)(uint64_t a, uint64_t b)
 118 {
 119 #define MULU(SHR) ((uint64_t) ((( \
 120         ((a >> SHR) & 0xffff) * ((b >> SHR) & 0xffff) \
 121     ) >> 0) & 0xffff) << SHR)
 122     return MULU(0) | MULU(16) | MULU(32) | MULU(48);
 123 #undef MULU
 124 }
 125
 126 uint64_t HELPER(iwmmxt_muluhw)(uint64_t a, uint64_t b)
 127 {
 128 #define MULU(SHR) ((uint64_t) ((( \
 129         ((a >> SHR) & 0xffff) * ((b >> SHR) & 0xffff) \
 130     ) >> 16) & 0xffff) << SHR)
 131     return MULU(0) | MULU(16) | MULU(32) | MULU(48);
 132 #undef MULU
 133 }
 134
 135 uint64_t HELPER(iwmmxt_macsw)(uint64_t a, uint64_t b)
 136 {
 137 #define MACS(SHR) ( \
 138         EXTEND16((a >> SHR) & 0xffff) * EXTEND16S((b >> SHR) & 0xffff))
 139     return (int64_t) (MACS(0) + MACS(16) + MACS(32) + MACS(48));
 140 #undef MACS
 141 }
 142
 143 uint64_t HELPER(iwmmxt_macuw)(uint64_t a, uint64_t b)
 144 {
 145 #define MACU(SHR) ( \
 146         (uint32_t) ((a >> SHR) & 0xffff) * \
 147         (uint32_t) ((b >> SHR) & 0xffff))
 148     return MACU(0) + MACU(16) + MACU(32) + MACU(48);
 149 #undef MACU
 150 }
 151
 152 #define NZBIT8(x, i) \
 153     SIMD8_SET(NBIT8((x) & 0xff), SIMD_NBIT, i) | \
 154     SIMD8_SET(ZBIT8((x) & 0xff), SIMD_ZBIT, i)
 155 #define NZBIT16(x, i) \
 156     SIMD16_SET(NBIT16((x) & 0xffff), SIMD_NBIT, i) | \
 157     SIMD16_SET(ZBIT16((x) & 0xffff), SIMD_ZBIT, i)
 158 #define NZBIT32(x, i) \
 159     SIMD32_SET(NBIT32((x) & 0xffffffff), SIMD_NBIT, i) | \
 160     SIMD32_SET(ZBIT32((x) & 0xffffffff), SIMD_ZBIT, i)
 161 #define NZBIT64(x) \
 162     SIMD64_SET(NBIT64(x), SIMD_NBIT) | \
 163     SIMD64_SET(ZBIT64(x), SIMD_ZBIT)
 164 #define IWMMXT_OP_UNPACK(S, SH0, SH1, SH2, SH3)                 \
 165 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, b)))(CPUARMState *env, \
 166                                                  uint64_t a, uint64_t b) \
 167 {                                                               \
 168     a =                                                         \
 169         (((a >> SH0) & 0xff) << 0) | (((b >> SH0) & 0xff) << 8) |       \
 170         (((a >> SH1) & 0xff) << 16) | (((b >> SH1) & 0xff) << 24) |     \
 171         (((a >> SH2) & 0xff) << 32) | (((b >> SH2) & 0xff) << 40) |     \
 172         (((a >> SH3) & 0xff) << 48) | (((b >> SH3) & 0xff) << 56);      \
 173     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 174         NZBIT8(a >> 0, 0) | NZBIT8(a >> 8, 1) |                 \
 175         NZBIT8(a >> 16, 2) | NZBIT8(a >> 24, 3) |               \
 176         NZBIT8(a >> 32, 4) | NZBIT8(a >> 40, 5) |               \
 177         NZBIT8(a >> 48, 6) | NZBIT8(a >> 56, 7);                \
 178     return a;                                                   \
 179 }                                                               \
 180 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, w)))(CPUARMState *env, \
 181                                         uint64_t a, uint64_t b) \
 182 {                                                               \
 183     a =                                                         \
 184         (((a >> SH0) & 0xffff) << 0) |                          \
 185         (((b >> SH0) & 0xffff) << 16) |                         \
 186         (((a >> SH2) & 0xffff) << 32) |                         \
 187         (((b >> SH2) & 0xffff) << 48);                          \
 188     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 189         NZBIT8(a >> 0, 0) | NZBIT8(a >> 16, 1) |                \
 190         NZBIT8(a >> 32, 2) | NZBIT8(a >> 48, 3);                \
 191     return a;                                                   \
 192 }                                                               \
 193 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, l)))(CPUARMState *env, \
 194                                         uint64_t a, uint64_t b) \
 195 {                                                               \
 196     a =                                                         \
 197         (((a >> SH0) & 0xffffffff) << 0) |                      \
 198         (((b >> SH0) & 0xffffffff) << 32);                      \
 199     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 200         NZBIT32(a >> 0, 0) | NZBIT32(a >> 32, 1);               \
 201     return a;                                                   \
 202 }                                                               \
 203 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, ub)))(CPUARMState *env, \
 204                                                   uint64_t x)   \
 205 {                                                               \
 206     x =                                                         \
 207         (((x >> SH0) & 0xff) << 0) |                            \
 208         (((x >> SH1) & 0xff) << 16) |                           \
 209         (((x >> SH2) & 0xff) << 32) |                           \
 210         (((x >> SH3) & 0xff) << 48);                            \
 211     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 212         NZBIT16(x >> 0, 0) | NZBIT16(x >> 16, 1) |              \
 213         NZBIT16(x >> 32, 2) | NZBIT16(x >> 48, 3);              \
 214     return x;                                                   \
 215 }                                                               \
 216 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, uw)))(CPUARMState *env, \
 217                                                   uint64_t x)   \
 218 {                                                               \
 219     x =                                                         \
 220         (((x >> SH0) & 0xffff) << 0) |                          \
 221         (((x >> SH2) & 0xffff) << 32);                          \
 222     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 223         NZBIT32(x >> 0, 0) | NZBIT32(x >> 32, 1);               \
 224     return x;                                                   \
 225 }                                                               \
 226 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, ul)))(CPUARMState *env, \
 227                                                   uint64_t x)   \
 228 {                                                               \
 229     x = (((x >> SH0) & 0xffffffff) << 0);                       \
 230     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] = NZBIT64(x >> 0);      \
 231     return x;                                                   \
 232 }                                                               \
 233 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, sb)))(CPUARMState *env, \
 234                                                   uint64_t x)   \
 235 {                                                               \
 236     x =                                                         \
 237         ((uint64_t) EXTEND8H((x >> SH0) & 0xff) << 0) |         \
 238         ((uint64_t) EXTEND8H((x >> SH1) & 0xff) << 16) |        \
 239         ((uint64_t) EXTEND8H((x >> SH2) & 0xff) << 32) |        \
 240         ((uint64_t) EXTEND8H((x >> SH3) & 0xff) << 48);         \
 241     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 242         NZBIT16(x >> 0, 0) | NZBIT16(x >> 16, 1) |              \
 243         NZBIT16(x >> 32, 2) | NZBIT16(x >> 48, 3);              \
 244     return x;                                                   \
 245 }                                                               \
 246 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, sw)))(CPUARMState *env, \
 247                                                   uint64_t x)   \
 248 {                                                               \
 249     x =                                                         \
 250         ((uint64_t) EXTEND16((x >> SH0) & 0xffff) << 0) |       \
 251         ((uint64_t) EXTEND16((x >> SH2) & 0xffff) << 32);       \
 252     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 253         NZBIT32(x >> 0, 0) | NZBIT32(x >> 32, 1);               \
 254     return x;                                                   \
 255 }                                                               \
 256 uint64_t HELPER(glue(iwmmxt_unpack, glue(S, sl)))(CPUARMState *env, \
 257                                                   uint64_t x)   \
 258 {                                                               \
 259     x = EXTEND32((x >> SH0) & 0xffffffff);                      \
 260     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] = NZBIT64(x >> 0);      \
 261     return x;                                                   \
 262 }
 263 IWMMXT_OP_UNPACK(l, 0, 8, 16, 24)
 264 IWMMXT_OP_UNPACK(h, 32, 40, 48, 56)
 265
 266 #define IWMMXT_OP_CMP(SUFF, Tb, Tw, Tl, O)                      \
 267 uint64_t HELPER(glue(iwmmxt_, glue(SUFF, b)))(CPUARMState *env,    \
 268                                         uint64_t a, uint64_t b) \
 269 {                                                               \
 270     a =                                                         \
 271         CMP(0, Tb, O, 0xff) | CMP(8, Tb, O, 0xff) |             \
 272         CMP(16, Tb, O, 0xff) | CMP(24, Tb, O, 0xff) |           \
 273         CMP(32, Tb, O, 0xff) | CMP(40, Tb, O, 0xff) |           \
 274         CMP(48, Tb, O, 0xff) | CMP(56, Tb, O, 0xff);            \
 275     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 276         NZBIT8(a >> 0, 0) | NZBIT8(a >> 8, 1) |                 \
 277         NZBIT8(a >> 16, 2) | NZBIT8(a >> 24, 3) |               \
 278         NZBIT8(a >> 32, 4) | NZBIT8(a >> 40, 5) |               \
 279         NZBIT8(a >> 48, 6) | NZBIT8(a >> 56, 7);                \
 280     return a;                                                   \
 281 }                                                               \
 282 uint64_t HELPER(glue(iwmmxt_, glue(SUFF, w)))(CPUARMState *env,    \
 283                                         uint64_t a, uint64_t b) \
 284 {                                                               \
 285     a = CMP(0, Tw, O, 0xffff) | CMP(16, Tw, O, 0xffff) |        \
 286         CMP(32, Tw, O, 0xffff) | CMP(48, Tw, O, 0xffff);        \
 287     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 288         NZBIT16(a >> 0, 0) | NZBIT16(a >> 16, 1) |              \
 289         NZBIT16(a >> 32, 2) | NZBIT16(a >> 48, 3);              \
 290     return a;                                                   \
 291 }                                                               \
 292 uint64_t HELPER(glue(iwmmxt_, glue(SUFF, l)))(CPUARMState *env,    \
 293                                         uint64_t a, uint64_t b) \
 294 {                                                               \
 295     a = CMP(0, Tl, O, 0xffffffff) |                             \
 296         CMP(32, Tl, O, 0xffffffff);                             \
 297     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                       \
 298         NZBIT32(a >> 0, 0) | NZBIT32(a >> 32, 1);               \
 299     return a;                                                   \
 300 }
 301 #define CMP(SHR, TYPE, OPER, MASK) ((((TYPE) ((a >> SHR) & MASK) OPER \
 302             (TYPE) ((b >> SHR) & MASK)) ? (uint64_t) MASK : 0) << SHR)
 303 IWMMXT_OP_CMP(cmpeq, uint8_t, uint16_t, uint32_t, ==)
 304 IWMMXT_OP_CMP(cmpgts, int8_t, int16_t, int32_t, >)
 305 IWMMXT_OP_CMP(cmpgtu, uint8_t, uint16_t, uint32_t, >)
 306 #undef CMP
 307 #define CMP(SHR, TYPE, OPER, MASK) ((((TYPE) ((a >> SHR) & MASK) OPER \
 308             (TYPE) ((b >> SHR) & MASK)) ? a : b) & ((uint64_t) MASK << SHR))
 309 IWMMXT_OP_CMP(mins, int8_t, int16_t, int32_t, <)
 310 IWMMXT_OP_CMP(minu, uint8_t, uint16_t, uint32_t, <)
 311 IWMMXT_OP_CMP(maxs, int8_t, int16_t, int32_t, >)
 312 IWMMXT_OP_CMP(maxu, uint8_t, uint16_t, uint32_t, >)
 313 #undef CMP
 314 #define CMP(SHR, TYPE, OPER, MASK) ((uint64_t) (((TYPE) ((a >> SHR) & MASK) \
 315             OPER (TYPE) ((b >> SHR) & MASK)) & MASK) << SHR)
 316 IWMMXT_OP_CMP(subn, uint8_t, uint16_t, uint32_t, -)
 317 IWMMXT_OP_CMP(addn, uint8_t, uint16_t, uint32_t, +)
 318 #undef CMP
 319 /* TODO Signed- and Unsigned-Saturation */
 320 #define CMP(SHR, TYPE, OPER, MASK) ((uint64_t) (((TYPE) ((a >> SHR) & MASK) \
 321             OPER (TYPE) ((b >> SHR) & MASK)) & MASK) << SHR)
 322 IWMMXT_OP_CMP(subu, uint8_t, uint16_t, uint32_t, -)
 323 IWMMXT_OP_CMP(addu, uint8_t, uint16_t, uint32_t, +)
 324 IWMMXT_OP_CMP(subs, int8_t, int16_t, int32_t, -)
 325 IWMMXT_OP_CMP(adds, int8_t, int16_t, int32_t, +)
 326 #undef CMP
 327 #undef IWMMXT_OP_CMP
 328
 329 #define AVGB(SHR) ((( \
 330         ((a >> SHR) & 0xff) + ((b >> SHR) & 0xff) + round) >> 1) << SHR)
 331 #define IWMMXT_OP_AVGB(r)                                                 \
 332 uint64_t HELPER(iwmmxt_avgb##r)(CPUARMState *env, uint64_t a, uint64_t b)    \
 333 {                                                                         \
 334     const int round = r;                                                  \
 335     a = AVGB(0) | AVGB(8) | AVGB(16) | AVGB(24) |                         \
 336         AVGB(32) | AVGB(40) | AVGB(48) | AVGB(56);                        \
 337     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                                 \
 338         SIMD8_SET(ZBIT8((a >> 0) & 0xff), SIMD_ZBIT, 0) |                 \
 339         SIMD8_SET(ZBIT8((a >> 8) & 0xff), SIMD_ZBIT, 1) |                 \
 340         SIMD8_SET(ZBIT8((a >> 16) & 0xff), SIMD_ZBIT, 2) |                \
 341         SIMD8_SET(ZBIT8((a >> 24) & 0xff), SIMD_ZBIT, 3) |                \
 342         SIMD8_SET(ZBIT8((a >> 32) & 0xff), SIMD_ZBIT, 4) |                \
 343         SIMD8_SET(ZBIT8((a >> 40) & 0xff), SIMD_ZBIT, 5) |                \
 344         SIMD8_SET(ZBIT8((a >> 48) & 0xff), SIMD_ZBIT, 6) |                \
 345         SIMD8_SET(ZBIT8((a >> 56) & 0xff), SIMD_ZBIT, 7);                 \
 346     return a;                                                             \
 347 }
 348 IWMMXT_OP_AVGB(0)
 349 IWMMXT_OP_AVGB(1)
 350 #undef IWMMXT_OP_AVGB
 351 #undef AVGB
 352
 353 #define AVGW(SHR) ((( \
 354         ((a >> SHR) & 0xffff) + ((b >> SHR) & 0xffff) + round) >> 1) << SHR)
 355 #define IWMMXT_OP_AVGW(r)                                               \
 356 uint64_t HELPER(iwmmxt_avgw##r)(CPUARMState *env, uint64_t a, uint64_t b)  \
 357 {                                                                       \
 358     const int round = r;                                                \
 359     a = AVGW(0) | AVGW(16) | AVGW(32) | AVGW(48);                       \
 360     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =                               \
 361         SIMD16_SET(ZBIT16((a >> 0) & 0xffff), SIMD_ZBIT, 0) |           \
 362         SIMD16_SET(ZBIT16((a >> 16) & 0xffff), SIMD_ZBIT, 1) |          \
 363         SIMD16_SET(ZBIT16((a >> 32) & 0xffff), SIMD_ZBIT, 2) |          \
 364         SIMD16_SET(ZBIT16((a >> 48) & 0xffff), SIMD_ZBIT, 3);           \
 365     return a;                                                           \
 366 }
 367 IWMMXT_OP_AVGW(0)
 368 IWMMXT_OP_AVGW(1)
 369 #undef IWMMXT_OP_AVGW
 370 #undef AVGW
 371
 372 uint64_t HELPER(iwmmxt_align)(uint64_t a, uint64_t b, uint32_t n)
 373 {
 374     a >>= n << 3;
 375     a |= b << (64 - (n << 3));
 376     return a;
 377 }
 378
 379 uint64_t HELPER(iwmmxt_insr)(uint64_t x, uint32_t a, uint32_t b, uint32_t n)
 380 {
 381     x &= ~((uint64_t) b << n);
 382     x |= (uint64_t) (a & b) << n;
 383     return x;
 384 }
 385
 386 uint32_t HELPER(iwmmxt_setpsr_nz)(uint64_t x)
 387 {
 388     return SIMD64_SET((x == 0), SIMD_ZBIT) |
 389            SIMD64_SET((x & (1ULL << 63)), SIMD_NBIT);
 390 }
 391
 392 uint64_t HELPER(iwmmxt_bcstb)(uint32_t arg)
 393 {
 394     arg &= 0xff;
 395     return
 396         ((uint64_t) arg << 0 ) | ((uint64_t) arg << 8 ) |
 397         ((uint64_t) arg << 16) | ((uint64_t) arg << 24) |
 398         ((uint64_t) arg << 32) | ((uint64_t) arg << 40) |
 399         ((uint64_t) arg << 48) | ((uint64_t) arg << 56);
 400 }
 401
 402 uint64_t HELPER(iwmmxt_bcstw)(uint32_t arg)
 403 {
 404     arg &= 0xffff;
 405     return
 406         ((uint64_t) arg << 0 ) | ((uint64_t) arg << 16) |
 407         ((uint64_t) arg << 32) | ((uint64_t) arg << 48);
 408 }
 409
 410 uint64_t HELPER(iwmmxt_bcstl)(uint32_t arg)
 411 {
 412     return arg | ((uint64_t) arg << 32);
 413 }
 414
 415 uint64_t HELPER(iwmmxt_addcb)(uint64_t x)
 416 {
 417     return
 418         ((x >> 0) & 0xff) + ((x >> 8) & 0xff) +
 419         ((x >> 16) & 0xff) + ((x >> 24) & 0xff) +
 420         ((x >> 32) & 0xff) + ((x >> 40) & 0xff) +
 421         ((x >> 48) & 0xff) + ((x >> 56) & 0xff);
 422 }
 423
 424 uint64_t HELPER(iwmmxt_addcw)(uint64_t x)
 425 {
 426     return
 427         ((x >> 0) & 0xffff) + ((x >> 16) & 0xffff) +
 428         ((x >> 32) & 0xffff) + ((x >> 48) & 0xffff);
 429 }
 430
 431 uint64_t HELPER(iwmmxt_addcl)(uint64_t x)
 432 {
 433     return (x & 0xffffffff) + (x >> 32);
 434 }
 435
 436 uint32_t HELPER(iwmmxt_msbb)(uint64_t x)
 437 {
 438     return
 439         ((x >> 7) & 0x01) | ((x >> 14) & 0x02) |
 440         ((x >> 21) & 0x04) | ((x >> 28) & 0x08) |
 441         ((x >> 35) & 0x10) | ((x >> 42) & 0x20) |
 442         ((x >> 49) & 0x40) | ((x >> 56) & 0x80);
 443 }
 444
 445 uint32_t HELPER(iwmmxt_msbw)(uint64_t x)
 446 {
 447     return
 448         ((x >> 15) & 0x01) | ((x >> 30) & 0x02) |
 449         ((x >> 45) & 0x04) | ((x >> 52) & 0x08);
 450 }
 451
 452 uint32_t HELPER(iwmmxt_msbl)(uint64_t x)
 453 {
 454     return ((x >> 31) & 0x01) | ((x >> 62) & 0x02);
 455 }
 456
 457 /* FIXME: Split wCASF setting into a separate op to avoid env use.  */
 458 uint64_t HELPER(iwmmxt_srlw)(CPUARMState *env, uint64_t x, uint32_t n)
 459 {
 460     x = (((x & (0xffffll << 0)) >> n) & (0xffffll << 0)) |
 461         (((x & (0xffffll << 16)) >> n) & (0xffffll << 16)) |
 462         (((x & (0xffffll << 32)) >> n) & (0xffffll << 32)) |
 463         (((x & (0xffffll << 48)) >> n) & (0xffffll << 48));
 464     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 465         NZBIT16(x >> 0, 0) | NZBIT16(x >> 16, 1) |
 466         NZBIT16(x >> 32, 2) | NZBIT16(x >> 48, 3);
 467     return x;
 468 }
 469
 470 uint64_t HELPER(iwmmxt_srll)(CPUARMState *env, uint64_t x, uint32_t n)
 471 {
 472     x = ((x & (0xffffffffll << 0)) >> n) |
 473         ((x >> n) & (0xffffffffll << 32));
 474     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 475         NZBIT32(x >> 0, 0) | NZBIT32(x >> 32, 1);
 476     return x;
 477 }
 478
 479 uint64_t HELPER(iwmmxt_srlq)(CPUARMState *env, uint64_t x, uint32_t n)
 480 {
 481     x >>= n;
 482     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] = NZBIT64(x);
 483     return x;
 484 }
 485
 486 uint64_t HELPER(iwmmxt_sllw)(CPUARMState *env, uint64_t x, uint32_t n)
 487 {
 488     x = (((x & (0xffffll << 0)) << n) & (0xffffll << 0)) |
 489         (((x & (0xffffll << 16)) << n) & (0xffffll << 16)) |
 490         (((x & (0xffffll << 32)) << n) & (0xffffll << 32)) |
 491         (((x & (0xffffll << 48)) << n) & (0xffffll << 48));
 492     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 493         NZBIT16(x >> 0, 0) | NZBIT16(x >> 16, 1) |
 494         NZBIT16(x >> 32, 2) | NZBIT16(x >> 48, 3);
 495     return x;
 496 }
 497
 498 uint64_t HELPER(iwmmxt_slll)(CPUARMState *env, uint64_t x, uint32_t n)
 499 {
 500     x = ((x << n) & (0xffffffffll << 0)) |
 501         ((x & (0xffffffffll << 32)) << n);
 502     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 503         NZBIT32(x >> 0, 0) | NZBIT32(x >> 32, 1);
 504     return x;
 505 }
 506
 507 uint64_t HELPER(iwmmxt_sllq)(CPUARMState *env, uint64_t x, uint32_t n)
 508 {
 509     x <<= n;
 510     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] = NZBIT64(x);
 511     return x;
 512 }
 513
 514 uint64_t HELPER(iwmmxt_sraw)(CPUARMState *env, uint64_t x, uint32_t n)
 515 {
 516     x = ((uint64_t) ((EXTEND16(x >> 0) >> n) & 0xffff) << 0) |
 517         ((uint64_t) ((EXTEND16(x >> 16) >> n) & 0xffff) << 16) |
 518         ((uint64_t) ((EXTEND16(x >> 32) >> n) & 0xffff) << 32) |
 519         ((uint64_t) ((EXTEND16(x >> 48) >> n) & 0xffff) << 48);
 520     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 521         NZBIT16(x >> 0, 0) | NZBIT16(x >> 16, 1) |
 522         NZBIT16(x >> 32, 2) | NZBIT16(x >> 48, 3);
 523     return x;
 524 }
 525
 526 uint64_t HELPER(iwmmxt_sral)(CPUARMState *env, uint64_t x, uint32_t n)
 527 {
 528     x = (((EXTEND32(x >> 0) >> n) & 0xffffffff) << 0) |
 529         (((EXTEND32(x >> 32) >> n) & 0xffffffff) << 32);
 530     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 531         NZBIT32(x >> 0, 0) | NZBIT32(x >> 32, 1);
 532     return x;
 533 }
 534
 535 uint64_t HELPER(iwmmxt_sraq)(CPUARMState *env, uint64_t x, uint32_t n)
 536 {
 537     x = (int64_t) x >> n;
 538     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] = NZBIT64(x);
 539     return x;
 540 }
 541
 542 uint64_t HELPER(iwmmxt_rorw)(CPUARMState *env, uint64_t x, uint32_t n)
 543 {
 544     x = ((((x & (0xffffll << 0)) >> n) |
 545           ((x & (0xffffll << 0)) << (16 - n))) & (0xffffll << 0)) |
 546         ((((x & (0xffffll << 16)) >> n) |
 547           ((x & (0xffffll << 16)) << (16 - n))) & (0xffffll << 16)) |
 548         ((((x & (0xffffll << 32)) >> n) |
 549           ((x & (0xffffll << 32)) << (16 - n))) & (0xffffll << 32)) |
 550         ((((x & (0xffffll << 48)) >> n) |
 551           ((x & (0xffffll << 48)) << (16 - n))) & (0xffffll << 48));
 552     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 553         NZBIT16(x >> 0, 0) | NZBIT16(x >> 16, 1) |
 554         NZBIT16(x >> 32, 2) | NZBIT16(x >> 48, 3);
 555     return x;
 556 }
 557
 558 uint64_t HELPER(iwmmxt_rorl)(CPUARMState *env, uint64_t x, uint32_t n)
 559 {
 560     x = ((x & (0xffffffffll << 0)) >> n) |
 561         ((x >> n) & (0xffffffffll << 32)) |
 562         ((x << (32 - n)) & (0xffffffffll << 0)) |
 563         ((x & (0xffffffffll << 32)) << (32 - n));
 564     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 565         NZBIT32(x >> 0, 0) | NZBIT32(x >> 32, 1);
 566     return x;
 567 }
 568
 569 uint64_t HELPER(iwmmxt_rorq)(CPUARMState *env, uint64_t x, uint32_t n)
 570 {
 571     x = ror64(x, n);
 572     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] = NZBIT64(x);
 573     return x;
 574 }
 575
 576 uint64_t HELPER(iwmmxt_shufh)(CPUARMState *env, uint64_t x, uint32_t n)
 577 {
 578     x = (((x >> ((n << 4) & 0x30)) & 0xffff) << 0) |
 579         (((x >> ((n << 2) & 0x30)) & 0xffff) << 16) |
 580         (((x >> ((n << 0) & 0x30)) & 0xffff) << 32) |
 581         (((x >> ((n >> 2) & 0x30)) & 0xffff) << 48);
 582     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 583         NZBIT16(x >> 0, 0) | NZBIT16(x >> 16, 1) |
 584         NZBIT16(x >> 32, 2) | NZBIT16(x >> 48, 3);
 585     return x;
 586 }
 587
 588 /* TODO: Unsigned-Saturation */
 589 uint64_t HELPER(iwmmxt_packuw)(CPUARMState *env, uint64_t a, uint64_t b)
 590 {
 591     a = (((a >> 0) & 0xff) << 0) | (((a >> 16) & 0xff) << 8) |
 592         (((a >> 32) & 0xff) << 16) | (((a >> 48) & 0xff) << 24) |
 593         (((b >> 0) & 0xff) << 32) | (((b >> 16) & 0xff) << 40) |
 594         (((b >> 32) & 0xff) << 48) | (((b >> 48) & 0xff) << 56);
 595     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 596         NZBIT8(a >> 0, 0) | NZBIT8(a >> 8, 1) |
 597         NZBIT8(a >> 16, 2) | NZBIT8(a >> 24, 3) |
 598         NZBIT8(a >> 32, 4) | NZBIT8(a >> 40, 5) |
 599         NZBIT8(a >> 48, 6) | NZBIT8(a >> 56, 7);
 600     return a;
 601 }
 602
 603 uint64_t HELPER(iwmmxt_packul)(CPUARMState *env, uint64_t a, uint64_t b)
 604 {
 605     a = (((a >> 0) & 0xffff) << 0) | (((a >> 32) & 0xffff) << 16) |
 606         (((b >> 0) & 0xffff) << 32) | (((b >> 32) & 0xffff) << 48);
 607     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 608         NZBIT16(a >> 0, 0) | NZBIT16(a >> 16, 1) |
 609         NZBIT16(a >> 32, 2) | NZBIT16(a >> 48, 3);
 610     return a;
 611 }
 612
 613 uint64_t HELPER(iwmmxt_packuq)(CPUARMState *env, uint64_t a, uint64_t b)
 614 {
 615     a = (a & 0xffffffff) | ((b & 0xffffffff) << 32);
 616     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 617         NZBIT32(a >> 0, 0) | NZBIT32(a >> 32, 1);
 618     return a;
 619 }
 620
 621 /* TODO: Signed-Saturation */
 622 uint64_t HELPER(iwmmxt_packsw)(CPUARMState *env, uint64_t a, uint64_t b)
 623 {
 624     a = (((a >> 0) & 0xff) << 0) | (((a >> 16) & 0xff) << 8) |
 625         (((a >> 32) & 0xff) << 16) | (((a >> 48) & 0xff) << 24) |
 626         (((b >> 0) & 0xff) << 32) | (((b >> 16) & 0xff) << 40) |
 627         (((b >> 32) & 0xff) << 48) | (((b >> 48) & 0xff) << 56);
 628     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 629         NZBIT8(a >> 0, 0) | NZBIT8(a >> 8, 1) |
 630         NZBIT8(a >> 16, 2) | NZBIT8(a >> 24, 3) |
 631         NZBIT8(a >> 32, 4) | NZBIT8(a >> 40, 5) |
 632         NZBIT8(a >> 48, 6) | NZBIT8(a >> 56, 7);
 633     return a;
 634 }
 635
 636 uint64_t HELPER(iwmmxt_packsl)(CPUARMState *env, uint64_t a, uint64_t b)
 637 {
 638     a = (((a >> 0) & 0xffff) << 0) | (((a >> 32) & 0xffff) << 16) |
 639         (((b >> 0) & 0xffff) << 32) | (((b >> 32) & 0xffff) << 48);
 640     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 641         NZBIT16(a >> 0, 0) | NZBIT16(a >> 16, 1) |
 642         NZBIT16(a >> 32, 2) | NZBIT16(a >> 48, 3);
 643     return a;
 644 }
 645
 646 uint64_t HELPER(iwmmxt_packsq)(CPUARMState *env, uint64_t a, uint64_t b)
 647 {
 648     a = (a & 0xffffffff) | ((b & 0xffffffff) << 32);
 649     env->iwmmxt.cregs[ARM_IWMMXT_wCASF] =
 650         NZBIT32(a >> 0, 0) | NZBIT32(a >> 32, 1);
 651     return a;
 652 }
 653
 654 uint64_t HELPER(iwmmxt_muladdsl)(uint64_t c, uint32_t a, uint32_t b)
 655 {
 656     return c + ((int32_t) EXTEND32(a) * (int32_t) EXTEND32(b));
 657 }
 658
 659 uint64_t HELPER(iwmmxt_muladdsw)(uint64_t c, uint32_t a, uint32_t b)
 660 {
 661     c += EXTEND32(EXTEND16S((a >> 0) & 0xffff) *
 662                   EXTEND16S((b >> 0) & 0xffff));
 663     c += EXTEND32(EXTEND16S((a >> 16) & 0xffff) *
 664                   EXTEND16S((b >> 16) & 0xffff));
 665     return c;
 666 }
 667
 668 uint64_t HELPER(iwmmxt_muladdswl)(uint64_t c, uint32_t a, uint32_t b)
 669 {
 670     return c + (EXTEND32(EXTEND16S(a & 0xffff) *
 671                          EXTEND16S(b & 0xffff)));
 672 }