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 licenced 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))(CPUState *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)(CPUState *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)(CPUState *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)(CPUState *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)(CPUState *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)(CPUState *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)(CPUState *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)(CPUState *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)(CPUState *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 bits 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 bits inputs values is more
 551  * tricky than with 32 bits 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 bits 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 bits inputs values is more
 612  * tricky than with 32 bits 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)(CPUState *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)(CPUState *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)(CPUState *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)(CPUState *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 /* FIXME: This is wrong.  */
 792 #define NEON_FN(dest, src1, src2) do { \
 793     int8_t tmp; \
 794     tmp = (int8_t)src2; \
 795     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 796         if (src1) { \
 797             SET_QC(); \
 798             dest = ~0; \
 799         } else { \
 800             dest = 0; \
 801         } \
 802     } else if (tmp < -(ssize_t)sizeof(src1) * 8) { \
 803         dest = 0; \
 804     } else if (tmp == -(ssize_t)sizeof(src1) * 8) { \
 805         dest = src1 >> (sizeof(src1) * 8 - 1); \
 806     } else if (tmp < 0) { \
 807         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 808     } else { \
 809         dest = src1 << tmp; \
 810         if ((dest >> tmp) != src1) { \
 811             SET_QC(); \
 812             dest = ~0; \
 813         } \
 814     }} while (0)
 815 NEON_VOP_ENV(qrshl_u8, neon_u8, 4)
 816 NEON_VOP_ENV(qrshl_u16, neon_u16, 2)
 817 #undef NEON_FN
 818
 819 /* The addition of the rounding constant may overflow, so we use an
 820  * intermediate 64 bits accumulator.  */
 821 uint32_t HELPER(neon_qrshl_u32)(CPUState *env, uint32_t val, uint32_t shiftop)
 822 {
 823     uint32_t dest;
 824     int8_t shift = (int8_t)shiftop;
 825     if (shift >= 32) {
 826         if (val) {
 827             SET_QC();
 828             dest = ~0;
 829         } else {
 830             dest = 0;
 831         }
 832     } else if (shift < -32) {
 833         dest = 0;
 834     } else if (shift == -32) {
 835         dest = val >> 31;
 836     } else if (shift < 0) {
 837         uint64_t big_dest = ((uint64_t)val + (1 << (-1 - shift)));
 838         dest = big_dest >> -shift;
 839     } else {
 840         dest = val << shift;
 841         if ((dest >> shift) != val) {
 842             SET_QC();
 843             dest = ~0;
 844         }
 845     }
 846     return dest;
 847 }
 848
 849 /* Handling addition overflow with 64 bits inputs values is more
 850  * tricky than with 32 bits values.  */
 851 uint64_t HELPER(neon_qrshl_u64)(CPUState *env, uint64_t val, uint64_t shiftop)
 852 {
 853     int8_t shift = (int8_t)shiftop;
 854     if (shift >= 64) {
 855         if (val) {
 856             SET_QC();
 857             val = ~0;
 858         }
 859     } else if (shift < -64) {
 860         val = 0;
 861     } else if (shift == -64) {
 862         val >>= 63;
 863     } else if (shift < 0) {
 864         val >>= (-shift - 1);
 865         if (val == UINT64_MAX) {
 866             /* In this case, it means that the rounding constant is 1,
 867              * and the addition would overflow. Return the actual
 868              * result directly.  */
 869             val = 0x8000000000000000ULL;
 870         } else {
 871             val++;
 872             val >>= 1;
 873         }
 874     } else { \
 875         uint64_t tmp = val;
 876         val <<= shift;
 877         if ((val >> shift) != tmp) {
 878             SET_QC();
 879             val = ~0;
 880         }
 881     }
 882     return val;
 883 }
 884
 885 #define NEON_FN(dest, src1, src2) do { \
 886     int8_t tmp; \
 887     tmp = (int8_t)src2; \
 888     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 889         if (src1) { \
 890             SET_QC(); \
 891             dest = (1 << (sizeof(src1) * 8 - 1)); \
 892             if (src1 > 0) { \
 893                 dest--; \
 894             } \
 895         } else { \
 896             dest = 0; \
 897         } \
 898     } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 899         dest = 0; \
 900     } else if (tmp < 0) { \
 901         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 902     } else { \
 903         dest = src1 << tmp; \
 904         if ((dest >> tmp) != src1) { \
 905             SET_QC(); \
 906             dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
 907             if (src1 > 0) { \
 908                 dest--; \
 909             } \
 910         } \
 911     }} while (0)
 912 NEON_VOP_ENV(qrshl_s8, neon_s8, 4)
 913 NEON_VOP_ENV(qrshl_s16, neon_s16, 2)
 914 #undef NEON_FN
 915
 916 /* The addition of the rounding constant may overflow, so we use an
 917  * intermediate 64 bits accumulator.  */
 918 uint32_t HELPER(neon_qrshl_s32)(CPUState *env, uint32_t valop, uint32_t shiftop)
 919 {
 920     int32_t dest;
 921     int32_t val = (int32_t)valop;
 922     int8_t shift = (int8_t)shiftop;
 923     if (shift >= 32) {
 924         if (val) {
 925             SET_QC();
 926             dest = (val >> 31) ^ ~SIGNBIT;
 927         } else {
 928             dest = 0;
 929         }
 930     } else if (shift <= -32) {
 931         dest = 0;
 932     } else if (shift < 0) {
 933         int64_t big_dest = ((int64_t)val + (1 << (-1 - shift)));
 934         dest = big_dest >> -shift;
 935     } else {
 936         dest = val << shift;
 937         if ((dest >> shift) != val) {
 938             SET_QC();
 939             dest = (val >> 31) ^ ~SIGNBIT;
 940         }
 941     }
 942     return dest;
 943 }
 944
 945 /* Handling addition overflow with 64 bits inputs values is more
 946  * tricky than with 32 bits values.  */
 947 uint64_t HELPER(neon_qrshl_s64)(CPUState *env, uint64_t valop, uint64_t shiftop)
 948 {
 949     int8_t shift = (uint8_t)shiftop;
 950     int64_t val = valop;
 951
 952     if (shift >= 64) {
 953         if (val) {
 954             SET_QC();
 955             val = (val >> 63) ^ ~SIGNBIT64;
 956         }
 957     } else if (shift <= -64) {
 958         val = 0;
 959     } else if (shift < 0) {
 960         val >>= (-shift - 1);
 961         if (val == INT64_MAX) {
 962             /* In this case, it means that the rounding constant is 1,
 963              * and the addition would overflow. Return the actual
 964              * result directly.  */
 965             val = 0x4000000000000000ULL;
 966         } else {
 967             val++;
 968             val >>= 1;
 969         }
 970     } else {
 971         int64_t tmp = val;
 972         val <<= shift;
 973         if ((val >> shift) != tmp) {
 974             SET_QC();
 975             val = (tmp >> 63) ^ ~SIGNBIT64;
 976         }
 977     }
 978     return val;
 979 }
 980
 981 uint32_t HELPER(neon_add_u8)(uint32_t a, uint32_t b)
 982 {
 983     uint32_t mask;
 984     mask = (a ^ b) & 0x80808080u;
 985     a &= ~0x80808080u;
 986     b &= ~0x80808080u;
 987     return (a + b) ^ mask;
 988 }
 989
 990 uint32_t HELPER(neon_add_u16)(uint32_t a, uint32_t b)
 991 {
 992     uint32_t mask;
 993     mask = (a ^ b) & 0x80008000u;
 994     a &= ~0x80008000u;
 995     b &= ~0x80008000u;
 996     return (a + b) ^ mask;
 997 }
 998
 999 #define NEON_FN(dest, src1, src2) dest = src1 + src2
1000 NEON_POP(padd_u8, neon_u8, 4)
1001 NEON_POP(padd_u16, neon_u16, 2)
1002 #undef NEON_FN
1003
1004 #define NEON_FN(dest, src1, src2) dest = src1 - src2
1005 NEON_VOP(sub_u8, neon_u8, 4)
1006 NEON_VOP(sub_u16, neon_u16, 2)
1007 #undef NEON_FN
1008
1009 #define NEON_FN(dest, src1, src2) dest = src1 * src2
1010 NEON_VOP(mul_u8, neon_u8, 4)
1011 NEON_VOP(mul_u16, neon_u16, 2)
1012 #undef NEON_FN
1013
1014 /* Polynomial multiplication is like integer multiplication except the
1015    partial products are XORed, not added.  */
1016 uint32_t HELPER(neon_mul_p8)(uint32_t op1, uint32_t op2)
1017 {
1018     uint32_t mask;
1019     uint32_t result;
1020     result = 0;
1021     while (op1) {
1022         mask = 0;
1023         if (op1 & 1)
1024             mask |= 0xff;
1025         if (op1 & (1 << 8))
1026             mask |= (0xff << 8);
1027         if (op1 & (1 << 16))
1028             mask |= (0xff << 16);
1029         if (op1 & (1 << 24))
1030             mask |= (0xff << 24);
1031         result ^= op2 & mask;
1032         op1 = (op1 >> 1) & 0x7f7f7f7f;
1033         op2 = (op2 << 1) & 0xfefefefe;
1034     }
1035     return result;
1036 }
1037
1038 uint64_t HELPER(neon_mull_p8)(uint32_t op1, uint32_t op2)
1039 {
1040     uint64_t result = 0;
1041     uint64_t mask;
1042     uint64_t op2ex = op2;
1043     op2ex = (op2ex & 0xff) |
1044         ((op2ex & 0xff00) << 8) |
1045         ((op2ex & 0xff0000) << 16) |
1046         ((op2ex & 0xff000000) << 24);
1047     while (op1) {
1048         mask = 0;
1049         if (op1 & 1) {
1050             mask |= 0xffff;
1051         }
1052         if (op1 & (1 << 8)) {
1053             mask |= (0xffffU << 16);
1054         }
1055         if (op1 & (1 << 16)) {
1056             mask |= (0xffffULL << 32);
1057         }
1058         if (op1 & (1 << 24)) {
1059             mask |= (0xffffULL << 48);
1060         }
1061         result ^= op2ex & mask;
1062         op1 = (op1 >> 1) & 0x7f7f7f7f;
1063         op2ex <<= 1;
1064     }
1065     return result;
1066 }
1067
1068 #define NEON_FN(dest, src1, src2) dest = (src1 & src2) ? -1 : 0
1069 NEON_VOP(tst_u8, neon_u8, 4)
1070 NEON_VOP(tst_u16, neon_u16, 2)
1071 NEON_VOP(tst_u32, neon_u32, 1)
1072 #undef NEON_FN
1073
1074 #define NEON_FN(dest, src1, src2) dest = (src1 == src2) ? -1 : 0
1075 NEON_VOP(ceq_u8, neon_u8, 4)
1076 NEON_VOP(ceq_u16, neon_u16, 2)
1077 NEON_VOP(ceq_u32, neon_u32, 1)
1078 #undef NEON_FN
1079
1080 #define NEON_FN(dest, src, dummy) dest = (src < 0) ? -src : src
1081 NEON_VOP1(abs_s8, neon_s8, 4)
1082 NEON_VOP1(abs_s16, neon_s16, 2)
1083 #undef NEON_FN
1084
1085 /* Count Leading Sign/Zero Bits.  */
1086 static inline int do_clz8(uint8_t x)
1087 {
1088     int n;
1089     for (n = 8; x; n--)
1090         x >>= 1;
1091     return n;
1092 }
1093
1094 static inline int do_clz16(uint16_t x)
1095 {
1096     int n;
1097     for (n = 16; x; n--)
1098         x >>= 1;
1099     return n;
1100 }
1101
1102 #define NEON_FN(dest, src, dummy) dest = do_clz8(src)
1103 NEON_VOP1(clz_u8, neon_u8, 4)
1104 #undef NEON_FN
1105
1106 #define NEON_FN(dest, src, dummy) dest = do_clz16(src)
1107 NEON_VOP1(clz_u16, neon_u16, 2)
1108 #undef NEON_FN
1109
1110 #define NEON_FN(dest, src, dummy) dest = do_clz8((src < 0) ? ~src : src) - 1
1111 NEON_VOP1(cls_s8, neon_s8, 4)
1112 #undef NEON_FN
1113
1114 #define NEON_FN(dest, src, dummy) dest = do_clz16((src < 0) ? ~src : src) - 1
1115 NEON_VOP1(cls_s16, neon_s16, 2)
1116 #undef NEON_FN
1117
1118 uint32_t HELPER(neon_cls_s32)(uint32_t x)
1119 {
1120     int count;
1121     if ((int32_t)x < 0)
1122         x = ~x;
1123     for (count = 32; x; count--)
1124         x = x >> 1;
1125     return count - 1;
1126 }
1127
1128 /* Bit count.  */
1129 uint32_t HELPER(neon_cnt_u8)(uint32_t x)
1130 {
1131     x = (x & 0x55555555) + ((x >>  1) & 0x55555555);
1132     x = (x & 0x33333333) + ((x >>  2) & 0x33333333);
1133     x = (x & 0x0f0f0f0f) + ((x >>  4) & 0x0f0f0f0f);
1134     return x;
1135 }
1136
1137 #define NEON_QDMULH16(dest, src1, src2, round) do { \
1138     uint32_t tmp = (int32_t)(int16_t) src1 * (int16_t) src2; \
1139     if ((tmp ^ (tmp << 1)) & SIGNBIT) { \
1140         SET_QC(); \
1141         tmp = (tmp >> 31) ^ ~SIGNBIT; \
1142     } else { \
1143         tmp <<= 1; \
1144     } \
1145     if (round) { \
1146         int32_t old = tmp; \
1147         tmp += 1 << 15; \
1148         if ((int32_t)tmp < old) { \
1149             SET_QC(); \
1150             tmp = SIGNBIT - 1; \
1151         } \
1152     } \
1153     dest = tmp >> 16; \
1154     } while(0)
1155 #define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 0)
1156 NEON_VOP_ENV(qdmulh_s16, neon_s16, 2)
1157 #undef NEON_FN
1158 #define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 1)
1159 NEON_VOP_ENV(qrdmulh_s16, neon_s16, 2)
1160 #undef NEON_FN
1161 #undef NEON_QDMULH16
1162
1163 #define NEON_QDMULH32(dest, src1, src2, round) do { \
1164     uint64_t tmp = (int64_t)(int32_t) src1 * (int32_t) src2; \
1165     if ((tmp ^ (tmp << 1)) & SIGNBIT64) { \
1166         SET_QC(); \
1167         tmp = (tmp >> 63) ^ ~SIGNBIT64; \
1168     } else { \
1169         tmp <<= 1; \
1170     } \
1171     if (round) { \
1172         int64_t old = tmp; \
1173         tmp += (int64_t)1 << 31; \
1174         if ((int64_t)tmp < old) { \
1175             SET_QC(); \
1176             tmp = SIGNBIT64 - 1; \
1177         } \
1178     } \
1179     dest = tmp >> 32; \
1180     } while(0)
1181 #define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 0)
1182 NEON_VOP_ENV(qdmulh_s32, neon_s32, 1)
1183 #undef NEON_FN
1184 #define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 1)
1185 NEON_VOP_ENV(qrdmulh_s32, neon_s32, 1)
1186 #undef NEON_FN
1187 #undef NEON_QDMULH32
1188
1189 uint32_t HELPER(neon_narrow_u8)(uint64_t x)
1190 {
1191     return (x & 0xffu) | ((x >> 8) & 0xff00u) | ((x >> 16) & 0xff0000u)
1192            | ((x >> 24) & 0xff000000u);
1193 }
1194
1195 uint32_t HELPER(neon_narrow_u16)(uint64_t x)
1196 {
1197     return (x & 0xffffu) | ((x >> 16) & 0xffff0000u);
1198 }
1199
1200 uint32_t HELPER(neon_narrow_high_u8)(uint64_t x)
1201 {
1202     return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00)
1203             | ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000);
1204 }
1205
1206 uint32_t HELPER(neon_narrow_high_u16)(uint64_t x)
1207 {
1208     return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000);
1209 }
1210
1211 uint32_t HELPER(neon_narrow_round_high_u8)(uint64_t x)
1212 {
1213     x &= 0xff80ff80ff80ff80ull;
1214     x += 0x0080008000800080ull;
1215     return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00)
1216             | ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000);
1217 }
1218
1219 uint32_t HELPER(neon_narrow_round_high_u16)(uint64_t x)
1220 {
1221     x &= 0xffff8000ffff8000ull;
1222     x += 0x0000800000008000ull;
1223     return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000);
1224 }
1225
1226 uint32_t HELPER(neon_unarrow_sat8)(CPUState *env, uint64_t x)
1227 {
1228     uint16_t s;
1229     uint8_t d;
1230     uint32_t res = 0;
1231 #define SAT8(n) \
1232     s = x >> n; \
1233     if (s & 0x8000) { \
1234         SET_QC(); \
1235     } else { \
1236         if (s > 0xff) { \
1237             d = 0xff; \
1238             SET_QC(); \
1239         } else  { \
1240             d = s; \
1241         } \
1242         res |= (uint32_t)d << (n / 2); \
1243     }
1244
1245     SAT8(0);
1246     SAT8(16);
1247     SAT8(32);
1248     SAT8(48);
1249 #undef SAT8
1250     return res;
1251 }
1252
1253 uint32_t HELPER(neon_narrow_sat_u8)(CPUState *env, uint64_t x)
1254 {
1255     uint16_t s;
1256     uint8_t d;
1257     uint32_t res = 0;
1258 #define SAT8(n) \
1259     s = x >> n; \
1260     if (s > 0xff) { \
1261         d = 0xff; \
1262         SET_QC(); \
1263     } else  { \
1264         d = s; \
1265     } \
1266     res |= (uint32_t)d << (n / 2);
1267
1268     SAT8(0);
1269     SAT8(16);
1270     SAT8(32);
1271     SAT8(48);
1272 #undef SAT8
1273     return res;
1274 }
1275
1276 uint32_t HELPER(neon_narrow_sat_s8)(CPUState *env, uint64_t x)
1277 {
1278     int16_t s;
1279     uint8_t d;
1280     uint32_t res = 0;
1281 #define SAT8(n) \
1282     s = x >> n; \
1283     if (s != (int8_t)s) { \
1284         d = (s >> 15) ^ 0x7f; \
1285         SET_QC(); \
1286     } else  { \
1287         d = s; \
1288     } \
1289     res |= (uint32_t)d << (n / 2);
1290
1291     SAT8(0);
1292     SAT8(16);
1293     SAT8(32);
1294     SAT8(48);
1295 #undef SAT8
1296     return res;
1297 }
1298
1299 uint32_t HELPER(neon_unarrow_sat16)(CPUState *env, uint64_t x)
1300 {
1301     uint32_t high;
1302     uint32_t low;
1303     low = x;
1304     if (low & 0x80000000) {
1305         low = 0;
1306         SET_QC();
1307     } else if (low > 0xffff) {
1308         low = 0xffff;
1309         SET_QC();
1310     }
1311     high = x >> 32;
1312     if (high & 0x80000000) {
1313         high = 0;
1314         SET_QC();
1315     } else if (high > 0xffff) {
1316         high = 0xffff;
1317         SET_QC();
1318     }
1319     return low | (high << 16);
1320 }
1321
1322 uint32_t HELPER(neon_narrow_sat_u16)(CPUState *env, uint64_t x)
1323 {
1324     uint32_t high;
1325     uint32_t low;
1326     low = x;
1327     if (low > 0xffff) {
1328         low = 0xffff;
1329         SET_QC();
1330     }
1331     high = x >> 32;
1332     if (high > 0xffff) {
1333         high = 0xffff;
1334         SET_QC();
1335     }
1336     return low | (high << 16);
1337 }
1338
1339 uint32_t HELPER(neon_narrow_sat_s16)(CPUState *env, uint64_t x)
1340 {
1341     int32_t low;
1342     int32_t high;
1343     low = x;
1344     if (low != (int16_t)low) {
1345         low = (low >> 31) ^ 0x7fff;
1346         SET_QC();
1347     }
1348     high = x >> 32;
1349     if (high != (int16_t)high) {
1350         high = (high >> 31) ^ 0x7fff;
1351         SET_QC();
1352     }
1353     return (uint16_t)low | (high << 16);
1354 }
1355
1356 uint32_t HELPER(neon_unarrow_sat32)(CPUState *env, uint64_t x)
1357 {
1358     if (x & 0x8000000000000000ull) {
1359         SET_QC();
1360         return 0;
1361     }
1362     if (x > 0xffffffffu) {
1363         SET_QC();
1364         return 0xffffffffu;
1365     }
1366     return x;
1367 }
1368
1369 uint32_t HELPER(neon_narrow_sat_u32)(CPUState *env, uint64_t x)
1370 {
1371     if (x > 0xffffffffu) {
1372         SET_QC();
1373         return 0xffffffffu;
1374     }
1375     return x;
1376 }
1377
1378 uint32_t HELPER(neon_narrow_sat_s32)(CPUState *env, uint64_t x)
1379 {
1380     if ((int64_t)x != (int32_t)x) {
1381         SET_QC();
1382         return ((int64_t)x >> 63) ^ 0x7fffffff;
1383     }
1384     return x;
1385 }
1386
1387 uint64_t HELPER(neon_widen_u8)(uint32_t x)
1388 {
1389     uint64_t tmp;
1390     uint64_t ret;
1391     ret = (uint8_t)x;
1392     tmp = (uint8_t)(x >> 8);
1393     ret |= tmp << 16;
1394     tmp = (uint8_t)(x >> 16);
1395     ret |= tmp << 32;
1396     tmp = (uint8_t)(x >> 24);
1397     ret |= tmp << 48;
1398     return ret;
1399 }
1400
1401 uint64_t HELPER(neon_widen_s8)(uint32_t x)
1402 {
1403     uint64_t tmp;
1404     uint64_t ret;
1405     ret = (uint16_t)(int8_t)x;
1406     tmp = (uint16_t)(int8_t)(x >> 8);
1407     ret |= tmp << 16;
1408     tmp = (uint16_t)(int8_t)(x >> 16);
1409     ret |= tmp << 32;
1410     tmp = (uint16_t)(int8_t)(x >> 24);
1411     ret |= tmp << 48;
1412     return ret;
1413 }
1414
1415 uint64_t HELPER(neon_widen_u16)(uint32_t x)
1416 {
1417     uint64_t high = (uint16_t)(x >> 16);
1418     return ((uint16_t)x) | (high << 32);
1419 }
1420
1421 uint64_t HELPER(neon_widen_s16)(uint32_t x)
1422 {
1423     uint64_t high = (int16_t)(x >> 16);
1424     return ((uint32_t)(int16_t)x) | (high << 32);
1425 }
1426
1427 uint64_t HELPER(neon_addl_u16)(uint64_t a, uint64_t b)
1428 {
1429     uint64_t mask;
1430     mask = (a ^ b) & 0x8000800080008000ull;
1431     a &= ~0x8000800080008000ull;
1432     b &= ~0x8000800080008000ull;
1433     return (a + b) ^ mask;
1434 }
1435
1436 uint64_t HELPER(neon_addl_u32)(uint64_t a, uint64_t b)
1437 {
1438     uint64_t mask;
1439     mask = (a ^ b) & 0x8000000080000000ull;
1440     a &= ~0x8000000080000000ull;
1441     b &= ~0x8000000080000000ull;
1442     return (a + b) ^ mask;
1443 }
1444
1445 uint64_t HELPER(neon_paddl_u16)(uint64_t a, uint64_t b)
1446 {
1447     uint64_t tmp;
1448     uint64_t tmp2;
1449
1450     tmp = a & 0x0000ffff0000ffffull;
1451     tmp += (a >> 16) & 0x0000ffff0000ffffull;
1452     tmp2 = b & 0xffff0000ffff0000ull;
1453     tmp2 += (b << 16) & 0xffff0000ffff0000ull;
1454     return    ( tmp         & 0xffff)
1455             | ((tmp  >> 16) & 0xffff0000ull)
1456             | ((tmp2 << 16) & 0xffff00000000ull)
1457             | ( tmp2        & 0xffff000000000000ull);
1458 }
1459
1460 uint64_t HELPER(neon_paddl_u32)(uint64_t a, uint64_t b)
1461 {
1462     uint32_t low = a + (a >> 32);
1463     uint32_t high = b + (b >> 32);
1464     return low + ((uint64_t)high << 32);
1465 }
1466
1467 uint64_t HELPER(neon_subl_u16)(uint64_t a, uint64_t b)
1468 {
1469     uint64_t mask;
1470     mask = (a ^ ~b) & 0x8000800080008000ull;
1471     a |= 0x8000800080008000ull;
1472     b &= ~0x8000800080008000ull;
1473     return (a - b) ^ mask;
1474 }
1475
1476 uint64_t HELPER(neon_subl_u32)(uint64_t a, uint64_t b)
1477 {
1478     uint64_t mask;
1479     mask = (a ^ ~b) & 0x8000000080000000ull;
1480     a |= 0x8000000080000000ull;
1481     b &= ~0x8000000080000000ull;
1482     return (a - b) ^ mask;
1483 }
1484
1485 uint64_t HELPER(neon_addl_saturate_s32)(CPUState *env, uint64_t a, uint64_t b)
1486 {
1487     uint32_t x, y;
1488     uint32_t low, high;
1489
1490     x = a;
1491     y = b;
1492     low = x + y;
1493     if (((low ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) {
1494         SET_QC();
1495         low = ((int32_t)x >> 31) ^ ~SIGNBIT;
1496     }
1497     x = a >> 32;
1498     y = b >> 32;
1499     high = x + y;
1500     if (((high ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) {
1501         SET_QC();
1502         high = ((int32_t)x >> 31) ^ ~SIGNBIT;
1503     }
1504     return low | ((uint64_t)high << 32);
1505 }
1506
1507 uint64_t HELPER(neon_addl_saturate_s64)(CPUState *env, uint64_t a, uint64_t b)
1508 {
1509     uint64_t result;
1510
1511     result = a + b;
1512     if (((result ^ a) & SIGNBIT64) && !((a ^ b) & SIGNBIT64)) {
1513         SET_QC();
1514         result = ((int64_t)a >> 63) ^ ~SIGNBIT64;
1515     }
1516     return result;
1517 }
1518
1519 /* We have to do the arithmetic in a larger type than
1520  * the input type, because for example with a signed 32 bit
1521  * op the absolute difference can overflow a signed 32 bit value.
1522  */
1523 #define DO_ABD(dest, x, y, intype, arithtype) do {            \
1524     arithtype tmp_x = (intype)(x);                            \
1525     arithtype tmp_y = (intype)(y);                            \
1526     dest = ((tmp_x > tmp_y) ? tmp_x - tmp_y : tmp_y - tmp_x); \
1527     } while(0)
1528
1529 uint64_t HELPER(neon_abdl_u16)(uint32_t a, uint32_t b)
1530 {
1531     uint64_t tmp;
1532     uint64_t result;
1533     DO_ABD(result, a, b, uint8_t, uint32_t);
1534     DO_ABD(tmp, a >> 8, b >> 8, uint8_t, uint32_t);
1535     result |= tmp << 16;
1536     DO_ABD(tmp, a >> 16, b >> 16, uint8_t, uint32_t);
1537     result |= tmp << 32;
1538     DO_ABD(tmp, a >> 24, b >> 24, uint8_t, uint32_t);
1539     result |= tmp << 48;
1540     return result;
1541 }
1542
1543 uint64_t HELPER(neon_abdl_s16)(uint32_t a, uint32_t b)
1544 {
1545     uint64_t tmp;
1546     uint64_t result;
1547     DO_ABD(result, a, b, int8_t, int32_t);
1548     DO_ABD(tmp, a >> 8, b >> 8, int8_t, int32_t);
1549     result |= tmp << 16;
1550     DO_ABD(tmp, a >> 16, b >> 16, int8_t, int32_t);
1551     result |= tmp << 32;
1552     DO_ABD(tmp, a >> 24, b >> 24, int8_t, int32_t);
1553     result |= tmp << 48;
1554     return result;
1555 }
1556
1557 uint64_t HELPER(neon_abdl_u32)(uint32_t a, uint32_t b)
1558 {
1559     uint64_t tmp;
1560     uint64_t result;
1561     DO_ABD(result, a, b, uint16_t, uint32_t);
1562     DO_ABD(tmp, a >> 16, b >> 16, uint16_t, uint32_t);
1563     return result | (tmp << 32);
1564 }
1565
1566 uint64_t HELPER(neon_abdl_s32)(uint32_t a, uint32_t b)
1567 {
1568     uint64_t tmp;
1569     uint64_t result;
1570     DO_ABD(result, a, b, int16_t, int32_t);
1571     DO_ABD(tmp, a >> 16, b >> 16, int16_t, int32_t);
1572     return result | (tmp << 32);
1573 }
1574
1575 uint64_t HELPER(neon_abdl_u64)(uint32_t a, uint32_t b)
1576 {
1577     uint64_t result;
1578     DO_ABD(result, a, b, uint32_t, uint64_t);
1579     return result;
1580 }
1581
1582 uint64_t HELPER(neon_abdl_s64)(uint32_t a, uint32_t b)
1583 {
1584     uint64_t result;
1585     DO_ABD(result, a, b, int32_t, int64_t);
1586     return result;
1587 }
1588 #undef DO_ABD
1589
1590 /* Widening multiply. Named type is the source type.  */
1591 #define DO_MULL(dest, x, y, type1, type2) do { \
1592     type1 tmp_x = x; \
1593     type1 tmp_y = y; \
1594     dest = (type2)((type2)tmp_x * (type2)tmp_y); \
1595     } while(0)
1596
1597 uint64_t HELPER(neon_mull_u8)(uint32_t a, uint32_t b)
1598 {
1599     uint64_t tmp;
1600     uint64_t result;
1601
1602     DO_MULL(result, a, b, uint8_t, uint16_t);
1603     DO_MULL(tmp, a >> 8, b >> 8, uint8_t, uint16_t);
1604     result |= tmp << 16;
1605     DO_MULL(tmp, a >> 16, b >> 16, uint8_t, uint16_t);
1606     result |= tmp << 32;
1607     DO_MULL(tmp, a >> 24, b >> 24, uint8_t, uint16_t);
1608     result |= tmp << 48;
1609     return result;
1610 }
1611
1612 uint64_t HELPER(neon_mull_s8)(uint32_t a, uint32_t b)
1613 {
1614     uint64_t tmp;
1615     uint64_t result;
1616
1617     DO_MULL(result, a, b, int8_t, uint16_t);
1618     DO_MULL(tmp, a >> 8, b >> 8, int8_t, uint16_t);
1619     result |= tmp << 16;
1620     DO_MULL(tmp, a >> 16, b >> 16, int8_t, uint16_t);
1621     result |= tmp << 32;
1622     DO_MULL(tmp, a >> 24, b >> 24, int8_t, uint16_t);
1623     result |= tmp << 48;
1624     return result;
1625 }
1626
1627 uint64_t HELPER(neon_mull_u16)(uint32_t a, uint32_t b)
1628 {
1629     uint64_t tmp;
1630     uint64_t result;
1631
1632     DO_MULL(result, a, b, uint16_t, uint32_t);
1633     DO_MULL(tmp, a >> 16, b >> 16, uint16_t, uint32_t);
1634     return result | (tmp << 32);
1635 }
1636
1637 uint64_t HELPER(neon_mull_s16)(uint32_t a, uint32_t b)
1638 {
1639     uint64_t tmp;
1640     uint64_t result;
1641
1642     DO_MULL(result, a, b, int16_t, uint32_t);
1643     DO_MULL(tmp, a >> 16, b >> 16, int16_t, uint32_t);
1644     return result | (tmp << 32);
1645 }
1646
1647 uint64_t HELPER(neon_negl_u16)(uint64_t x)
1648 {
1649     uint16_t tmp;
1650     uint64_t result;
1651     result = (uint16_t)-x;
1652     tmp = -(x >> 16);
1653     result |= (uint64_t)tmp << 16;
1654     tmp = -(x >> 32);
1655     result |= (uint64_t)tmp << 32;
1656     tmp = -(x >> 48);
1657     result |= (uint64_t)tmp << 48;
1658     return result;
1659 }
1660
1661 uint64_t HELPER(neon_negl_u32)(uint64_t x)
1662 {
1663     uint32_t low = -x;
1664     uint32_t high = -(x >> 32);
1665     return low | ((uint64_t)high << 32);
1666 }
1667
1668 /* FIXME:  There should be a native op for this.  */
1669 uint64_t HELPER(neon_negl_u64)(uint64_t x)
1670 {
1671     return -x;
1672 }
1673
1674 /* Saturnating sign manuipulation.  */
1675 /* ??? Make these use NEON_VOP1 */
1676 #define DO_QABS8(x) do { \
1677     if (x == (int8_t)0x80) { \
1678         x = 0x7f; \
1679         SET_QC(); \
1680     } else if (x < 0) { \
1681         x = -x; \
1682     }} while (0)
1683 uint32_t HELPER(neon_qabs_s8)(CPUState *env, uint32_t x)
1684 {
1685     neon_s8 vec;
1686     NEON_UNPACK(neon_s8, vec, x);
1687     DO_QABS8(vec.v1);
1688     DO_QABS8(vec.v2);
1689     DO_QABS8(vec.v3);
1690     DO_QABS8(vec.v4);
1691     NEON_PACK(neon_s8, x, vec);
1692     return x;
1693 }
1694 #undef DO_QABS8
1695
1696 #define DO_QNEG8(x) do { \
1697     if (x == (int8_t)0x80) { \
1698         x = 0x7f; \
1699         SET_QC(); \
1700     } else { \
1701         x = -x; \
1702     }} while (0)
1703 uint32_t HELPER(neon_qneg_s8)(CPUState *env, uint32_t x)
1704 {
1705     neon_s8 vec;
1706     NEON_UNPACK(neon_s8, vec, x);
1707     DO_QNEG8(vec.v1);
1708     DO_QNEG8(vec.v2);
1709     DO_QNEG8(vec.v3);
1710     DO_QNEG8(vec.v4);
1711     NEON_PACK(neon_s8, x, vec);
1712     return x;
1713 }
1714 #undef DO_QNEG8
1715
1716 #define DO_QABS16(x) do { \
1717     if (x == (int16_t)0x8000) { \
1718         x = 0x7fff; \
1719         SET_QC(); \
1720     } else if (x < 0) { \
1721         x = -x; \
1722     }} while (0)
1723 uint32_t HELPER(neon_qabs_s16)(CPUState *env, uint32_t x)
1724 {
1725     neon_s16 vec;
1726     NEON_UNPACK(neon_s16, vec, x);
1727     DO_QABS16(vec.v1);
1728     DO_QABS16(vec.v2);
1729     NEON_PACK(neon_s16, x, vec);
1730     return x;
1731 }
1732 #undef DO_QABS16
1733
1734 #define DO_QNEG16(x) do { \
1735     if (x == (int16_t)0x8000) { \
1736         x = 0x7fff; \
1737         SET_QC(); \
1738     } else { \
1739         x = -x; \
1740     }} while (0)
1741 uint32_t HELPER(neon_qneg_s16)(CPUState *env, uint32_t x)
1742 {
1743     neon_s16 vec;
1744     NEON_UNPACK(neon_s16, vec, x);
1745     DO_QNEG16(vec.v1);
1746     DO_QNEG16(vec.v2);
1747     NEON_PACK(neon_s16, x, vec);
1748     return x;
1749 }
1750 #undef DO_QNEG16
1751
1752 uint32_t HELPER(neon_qabs_s32)(CPUState *env, uint32_t x)
1753 {
1754     if (x == SIGNBIT) {
1755         SET_QC();
1756         x = ~SIGNBIT;
1757     } else if ((int32_t)x < 0) {
1758         x = -x;
1759     }
1760     return x;
1761 }
1762
1763 uint32_t HELPER(neon_qneg_s32)(CPUState *env, uint32_t x)
1764 {
1765     if (x == SIGNBIT) {
1766         SET_QC();
1767         x = ~SIGNBIT;
1768     } else {
1769         x = -x;
1770     }
1771     return x;
1772 }
1773
1774 /* NEON Float helpers.  */
1775 uint32_t HELPER(neon_min_f32)(uint32_t a, uint32_t b, void *fpstp)
1776 {
1777     float_status *fpst = fpstp;
1778     return float32_val(float32_min(make_float32(a), make_float32(b), fpst));
1779 }
1780
1781 uint32_t HELPER(neon_max_f32)(uint32_t a, uint32_t b, void *fpstp)
1782 {
1783     float_status *fpst = fpstp;
1784     return float32_val(float32_max(make_float32(a), make_float32(b), fpst));
1785 }
1786
1787 uint32_t HELPER(neon_abd_f32)(uint32_t a, uint32_t b, void *fpstp)
1788 {
1789     float_status *fpst = fpstp;
1790     float32 f0 = make_float32(a);
1791     float32 f1 = make_float32(b);
1792     return float32_val(float32_abs(float32_sub(f0, f1, fpst)));
1793 }
1794
1795 /* Floating point comparisons produce an integer result.
1796  * Note that EQ doesn't signal InvalidOp for QNaNs but GE and GT do.
1797  * Softfloat routines return 0/1, which we convert to the 0/-1 Neon requires.
1798  */
1799 uint32_t HELPER(neon_ceq_f32)(uint32_t a, uint32_t b, void *fpstp)
1800 {
1801     float_status *fpst = fpstp;
1802     return -float32_eq_quiet(make_float32(a), make_float32(b), fpst);
1803 }
1804
1805 uint32_t HELPER(neon_cge_f32)(uint32_t a, uint32_t b, void *fpstp)
1806 {
1807     float_status *fpst = fpstp;
1808     return -float32_le(make_float32(b), make_float32(a), fpst);
1809 }
1810
1811 uint32_t HELPER(neon_cgt_f32)(uint32_t a, uint32_t b, void *fpstp)
1812 {
1813     float_status *fpst = fpstp;
1814     return -float32_lt(make_float32(b), make_float32(a), fpst);
1815 }
1816
1817 uint32_t HELPER(neon_acge_f32)(uint32_t a, uint32_t b, void *fpstp)
1818 {
1819     float_status *fpst = fpstp;
1820     float32 f0 = float32_abs(make_float32(a));
1821     float32 f1 = float32_abs(make_float32(b));
1822     return -float32_le(f1, f0, fpst);
1823 }
1824
1825 uint32_t HELPER(neon_acgt_f32)(uint32_t a, uint32_t b, void *fpstp)
1826 {
1827     float_status *fpst = fpstp;
1828     float32 f0 = float32_abs(make_float32(a));
1829     float32 f1 = float32_abs(make_float32(b));
1830     return -float32_lt(f1, f0, fpst);
1831 }
1832
1833 #define ELEM(V, N, SIZE) (((V) >> ((N) * (SIZE))) & ((1ull << (SIZE)) - 1))
1834
1835 void HELPER(neon_qunzip8)(CPUState *env, uint32_t rd, uint32_t rm)
1836 {
1837     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1838     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1839     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1840     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1841     uint64_t d0 = ELEM(zd0, 0, 8) | (ELEM(zd0, 2, 8) << 8)
1842         | (ELEM(zd0, 4, 8) << 16) | (ELEM(zd0, 6, 8) << 24)
1843         | (ELEM(zd1, 0, 8) << 32) | (ELEM(zd1, 2, 8) << 40)
1844         | (ELEM(zd1, 4, 8) << 48) | (ELEM(zd1, 6, 8) << 56);
1845     uint64_t d1 = ELEM(zm0, 0, 8) | (ELEM(zm0, 2, 8) << 8)
1846         | (ELEM(zm0, 4, 8) << 16) | (ELEM(zm0, 6, 8) << 24)
1847         | (ELEM(zm1, 0, 8) << 32) | (ELEM(zm1, 2, 8) << 40)
1848         | (ELEM(zm1, 4, 8) << 48) | (ELEM(zm1, 6, 8) << 56);
1849     uint64_t m0 = ELEM(zd0, 1, 8) | (ELEM(zd0, 3, 8) << 8)
1850         | (ELEM(zd0, 5, 8) << 16) | (ELEM(zd0, 7, 8) << 24)
1851         | (ELEM(zd1, 1, 8) << 32) | (ELEM(zd1, 3, 8) << 40)
1852         | (ELEM(zd1, 5, 8) << 48) | (ELEM(zd1, 7, 8) << 56);
1853     uint64_t m1 = ELEM(zm0, 1, 8) | (ELEM(zm0, 3, 8) << 8)
1854         | (ELEM(zm0, 5, 8) << 16) | (ELEM(zm0, 7, 8) << 24)
1855         | (ELEM(zm1, 1, 8) << 32) | (ELEM(zm1, 3, 8) << 40)
1856         | (ELEM(zm1, 5, 8) << 48) | (ELEM(zm1, 7, 8) << 56);
1857     env->vfp.regs[rm] = make_float64(m0);
1858     env->vfp.regs[rm + 1] = make_float64(m1);
1859     env->vfp.regs[rd] = make_float64(d0);
1860     env->vfp.regs[rd + 1] = make_float64(d1);
1861 }
1862
1863 void HELPER(neon_qunzip16)(CPUState *env, uint32_t rd, uint32_t rm)
1864 {
1865     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1866     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1867     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1868     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1869     uint64_t d0 = ELEM(zd0, 0, 16) | (ELEM(zd0, 2, 16) << 16)
1870         | (ELEM(zd1, 0, 16) << 32) | (ELEM(zd1, 2, 16) << 48);
1871     uint64_t d1 = ELEM(zm0, 0, 16) | (ELEM(zm0, 2, 16) << 16)
1872         | (ELEM(zm1, 0, 16) << 32) | (ELEM(zm1, 2, 16) << 48);
1873     uint64_t m0 = ELEM(zd0, 1, 16) | (ELEM(zd0, 3, 16) << 16)
1874         | (ELEM(zd1, 1, 16) << 32) | (ELEM(zd1, 3, 16) << 48);
1875     uint64_t m1 = ELEM(zm0, 1, 16) | (ELEM(zm0, 3, 16) << 16)
1876         | (ELEM(zm1, 1, 16) << 32) | (ELEM(zm1, 3, 16) << 48);
1877     env->vfp.regs[rm] = make_float64(m0);
1878     env->vfp.regs[rm + 1] = make_float64(m1);
1879     env->vfp.regs[rd] = make_float64(d0);
1880     env->vfp.regs[rd + 1] = make_float64(d1);
1881 }
1882
1883 void HELPER(neon_qunzip32)(CPUState *env, uint32_t rd, uint32_t rm)
1884 {
1885     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1886     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1887     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1888     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1889     uint64_t d0 = ELEM(zd0, 0, 32) | (ELEM(zd1, 0, 32) << 32);
1890     uint64_t d1 = ELEM(zm0, 0, 32) | (ELEM(zm1, 0, 32) << 32);
1891     uint64_t m0 = ELEM(zd0, 1, 32) | (ELEM(zd1, 1, 32) << 32);
1892     uint64_t m1 = ELEM(zm0, 1, 32) | (ELEM(zm1, 1, 32) << 32);
1893     env->vfp.regs[rm] = make_float64(m0);
1894     env->vfp.regs[rm + 1] = make_float64(m1);
1895     env->vfp.regs[rd] = make_float64(d0);
1896     env->vfp.regs[rd + 1] = make_float64(d1);
1897 }
1898
1899 void HELPER(neon_unzip8)(CPUState *env, uint32_t rd, uint32_t rm)
1900 {
1901     uint64_t zm = float64_val(env->vfp.regs[rm]);
1902     uint64_t zd = float64_val(env->vfp.regs[rd]);
1903     uint64_t d0 = ELEM(zd, 0, 8) | (ELEM(zd, 2, 8) << 8)
1904         | (ELEM(zd, 4, 8) << 16) | (ELEM(zd, 6, 8) << 24)
1905         | (ELEM(zm, 0, 8) << 32) | (ELEM(zm, 2, 8) << 40)
1906         | (ELEM(zm, 4, 8) << 48) | (ELEM(zm, 6, 8) << 56);
1907     uint64_t m0 = ELEM(zd, 1, 8) | (ELEM(zd, 3, 8) << 8)
1908         | (ELEM(zd, 5, 8) << 16) | (ELEM(zd, 7, 8) << 24)
1909         | (ELEM(zm, 1, 8) << 32) | (ELEM(zm, 3, 8) << 40)
1910         | (ELEM(zm, 5, 8) << 48) | (ELEM(zm, 7, 8) << 56);
1911     env->vfp.regs[rm] = make_float64(m0);
1912     env->vfp.regs[rd] = make_float64(d0);
1913 }
1914
1915 void HELPER(neon_unzip16)(CPUState *env, uint32_t rd, uint32_t rm)
1916 {
1917     uint64_t zm = float64_val(env->vfp.regs[rm]);
1918     uint64_t zd = float64_val(env->vfp.regs[rd]);
1919     uint64_t d0 = ELEM(zd, 0, 16) | (ELEM(zd, 2, 16) << 16)
1920         | (ELEM(zm, 0, 16) << 32) | (ELEM(zm, 2, 16) << 48);
1921     uint64_t m0 = ELEM(zd, 1, 16) | (ELEM(zd, 3, 16) << 16)
1922         | (ELEM(zm, 1, 16) << 32) | (ELEM(zm, 3, 16) << 48);
1923     env->vfp.regs[rm] = make_float64(m0);
1924     env->vfp.regs[rd] = make_float64(d0);
1925 }
1926
1927 void HELPER(neon_qzip8)(CPUState *env, uint32_t rd, uint32_t rm)
1928 {
1929     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1930     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1931     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1932     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1933     uint64_t d0 = ELEM(zd0, 0, 8) | (ELEM(zm0, 0, 8) << 8)
1934         | (ELEM(zd0, 1, 8) << 16) | (ELEM(zm0, 1, 8) << 24)
1935         | (ELEM(zd0, 2, 8) << 32) | (ELEM(zm0, 2, 8) << 40)
1936         | (ELEM(zd0, 3, 8) << 48) | (ELEM(zm0, 3, 8) << 56);
1937     uint64_t d1 = ELEM(zd0, 4, 8) | (ELEM(zm0, 4, 8) << 8)
1938         | (ELEM(zd0, 5, 8) << 16) | (ELEM(zm0, 5, 8) << 24)
1939         | (ELEM(zd0, 6, 8) << 32) | (ELEM(zm0, 6, 8) << 40)
1940         | (ELEM(zd0, 7, 8) << 48) | (ELEM(zm0, 7, 8) << 56);
1941     uint64_t m0 = ELEM(zd1, 0, 8) | (ELEM(zm1, 0, 8) << 8)
1942         | (ELEM(zd1, 1, 8) << 16) | (ELEM(zm1, 1, 8) << 24)
1943         | (ELEM(zd1, 2, 8) << 32) | (ELEM(zm1, 2, 8) << 40)
1944         | (ELEM(zd1, 3, 8) << 48) | (ELEM(zm1, 3, 8) << 56);
1945     uint64_t m1 = ELEM(zd1, 4, 8) | (ELEM(zm1, 4, 8) << 8)
1946         | (ELEM(zd1, 5, 8) << 16) | (ELEM(zm1, 5, 8) << 24)
1947         | (ELEM(zd1, 6, 8) << 32) | (ELEM(zm1, 6, 8) << 40)
1948         | (ELEM(zd1, 7, 8) << 48) | (ELEM(zm1, 7, 8) << 56);
1949     env->vfp.regs[rm] = make_float64(m0);
1950     env->vfp.regs[rm + 1] = make_float64(m1);
1951     env->vfp.regs[rd] = make_float64(d0);
1952     env->vfp.regs[rd + 1] = make_float64(d1);
1953 }
1954
1955 void HELPER(neon_qzip16)(CPUState *env, uint32_t rd, uint32_t rm)
1956 {
1957     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1958     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1959     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1960     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1961     uint64_t d0 = ELEM(zd0, 0, 16) | (ELEM(zm0, 0, 16) << 16)
1962         | (ELEM(zd0, 1, 16) << 32) | (ELEM(zm0, 1, 16) << 48);
1963     uint64_t d1 = ELEM(zd0, 2, 16) | (ELEM(zm0, 2, 16) << 16)
1964         | (ELEM(zd0, 3, 16) << 32) | (ELEM(zm0, 3, 16) << 48);
1965     uint64_t m0 = ELEM(zd1, 0, 16) | (ELEM(zm1, 0, 16) << 16)
1966         | (ELEM(zd1, 1, 16) << 32) | (ELEM(zm1, 1, 16) << 48);
1967     uint64_t m1 = ELEM(zd1, 2, 16) | (ELEM(zm1, 2, 16) << 16)
1968         | (ELEM(zd1, 3, 16) << 32) | (ELEM(zm1, 3, 16) << 48);
1969     env->vfp.regs[rm] = make_float64(m0);
1970     env->vfp.regs[rm + 1] = make_float64(m1);
1971     env->vfp.regs[rd] = make_float64(d0);
1972     env->vfp.regs[rd + 1] = make_float64(d1);
1973 }
1974
1975 void HELPER(neon_qzip32)(CPUState *env, uint32_t rd, uint32_t rm)
1976 {
1977     uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1978     uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1979     uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1980     uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1981     uint64_t d0 = ELEM(zd0, 0, 32) | (ELEM(zm0, 0, 32) << 32);
1982     uint64_t d1 = ELEM(zd0, 1, 32) | (ELEM(zm0, 1, 32) << 32);
1983     uint64_t m0 = ELEM(zd1, 0, 32) | (ELEM(zm1, 0, 32) << 32);
1984     uint64_t m1 = ELEM(zd1, 1, 32) | (ELEM(zm1, 1, 32) << 32);
1985     env->vfp.regs[rm] = make_float64(m0);
1986     env->vfp.regs[rm + 1] = make_float64(m1);
1987     env->vfp.regs[rd] = make_float64(d0);
1988     env->vfp.regs[rd + 1] = make_float64(d1);
1989 }
1990
1991 void HELPER(neon_zip8)(CPUState *env, uint32_t rd, uint32_t rm)
1992 {
1993     uint64_t zm = float64_val(env->vfp.regs[rm]);
1994     uint64_t zd = float64_val(env->vfp.regs[rd]);
1995     uint64_t d0 = ELEM(zd, 0, 8) | (ELEM(zm, 0, 8) << 8)
1996         | (ELEM(zd, 1, 8) << 16) | (ELEM(zm, 1, 8) << 24)
1997         | (ELEM(zd, 2, 8) << 32) | (ELEM(zm, 2, 8) << 40)
1998         | (ELEM(zd, 3, 8) << 48) | (ELEM(zm, 3, 8) << 56);
1999     uint64_t m0 = ELEM(zd, 4, 8) | (ELEM(zm, 4, 8) << 8)
2000         | (ELEM(zd, 5, 8) << 16) | (ELEM(zm, 5, 8) << 24)
2001         | (ELEM(zd, 6, 8) << 32) | (ELEM(zm, 6, 8) << 40)
2002         | (ELEM(zd, 7, 8) << 48) | (ELEM(zm, 7, 8) << 56);
2003     env->vfp.regs[rm] = make_float64(m0);
2004     env->vfp.regs[rd] = make_float64(d0);
2005 }
2006
2007 void HELPER(neon_zip16)(CPUState *env, uint32_t rd, uint32_t rm)
2008 {
2009     uint64_t zm = float64_val(env->vfp.regs[rm]);
2010     uint64_t zd = float64_val(env->vfp.regs[rd]);
2011     uint64_t d0 = ELEM(zd, 0, 16) | (ELEM(zm, 0, 16) << 16)
2012         | (ELEM(zd, 1, 16) << 32) | (ELEM(zm, 1, 16) << 48);
2013     uint64_t m0 = ELEM(zd, 2, 16) | (ELEM(zm, 2, 16) << 16)
2014         | (ELEM(zd, 3, 16) << 32) | (ELEM(zm, 3, 16) << 48);
2015     env->vfp.regs[rm] = make_float64(m0);
2016     env->vfp.regs[rd] = make_float64(d0);
2017 }