target-arm/neon_helper.c

   1 /*
   2  * ARM NEON vector operations.
   3  *
   4  * Copyright (c) 2007, 2008 CodeSourcery.
   5  * Written by Paul Brook
   6  *
   7  * This code is licensed under the GNU GPL v2.
   8  */
   9 #include <stdlib.h>
  10 #include <stdio.h>
  11
  12 #include "cpu.h"
  13 #include "exec-all.h"
  14 #include "helper.h"
  15
  16 #define SIGNBIT (uint32_t)0x80000000
  17 #define SIGNBIT64 ((uint64_t)1 << 63)
  18
  19 #define SET_QC() env->vfp.xregs[ARM_VFP_FPSCR] |= CPSR_Q
  20
  21 #define NEON_TYPE1(name, type) \
  22 typedef struct \
  23 { \
  24     type v1; \
  25 } neon_##name;
  26 #ifdef HOST_WORDS_BIGENDIAN
  27 #define NEON_TYPE2(name, type) \
  28 typedef struct \
  29 { \
  30     type v2; \
  31     type v1; \
  32 } neon_##name;
  33 #define NEON_TYPE4(name, type) \
  34 typedef struct \
  35 { \
  36     type v4; \
  37     type v3; \
  38     type v2; \
  39     type v1; \
  40 } neon_##name;
  41 #else
  42 #define NEON_TYPE2(name, type) \
  43 typedef struct \
  44 { \
  45     type v1; \
  46     type v2; \
  47 } neon_##name;
  48 #define NEON_TYPE4(name, type) \
  49 typedef struct \
  50 { \
  51     type v1; \
  52     type v2; \
  53     type v3; \
  54     type v4; \
  55 } neon_##name;
  56 #endif
  57
  58 NEON_TYPE4(s8, int8_t)
  59 NEON_TYPE4(u8, uint8_t)
  60 NEON_TYPE2(s16, int16_t)
  61 NEON_TYPE2(u16, uint16_t)
  62 NEON_TYPE1(s32, int32_t)
  63 NEON_TYPE1(u32, uint32_t)
  64 #undef NEON_TYPE4
  65 #undef NEON_TYPE2
  66 #undef NEON_TYPE1
  67
  68 /* Copy from a uint32_t to a vector structure type.  */
  69 #define NEON_UNPACK(vtype, dest, val) do { \
  70     union { \
  71         vtype v; \
  72         uint32_t i; \
  73     } conv_u; \
  74     conv_u.i = (val); \
  75     dest = conv_u.v; \
  76     } while(0)
  77
  78 /* Copy from a vector structure type to a uint32_t.  */
  79 #define NEON_PACK(vtype, dest, val) do { \
  80     union { \
  81         vtype v; \
  82         uint32_t i; \
  83     } conv_u; \
  84     conv_u.v = (val); \
  85     dest = conv_u.i; \
  86     } while(0)
  87
  88 #define NEON_DO1 \
  89     NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1);
  90 #define NEON_DO2 \
  91     NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); \
  92     NEON_FN(vdest.v2, vsrc1.v2, vsrc2.v2);
  93 #define NEON_DO4 \
  94     NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); \
  95     NEON_FN(vdest.v2, vsrc1.v2, vsrc2.v2); \
  96     NEON_FN(vdest.v3, vsrc1.v3, vsrc2.v3); \
  97     NEON_FN(vdest.v4, vsrc1.v4, vsrc2.v4);
  98
  99 #define NEON_VOP_BODY(vtype, n) \
 100 { \
 101     uint32_t res; \
 102     vtype vsrc1; \
 103     vtype vsrc2; \
 104     vtype vdest; \
 105     NEON_UNPACK(vtype, vsrc1, arg1); \
 106     NEON_UNPACK(vtype, vsrc2, arg2); \
 107     NEON_DO##n; \
 108     NEON_PACK(vtype, res, vdest); \
 109     return res; \
 110 }
 111
 112 #define NEON_VOP(name, vtype, n) \
 113 uint32_t HELPER(glue(neon_,name))(uint32_t arg1, uint32_t arg2) \
 114 NEON_VOP_BODY(vtype, n)
 115
 116 #define NEON_VOP_ENV(name, vtype, n) \
 117 uint32_t HELPER(glue(neon_,name))(CPUARMState *env, uint32_t arg1, uint32_t arg2) \
 118 NEON_VOP_BODY(vtype, n)
 119
 120 /* Pairwise operations.  */
 121 /* For 32-bit elements each segment only contains a single element, so
 122    the elementwise and pairwise operations are the same.  */
 123 #define NEON_PDO2 \
 124     NEON_FN(vdest.v1, vsrc1.v1, vsrc1.v2); \
 125     NEON_FN(vdest.v2, vsrc2.v1, vsrc2.v2);
 126 #define NEON_PDO4 \
 127     NEON_FN(vdest.v1, vsrc1.v1, vsrc1.v2); \
 128     NEON_FN(vdest.v2, vsrc1.v3, vsrc1.v4); \
 129     NEON_FN(vdest.v3, vsrc2.v1, vsrc2.v2); \
 130     NEON_FN(vdest.v4, vsrc2.v3, vsrc2.v4); \
 131
 132 #define NEON_POP(name, vtype, n) \
 133 uint32_t HELPER(glue(neon_,name))(uint32_t arg1, uint32_t arg2) \
 134 { \
 135     uint32_t res; \
 136     vtype vsrc1; \
 137     vtype vsrc2; \
 138     vtype vdest; \
 139     NEON_UNPACK(vtype, vsrc1, arg1); \
 140     NEON_UNPACK(vtype, vsrc2, arg2); \
 141     NEON_PDO##n; \
 142     NEON_PACK(vtype, res, vdest); \
 143     return res; \
 144 }
 145
 146 /* Unary operators.  */
 147 #define NEON_VOP1(name, vtype, n) \
 148 uint32_t HELPER(glue(neon_,name))(uint32_t arg) \
 149 { \
 150     vtype vsrc1; \
 151     vtype vdest; \
 152     NEON_UNPACK(vtype, vsrc1, arg); \
 153     NEON_DO##n; \
 154     NEON_PACK(vtype, arg, vdest); \
 155     return arg; \
 156 }
 157
 158
 159 #define NEON_USAT(dest, src1, src2, type) do { \
 160     uint32_t tmp = (uint32_t)src1 + (uint32_t)src2; \
 161     if (tmp != (type)tmp) { \
 162         SET_QC(); \
 163         dest = ~0; \
 164     } else { \
 165         dest = tmp; \
 166     }} while(0)
 167 #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint8_t)
 168 NEON_VOP_ENV(qadd_u8, neon_u8, 4)
 169 #undef NEON_FN
 170 #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint16_t)
 171 NEON_VOP_ENV(qadd_u16, neon_u16, 2)
 172 #undef NEON_FN
 173 #undef NEON_USAT
 174
 175 uint32_t HELPER(neon_qadd_u32)(CPUARMState *env, uint32_t a, uint32_t b)
 176 {
 177     uint32_t res = a + b;
 178     if (res < a) {
 179         SET_QC();
 180         res = ~0;
 181     }
 182     return res;
 183 }
 184
 185 uint64_t HELPER(neon_qadd_u64)(CPUARMState *env, uint64_t src1, uint64_t src2)
 186 {
 187     uint64_t res;
 188
 189     res = src1 + src2;
 190     if (res < src1) {
 191         SET_QC();
 192         res = ~(uint64_t)0;
 193     }
 194     return res;
 195 }
 196
 197 #define NEON_SSAT(dest, src1, src2, type) do { \
 198     int32_t tmp = (uint32_t)src1 + (uint32_t)src2; \
 199     if (tmp != (type)tmp) { \
 200         SET_QC(); \
 201         if (src2 > 0) { \
 202             tmp = (1 << (sizeof(type) * 8 - 1)) - 1; \
 203         } else { \
 204             tmp = 1 << (sizeof(type) * 8 - 1); \
 205         } \
 206     } \
 207     dest = tmp; \
 208     } while(0)
 209 #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int8_t)
 210 NEON_VOP_ENV(qadd_s8, neon_s8, 4)
 211 #undef NEON_FN
 212 #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int16_t)
 213 NEON_VOP_ENV(qadd_s16, neon_s16, 2)
 214 #undef NEON_FN
 215 #undef NEON_SSAT
 216
 217 uint32_t HELPER(neon_qadd_s32)(CPUARMState *env, uint32_t a, uint32_t b)
 218 {
 219     uint32_t res = a + b;
 220     if (((res ^ a) & SIGNBIT) && !((a ^ b) & SIGNBIT)) {
 221         SET_QC();
 222         res = ~(((int32_t)a >> 31) ^ SIGNBIT);
 223     }
 224     return res;
 225 }
 226
 227 uint64_t HELPER(neon_qadd_s64)(CPUARMState *env, uint64_t src1, uint64_t src2)
 228 {
 229     uint64_t res;
 230
 231     res = src1 + src2;
 232     if (((res ^ src1) & SIGNBIT64) && !((src1 ^ src2) & SIGNBIT64)) {
 233         SET_QC();
 234         res = ((int64_t)src1 >> 63) ^ ~SIGNBIT64;
 235     }
 236     return res;
 237 }
 238
 239 #define NEON_USAT(dest, src1, src2, type) do { \
 240     uint32_t tmp = (uint32_t)src1 - (uint32_t)src2; \
 241     if (tmp != (type)tmp) { \
 242         SET_QC(); \
 243         dest = 0; \
 244     } else { \
 245         dest = tmp; \
 246     }} while(0)
 247 #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint8_t)
 248 NEON_VOP_ENV(qsub_u8, neon_u8, 4)
 249 #undef NEON_FN
 250 #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint16_t)
 251 NEON_VOP_ENV(qsub_u16, neon_u16, 2)
 252 #undef NEON_FN
 253 #undef NEON_USAT
 254
 255 uint32_t HELPER(neon_qsub_u32)(CPUARMState *env, uint32_t a, uint32_t b)
 256 {
 257     uint32_t res = a - b;
 258     if (res > a) {
 259         SET_QC();
 260         res = 0;
 261     }
 262     return res;
 263 }
 264
 265 uint64_t HELPER(neon_qsub_u64)(CPUARMState *env, uint64_t src1, uint64_t src2)
 266 {
 267     uint64_t res;
 268
 269     if (src1 < src2) {
 270         SET_QC();
 271         res = 0;
 272     } else {
 273         res = src1 - src2;
 274     }
 275     return res;
 276 }
 277
 278 #define NEON_SSAT(dest, src1, src2, type) do { \
 279     int32_t tmp = (uint32_t)src1 - (uint32_t)src2; \
 280     if (tmp != (type)tmp) { \
 281         SET_QC(); \
 282         if (src2 < 0) { \
 283             tmp = (1 << (sizeof(type) * 8 - 1)) - 1; \
 284         } else { \
 285             tmp = 1 << (sizeof(type) * 8 - 1); \
 286         } \
 287     } \
 288     dest = tmp; \
 289     } while(0)
 290 #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int8_t)
 291 NEON_VOP_ENV(qsub_s8, neon_s8, 4)
 292 #undef NEON_FN
 293 #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int16_t)
 294 NEON_VOP_ENV(qsub_s16, neon_s16, 2)
 295 #undef NEON_FN
 296 #undef NEON_SSAT
 297
 298 uint32_t HELPER(neon_qsub_s32)(CPUARMState *env, uint32_t a, uint32_t b)
 299 {
 300     uint32_t res = a - b;
 301     if (((res ^ a) & SIGNBIT) && ((a ^ b) & SIGNBIT)) {
 302         SET_QC();
 303         res = ~(((int32_t)a >> 31) ^ SIGNBIT);
 304     }
 305     return res;
 306 }
 307
 308 uint64_t HELPER(neon_qsub_s64)(CPUARMState *env, uint64_t src1, uint64_t src2)
 309 {
 310     uint64_t res;
 311
 312     res = src1 - src2;
 313     if (((res ^ src1) & SIGNBIT64) && ((src1 ^ src2) & SIGNBIT64)) {
 314         SET_QC();
 315         res = ((int64_t)src1 >> 63) ^ ~SIGNBIT64;
 316     }
 317     return res;
 318 }
 319
 320 #define NEON_FN(dest, src1, src2) dest = (src1 + src2) >> 1
 321 NEON_VOP(hadd_s8, neon_s8, 4)
 322 NEON_VOP(hadd_u8, neon_u8, 4)
 323 NEON_VOP(hadd_s16, neon_s16, 2)
 324 NEON_VOP(hadd_u16, neon_u16, 2)
 325 #undef NEON_FN
 326
 327 int32_t HELPER(neon_hadd_s32)(int32_t src1, int32_t src2)
 328 {
 329     int32_t dest;
 330
 331     dest = (src1 >> 1) + (src2 >> 1);
 332     if (src1 & src2 & 1)
 333         dest++;
 334     return dest;
 335 }
 336
 337 uint32_t HELPER(neon_hadd_u32)(uint32_t src1, uint32_t src2)
 338 {
 339     uint32_t dest;
 340
 341     dest = (src1 >> 1) + (src2 >> 1);
 342     if (src1 & src2 & 1)
 343         dest++;
 344     return dest;
 345 }
 346
 347 #define NEON_FN(dest, src1, src2) dest = (src1 + src2 + 1) >> 1
 348 NEON_VOP(rhadd_s8, neon_s8, 4)
 349 NEON_VOP(rhadd_u8, neon_u8, 4)
 350 NEON_VOP(rhadd_s16, neon_s16, 2)
 351 NEON_VOP(rhadd_u16, neon_u16, 2)
 352 #undef NEON_FN
 353
 354 int32_t HELPER(neon_rhadd_s32)(int32_t src1, int32_t src2)
 355 {
 356     int32_t dest;
 357
 358     dest = (src1 >> 1) + (src2 >> 1);
 359     if ((src1 | src2) & 1)
 360         dest++;
 361     return dest;
 362 }
 363
 364 uint32_t HELPER(neon_rhadd_u32)(uint32_t src1, uint32_t src2)
 365 {
 366     uint32_t dest;
 367
 368     dest = (src1 >> 1) + (src2 >> 1);
 369     if ((src1 | src2) & 1)
 370         dest++;
 371     return dest;
 372 }
 373
 374 #define NEON_FN(dest, src1, src2) dest = (src1 - src2) >> 1
 375 NEON_VOP(hsub_s8, neon_s8, 4)
 376 NEON_VOP(hsub_u8, neon_u8, 4)
 377 NEON_VOP(hsub_s16, neon_s16, 2)
 378 NEON_VOP(hsub_u16, neon_u16, 2)
 379 #undef NEON_FN
 380
 381 int32_t HELPER(neon_hsub_s32)(int32_t src1, int32_t src2)
 382 {
 383     int32_t dest;
 384
 385     dest = (src1 >> 1) - (src2 >> 1);
 386     if ((~src1) & src2 & 1)
 387         dest--;
 388     return dest;
 389 }
 390
 391 uint32_t HELPER(neon_hsub_u32)(uint32_t src1, uint32_t src2)
 392 {
 393     uint32_t dest;
 394
 395     dest = (src1 >> 1) - (src2 >> 1);
 396     if ((~src1) & src2 & 1)
 397         dest--;
 398     return dest;
 399 }
 400
 401 #define NEON_FN(dest, src1, src2) dest = (src1 > src2) ? ~0 : 0
 402 NEON_VOP(cgt_s8, neon_s8, 4)
 403 NEON_VOP(cgt_u8, neon_u8, 4)
 404 NEON_VOP(cgt_s16, neon_s16, 2)
 405 NEON_VOP(cgt_u16, neon_u16, 2)
 406 NEON_VOP(cgt_s32, neon_s32, 1)
 407 NEON_VOP(cgt_u32, neon_u32, 1)
 408 #undef NEON_FN
 409
 410 #define NEON_FN(dest, src1, src2) dest = (src1 >= src2) ? ~0 : 0
 411 NEON_VOP(cge_s8, neon_s8, 4)
 412 NEON_VOP(cge_u8, neon_u8, 4)
 413 NEON_VOP(cge_s16, neon_s16, 2)
 414 NEON_VOP(cge_u16, neon_u16, 2)
 415 NEON_VOP(cge_s32, neon_s32, 1)
 416 NEON_VOP(cge_u32, neon_u32, 1)
 417 #undef NEON_FN
 418
 419 #define NEON_FN(dest, src1, src2) dest = (src1 < src2) ? src1 : src2
 420 NEON_VOP(min_s8, neon_s8, 4)
 421 NEON_VOP(min_u8, neon_u8, 4)
 422 NEON_VOP(min_s16, neon_s16, 2)
 423 NEON_VOP(min_u16, neon_u16, 2)
 424 NEON_VOP(min_s32, neon_s32, 1)
 425 NEON_VOP(min_u32, neon_u32, 1)
 426 NEON_POP(pmin_s8, neon_s8, 4)
 427 NEON_POP(pmin_u8, neon_u8, 4)
 428 NEON_POP(pmin_s16, neon_s16, 2)
 429 NEON_POP(pmin_u16, neon_u16, 2)
 430 #undef NEON_FN
 431
 432 #define NEON_FN(dest, src1, src2) dest = (src1 > src2) ? src1 : src2
 433 NEON_VOP(max_s8, neon_s8, 4)
 434 NEON_VOP(max_u8, neon_u8, 4)
 435 NEON_VOP(max_s16, neon_s16, 2)
 436 NEON_VOP(max_u16, neon_u16, 2)
 437 NEON_VOP(max_s32, neon_s32, 1)
 438 NEON_VOP(max_u32, neon_u32, 1)
 439 NEON_POP(pmax_s8, neon_s8, 4)
 440 NEON_POP(pmax_u8, neon_u8, 4)
 441 NEON_POP(pmax_s16, neon_s16, 2)
 442 NEON_POP(pmax_u16, neon_u16, 2)
 443 #undef NEON_FN
 444
 445 #define NEON_FN(dest, src1, src2) \
 446     dest = (src1 > src2) ? (src1 - src2) : (src2 - src1)
 447 NEON_VOP(abd_s8, neon_s8, 4)
 448 NEON_VOP(abd_u8, neon_u8, 4)
 449 NEON_VOP(abd_s16, neon_s16, 2)
 450 NEON_VOP(abd_u16, neon_u16, 2)
 451 NEON_VOP(abd_s32, neon_s32, 1)
 452 NEON_VOP(abd_u32, neon_u32, 1)
 453 #undef NEON_FN
 454
 455 #define NEON_FN(dest, src1, src2) do { \
 456     int8_t tmp; \
 457     tmp = (int8_t)src2; \
 458     if (tmp >= (ssize_t)sizeof(src1) * 8 || \
 459         tmp <= -(ssize_t)sizeof(src1) * 8) { \
 460         dest = 0; \
 461     } else if (tmp < 0) { \
 462         dest = src1 >> -tmp; \
 463     } else { \
 464         dest = src1 << tmp; \
 465     }} while (0)
 466 NEON_VOP(shl_u8, neon_u8, 4)
 467 NEON_VOP(shl_u16, neon_u16, 2)
 468 NEON_VOP(shl_u32, neon_u32, 1)
 469 #undef NEON_FN
 470
 471 uint64_t HELPER(neon_shl_u64)(uint64_t val, uint64_t shiftop)
 472 {
 473     int8_t shift = (int8_t)shiftop;
 474     if (shift >= 64 || shift <= -64) {
 475         val = 0;
 476     } else if (shift < 0) {
 477         val >>= -shift;
 478     } else {
 479         val <<= shift;
 480     }
 481     return val;
 482 }
 483
 484 #define NEON_FN(dest, src1, src2) do { \
 485     int8_t tmp; \
 486     tmp = (int8_t)src2; \
 487     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 488         dest = 0; \
 489     } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 490         dest = src1 >> (sizeof(src1) * 8 - 1); \
 491     } else if (tmp < 0) { \
 492         dest = src1 >> -tmp; \
 493     } else { \
 494         dest = src1 << tmp; \
 495     }} while (0)
 496 NEON_VOP(shl_s8, neon_s8, 4)
 497 NEON_VOP(shl_s16, neon_s16, 2)
 498 NEON_VOP(shl_s32, neon_s32, 1)
 499 #undef NEON_FN
 500
 501 uint64_t HELPER(neon_shl_s64)(uint64_t valop, uint64_t shiftop)
 502 {
 503     int8_t shift = (int8_t)shiftop;
 504     int64_t val = valop;
 505     if (shift >= 64) {
 506         val = 0;
 507     } else if (shift <= -64) {
 508         val >>= 63;
 509     } else if (shift < 0) {
 510         val >>= -shift;
 511     } else {
 512         val <<= shift;
 513     }
 514     return val;
 515 }
 516
 517 #define NEON_FN(dest, src1, src2) do { \
 518     int8_t tmp; \
 519     tmp = (int8_t)src2; \
 520     if ((tmp >= (ssize_t)sizeof(src1) * 8) \
 521         || (tmp <= -(ssize_t)sizeof(src1) * 8)) { \
 522         dest = 0; \
 523     } else if (tmp < 0) { \
 524         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 525     } else { \
 526         dest = src1 << tmp; \
 527     }} while (0)
 528 NEON_VOP(rshl_s8, neon_s8, 4)
 529 NEON_VOP(rshl_s16, neon_s16, 2)
 530 #undef NEON_FN
 531
 532 /* The addition of the rounding constant may overflow, so we use an
 533  * intermediate 64 bit accumulator.  */
 534 uint32_t HELPER(neon_rshl_s32)(uint32_t valop, uint32_t shiftop)
 535 {
 536     int32_t dest;
 537     int32_t val = (int32_t)valop;
 538     int8_t shift = (int8_t)shiftop;
 539     if ((shift >= 32) || (shift <= -32)) {
 540         dest = 0;
 541     } else if (shift < 0) {
 542         int64_t big_dest = ((int64_t)val + (1 << (-1 - shift)));
 543         dest = big_dest >> -shift;
 544     } else {
 545         dest = val << shift;
 546     }
 547     return dest;
 548 }
 549
 550 /* Handling addition overflow with 64 bit input values is more
 551  * tricky than with 32 bit values.  */
 552 uint64_t HELPER(neon_rshl_s64)(uint64_t valop, uint64_t shiftop)
 553 {
 554     int8_t shift = (int8_t)shiftop;
 555     int64_t val = valop;
 556     if ((shift >= 64) || (shift <= -64)) {
 557         val = 0;
 558     } else if (shift < 0) {
 559         val >>= (-shift - 1);
 560         if (val == INT64_MAX) {
 561             /* In this case, it means that the rounding constant is 1,
 562              * and the addition would overflow. Return the actual
 563              * result directly.  */
 564             val = 0x4000000000000000LL;
 565         } else {
 566             val++;
 567             val >>= 1;
 568         }
 569     } else {
 570         val <<= shift;
 571     }
 572     return val;
 573 }
 574
 575 #define NEON_FN(dest, src1, src2) do { \
 576     int8_t tmp; \
 577     tmp = (int8_t)src2; \
 578     if (tmp >= (ssize_t)sizeof(src1) * 8 || \
 579         tmp < -(ssize_t)sizeof(src1) * 8) { \
 580         dest = 0; \
 581     } else if (tmp == -(ssize_t)sizeof(src1) * 8) { \
 582         dest = src1 >> (-tmp - 1); \
 583     } else if (tmp < 0) { \
 584         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 585     } else { \
 586         dest = src1 << tmp; \
 587     }} while (0)
 588 NEON_VOP(rshl_u8, neon_u8, 4)
 589 NEON_VOP(rshl_u16, neon_u16, 2)
 590 #undef NEON_FN
 591
 592 /* The addition of the rounding constant may overflow, so we use an
 593  * intermediate 64 bit accumulator.  */
 594 uint32_t HELPER(neon_rshl_u32)(uint32_t val, uint32_t shiftop)
 595 {
 596     uint32_t dest;
 597     int8_t shift = (int8_t)shiftop;
 598     if (shift >= 32 || shift < -32) {
 599         dest = 0;
 600     } else if (shift == -32) {
 601         dest = val >> 31;
 602     } else if (shift < 0) {
 603         uint64_t big_dest = ((uint64_t)val + (1 << (-1 - shift)));
 604         dest = big_dest >> -shift;
 605     } else {
 606         dest = val << shift;
 607     }
 608     return dest;
 609 }
 610
 611 /* Handling addition overflow with 64 bit input values is more
 612  * tricky than with 32 bit values.  */
 613 uint64_t HELPER(neon_rshl_u64)(uint64_t val, uint64_t shiftop)
 614 {
 615     int8_t shift = (uint8_t)shiftop;
 616     if (shift >= 64 || shift < -64) {
 617         val = 0;
 618     } else if (shift == -64) {
 619         /* Rounding a 1-bit result just preserves that bit.  */
 620         val >>= 63;
 621     } else if (shift < 0) {
 622         val >>= (-shift - 1);
 623         if (val == UINT64_MAX) {
 624             /* In this case, it means that the rounding constant is 1,
 625              * and the addition would overflow. Return the actual
 626              * result directly.  */
 627             val = 0x8000000000000000ULL;
 628         } else {
 629             val++;
 630             val >>= 1;
 631         }
 632     } else {
 633         val <<= shift;
 634     }
 635     return val;
 636 }
 637
 638 #define NEON_FN(dest, src1, src2) do { \
 639     int8_t tmp; \
 640     tmp = (int8_t)src2; \
 641     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 642         if (src1) { \
 643             SET_QC(); \
 644             dest = ~0; \
 645         } else { \
 646             dest = 0; \
 647         } \
 648     } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 649         dest = 0; \
 650     } else if (tmp < 0) { \
 651         dest = src1 >> -tmp; \
 652     } else { \
 653         dest = src1 << tmp; \
 654         if ((dest >> tmp) != src1) { \
 655             SET_QC(); \
 656             dest = ~0; \
 657         } \
 658     }} while (0)
 659 NEON_VOP_ENV(qshl_u8, neon_u8, 4)
 660 NEON_VOP_ENV(qshl_u16, neon_u16, 2)
 661 NEON_VOP_ENV(qshl_u32, neon_u32, 1)
 662 #undef NEON_FN
 663
 664 uint64_t HELPER(neon_qshl_u64)(CPUARMState *env, uint64_t val, uint64_t shiftop)
 665 {
 666     int8_t shift = (int8_t)shiftop;
 667     if (shift >= 64) {
 668         if (val) {
 669             val = ~(uint64_t)0;
 670             SET_QC();
 671         }
 672     } else if (shift <= -64) {
 673         val = 0;
 674     } else if (shift < 0) {
 675         val >>= -shift;
 676     } else {
 677         uint64_t tmp = val;
 678         val <<= shift;
 679         if ((val >> shift) != tmp) {
 680             SET_QC();
 681             val = ~(uint64_t)0;
 682         }
 683     }
 684     return val;
 685 }
 686
 687 #define NEON_FN(dest, src1, src2) do { \
 688     int8_t tmp; \
 689     tmp = (int8_t)src2; \
 690     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 691         if (src1) { \
 692             SET_QC(); \
 693             dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
 694             if (src1 > 0) { \
 695                 dest--; \
 696             } \
 697         } else { \
 698             dest = src1; \
 699         } \
 700     } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 701         dest = src1 >> 31; \
 702     } else if (tmp < 0) { \
 703         dest = src1 >> -tmp; \
 704     } else { \
 705         dest = src1 << tmp; \
 706         if ((dest >> tmp) != src1) { \
 707             SET_QC(); \
 708             dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
 709             if (src1 > 0) { \
 710                 dest--; \
 711             } \
 712         } \
 713     }} while (0)
 714 NEON_VOP_ENV(qshl_s8, neon_s8, 4)
 715 NEON_VOP_ENV(qshl_s16, neon_s16, 2)
 716 NEON_VOP_ENV(qshl_s32, neon_s32, 1)
 717 #undef NEON_FN
 718
 719 uint64_t HELPER(neon_qshl_s64)(CPUARMState *env, uint64_t valop, uint64_t shiftop)
 720 {
 721     int8_t shift = (uint8_t)shiftop;
 722     int64_t val = valop;
 723     if (shift >= 64) {
 724         if (val) {
 725             SET_QC();
 726             val = (val >> 63) ^ ~SIGNBIT64;
 727         }
 728     } else if (shift <= -64) {
 729         val >>= 63;
 730     } else if (shift < 0) {
 731         val >>= -shift;
 732     } else {
 733         int64_t tmp = val;
 734         val <<= shift;
 735         if ((val >> shift) != tmp) {
 736             SET_QC();
 737             val = (tmp >> 63) ^ ~SIGNBIT64;
 738         }
 739     }
 740     return val;
 741 }
 742
 743 #define NEON_FN(dest, src1, src2) do { \
 744     if (src1 & (1 << (sizeof(src1) * 8 - 1))) { \
 745         SET_QC(); \
 746         dest = 0; \
 747     } else { \
 748         int8_t tmp; \
 749         tmp = (int8_t)src2; \
 750         if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 751             if (src1) { \
 752                 SET_QC(); \
 753                 dest = ~0; \
 754             } else { \
 755                 dest = 0; \
 756             } \
 757         } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 758             dest = 0; \
 759         } else if (tmp < 0) { \
 760             dest = src1 >> -tmp; \
 761         } else { \
 762             dest = src1 << tmp; \
 763             if ((dest >> tmp) != src1) { \
 764                 SET_QC(); \
 765                 dest = ~0; \
 766             } \
 767         } \
 768     }} while (0)
 769 NEON_VOP_ENV(qshlu_s8, neon_u8, 4)
 770 NEON_VOP_ENV(qshlu_s16, neon_u16, 2)
 771 #undef NEON_FN
 772
 773 uint32_t HELPER(neon_qshlu_s32)(CPUARMState *env, uint32_t valop, uint32_t shiftop)
 774 {
 775     if ((int32_t)valop < 0) {
 776         SET_QC();
 777         return 0;
 778     }
 779     return helper_neon_qshl_u32(env, valop, shiftop);
 780 }
 781
 782 uint64_t HELPER(neon_qshlu_s64)(CPUARMState *env, uint64_t valop, uint64_t shiftop)
 783 {
 784     if ((int64_t)valop < 0) {
 785         SET_QC();
 786         return 0;
 787     }
 788     return helper_neon_qshl_u64(env, valop, shiftop);
 789 }
 790
 791 #define NEON_FN(dest, src1, src2) do { \
 792     int8_t tmp; \
 793     tmp = (int8_t)src2; \
 794     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 795         if (src1) { \
 796             SET_QC(); \
 797             dest = ~0; \
 798         } else { \
 799             dest = 0; \
 800         } \
 801     } else if (tmp < -(ssize_t)sizeof(src1) * 8) { \
 802         dest = 0; \
 803     } else if (tmp == -(ssize_t)sizeof(src1) * 8) { \
 804         dest = src1 >> (sizeof(src1) * 8 - 1); \
 805     } else if (tmp < 0) { \
 806         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 807     } else { \
 808         dest = src1 << tmp; \
 809         if ((dest >> tmp) != src1) { \
 810             SET_QC(); \
 811             dest = ~0; \
 812         } \
 813     }} while (0)
 814 NEON_VOP_ENV(qrshl_u8, neon_u8, 4)
 815 NEON_VOP_ENV(qrshl_u16, neon_u16, 2)
 816 #undef NEON_FN
 817
 818 /* The addition of the rounding constant may overflow, so we use an
 819  * intermediate 64 bit accumulator.  */
 820 uint32_t HELPER(neon_qrshl_u32)(CPUARMState *env, uint32_t val, uint32_t shiftop)
 821 {
 822     uint32_t dest;
 823     int8_t shift = (int8_t)shiftop;
 824     if (shift >= 32) {
 825         if (val) {
 826             SET_QC();
 827             dest = ~0;
 828         } else {
 829             dest = 0;
 830         }
 831     } else if (shift < -32) {
 832         dest = 0;
 833     } else if (shift == -32) {
 834         dest = val >> 31;
 835     } else if (shift < 0) {
 836         uint64_t big_dest = ((uint64_t)val + (1 << (-1 - shift)));
 837         dest = big_dest >> -shift;
 838     } else {
 839         dest = val << shift;
 840         if ((dest >> shift) != val) {
 841             SET_QC();
 842             dest = ~0;
 843         }
 844     }
 845     return dest;
 846 }
 847
 848 /* Handling addition overflow with 64 bit input values is more
 849  * tricky than with 32 bit values.  */
 850 uint64_t HELPER(neon_qrshl_u64)(CPUARMState *env, uint64_t val, uint64_t shiftop)
 851 {
 852     int8_t shift = (int8_t)shiftop;
 853     if (shift >= 64) {
 854         if (val) {
 855             SET_QC();
 856             val = ~0;
 857         }
 858     } else if (shift < -64) {
 859         val = 0;
 860     } else if (shift == -64) {
 861         val >>= 63;
 862     } else if (shift < 0) {
 863         val >>= (-shift - 1);
 864         if (val == UINT64_MAX) {
 865             /* In this case, it means that the rounding constant is 1,
 866              * and the addition would overflow. Return the actual
 867              * result directly.  */
 868             val = 0x8000000000000000ULL;
 869         } else {
 870             val++;
 871             val >>= 1;
 872         }
 873     } else { \
 874         uint64_t tmp = val;
 875         val <<= shift;
 876         if ((val >> shift) != tmp) {
 877             SET_QC();
 878             val = ~0;
 879         }
 880     }
 881     return val;
 882 }
 883
 884 #define NEON_FN(dest, src1, src2) do { \
 885     int8_t tmp; \
 886     tmp = (int8_t)src2; \
 887     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 888         if (src1) { \
 889             SET_QC(); \
 890             dest = (1 << (sizeof(src1) * 8 - 1)); \
 891             if (src1 > 0) { \
 892                 dest--; \
 893             } \
 894         } else { \
 895             dest = 0; \
 896         } \
 897     } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 898         dest = 0; \
 899     } else if (tmp < 0) { \
 900         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 901     } else { \
 902         dest = src1 << tmp; \
 903         if ((dest >> tmp) != src1) { \
 904             SET_QC(); \
 905             dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
 906             if (src1 > 0) { \
 907                 dest--; \
 908             } \
 909         } \
 910     }} while (0)
 911 NEON_VOP_ENV(qrshl_s8, neon_s8, 4)
 912 NEON_VOP_ENV(qrshl_s16, neon_s16, 2)
 913 #undef NEON_FN
 914
 915 /* The addition of the rounding constant may overflow, so we use an
 916  * intermediate 64 bit accumulator.  */
 917 uint32_t HELPER(neon_qrshl_s32)(CPUARMState *env, uint32_t valop, uint32_t shiftop)
 918 {
 919     int32_t dest;
 920     int32_t val = (int32_t)valop;
 921     int8_t shift = (int8_t)shiftop;
 922     if (shift >= 32) {
 923         if (val) {
 924             SET_QC();
 925             dest = (val >> 31) ^ ~SIGNBIT;
 926         } else {
 927             dest = 0;
 928         }
 929     } else if (shift <= -32) {
 930         dest = 0;
 931     } else if (shift < 0) {
 932         int64_t big_dest = ((int64_t)val + (1 << (-1 - shift)));
 933         dest = big_dest >> -shift;
 934     } else {
 935         dest = val << shift;
 936         if ((dest >> shift) != val) {
 937             SET_QC();
 938             dest = (val >> 31) ^ ~SIGNBIT;
 939         }
 940     }
 941     return dest;
 942 }
 943
 944 /* Handling addition overflow with 64 bit input values is more
 945  * tricky than with 32 bit values.  */
 946 uint64_t HELPER(neon_qrshl_s64)(CPUARMState *env, uint64_t valop, uint64_t shiftop)
 947 {
 948     int8_t shift = (uint8_t)shiftop;
 949     int64_t val = valop;
 950
 951     if (shift >= 64) {
 952         if (val) {
 953             SET_QC();
 954             val = (val >> 63) ^ ~SIGNBIT64;
 955         }
 956     } else if (shift <= -64) {
 957         val = 0;
 958     } else if (shift < 0) {
 959         val >>= (-shift - 1);
 960         if (val == INT64_MAX) {
 961             /* In this case, it means that the rounding constant is 1,
 962              * and the addition would overflow. Return the actual
 963              * result directly.  */
 964             val = 0x4000000000000000ULL;
 965         } else {
 966             val++;
 967             val >>= 1;
 968         }
 969     } else {
 970         int64_t tmp = val;
 971         val <<= shift;
 972         if ((val >> shift) != tmp) {
 973             SET_QC();
 974             val = (tmp >> 63) ^ ~SIGNBIT64;
 975         }
 976     }
 977     return val;
 978 }
 979
 980 uint32_t HELPER(neon_add_u8)(uint32_t a, uint32_t b)
 981 {
 982     uint32_t mask;
 983     mask = (a ^ b) & 0x80808080u;
 984     a &= ~0x80808080u;
 985     b &= ~0x80808080u;
 986     return (a + b) ^ mask;
 987 }
 988
 989 uint32_t HELPER(neon_add_u16)(uint32_t a, uint32_t b)
 990 {
 991     uint32_t mask;
 992     mask = (a ^ b) & 0x80008000u;
 993     a &= ~0x80008000u;
 994     b &= ~0x80008000u;
 995     return (a + b) ^ mask;
 996 }
 997
 998 #define NEON_FN(dest, src1, src2) dest = src1 + src2
 999 NEON_POP(padd_u8, neon_u8, 4)
1000 NEON_POP(padd_u16, neon_u16, 2)
1001 #undef NEON_FN
1002
1003 #define NEON_FN(dest, src1, src2) dest = src1 - src2
1004 NEON_VOP(sub_u8, neon_u8, 4)
1005 NEON_VOP(sub_u16, neon_u16, 2)
1006 #undef NEON_FN
1007
1008 #define NEON_FN(dest, src1, src2) dest = src1 * src2
1009 NEON_VOP(mul_u8, neon_u8, 4)
1010 NEON_VOP(mul_u16, neon_u16, 2)
1011 #undef NEON_FN
1012
1013 /* Polynomial multiplication is like integer multiplication except the
1014    partial products are XORed, not added.  */
1015 uint32_t HELPER(neon_mul_p8)(uint32_t op1, uint32_t op2)
1016 {
1017     uint32_t mask;
1018     uint32_t result;
1019     result = 0;
1020     while (op1) {
1021         mask = 0;
1022         if (op1 & 1)
1023             mask |= 0xff;
1024         if (op1 & (1 << 8))
1025             mask |= (0xff << 8);
1026         if (op1 & (1 << 16))
1027             mask |= (0xff << 16);
1028         if (op1 & (1 << 24))
1029             mask |= (0xff << 24);
1030         result ^= op2 & mask;
1031         op1 = (op1 >> 1) & 0x7f7f7f7f;
1032         op2 = (op2 << 1) & 0xfefefefe;
1033     }
1034     return result;
1035 }
1036
1037 uint64_t HELPER(neon_mull_p8)(uint32_t op1, uint32_t op2)
1038 {
1039     uint64_t result = 0;
1040     uint64_t mask;
1041     uint64_t op2ex = op2;
1042     op2ex = (op2ex & 0xff) |
1043         ((op2ex & 0xff00) << 8) |
1044         ((op2ex & 0xff0000) << 16) |
1045         ((op2ex & 0xff000000) << 24);
1046     while (op1) {
1047         mask = 0;
1048         if (op1 & 1) {
1049             mask |= 0xffff;
1050         }
1051         if (op1 & (1 << 8)) {
1052             mask |= (0xffffU << 16);
1053         }
1054         if (op1 & (1 << 16)) {
1055             mask |= (0xffffULL << 32);
1056         }
1057         if (op1 & (1 << 24)) {
1058             mask |= (0xffffULL << 48);
1059         }
1060         result ^= op2ex & mask;
1061         op1 = (op1 >> 1) & 0x7f7f7f7f;
1062         op2ex <<= 1;
1063     }
1064     return result;
1065 }
1066
1067 #define NEON_FN(dest, src1, src2) dest = (src1 & src2) ? -1 : 0
1068 NEON_VOP(tst_u8, neon_u8, 4)
1069 NEON_VOP(tst_u16, neon_u16, 2)
1070 NEON_VOP(tst_u32, neon_u32, 1)
1071 #undef NEON_FN
1072
1073 #define NEON_FN(dest, src1, src2) dest = (src1 == src2) ? -1 : 0
1074 NEON_VOP(ceq_u8, neon_u8, 4)
1075 NEON_VOP(ceq_u16, neon_u16, 2)
1076 NEON_VOP(ceq_u32, neon_u32, 1)
1077 #undef NEON_FN
1078
1079 #define NEON_FN(dest, src, dummy) dest = (src < 0) ? -src : src
1080 NEON_VOP1(abs_s8, neon_s8, 4)
1081 NEON_VOP1(abs_s16, neon_s16, 2)
1082 #undef NEON_FN
1083
1084 /* Count Leading Sign/Zero Bits.  */
1085 static inline int do_clz8(uint8_t x)
1086 {
1087     int n;
1088     for (n = 8; x; n--)
1089         x >>= 1;
1090     return n;
1091 }
1092
1093 static inline int do_clz16(uint16_t x)
1094 {
1095     int n;
1096     for (n = 16; x; n--)
1097         x >>= 1;
1098     return n;
1099 }
1100
1101 #define NEON_FN(dest, src, dummy) dest = do_clz8(src)
1102 NEON_VOP1(clz_u8, neon_u8, 4)
1103 #undef NEON_FN
1104
1105 #define NEON_FN(dest, src, dummy) dest = do_clz16(src)
1106 NEON_VOP1(clz_u16, neon_u16, 2)
1107 #undef NEON_FN
1108
1109 #define NEON_FN(dest, src, dummy) dest = do_clz8((src < 0) ? ~src : src) - 1
1110 NEON_VOP1(cls_s8, neon_s8, 4)
1111 #undef NEON_FN
1112
1113 #define NEON_FN(dest, src, dummy) dest = do_clz16((src < 0) ? ~src : src) - 1
1114 NEON_VOP1(cls_s16, neon_s16, 2)
1115 #undef NEON_FN
1116
1117 uint32_t HELPER(neon_cls_s32)(uint32_t x)
1118 {
1119     int count;
1120     if ((int32_t)x < 0)
1121         x = ~x;
1122     for (count = 32; x; count--)
1123         x = x >> 1;
1124     return count - 1;
1125 }
1126
1127 /* Bit count.  */
1128 uint32_t HELPER(neon_cnt_u8)(uint32_t x)
1129 {
1130     x = (x & 0x55555555) + ((x >>  1) & 0x55555555);
1131     x = (x & 0x33333333) + ((x >>  2) & 0x33333333);
1132     x = (x & 0x0f0f0f0f) + ((x >>  4) & 0x0f0f0f0f);
1133     return x;
1134 }
1135
1136 #define NEON_QDMULH16(dest, src1, src2, round) do { \
1137     uint32_t tmp = (int32_t)(int16_t) src1 * (int16_t) src2; \
1138     if ((tmp ^ (tmp << 1)) & SIGNBIT) { \
1139         SET_QC(); \
1140         tmp = (tmp >> 31) ^ ~SIGNBIT; \
1141     } else { \
1142         tmp <<= 1; \
1143     } \
1144     if (round) { \
1145         int32_t old = tmp; \
1146         tmp += 1 << 15; \
1147         if ((int32_t)tmp < old) { \
1148             SET_QC(); \
1149             tmp = SIGNBIT - 1; \
1150         } \
1151     } \
1152     dest = tmp >> 16; \
1153     } while(0)
1154 #define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 0)
1155 NEON_VOP_ENV(qdmulh_s16, neon_s16, 2)
1156 #undef NEON_FN
1157 #define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 1)
1158 NEON_VOP_ENV(qrdmulh_s16, neon_s16, 2)
1159 #undef NEON_FN
1160 #undef NEON_QDMULH16
1161
1162 #define NEON_QDMULH32(dest, src1, src2, round) do { \
1163     uint64_t tmp = (int64_t)(int32_t) src1 * (int32_t) src2; \
1164     if ((tmp ^ (tmp << 1)) & SIGNBIT64) { \
1165         SET_QC(); \
1166         tmp = (tmp >> 63) ^ ~SIGNBIT64; \
1167     } else { \
1168         tmp <<= 1; \
1169     } \
1170     if (round) { \
1171         int64_t old = tmp; \
1172         tmp += (int64_t)1 << 31; \
1173         if ((int64_t)tmp < old) { \
1174             SET_QC(); \
1175             tmp = SIGNBIT64 - 1; \
1176         } \
1177     } \
1178     dest = tmp >> 32; \
1179     } while(0)
1180 #define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 0)
1181 NEON_VOP_ENV(qdmulh_s32, neon_s32, 1)
1182 #undef NEON_FN
1183 #define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 1)
1184 NEON_VOP_ENV(qrdmulh_s32, neon_s32, 1)
1185 #undef NEON_FN
1186 #undef NEON_QDMULH32
1187
1188 uint32_t HELPER(neon_narrow_u8)(uint64_t x)
1189 {
1190     return (x & 0xffu) | ((x >> 8) & 0xff00u) | ((x >> 16) & 0xff0000u)
1191            | ((x >> 24) & 0xff000000u);
1192 }
1193
1194 uint32_t HELPER(neon_narrow_u16)(uint64_t x)
1195 {
1196     return (x & 0xffffu) | ((x >> 16) & 0xffff0000u);
1197 }
1198
1199 uint32_t HELPER(neon_narrow_high_u8)(uint64_t x)
1200 {
1201     return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00)
1202             | ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000);
1203 }
1204
1205 uint32_t HELPER(neon_narrow_high_u16)(uint64_t x)
1206 {
1207     return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000);
1208 }
1209
1210 uint32_t HELPER(neon_narrow_round_high_u8)(uint64_t x)
1211 {
1212     x &= 0xff80ff80ff80ff80ull;
1213     x += 0x0080008000800080ull;
1214     return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00)
1215             | ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000);
1216 }
1217
1218 uint32_t HELPER(neon_narrow_round_high_u16)(uint64_t x)
1219 {
1220     x &= 0xffff8000ffff8000ull;
1221     x += 0x0000800000008000ull;
1222     return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000);
1223 }
1224
1225 uint32_t HELPER(neon_unarrow_sat8)(CPUARMState *env, uint64_t x)
1226 {
1227     uint16_t s;
1228     uint8_t d;
1229     uint32_t res = 0;
1230 #define SAT8(n) \
1231     s = x >> n; \
1232     if (s & 0x8000) { \
1233         SET_QC(); \
1234     } else { \
1235         if (s > 0xff) { \
1236             d = 0xff; \
1237             SET_QC(); \
1238         } else  { \
1239             d = s; \
1240         } \
1241         res |= (uint32_t)d << (n / 2); \
1242     }
1243
1244     SAT8(0);
1245     SAT8(16);
1246     SAT8(32);
1247     SAT8(48);
1248 #undef SAT8
1249     return res;
1250 }
1251
1252 uint32_t HELPER(neon_narrow_sat_u8)(CPUARMState *env, uint64_t x)
1253 {
1254     uint16_t s;
1255     uint8_t d;
1256     uint32_t res = 0;
1257 #define SAT8(n) \
1258     s = x >> n; \
1259     if (s > 0xff) { \
1260         d = 0xff; \
1261         SET_QC(); \
1262     } else  { \
1263         d = s; \
1264     } \
1265     res |= (uint32_t)d << (n / 2);
1266
1267     SAT8(0);
1268     SAT8(16);
1269     SAT8(32);
1270     SAT8(48);
1271 #undef SAT8
1272     return res;
1273 }
1274
1275 uint32_t HELPER(neon_narrow_sat_s8)(CPUARMState *env, uint64_t x)
1276 {
1277     int16_t s;
1278     uint8_t d;
1279     uint32_t res = 0;
1280 #define SAT8(n) \
1281     s = x >> n; \
1282     if (s != (int8_t)s) { \
1283         d = (s >> 15) ^ 0x7f; \
1284         SET_QC(); \
1285     } else  { \
1286         d = s; \
1287     } \
1288     res |= (uint32_t)d << (n / 2);
1289
1290     SAT8(0);
1291     SAT8(16);
1292     SAT8(32);
1293     SAT8(48);
1294 #undef SAT8
1295     return res;
1296 }
1297
1298 uint32_t HELPER(neon_unarrow_sat16)(CPUARMState *env, uint64_t x)
1299 {
1300     uint32_t high;
1301     uint32_t low;
1302     low = x;
1303     if (low & 0x80000000) {
1304         low = 0;
1305         SET_QC();
1306     } else if (low > 0xffff) {
1307         low = 0xffff;
1308         SET_QC();
1309     }
1310     high = x >> 32;
1311     if (high & 0x80000000) {
1312         high = 0;
1313         SET_QC();
1314     } else if (high > 0xffff) {
1315         high = 0xffff;
1316         SET_QC();
1317     }
1318     return low | (high << 16);
1319 }
1320
1321 uint32_t HELPER(neon_narrow_sat_u16)(CPUARMState *env, uint64_t x)
1322 {
1323     uint32_t high;
1324     uint32_t low;
1325     low = x;
1326     if (low > 0xffff) {
1327         low = 0xffff;
1328         SET_QC();
1329     }
1330     high = x >> 32;
1331     if (high > 0xffff) {
1332         high = 0xffff;
1333         SET_QC();
1334     }
1335     return low | (high << 16);
1336 }
1337
1338 uint32_t HELPER(neon_narrow_sat_s16)(CPUARMState *env, uint64_t x)
1339 {
1340     int32_t low;
1341     int32_t high;
1342     low = x;
1343     if (low != (int16_t)low) {
1344         low = (low >> 31) ^ 0x7fff;
1345         SET_QC();
1346     }
1347     high = x >> 32;
1348     if (high != (int16_t)high) {
1349         high = (high >> 31) ^ 0x7fff;
1350         SET_QC();
1351     }
1352     return (uint16_t)low | (high << 16);
1353 }
1354
1355 uint32_t HELPER(neon_unarrow_sat32)(CPUARMState *env, uint64_t x)
1356 {
1357     if (x & 0x8000000000000000ull) {
1358         SET_QC();
1359         return 0;
1360     }
1361     if (x > 0xffffffffu) {
1362         SET_QC();
1363         return 0xffffffffu;
1364     }
1365     return x;
1366 }
1367
1368 uint32_t HELPER(neon_narrow_sat_u32)(CPUARMState *env, uint64_t x)
1369 {
1370     if (x > 0xffffffffu) {
1371         SET_QC();
1372         return 0xffffffffu;
1373     }
1374     return x;
1375 }
1376
1377 uint32_t HELPER(neon_narrow_sat_s32)(CPUARMState *env, uint64_t x)
1378 {
1379     if ((int64_t)x != (int32_t)x) {
1380         SET_QC();
1381         return ((int64_t)x >> 63) ^ 0x7fffffff;
1382     }
1383     return x;
1384 }
1385
1386 uint64_t HELPER(neon_widen_u8)(uint32_t x)
1387 {
1388     uint64_t tmp;
1389     uint64_t ret;
1390     ret = (uint8_t)x;
1391     tmp = (uint8_t)(x >> 8);
1392     ret |= tmp << 16;
1393     tmp = (uint8_t)(x >> 16);
1394     ret |= tmp << 32;
1395     tmp = (uint8_t)(x >> 24);
1396     ret |= tmp << 48;
1397     return ret;
1398 }
1399
1400 uint64_t HELPER(neon_widen_s8)(uint32_t x)
1401 {
1402     uint64_t tmp;
1403     uint64_t ret;
1404     ret = (uint16_t)(int8_t)x;
1405     tmp = (uint16_t)(int8_t)(x >> 8);
1406     ret |= tmp << 16;
1407     tmp = (uint16_t)(int8_t)(x >> 16);
1408     ret |= tmp << 32;
1409     tmp = (uint16_t)(int8_t)(x >> 24);
1410     ret |= tmp << 48;
1411     return ret;
1412 }
1413
1414 uint64_t HELPER(neon_widen_u16)(uint32_t x)
1415 {
1416     uint64_t high = (uint16_t)(x >> 16);
1417     return ((uint16_t)x) | (high << 32);
1418 }
1419
1420 uint64_t HELPER(neon_widen_s16)(uint32_t x)
1421 {
1422     uint64_t high = (int16_t)(x >> 16);
1423     return ((uint32_t)(int16_t)x) | (high << 32);
1424 }
1425
1426 uint64_t HELPER(neon_addl_u16)(uint64_t a, uint64_t b)
1427 {
1428     uint64_t mask;
1429     mask = (a ^ b) & 0x8000800080008000ull;
1430     a &= ~0x8000800080008000ull;
1431     b &= ~0x8000800080008000ull;
1432     return (a + b) ^ mask;
1433 }
1434
1435 uint64_t HELPER(neon_addl_u32)(uint64_t a, uint64_t b)
1436 {
1437     uint64_t mask;
1438     mask = (a ^ b) & 0x8000000080000000ull;
1439     a &= ~0x8000000080000000ull;
1440     b &= ~0x8000000080000000ull;
1441     return (a + b) ^ mask;
1442 }
1443
1444 uint64_t HELPER(neon_paddl_u16)(uint64_t a, uint64_t b)
1445 {
1446     uint64_t tmp;
1447     uint64_t tmp2;
1448
1449     tmp = a & 0x0000ffff0000ffffull;
1450     tmp += (a >> 16) & 0x0000ffff0000ffffull;
1451     tmp2 = b & 0xffff0000ffff0000ull;
1452     tmp2 += (b << 16) & 0xffff0000ffff0000ull;
1453     return    ( tmp         & 0xffff)
1454             | ((tmp  >> 16) & 0xffff0000ull)
1455             | ((tmp2 << 16) & 0xffff00000000ull)
1456             | ( tmp2        & 0xffff000000000000ull);
1457 }
1458
1459 uint64_t HELPER(neon_paddl_u32)(uint64_t a, uint64_t b)
1460 {
1461     uint32_t low = a + (a >> 32);
1462     uint32_t high = b + (b >> 32);
1463     return low + ((uint64_t)high << 32);
1464 }
1465
1466 uint64_t HELPER(neon_subl_u16)(uint64_t a, uint64_t b)
1467 {
1468     uint64_t mask;
1469     mask = (a ^ ~b) & 0x8000800080008000ull;
1470     a |= 0x8000800080008000ull;
1471     b &= ~0x8000800080008000ull;
1472     return (a - b) ^ mask;
1473 }
1474
1475 uint64_t HELPER(neon_subl_u32)(uint64_t a, uint64_t b)
1476 {
1477     uint64_t mask;
1478     mask = (a ^ ~b) & 0x8000000080000000ull;
1479     a |= 0x8000000080000000ull;
1480     b &= ~0x8000000080000000ull;
1481     return (a - b) ^ mask;
1482 }
1483
1484 uint64_t HELPER(neon_addl_saturate_s32)(CPUARMState *env, uint64_t a, uint64_t b)
1485 {
1486     uint32_t x, y;
1487     uint32_t low, high;
1488
1489     x = a;
1490     y = b;
1491     low = x + y;
1492     if (((low ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) {
1493         SET_QC();
1494         low = ((int32_t)x >> 31) ^ ~SIGNBIT;
1495     }
1496     x = a >> 32;
1497     y = b >> 32;
1498     high = x + y;
1499     if (((high ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) {
1500         SET_QC();
1501         high = ((int32_t)x >> 31) ^ ~SIGNBIT;
1502     }
1503     return low | ((uint64_t)high << 32);
1504 }
1505
1506 uint64_t HELPER(neon_addl_saturate_s64)(CPUARMState *env, uint64_t a, uint64_t b)
1507 {
1508     uint64_t result;
1509
1510     result = a + b;
1511     if (((result ^ a) & SIGNBIT64) && !((a ^ b) & SIGNBIT64)) {
1512         SET_QC();
1513         result = ((int64_t)a >> 63) ^ ~SIGNBIT64;
1514     }
1515     return result;
1516 }
1517
1518 /* We have to do the arithmetic in a larger type than
1519  * the input type, because for example with a signed 32 bit
1520  * op the absolute difference can overflow a signed 32 bit value.
1521  */
1522 #define DO_ABD(dest, x, y, intype, arithtype) do {            \
1523     arithtype tmp_x = (intype)(x);                            \
1524     arithtype tmp_y = (intype)(y);                            \
1525     dest = ((tmp_x > tmp_y) ? tmp_x - tmp_y : tmp_y - tmp_x); \
1526     } while(0)
1527
1528 uint64_t HELPER(neon_abdl_u16)(uint32_t a, uint32_t b)
1529 {
1530     uint64_t tmp;
1531     uint64_t result;
1532     DO_ABD(result, a, b, uint8_t, uint32_t);
1533     DO_ABD(tmp, a >> 8, b >> 8, uint8_t, uint32_t);
1534     result |= tmp << 16;
1535     DO_ABD(tmp, a >> 16, b >> 16, uint8_t, uint32_t);
1536     result |= tmp << 32;
1537     DO_ABD(tmp, a >> 24, b >> 24, uint8_t, uint32_t);
1538     result |= tmp << 48;
1539     return result;
1540 }
1541
1542 uint64_t HELPER(neon_abdl_s16)(uint32_t a, uint32_t b)
1543 {
1544     uint64_t tmp;
1545     uint64_t result;
1546     DO_ABD(result, a, b, int8_t, int32_t);
1547     DO_ABD(tmp, a >> 8, b >> 8, int8_t, int32_t);
1548     result |= tmp << 16;
1549     DO_ABD(tmp, a >> 16, b >> 16, int8_t, int32_t);
1550     result |= tmp << 32;
1551     DO_ABD(tmp, a >> 24, b >> 24, int8_t, int32_t);
1552     result |= tmp << 48;
1553     return result;
1554 }
1555
1556 uint64_t HELPER(neon_abdl_u32)(uint32_t a, uint32_t b)
1557 {
1558     uint64_t tmp;
1559     uint64_t result;
1560     DO_ABD(result, a, b, uint16_t, uint32_t);
1561     DO_ABD(tmp, a >> 16, b >> 16, uint16_t, uint32_t);
1562     return result | (tmp << 32);
1563 }
1564
1565 uint64_t HELPER(neon_abdl_s32)(uint32_t a, uint32_t b)
1566 {
1567     uint64_t tmp;
1568     uint64_t result;
1569     DO_ABD(result, a, b, int16_t, int32_t);
1570     DO_ABD(tmp, a >> 16, b >> 16, int16_t, int32_t);
1571     return result | (tmp << 32);
1572 }
1573
1574 uint64_t HELPER(neon_abdl_u64)(uint32_t a, uint32_t b)
1575 {
1576     uint64_t result;
1577     DO_ABD(result, a, b, uint32_t, uint64_t);
1578     return result;
1579 }
1580
1581 uint64_t HELPER(neon_abdl_s64)(uint32_t a, uint32_t b)
1582 {
1583     uint64_t result;
1584     DO_ABD(result, a, b, int32_t, int64_t);
1585     return result;
1586 }
1587 #undef DO_ABD
1588
1589 /* Widening multiply. Named type is the source type.  */
1590 #define DO_MULL(dest, x, y, type1, type2) do { \
1591     type1 tmp_x = x; \
1592     type1 tmp_y = y; \
1593     dest = (type2)((type2)tmp_x * (type2)tmp_y); \
1594     } while(0)
1595
1596 uint64_t HELPER(neon_mull_u8)(uint32_t a, uint32_t b)
1597 {
1598     uint64_t tmp;
1599     uint64_t result;
1600
1601     DO_MULL(result, a, b, uint8_t, uint16_t);
1602     DO_MULL(tmp, a >> 8, b >> 8, uint8_t, uint16_t);
1603     result |= tmp << 16;
1604     DO_MULL(tmp, a >> 16, b >> 16, uint8_t, uint16_t);
1605     result |= tmp << 32;
1606     DO_MULL(tmp, a >> 24, b >> 24, uint8_t, uint16_t);
1607     result |= tmp << 48;
1608     return result;
1609 }
1610
1611 uint64_t HELPER(neon_mull_s8)(uint32_t a, uint32_t b)
1612 {
1613     uint64_t tmp;
1614     uint64_t result;
1615
1616     DO_MULL(result, a, b, int8_t, uint16_t);
1617     DO_MULL(tmp, a >> 8, b >> 8, int8_t, uint16_t);
1618     result |= tmp << 16;
1619     DO_MULL(tmp, a >> 16, b >> 16, int8_t, uint16_t);
1620     result |= tmp << 32;
1621     DO_MULL(tmp, a >> 24, b >> 24, int8_t, uint16_t);
1622     result |= tmp << 48;
1623     return result;
1624 }
1625
1626 uint64_t HELPER(neon_mull_u16)(uint32_t a, uint32_t b)
1627 {
1628     uint64_t tmp;
1629     uint64_t result;
1630
1631     DO_MULL(result, a, b, uint16_t, uint32_t);
1632     DO_MULL(tmp, a >> 16, b >> 16, uint16_t, uint32_t);
1633     return result | (tmp << 32);
1634 }
1635
1636 uint64_t HELPER(neon_mull_s16)(uint32_t a, uint32_t b)
1637 {
1638     uint64_t tmp;
1639     uint64_t result;
1640
1641     DO_MULL(result, a, b, int16_t, uint32_t);
1642     DO_MULL(tmp, a >> 16, b >> 16, int16_t, uint32_t);
1643     return result | (tmp << 32);
1644 }
1645
1646 uint64_t HELPER(neon_negl_u16)(uint64_t x)
1647 {
1648     uint16_t tmp;
1649     uint64_t result;
1650     result = (uint16_t)-x;
1651     tmp = -(x >> 16);
1652     result |= (uint64_t)tmp << 16;
1653     tmp = -(x >> 32);
1654     result |= (uint64_t)tmp << 32;
1655     tmp = -(x >> 48);
1656     result |= (uint64_t)tmp << 48;
1657     return result;
1658 }
1659
1660 uint64_t HELPER(neon_negl_u32)(uint64_t x)
1661 {
1662     uint32_t low = -x;
1663     uint32_t high = -(x >> 32);
1664     return low | ((uint64_t)high << 32);
1665 }
1666
1667 /* Saturating sign manipulation.  */
1668 /* ??? Make these use NEON_VOP1 */
1669 #define DO_QABS8(x) do { \
1670     if (x == (int8_t)0x80) { \
1671         x = 0x7f; \
1672         SET_QC(); \
1673     } else if (x < 0) { \
1674         x = -x; \
1675     }} while (0)
1676 uint32_t HELPER(neon_qabs_s8)(CPUARMState *env, uint32_t x)
1677 {
1678     neon_s8 vec;
1679     NEON_UNPACK(neon_s8, vec, x);
1680     DO_QABS8(vec.v1);
1681     DO_QABS8(vec.v2);
1682     DO_QABS8(vec.v3);
1683     DO_QABS8(vec.v4);
1684     NEON_PACK(neon_s8, x, vec);
1685     return x;
1686 }
1687 #undef DO_QABS8
1688
1689 #define DO_QNEG8(x) do { \
1690     if (x == (int8_t)0x80) { \
1691         x = 0x7f; \
1692         SET_QC(); \
1693     } else { \
1694         x = -x; \
1695     }} while (0)
1696 uint32_t HELPER(neon_qneg_s8)(CPUARMState *env, uint32_t x)
1697 {
1698     neon_s8 vec;
1699     NEON_UNPACK(neon_s8, vec, x);
1700     DO_QNEG8(vec.v1);
1701     DO_QNEG8(vec.v2);
1702     DO_QNEG8(vec.v3);
1703     DO_QNEG8(vec.v4);
1704     NEON_PACK(neon_s8, x, vec);
1705     return x;
1706 }
1707 #undef DO_QNEG8
1708
1709 #define DO_QABS16(x) do { \
1710     if (x == (int16_t)0x8000) { \
1711         x = 0x7fff; \
1712         SET_QC(); \
1713     } else if (x < 0) { \
1714         x = -x; \
1715     }} while (0)
1716 uint32_t HELPER(neon_qabs_s16)(CPUARMState *env, uint32_t x)
1717 {
1718     neon_s16 vec;
1719     NEON_UNPACK(neon_s16, vec, x);
1720     DO_QABS16(vec.v1);
1721     DO_QABS16(vec.v2);
1722     NEON_PACK(neon_s16, x, vec);
1723     return x;
1724 }
1725 #undef DO_QABS16
1726
1727 #define DO_QNEG16(x) do { \
1728     if (x == (int16_t)0x8000) { \
1729         x = 0x7fff; \
1730         SET_QC(); \
1731     } else { \
1732         x = -x; \
1733     }} while (0)
1734 uint32_t HELPER(neon_qneg_s16)(CPUARMState *env, uint32_t x)
1735 {
1736     neon_s16 vec;
1737     NEON_UNPACK(neon_s16, vec, x);
1738     DO_QNEG16(vec.v1);
1739     DO_QNEG16(vec.v2);
1740     NEON_PACK(neon_s16, x, vec);
1741     return x;
1742 }
1743 #undef DO_QNEG16
1744
1745 uint32_t HELPER(neon_qabs_s32)(CPUARMState *env, uint32_t x)
1746 {
1747     if (x == SIGNBIT) {
1748         SET_QC();
1749         x = ~SIGNBIT;
1750     } else if ((int32_t)x < 0) {
1751         x = -x;
1752     }
1753     return x;
1754 }
1755
1756 uint32_t HELPER(neon_qneg_s32)(CPUARMState *env, uint32_t x)
1757 {
1758     if (x == SIGNBIT) {
1759         SET_QC();
1760         x = ~SIGNBIT;
1761     } else {
1762         x = -x;
1763     }
1764     return x;
1765 }
1766
1767 /* NEON Float helpers.  */
1768 uint32_t HELPER(neon_min_f32)(uint32_t a, uint32_t b, void *fpstp)
1769 {
1770     float_status *fpst = fpstp;
1771     return float32_val(float32_min(make_float32(a), make_float32(b), fpst));
1772 }
1773
1774 uint32_t HELPER(neon_max_f32)(uint32_t a, uint32_t b, void *fpstp)
1775 {
1776     float_status *fpst = fpstp;
1777     return float32_val(float32_max(make_float32(a), make_float32(b), fpst));
1778 }
1779
1780 uint32_t HELPER(neon_abd_f32)(uint32_t a, uint32_t b, void *fpstp)
1781 {
1782     float_status *fpst = fpstp;
1783     float32 f0 = make_float32(a);
1784     float32 f1 = make_float32(b);
1785     return float32_val(float32_abs(float32_sub(f0, f1, fpst)));
1786 }
1787
1788 /* Floating point comparisons produce an integer result.
1789  * Note that EQ doesn't signal InvalidOp for QNaNs but GE and GT do.
1790  * Softfloat routines return 0/1, which we convert to the 0/-1 Neon requires.
1791  */
1792 uint32_t HELPER(neon_ceq_f32)(uint32_t a, uint32_t b, void *fpstp)
1793 {
1794     float_status *fpst = fpstp;
1795     return -float32_eq_quiet(make_float32(a), make_float32(b), fpst);
1796 }
1797
1798 uint32_t HELPER(neon_cge_f32)(uint32_t a, uint32_t b, void *fpstp)
1799 {
1800     float_status *fpst = fpstp;
1801     return -float32_le(make_float32(b), make_float32(a), fpst);
1802 }
1803
1804 uint32_t HELPER(neon_cgt_f32)(uint32_t a, uint32_t b, void *fpstp)
1805 {
1806     float_status *fpst = fpstp;
1807     return -float32_lt(make_float32(b), make_float32(a), fpst);
1808 }
1809
1810 uint32_t HELPER(neon_acge_f32)(uint32_t a, uint32_t b, void *fpstp)
1811 {
1812     float_status *fpst = fpstp;
1813     float32 f0 = float32_abs(make_float32(a));
1814     float32 f1 = float32_abs(make_float32(b));
1815     return -float32_le(f1, f0, fpst);
1816 }
1817
1818 uint32_t HELPER(neon_acgt_f32)(uint32_t a, uint32_t b, void *fpstp)
1819 {
1820     float_status *fpst = fpstp;
1821     float32 f0 = float32_abs(make_float32(a));
1822     float32 f1 = float32_abs(make_float32(b));
1823     return -float32_lt(f1, f0, fpst);
1824 }
1825
1826 #define ELEM(V, N, SIZE) (((V) >> ((N) * (SIZE))) & ((1ull << (SIZE)) - 1))
1827
1828 void HELPER(neon_qunzip8)(CPUARMState *env, uint32_t rd, uint32_t rm)
1829 {
1830     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1831     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1832     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1833     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1834     uint64_t d0 = ELEM(zd0, 0, 8) | (ELEM(zd0, 2, 8) << 8)
1835         | (ELEM(zd0, 4, 8) << 16) | (ELEM(zd0, 6, 8) << 24)
1836         | (ELEM(zd1, 0, 8) << 32) | (ELEM(zd1, 2, 8) << 40)
1837         | (ELEM(zd1, 4, 8) << 48) | (ELEM(zd1, 6, 8) << 56);
1838     uint64_t d1 = ELEM(zm0, 0, 8) | (ELEM(zm0, 2, 8) << 8)
1839         | (ELEM(zm0, 4, 8) << 16) | (ELEM(zm0, 6, 8) << 24)
1840         | (ELEM(zm1, 0, 8) << 32) | (ELEM(zm1, 2, 8) << 40)
1841         | (ELEM(zm1, 4, 8) << 48) | (ELEM(zm1, 6, 8) << 56);
1842     uint64_t m0 = ELEM(zd0, 1, 8) | (ELEM(zd0, 3, 8) << 8)
1843         | (ELEM(zd0, 5, 8) << 16) | (ELEM(zd0, 7, 8) << 24)
1844         | (ELEM(zd1, 1, 8) << 32) | (ELEM(zd1, 3, 8) << 40)
1845         | (ELEM(zd1, 5, 8) << 48) | (ELEM(zd1, 7, 8) << 56);
1846     uint64_t m1 = ELEM(zm0, 1, 8) | (ELEM(zm0, 3, 8) << 8)
1847         | (ELEM(zm0, 5, 8) << 16) | (ELEM(zm0, 7, 8) << 24)
1848         | (ELEM(zm1, 1, 8) << 32) | (ELEM(zm1, 3, 8) << 40)
1849         | (ELEM(zm1, 5, 8) << 48) | (ELEM(zm1, 7, 8) << 56);
1850     env->vfp.regs[rm] = make_float64(m0);
1851     env->vfp.regs[rm + 1] = make_float64(m1);
1852     env->vfp.regs[rd] = make_float64(d0);
1853     env->vfp.regs[rd + 1] = make_float64(d1);
1854 }
1855
1856 void HELPER(neon_qunzip16)(CPUARMState *env, uint32_t rd, uint32_t rm)
1857 {
1858     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1859     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1860     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1861     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1862     uint64_t d0 = ELEM(zd0, 0, 16) | (ELEM(zd0, 2, 16) << 16)
1863         | (ELEM(zd1, 0, 16) << 32) | (ELEM(zd1, 2, 16) << 48);
1864     uint64_t d1 = ELEM(zm0, 0, 16) | (ELEM(zm0, 2, 16) << 16)
1865         | (ELEM(zm1, 0, 16) << 32) | (ELEM(zm1, 2, 16) << 48);
1866     uint64_t m0 = ELEM(zd0, 1, 16) | (ELEM(zd0, 3, 16) << 16)
1867         | (ELEM(zd1, 1, 16) << 32) | (ELEM(zd1, 3, 16) << 48);
1868     uint64_t m1 = ELEM(zm0, 1, 16) | (ELEM(zm0, 3, 16) << 16)
1869         | (ELEM(zm1, 1, 16) << 32) | (ELEM(zm1, 3, 16) << 48);
1870     env->vfp.regs[rm] = make_float64(m0);
1871     env->vfp.regs[rm + 1] = make_float64(m1);
1872     env->vfp.regs[rd] = make_float64(d0);
1873     env->vfp.regs[rd + 1] = make_float64(d1);
1874 }
1875
1876 void HELPER(neon_qunzip32)(CPUARMState *env, uint32_t rd, uint32_t rm)
1877 {
1878     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1879     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1880     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1881     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1882     uint64_t d0 = ELEM(zd0, 0, 32) | (ELEM(zd1, 0, 32) << 32);
1883     uint64_t d1 = ELEM(zm0, 0, 32) | (ELEM(zm1, 0, 32) << 32);
1884     uint64_t m0 = ELEM(zd0, 1, 32) | (ELEM(zd1, 1, 32) << 32);
1885     uint64_t m1 = ELEM(zm0, 1, 32) | (ELEM(zm1, 1, 32) << 32);
1886     env->vfp.regs[rm] = make_float64(m0);
1887     env->vfp.regs[rm + 1] = make_float64(m1);
1888     env->vfp.regs[rd] = make_float64(d0);
1889     env->vfp.regs[rd + 1] = make_float64(d1);
1890 }
1891
1892 void HELPER(neon_unzip8)(CPUARMState *env, uint32_t rd, uint32_t rm)
1893 {
1894     uint64_t zm = float64_val(env->vfp.regs[rm]);
1895     uint64_t zd = float64_val(env->vfp.regs[rd]);
1896     uint64_t d0 = ELEM(zd, 0, 8) | (ELEM(zd, 2, 8) << 8)
1897         | (ELEM(zd, 4, 8) << 16) | (ELEM(zd, 6, 8) << 24)
1898         | (ELEM(zm, 0, 8) << 32) | (ELEM(zm, 2, 8) << 40)
1899         | (ELEM(zm, 4, 8) << 48) | (ELEM(zm, 6, 8) << 56);
1900     uint64_t m0 = ELEM(zd, 1, 8) | (ELEM(zd, 3, 8) << 8)
1901         | (ELEM(zd, 5, 8) << 16) | (ELEM(zd, 7, 8) << 24)
1902         | (ELEM(zm, 1, 8) << 32) | (ELEM(zm, 3, 8) << 40)
1903         | (ELEM(zm, 5, 8) << 48) | (ELEM(zm, 7, 8) << 56);
1904     env->vfp.regs[rm] = make_float64(m0);
1905     env->vfp.regs[rd] = make_float64(d0);
1906 }
1907
1908 void HELPER(neon_unzip16)(CPUARMState *env, uint32_t rd, uint32_t rm)
1909 {
1910     uint64_t zm = float64_val(env->vfp.regs[rm]);
1911     uint64_t zd = float64_val(env->vfp.regs[rd]);
1912     uint64_t d0 = ELEM(zd, 0, 16) | (ELEM(zd, 2, 16) << 16)
1913         | (ELEM(zm, 0, 16) << 32) | (ELEM(zm, 2, 16) << 48);
1914     uint64_t m0 = ELEM(zd, 1, 16) | (ELEM(zd, 3, 16) << 16)
1915         | (ELEM(zm, 1, 16) << 32) | (ELEM(zm, 3, 16) << 48);
1916     env->vfp.regs[rm] = make_float64(m0);
1917     env->vfp.regs[rd] = make_float64(d0);
1918 }
1919
1920 void HELPER(neon_qzip8)(CPUARMState *env, uint32_t rd, uint32_t rm)
1921 {
1922     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1923     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1924     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1925     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1926     uint64_t d0 = ELEM(zd0, 0, 8) | (ELEM(zm0, 0, 8) << 8)
1927         | (ELEM(zd0, 1, 8) << 16) | (ELEM(zm0, 1, 8) << 24)
1928         | (ELEM(zd0, 2, 8) << 32) | (ELEM(zm0, 2, 8) << 40)
1929         | (ELEM(zd0, 3, 8) << 48) | (ELEM(zm0, 3, 8) << 56);
1930     uint64_t d1 = ELEM(zd0, 4, 8) | (ELEM(zm0, 4, 8) << 8)
1931         | (ELEM(zd0, 5, 8) << 16) | (ELEM(zm0, 5, 8) << 24)
1932         | (ELEM(zd0, 6, 8) << 32) | (ELEM(zm0, 6, 8) << 40)
1933         | (ELEM(zd0, 7, 8) << 48) | (ELEM(zm0, 7, 8) << 56);
1934     uint64_t m0 = ELEM(zd1, 0, 8) | (ELEM(zm1, 0, 8) << 8)
1935         | (ELEM(zd1, 1, 8) << 16) | (ELEM(zm1, 1, 8) << 24)
1936         | (ELEM(zd1, 2, 8) << 32) | (ELEM(zm1, 2, 8) << 40)
1937         | (ELEM(zd1, 3, 8) << 48) | (ELEM(zm1, 3, 8) << 56);
1938     uint64_t m1 = ELEM(zd1, 4, 8) | (ELEM(zm1, 4, 8) << 8)
1939         | (ELEM(zd1, 5, 8) << 16) | (ELEM(zm1, 5, 8) << 24)
1940         | (ELEM(zd1, 6, 8) << 32) | (ELEM(zm1, 6, 8) << 40)
1941         | (ELEM(zd1, 7, 8) << 48) | (ELEM(zm1, 7, 8) << 56);
1942     env->vfp.regs[rm] = make_float64(m0);
1943     env->vfp.regs[rm + 1] = make_float64(m1);
1944     env->vfp.regs[rd] = make_float64(d0);
1945     env->vfp.regs[rd + 1] = make_float64(d1);
1946 }
1947
1948 void HELPER(neon_qzip16)(CPUARMState *env, uint32_t rd, uint32_t rm)
1949 {
1950     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1951     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1952     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1953     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1954     uint64_t d0 = ELEM(zd0, 0, 16) | (ELEM(zm0, 0, 16) << 16)
1955         | (ELEM(zd0, 1, 16) << 32) | (ELEM(zm0, 1, 16) << 48);
1956     uint64_t d1 = ELEM(zd0, 2, 16) | (ELEM(zm0, 2, 16) << 16)
1957         | (ELEM(zd0, 3, 16) << 32) | (ELEM(zm0, 3, 16) << 48);
1958     uint64_t m0 = ELEM(zd1, 0, 16) | (ELEM(zm1, 0, 16) << 16)
1959         | (ELEM(zd1, 1, 16) << 32) | (ELEM(zm1, 1, 16) << 48);
1960     uint64_t m1 = ELEM(zd1, 2, 16) | (ELEM(zm1, 2, 16) << 16)
1961         | (ELEM(zd1, 3, 16) << 32) | (ELEM(zm1, 3, 16) << 48);
1962     env->vfp.regs[rm] = make_float64(m0);
1963     env->vfp.regs[rm + 1] = make_float64(m1);
1964     env->vfp.regs[rd] = make_float64(d0);
1965     env->vfp.regs[rd + 1] = make_float64(d1);
1966 }
1967
1968 void HELPER(neon_qzip32)(CPUARMState *env, uint32_t rd, uint32_t rm)
1969 {
1970     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1971     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1972     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1973     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1974     uint64_t d0 = ELEM(zd0, 0, 32) | (ELEM(zm0, 0, 32) << 32);
1975     uint64_t d1 = ELEM(zd0, 1, 32) | (ELEM(zm0, 1, 32) << 32);
1976     uint64_t m0 = ELEM(zd1, 0, 32) | (ELEM(zm1, 0, 32) << 32);
1977     uint64_t m1 = ELEM(zd1, 1, 32) | (ELEM(zm1, 1, 32) << 32);
1978     env->vfp.regs[rm] = make_float64(m0);
1979     env->vfp.regs[rm + 1] = make_float64(m1);
1980     env->vfp.regs[rd] = make_float64(d0);
1981     env->vfp.regs[rd + 1] = make_float64(d1);
1982 }
1983
1984 void HELPER(neon_zip8)(CPUARMState *env, uint32_t rd, uint32_t rm)
1985 {
1986     uint64_t zm = float64_val(env->vfp.regs[rm]);
1987     uint64_t zd = float64_val(env->vfp.regs[rd]);
1988     uint64_t d0 = ELEM(zd, 0, 8) | (ELEM(zm, 0, 8) << 8)
1989         | (ELEM(zd, 1, 8) << 16) | (ELEM(zm, 1, 8) << 24)
1990         | (ELEM(zd, 2, 8) << 32) | (ELEM(zm, 2, 8) << 40)
1991         | (ELEM(zd, 3, 8) << 48) | (ELEM(zm, 3, 8) << 56);
1992     uint64_t m0 = ELEM(zd, 4, 8) | (ELEM(zm, 4, 8) << 8)
1993         | (ELEM(zd, 5, 8) << 16) | (ELEM(zm, 5, 8) << 24)
1994         | (ELEM(zd, 6, 8) << 32) | (ELEM(zm, 6, 8) << 40)
1995         | (ELEM(zd, 7, 8) << 48) | (ELEM(zm, 7, 8) << 56);
1996     env->vfp.regs[rm] = make_float64(m0);
1997     env->vfp.regs[rd] = make_float64(d0);
1998 }
1999
2000 void HELPER(neon_zip16)(CPUARMState *env, uint32_t rd, uint32_t rm)
2001 {
2002     uint64_t zm = float64_val(env->vfp.regs[rm]);
2003     uint64_t zd = float64_val(env->vfp.regs[rd]);
2004     uint64_t d0 = ELEM(zd, 0, 16) | (ELEM(zm, 0, 16) << 16)
2005         | (ELEM(zd, 1, 16) << 32) | (ELEM(zm, 1, 16) << 48);
2006     uint64_t m0 = ELEM(zd, 2, 16) | (ELEM(zm, 2, 16) << 16)
2007         | (ELEM(zd, 3, 16) << 32) | (ELEM(zm, 3, 16) << 48);
2008     env->vfp.regs[rm] = make_float64(m0);
2009     env->vfp.regs[rd] = make_float64(d0);
2010 }