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 "helpers.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 static float_status neon_float_status;
  22 #define NFS &neon_float_status
  23
  24 /* Helper routines to perform bitwise copies between float and int.  */
  25 static inline float32 vfp_itos(uint32_t i)
  26 {
  27     union {
  28         uint32_t i;
  29         float32 s;
  30     } v;
  31
  32     v.i = i;
  33     return v.s;
  34 }
  35
  36 static inline uint32_t vfp_stoi(float32 s)
  37 {
  38     union {
  39         uint32_t i;
  40         float32 s;
  41     } v;
  42
  43     v.s = s;
  44     return v.i;
  45 }
  46
  47 #define NEON_TYPE1(name, type) \
  48 typedef struct \
  49 { \
  50     type v1; \
  51 } neon_##name;
  52 #ifdef HOST_WORDS_BIGENDIAN
  53 #define NEON_TYPE2(name, type) \
  54 typedef struct \
  55 { \
  56     type v2; \
  57     type v1; \
  58 } neon_##name;
  59 #define NEON_TYPE4(name, type) \
  60 typedef struct \
  61 { \
  62     type v4; \
  63     type v3; \
  64     type v2; \
  65     type v1; \
  66 } neon_##name;
  67 #else
  68 #define NEON_TYPE2(name, type) \
  69 typedef struct \
  70 { \
  71     type v1; \
  72     type v2; \
  73 } neon_##name;
  74 #define NEON_TYPE4(name, type) \
  75 typedef struct \
  76 { \
  77     type v1; \
  78     type v2; \
  79     type v3; \
  80     type v4; \
  81 } neon_##name;
  82 #endif
  83
  84 NEON_TYPE4(s8, int8_t)
  85 NEON_TYPE4(u8, uint8_t)
  86 NEON_TYPE2(s16, int16_t)
  87 NEON_TYPE2(u16, uint16_t)
  88 NEON_TYPE1(s32, int32_t)
  89 NEON_TYPE1(u32, uint32_t)
  90 #undef NEON_TYPE4
  91 #undef NEON_TYPE2
  92 #undef NEON_TYPE1
  93
  94 /* Copy from a uint32_t to a vector structure type.  */
  95 #define NEON_UNPACK(vtype, dest, val) do { \
  96     union { \
  97         vtype v; \
  98         uint32_t i; \
  99     } conv_u; \
 100     conv_u.i = (val); \
 101     dest = conv_u.v; \
 102     } while(0)
 103
 104 /* Copy from a vector structure type to a uint32_t.  */
 105 #define NEON_PACK(vtype, dest, val) do { \
 106     union { \
 107         vtype v; \
 108         uint32_t i; \
 109     } conv_u; \
 110     conv_u.v = (val); \
 111     dest = conv_u.i; \
 112     } while(0)
 113
 114 #define NEON_DO1 \
 115     NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1);
 116 #define NEON_DO2 \
 117     NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); \
 118     NEON_FN(vdest.v2, vsrc1.v2, vsrc2.v2);
 119 #define NEON_DO4 \
 120     NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); \
 121     NEON_FN(vdest.v2, vsrc1.v2, vsrc2.v2); \
 122     NEON_FN(vdest.v3, vsrc1.v3, vsrc2.v3); \
 123     NEON_FN(vdest.v4, vsrc1.v4, vsrc2.v4);
 124
 125 #define NEON_VOP_BODY(vtype, n) \
 126 { \
 127     uint32_t res; \
 128     vtype vsrc1; \
 129     vtype vsrc2; \
 130     vtype vdest; \
 131     NEON_UNPACK(vtype, vsrc1, arg1); \
 132     NEON_UNPACK(vtype, vsrc2, arg2); \
 133     NEON_DO##n; \
 134     NEON_PACK(vtype, res, vdest); \
 135     return res; \
 136 }
 137
 138 #define NEON_VOP(name, vtype, n) \
 139 uint32_t HELPER(glue(neon_,name))(uint32_t arg1, uint32_t arg2) \
 140 NEON_VOP_BODY(vtype, n)
 141
 142 #define NEON_VOP_ENV(name, vtype, n) \
 143 uint32_t HELPER(glue(neon_,name))(CPUState *env, uint32_t arg1, uint32_t arg2) \
 144 NEON_VOP_BODY(vtype, n)
 145
 146 /* Pairwise operations.  */
 147 /* For 32-bit elements each segment only contains a single element, so
 148    the elementwise and pairwise operations are the same.  */
 149 #define NEON_PDO2 \
 150     NEON_FN(vdest.v1, vsrc1.v1, vsrc1.v2); \
 151     NEON_FN(vdest.v2, vsrc2.v1, vsrc2.v2);
 152 #define NEON_PDO4 \
 153     NEON_FN(vdest.v1, vsrc1.v1, vsrc1.v2); \
 154     NEON_FN(vdest.v2, vsrc1.v3, vsrc1.v4); \
 155     NEON_FN(vdest.v3, vsrc2.v1, vsrc2.v2); \
 156     NEON_FN(vdest.v4, vsrc2.v3, vsrc2.v4); \
 157
 158 #define NEON_POP(name, vtype, n) \
 159 uint32_t HELPER(glue(neon_,name))(uint32_t arg1, uint32_t arg2) \
 160 { \
 161     uint32_t res; \
 162     vtype vsrc1; \
 163     vtype vsrc2; \
 164     vtype vdest; \
 165     NEON_UNPACK(vtype, vsrc1, arg1); \
 166     NEON_UNPACK(vtype, vsrc2, arg2); \
 167     NEON_PDO##n; \
 168     NEON_PACK(vtype, res, vdest); \
 169     return res; \
 170 }
 171
 172 /* Unary operators.  */
 173 #define NEON_VOP1(name, vtype, n) \
 174 uint32_t HELPER(glue(neon_,name))(uint32_t arg) \
 175 { \
 176     vtype vsrc1; \
 177     vtype vdest; \
 178     NEON_UNPACK(vtype, vsrc1, arg); \
 179     NEON_DO##n; \
 180     NEON_PACK(vtype, arg, vdest); \
 181     return arg; \
 182 }
 183
 184
 185 #define NEON_USAT(dest, src1, src2, type) do { \
 186     uint32_t tmp = (uint32_t)src1 + (uint32_t)src2; \
 187     if (tmp != (type)tmp) { \
 188         SET_QC(); \
 189         dest = ~0; \
 190     } else { \
 191         dest = tmp; \
 192     }} while(0)
 193 #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint8_t)
 194 NEON_VOP_ENV(qadd_u8, neon_u8, 4)
 195 #undef NEON_FN
 196 #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint16_t)
 197 NEON_VOP_ENV(qadd_u16, neon_u16, 2)
 198 #undef NEON_FN
 199 #undef NEON_USAT
 200
 201 uint32_t HELPER(neon_qadd_u32)(CPUState *env, uint32_t a, uint32_t b)
 202 {
 203     uint32_t res = a + b;
 204     if (res < a) {
 205         SET_QC();
 206         res = ~0;
 207     }
 208     return res;
 209 }
 210
 211 uint64_t HELPER(neon_qadd_u64)(CPUState *env, uint64_t src1, uint64_t src2)
 212 {
 213     uint64_t res;
 214
 215     res = src1 + src2;
 216     if (res < src1) {
 217         SET_QC();
 218         res = ~(uint64_t)0;
 219     }
 220     return res;
 221 }
 222
 223 #define NEON_SSAT(dest, src1, src2, type) do { \
 224     int32_t tmp = (uint32_t)src1 + (uint32_t)src2; \
 225     if (tmp != (type)tmp) { \
 226         SET_QC(); \
 227         if (src2 > 0) { \
 228             tmp = (1 << (sizeof(type) * 8 - 1)) - 1; \
 229         } else { \
 230             tmp = 1 << (sizeof(type) * 8 - 1); \
 231         } \
 232     } \
 233     dest = tmp; \
 234     } while(0)
 235 #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int8_t)
 236 NEON_VOP_ENV(qadd_s8, neon_s8, 4)
 237 #undef NEON_FN
 238 #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int16_t)
 239 NEON_VOP_ENV(qadd_s16, neon_s16, 2)
 240 #undef NEON_FN
 241 #undef NEON_SSAT
 242
 243 uint32_t HELPER(neon_qadd_s32)(CPUState *env, uint32_t a, uint32_t b)
 244 {
 245     uint32_t res = a + b;
 246     if (((res ^ a) & SIGNBIT) && !((a ^ b) & SIGNBIT)) {
 247         SET_QC();
 248         res = ~(((int32_t)a >> 31) ^ SIGNBIT);
 249     }
 250     return res;
 251 }
 252
 253 uint64_t HELPER(neon_qadd_s64)(CPUState *env, uint64_t src1, uint64_t src2)
 254 {
 255     uint64_t res;
 256
 257     res = src1 + src2;
 258     if (((res ^ src1) & SIGNBIT64) && !((src1 ^ src2) & SIGNBIT64)) {
 259         SET_QC();
 260         res = ((int64_t)src1 >> 63) ^ ~SIGNBIT64;
 261     }
 262     return res;
 263 }
 264
 265 #define NEON_USAT(dest, src1, src2, type) do { \
 266     uint32_t tmp = (uint32_t)src1 - (uint32_t)src2; \
 267     if (tmp != (type)tmp) { \
 268         SET_QC(); \
 269         dest = 0; \
 270     } else { \
 271         dest = tmp; \
 272     }} while(0)
 273 #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint8_t)
 274 NEON_VOP_ENV(qsub_u8, neon_u8, 4)
 275 #undef NEON_FN
 276 #define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint16_t)
 277 NEON_VOP_ENV(qsub_u16, neon_u16, 2)
 278 #undef NEON_FN
 279 #undef NEON_USAT
 280
 281 uint32_t HELPER(neon_qsub_u32)(CPUState *env, uint32_t a, uint32_t b)
 282 {
 283     uint32_t res = a - b;
 284     if (res > a) {
 285         SET_QC();
 286         res = 0;
 287     }
 288     return res;
 289 }
 290
 291 uint64_t HELPER(neon_qsub_u64)(CPUState *env, uint64_t src1, uint64_t src2)
 292 {
 293     uint64_t res;
 294
 295     if (src1 < src2) {
 296         SET_QC();
 297         res = 0;
 298     } else {
 299         res = src1 - src2;
 300     }
 301     return res;
 302 }
 303
 304 #define NEON_SSAT(dest, src1, src2, type) do { \
 305     int32_t tmp = (uint32_t)src1 - (uint32_t)src2; \
 306     if (tmp != (type)tmp) { \
 307         SET_QC(); \
 308         if (src2 < 0) { \
 309             tmp = (1 << (sizeof(type) * 8 - 1)) - 1; \
 310         } else { \
 311             tmp = 1 << (sizeof(type) * 8 - 1); \
 312         } \
 313     } \
 314     dest = tmp; \
 315     } while(0)
 316 #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int8_t)
 317 NEON_VOP_ENV(qsub_s8, neon_s8, 4)
 318 #undef NEON_FN
 319 #define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int16_t)
 320 NEON_VOP_ENV(qsub_s16, neon_s16, 2)
 321 #undef NEON_FN
 322 #undef NEON_SSAT
 323
 324 uint32_t HELPER(neon_qsub_s32)(CPUState *env, uint32_t a, uint32_t b)
 325 {
 326     uint32_t res = a - b;
 327     if (((res ^ a) & SIGNBIT) && ((a ^ b) & SIGNBIT)) {
 328         SET_QC();
 329         res = ~(((int32_t)a >> 31) ^ SIGNBIT);
 330     }
 331     return res;
 332 }
 333
 334 uint64_t HELPER(neon_qsub_s64)(CPUState *env, uint64_t src1, uint64_t src2)
 335 {
 336     uint64_t res;
 337
 338     res = src1 - src2;
 339     if (((res ^ src1) & SIGNBIT64) && ((src1 ^ src2) & SIGNBIT64)) {
 340         SET_QC();
 341         res = ((int64_t)src1 >> 63) ^ ~SIGNBIT64;
 342     }
 343     return res;
 344 }
 345
 346 #define NEON_FN(dest, src1, src2) dest = (src1 + src2) >> 1
 347 NEON_VOP(hadd_s8, neon_s8, 4)
 348 NEON_VOP(hadd_u8, neon_u8, 4)
 349 NEON_VOP(hadd_s16, neon_s16, 2)
 350 NEON_VOP(hadd_u16, neon_u16, 2)
 351 #undef NEON_FN
 352
 353 int32_t HELPER(neon_hadd_s32)(int32_t src1, int32_t src2)
 354 {
 355     int32_t dest;
 356
 357     dest = (src1 >> 1) + (src2 >> 1);
 358     if (src1 & src2 & 1)
 359         dest++;
 360     return dest;
 361 }
 362
 363 uint32_t HELPER(neon_hadd_u32)(uint32_t src1, uint32_t src2)
 364 {
 365     uint32_t dest;
 366
 367     dest = (src1 >> 1) + (src2 >> 1);
 368     if (src1 & src2 & 1)
 369         dest++;
 370     return dest;
 371 }
 372
 373 #define NEON_FN(dest, src1, src2) dest = (src1 + src2 + 1) >> 1
 374 NEON_VOP(rhadd_s8, neon_s8, 4)
 375 NEON_VOP(rhadd_u8, neon_u8, 4)
 376 NEON_VOP(rhadd_s16, neon_s16, 2)
 377 NEON_VOP(rhadd_u16, neon_u16, 2)
 378 #undef NEON_FN
 379
 380 int32_t HELPER(neon_rhadd_s32)(int32_t src1, int32_t src2)
 381 {
 382     int32_t dest;
 383
 384     dest = (src1 >> 1) + (src2 >> 1);
 385     if ((src1 | src2) & 1)
 386         dest++;
 387     return dest;
 388 }
 389
 390 uint32_t HELPER(neon_rhadd_u32)(uint32_t src1, uint32_t src2)
 391 {
 392     uint32_t dest;
 393
 394     dest = (src1 >> 1) + (src2 >> 1);
 395     if ((src1 | src2) & 1)
 396         dest++;
 397     return dest;
 398 }
 399
 400 #define NEON_FN(dest, src1, src2) dest = (src1 - src2) >> 1
 401 NEON_VOP(hsub_s8, neon_s8, 4)
 402 NEON_VOP(hsub_u8, neon_u8, 4)
 403 NEON_VOP(hsub_s16, neon_s16, 2)
 404 NEON_VOP(hsub_u16, neon_u16, 2)
 405 #undef NEON_FN
 406
 407 int32_t HELPER(neon_hsub_s32)(int32_t src1, int32_t src2)
 408 {
 409     int32_t dest;
 410
 411     dest = (src1 >> 1) - (src2 >> 1);
 412     if ((~src1) & src2 & 1)
 413         dest--;
 414     return dest;
 415 }
 416
 417 uint32_t HELPER(neon_hsub_u32)(uint32_t src1, uint32_t src2)
 418 {
 419     uint32_t dest;
 420
 421     dest = (src1 >> 1) - (src2 >> 1);
 422     if ((~src1) & src2 & 1)
 423         dest--;
 424     return dest;
 425 }
 426
 427 #define NEON_FN(dest, src1, src2) dest = (src1 > src2) ? ~0 : 0
 428 NEON_VOP(cgt_s8, neon_s8, 4)
 429 NEON_VOP(cgt_u8, neon_u8, 4)
 430 NEON_VOP(cgt_s16, neon_s16, 2)
 431 NEON_VOP(cgt_u16, neon_u16, 2)
 432 NEON_VOP(cgt_s32, neon_s32, 1)
 433 NEON_VOP(cgt_u32, neon_u32, 1)
 434 #undef NEON_FN
 435
 436 #define NEON_FN(dest, src1, src2) dest = (src1 >= src2) ? ~0 : 0
 437 NEON_VOP(cge_s8, neon_s8, 4)
 438 NEON_VOP(cge_u8, neon_u8, 4)
 439 NEON_VOP(cge_s16, neon_s16, 2)
 440 NEON_VOP(cge_u16, neon_u16, 2)
 441 NEON_VOP(cge_s32, neon_s32, 1)
 442 NEON_VOP(cge_u32, neon_u32, 1)
 443 #undef NEON_FN
 444
 445 #define NEON_FN(dest, src1, src2) dest = (src1 < src2) ? src1 : src2
 446 NEON_VOP(min_s8, neon_s8, 4)
 447 NEON_VOP(min_u8, neon_u8, 4)
 448 NEON_VOP(min_s16, neon_s16, 2)
 449 NEON_VOP(min_u16, neon_u16, 2)
 450 NEON_VOP(min_s32, neon_s32, 1)
 451 NEON_VOP(min_u32, neon_u32, 1)
 452 NEON_POP(pmin_s8, neon_s8, 4)
 453 NEON_POP(pmin_u8, neon_u8, 4)
 454 NEON_POP(pmin_s16, neon_s16, 2)
 455 NEON_POP(pmin_u16, neon_u16, 2)
 456 #undef NEON_FN
 457
 458 #define NEON_FN(dest, src1, src2) dest = (src1 > src2) ? src1 : src2
 459 NEON_VOP(max_s8, neon_s8, 4)
 460 NEON_VOP(max_u8, neon_u8, 4)
 461 NEON_VOP(max_s16, neon_s16, 2)
 462 NEON_VOP(max_u16, neon_u16, 2)
 463 NEON_VOP(max_s32, neon_s32, 1)
 464 NEON_VOP(max_u32, neon_u32, 1)
 465 NEON_POP(pmax_s8, neon_s8, 4)
 466 NEON_POP(pmax_u8, neon_u8, 4)
 467 NEON_POP(pmax_s16, neon_s16, 2)
 468 NEON_POP(pmax_u16, neon_u16, 2)
 469 #undef NEON_FN
 470
 471 #define NEON_FN(dest, src1, src2) \
 472     dest = (src1 > src2) ? (src1 - src2) : (src2 - src1)
 473 NEON_VOP(abd_s8, neon_s8, 4)
 474 NEON_VOP(abd_u8, neon_u8, 4)
 475 NEON_VOP(abd_s16, neon_s16, 2)
 476 NEON_VOP(abd_u16, neon_u16, 2)
 477 NEON_VOP(abd_s32, neon_s32, 1)
 478 NEON_VOP(abd_u32, neon_u32, 1)
 479 #undef NEON_FN
 480
 481 #define NEON_FN(dest, src1, src2) do { \
 482     int8_t tmp; \
 483     tmp = (int8_t)src2; \
 484     if (tmp >= (ssize_t)sizeof(src1) * 8 || \
 485         tmp <= -(ssize_t)sizeof(src1) * 8) { \
 486         dest = 0; \
 487     } else if (tmp < 0) { \
 488         dest = src1 >> -tmp; \
 489     } else { \
 490         dest = src1 << tmp; \
 491     }} while (0)
 492 NEON_VOP(shl_u8, neon_u8, 4)
 493 NEON_VOP(shl_u16, neon_u16, 2)
 494 NEON_VOP(shl_u32, neon_u32, 1)
 495 #undef NEON_FN
 496
 497 uint64_t HELPER(neon_shl_u64)(uint64_t val, uint64_t shiftop)
 498 {
 499     int8_t shift = (int8_t)shiftop;
 500     if (shift >= 64 || shift <= -64) {
 501         val = 0;
 502     } else if (shift < 0) {
 503         val >>= -shift;
 504     } else {
 505         val <<= shift;
 506     }
 507     return val;
 508 }
 509
 510 #define NEON_FN(dest, src1, src2) do { \
 511     int8_t tmp; \
 512     tmp = (int8_t)src2; \
 513     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 514         dest = 0; \
 515     } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 516         dest = src1 >> (sizeof(src1) * 8 - 1); \
 517     } else if (tmp < 0) { \
 518         dest = src1 >> -tmp; \
 519     } else { \
 520         dest = src1 << tmp; \
 521     }} while (0)
 522 NEON_VOP(shl_s8, neon_s8, 4)
 523 NEON_VOP(shl_s16, neon_s16, 2)
 524 NEON_VOP(shl_s32, neon_s32, 1)
 525 #undef NEON_FN
 526
 527 uint64_t HELPER(neon_shl_s64)(uint64_t valop, uint64_t shiftop)
 528 {
 529     int8_t shift = (int8_t)shiftop;
 530     int64_t val = valop;
 531     if (shift >= 64) {
 532         val = 0;
 533     } else if (shift <= -64) {
 534         val >>= 63;
 535     } else if (shift < 0) {
 536         val >>= -shift;
 537     } else {
 538         val <<= shift;
 539     }
 540     return val;
 541 }
 542
 543 #define NEON_FN(dest, src1, src2) do { \
 544     int8_t tmp; \
 545     tmp = (int8_t)src2; \
 546     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 547         dest = 0; \
 548     } else if (tmp < -(ssize_t)sizeof(src1) * 8) { \
 549         dest = src1 >> (sizeof(src1) * 8 - 1); \
 550     } else if (tmp == -(ssize_t)sizeof(src1) * 8) { \
 551         dest = src1 >> (tmp - 1); \
 552         dest++; \
 553         dest >>= 1; \
 554     } else if (tmp < 0) { \
 555         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 556     } else { \
 557         dest = src1 << tmp; \
 558     }} while (0)
 559 NEON_VOP(rshl_s8, neon_s8, 4)
 560 NEON_VOP(rshl_s16, neon_s16, 2)
 561 NEON_VOP(rshl_s32, neon_s32, 1)
 562 #undef NEON_FN
 563
 564 uint64_t HELPER(neon_rshl_s64)(uint64_t valop, uint64_t shiftop)
 565 {
 566     int8_t shift = (int8_t)shiftop;
 567     int64_t val = valop;
 568     if (shift >= 64) {
 569         val = 0;
 570     } else if (shift < -64) {
 571         val >>= 63;
 572     } else if (shift == -63) {
 573         val >>= 63;
 574         val++;
 575         val >>= 1;
 576     } else if (shift < 0) {
 577         val = (val + ((int64_t)1 << (-1 - shift))) >> -shift;
 578     } else {
 579         val <<= shift;
 580     }
 581     return val;
 582 }
 583
 584 #define NEON_FN(dest, src1, src2) do { \
 585     int8_t tmp; \
 586     tmp = (int8_t)src2; \
 587     if (tmp >= (ssize_t)sizeof(src1) * 8 || \
 588         tmp < -(ssize_t)sizeof(src1) * 8) { \
 589         dest = 0; \
 590     } else if (tmp == -(ssize_t)sizeof(src1) * 8) { \
 591         dest = src1 >> (tmp - 1); \
 592     } else if (tmp < 0) { \
 593         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 594     } else { \
 595         dest = src1 << tmp; \
 596     }} while (0)
 597 NEON_VOP(rshl_u8, neon_u8, 4)
 598 NEON_VOP(rshl_u16, neon_u16, 2)
 599 NEON_VOP(rshl_u32, neon_u32, 1)
 600 #undef NEON_FN
 601
 602 uint64_t HELPER(neon_rshl_u64)(uint64_t val, uint64_t shiftop)
 603 {
 604     int8_t shift = (uint8_t)shiftop;
 605     if (shift >= 64 || shift < 64) {
 606         val = 0;
 607     } else if (shift == -64) {
 608         /* Rounding a 1-bit result just preserves that bit.  */
 609         val >>= 63;
 610     } if (shift < 0) {
 611         val = (val + ((uint64_t)1 << (-1 - shift))) >> -shift;
 612         val >>= -shift;
 613     } else {
 614         val <<= shift;
 615     }
 616     return val;
 617 }
 618
 619 #define NEON_FN(dest, src1, src2) do { \
 620     int8_t tmp; \
 621     tmp = (int8_t)src2; \
 622     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 623         if (src1) { \
 624             SET_QC(); \
 625             dest = ~0; \
 626         } else { \
 627             dest = 0; \
 628         } \
 629     } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 630         dest = 0; \
 631     } else if (tmp < 0) { \
 632         dest = src1 >> -tmp; \
 633     } else { \
 634         dest = src1 << tmp; \
 635         if ((dest >> tmp) != src1) { \
 636             SET_QC(); \
 637             dest = ~0; \
 638         } \
 639     }} while (0)
 640 NEON_VOP_ENV(qshl_u8, neon_u8, 4)
 641 NEON_VOP_ENV(qshl_u16, neon_u16, 2)
 642 NEON_VOP_ENV(qshl_u32, neon_u32, 1)
 643 #undef NEON_FN
 644
 645 uint64_t HELPER(neon_qshl_u64)(CPUState *env, uint64_t val, uint64_t shiftop)
 646 {
 647     int8_t shift = (int8_t)shiftop;
 648     if (shift >= 64) {
 649         if (val) {
 650             val = ~(uint64_t)0;
 651             SET_QC();
 652         }
 653     } else if (shift <= -64) {
 654         val = 0;
 655     } else if (shift < 0) {
 656         val >>= -shift;
 657     } else {
 658         uint64_t tmp = val;
 659         val <<= shift;
 660         if ((val >> shift) != tmp) {
 661             SET_QC();
 662             val = ~(uint64_t)0;
 663         }
 664     }
 665     return val;
 666 }
 667
 668 #define NEON_FN(dest, src1, src2) do { \
 669     int8_t tmp; \
 670     tmp = (int8_t)src2; \
 671     if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 672         if (src1) { \
 673             SET_QC(); \
 674             dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
 675             if (src1 > 0) { \
 676                 dest--; \
 677             } \
 678         } else { \
 679             dest = src1; \
 680         } \
 681     } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 682         dest = src1 >> 31; \
 683     } else if (tmp < 0) { \
 684         dest = src1 >> -tmp; \
 685     } else { \
 686         dest = src1 << tmp; \
 687         if ((dest >> tmp) != src1) { \
 688             SET_QC(); \
 689             dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
 690             if (src1 > 0) { \
 691                 dest--; \
 692             } \
 693         } \
 694     }} while (0)
 695 NEON_VOP_ENV(qshl_s8, neon_s8, 4)
 696 NEON_VOP_ENV(qshl_s16, neon_s16, 2)
 697 NEON_VOP_ENV(qshl_s32, neon_s32, 1)
 698 #undef NEON_FN
 699
 700 uint64_t HELPER(neon_qshl_s64)(CPUState *env, uint64_t valop, uint64_t shiftop)
 701 {
 702     int8_t shift = (uint8_t)shiftop;
 703     int64_t val = valop;
 704     if (shift >= 64) {
 705         if (val) {
 706             SET_QC();
 707             val = (val >> 63) ^ ~SIGNBIT64;
 708         }
 709     } else if (shift <= -64) {
 710         val >>= 63;
 711     } else if (shift < 0) {
 712         val >>= -shift;
 713     } else {
 714         int64_t tmp = val;
 715         val <<= shift;
 716         if ((val >> shift) != tmp) {
 717             SET_QC();
 718             val = (tmp >> 63) ^ ~SIGNBIT64;
 719         }
 720     }
 721     return val;
 722 }
 723
 724 #define NEON_FN(dest, src1, src2) do { \
 725     if (src1 & (1 << (sizeof(src1) * 8 - 1))) { \
 726         SET_QC(); \
 727         dest = 0; \
 728     } else { \
 729         int8_t tmp; \
 730         tmp = (int8_t)src2; \
 731         if (tmp >= (ssize_t)sizeof(src1) * 8) { \
 732             if (src1) { \
 733                 SET_QC(); \
 734                 dest = ~0; \
 735             } else { \
 736                 dest = 0; \
 737             } \
 738         } else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
 739             dest = 0; \
 740         } else if (tmp < 0) { \
 741             dest = src1 >> -tmp; \
 742         } else { \
 743             dest = src1 << tmp; \
 744             if ((dest >> tmp) != src1) { \
 745                 SET_QC(); \
 746                 dest = ~0; \
 747             } \
 748         } \
 749     }} while (0)
 750 NEON_VOP_ENV(qshlu_s8, neon_u8, 4)
 751 NEON_VOP_ENV(qshlu_s16, neon_u16, 2)
 752 #undef NEON_FN
 753
 754 uint32_t HELPER(neon_qshlu_s32)(CPUState *env, uint32_t valop, uint32_t shiftop)
 755 {
 756     if ((int32_t)valop < 0) {
 757         SET_QC();
 758         return 0;
 759     }
 760     return helper_neon_qshl_u32(env, valop, shiftop);
 761 }
 762
 763 uint64_t HELPER(neon_qshlu_s64)(CPUState *env, uint64_t valop, uint64_t shiftop)
 764 {
 765     if ((int64_t)valop < 0) {
 766         SET_QC();
 767         return 0;
 768     }
 769     return helper_neon_qshl_u64(env, valop, shiftop);
 770 }
 771
 772 /* FIXME: This is wrong.  */
 773 #define NEON_FN(dest, src1, src2) do { \
 774     int8_t tmp; \
 775     tmp = (int8_t)src2; \
 776     if (tmp < 0) { \
 777         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 778     } else { \
 779         dest = src1 << tmp; \
 780         if ((dest >> tmp) != src1) { \
 781             SET_QC(); \
 782             dest = ~0; \
 783         } \
 784     }} while (0)
 785 NEON_VOP_ENV(qrshl_u8, neon_u8, 4)
 786 NEON_VOP_ENV(qrshl_u16, neon_u16, 2)
 787 NEON_VOP_ENV(qrshl_u32, neon_u32, 1)
 788 #undef NEON_FN
 789
 790 uint64_t HELPER(neon_qrshl_u64)(CPUState *env, uint64_t val, uint64_t shiftop)
 791 {
 792     int8_t shift = (int8_t)shiftop;
 793     if (shift < 0) {
 794         val = (val + (1 << (-1 - shift))) >> -shift;
 795     } else { \
 796         uint64_t tmp = val;
 797         val <<= shift;
 798         if ((val >> shift) != tmp) {
 799             SET_QC();
 800             val = ~0;
 801         }
 802     }
 803     return val;
 804 }
 805
 806 #define NEON_FN(dest, src1, src2) do { \
 807     int8_t tmp; \
 808     tmp = (int8_t)src2; \
 809     if (tmp < 0) { \
 810         dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
 811     } else { \
 812         dest = src1 << tmp; \
 813         if ((dest >> tmp) != src1) { \
 814             SET_QC(); \
 815             dest = src1 >> 31; \
 816         } \
 817     }} while (0)
 818 NEON_VOP_ENV(qrshl_s8, neon_s8, 4)
 819 NEON_VOP_ENV(qrshl_s16, neon_s16, 2)
 820 NEON_VOP_ENV(qrshl_s32, neon_s32, 1)
 821 #undef NEON_FN
 822
 823 uint64_t HELPER(neon_qrshl_s64)(CPUState *env, uint64_t valop, uint64_t shiftop)
 824 {
 825     int8_t shift = (uint8_t)shiftop;
 826     int64_t val = valop;
 827
 828     if (shift < 0) {
 829         val = (val + (1 << (-1 - shift))) >> -shift;
 830     } else {
 831         int64_t tmp = val;;
 832         val <<= shift;
 833         if ((val >> shift) != tmp) {
 834             SET_QC();
 835             val = tmp >> 31;
 836         }
 837     }
 838     return val;
 839 }
 840
 841 uint32_t HELPER(neon_add_u8)(uint32_t a, uint32_t b)
 842 {
 843     uint32_t mask;
 844     mask = (a ^ b) & 0x80808080u;
 845     a &= ~0x80808080u;
 846     b &= ~0x80808080u;
 847     return (a + b) ^ mask;
 848 }
 849
 850 uint32_t HELPER(neon_add_u16)(uint32_t a, uint32_t b)
 851 {
 852     uint32_t mask;
 853     mask = (a ^ b) & 0x80008000u;
 854     a &= ~0x80008000u;
 855     b &= ~0x80008000u;
 856     return (a + b) ^ mask;
 857 }
 858
 859 #define NEON_FN(dest, src1, src2) dest = src1 + src2
 860 NEON_POP(padd_u8, neon_u8, 4)
 861 NEON_POP(padd_u16, neon_u16, 2)
 862 #undef NEON_FN
 863
 864 #define NEON_FN(dest, src1, src2) dest = src1 - src2
 865 NEON_VOP(sub_u8, neon_u8, 4)
 866 NEON_VOP(sub_u16, neon_u16, 2)
 867 #undef NEON_FN
 868
 869 #define NEON_FN(dest, src1, src2) dest = src1 * src2
 870 NEON_VOP(mul_u8, neon_u8, 4)
 871 NEON_VOP(mul_u16, neon_u16, 2)
 872 #undef NEON_FN
 873
 874 /* Polynomial multiplication is like integer multiplication except the
 875    partial products are XORed, not added.  */
 876 uint32_t HELPER(neon_mul_p8)(uint32_t op1, uint32_t op2)
 877 {
 878     uint32_t mask;
 879     uint32_t result;
 880     result = 0;
 881     while (op1) {
 882         mask = 0;
 883         if (op1 & 1)
 884             mask |= 0xff;
 885         if (op1 & (1 << 8))
 886             mask |= (0xff << 8);
 887         if (op1 & (1 << 16))
 888             mask |= (0xff << 16);
 889         if (op1 & (1 << 24))
 890             mask |= (0xff << 24);
 891         result ^= op2 & mask;
 892         op1 = (op1 >> 1) & 0x7f7f7f7f;
 893         op2 = (op2 << 1) & 0xfefefefe;
 894     }
 895     return result;
 896 }
 897
 898 #define NEON_FN(dest, src1, src2) dest = (src1 & src2) ? -1 : 0
 899 NEON_VOP(tst_u8, neon_u8, 4)
 900 NEON_VOP(tst_u16, neon_u16, 2)
 901 NEON_VOP(tst_u32, neon_u32, 1)
 902 #undef NEON_FN
 903
 904 #define NEON_FN(dest, src1, src2) dest = (src1 == src2) ? -1 : 0
 905 NEON_VOP(ceq_u8, neon_u8, 4)
 906 NEON_VOP(ceq_u16, neon_u16, 2)
 907 NEON_VOP(ceq_u32, neon_u32, 1)
 908 #undef NEON_FN
 909
 910 #define NEON_FN(dest, src, dummy) dest = (src < 0) ? -src : src
 911 NEON_VOP1(abs_s8, neon_s8, 4)
 912 NEON_VOP1(abs_s16, neon_s16, 2)
 913 #undef NEON_FN
 914
 915 /* Count Leading Sign/Zero Bits.  */
 916 static inline int do_clz8(uint8_t x)
 917 {
 918     int n;
 919     for (n = 8; x; n--)
 920         x >>= 1;
 921     return n;
 922 }
 923
 924 static inline int do_clz16(uint16_t x)
 925 {
 926     int n;
 927     for (n = 16; x; n--)
 928         x >>= 1;
 929     return n;
 930 }
 931
 932 #define NEON_FN(dest, src, dummy) dest = do_clz8(src)
 933 NEON_VOP1(clz_u8, neon_u8, 4)
 934 #undef NEON_FN
 935
 936 #define NEON_FN(dest, src, dummy) dest = do_clz16(src)
 937 NEON_VOP1(clz_u16, neon_u16, 2)
 938 #undef NEON_FN
 939
 940 #define NEON_FN(dest, src, dummy) dest = do_clz8((src < 0) ? ~src : src) - 1
 941 NEON_VOP1(cls_s8, neon_s8, 4)
 942 #undef NEON_FN
 943
 944 #define NEON_FN(dest, src, dummy) dest = do_clz16((src < 0) ? ~src : src) - 1
 945 NEON_VOP1(cls_s16, neon_s16, 2)
 946 #undef NEON_FN
 947
 948 uint32_t HELPER(neon_cls_s32)(uint32_t x)
 949 {
 950     int count;
 951     if ((int32_t)x < 0)
 952         x = ~x;
 953     for (count = 32; x; count--)
 954         x = x >> 1;
 955     return count - 1;
 956 }
 957
 958 /* Bit count.  */
 959 uint32_t HELPER(neon_cnt_u8)(uint32_t x)
 960 {
 961     x = (x & 0x55555555) + ((x >>  1) & 0x55555555);
 962     x = (x & 0x33333333) + ((x >>  2) & 0x33333333);
 963     x = (x & 0x0f0f0f0f) + ((x >>  4) & 0x0f0f0f0f);
 964     return x;
 965 }
 966
 967 #define NEON_QDMULH16(dest, src1, src2, round) do { \
 968     uint32_t tmp = (int32_t)(int16_t) src1 * (int16_t) src2; \
 969     if ((tmp ^ (tmp << 1)) & SIGNBIT) { \
 970         SET_QC(); \
 971         tmp = (tmp >> 31) ^ ~SIGNBIT; \
 972     } else { \
 973         tmp <<= 1; \
 974     } \
 975     if (round) { \
 976         int32_t old = tmp; \
 977         tmp += 1 << 15; \
 978         if ((int32_t)tmp < old) { \
 979             SET_QC(); \
 980             tmp = SIGNBIT - 1; \
 981         } \
 982     } \
 983     dest = tmp >> 16; \
 984     } while(0)
 985 #define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 0)
 986 NEON_VOP_ENV(qdmulh_s16, neon_s16, 2)
 987 #undef NEON_FN
 988 #define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 1)
 989 NEON_VOP_ENV(qrdmulh_s16, neon_s16, 2)
 990 #undef NEON_FN
 991 #undef NEON_QDMULH16
 992
 993 #define NEON_QDMULH32(dest, src1, src2, round) do { \
 994     uint64_t tmp = (int64_t)(int32_t) src1 * (int32_t) src2; \
 995     if ((tmp ^ (tmp << 1)) & SIGNBIT64) { \
 996         SET_QC(); \
 997         tmp = (tmp >> 63) ^ ~SIGNBIT64; \
 998     } else { \
 999         tmp <<= 1; \
1000     } \
1001     if (round) { \
1002         int64_t old = tmp; \
1003         tmp += (int64_t)1 << 31; \
1004         if ((int64_t)tmp < old) { \
1005             SET_QC(); \
1006             tmp = SIGNBIT64 - 1; \
1007         } \
1008     } \
1009     dest = tmp >> 32; \
1010     } while(0)
1011 #define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 0)
1012 NEON_VOP_ENV(qdmulh_s32, neon_s32, 1)
1013 #undef NEON_FN
1014 #define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 1)
1015 NEON_VOP_ENV(qrdmulh_s32, neon_s32, 1)
1016 #undef NEON_FN
1017 #undef NEON_QDMULH32
1018
1019 uint32_t HELPER(neon_narrow_u8)(uint64_t x)
1020 {
1021     return (x & 0xffu) | ((x >> 8) & 0xff00u) | ((x >> 16) & 0xff0000u)
1022            | ((x >> 24) & 0xff000000u);
1023 }
1024
1025 uint32_t HELPER(neon_narrow_u16)(uint64_t x)
1026 {
1027     return (x & 0xffffu) | ((x >> 16) & 0xffff0000u);
1028 }
1029
1030 uint32_t HELPER(neon_narrow_high_u8)(uint64_t x)
1031 {
1032     return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00)
1033             | ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000);
1034 }
1035
1036 uint32_t HELPER(neon_narrow_high_u16)(uint64_t x)
1037 {
1038     return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000);
1039 }
1040
1041 uint32_t HELPER(neon_narrow_round_high_u8)(uint64_t x)
1042 {
1043     x &= 0xff80ff80ff80ff80ull;
1044     x += 0x0080008000800080ull;
1045     return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00)
1046             | ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000);
1047 }
1048
1049 uint32_t HELPER(neon_narrow_round_high_u16)(uint64_t x)
1050 {
1051     x &= 0xffff8000ffff8000ull;
1052     x += 0x0000800000008000ull;
1053     return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000);
1054 }
1055
1056 uint32_t HELPER(neon_unarrow_sat8)(CPUState *env, uint64_t x)
1057 {
1058     uint16_t s;
1059     uint8_t d;
1060     uint32_t res = 0;
1061 #define SAT8(n) \
1062     s = x >> n; \
1063     if (s & 0x8000) { \
1064         SET_QC(); \
1065     } else { \
1066         if (s > 0xff) { \
1067             d = 0xff; \
1068             SET_QC(); \
1069         } else  { \
1070             d = s; \
1071         } \
1072         res |= (uint32_t)d << (n / 2); \
1073     }
1074
1075     SAT8(0);
1076     SAT8(16);
1077     SAT8(32);
1078     SAT8(48);
1079 #undef SAT8
1080     return res;
1081 }
1082
1083 uint32_t HELPER(neon_narrow_sat_u8)(CPUState *env, uint64_t x)
1084 {
1085     uint16_t s;
1086     uint8_t d;
1087     uint32_t res = 0;
1088 #define SAT8(n) \
1089     s = x >> n; \
1090     if (s > 0xff) { \
1091         d = 0xff; \
1092         SET_QC(); \
1093     } else  { \
1094         d = s; \
1095     } \
1096     res |= (uint32_t)d << (n / 2);
1097
1098     SAT8(0);
1099     SAT8(16);
1100     SAT8(32);
1101     SAT8(48);
1102 #undef SAT8
1103     return res;
1104 }
1105
1106 uint32_t HELPER(neon_narrow_sat_s8)(CPUState *env, uint64_t x)
1107 {
1108     int16_t s;
1109     uint8_t d;
1110     uint32_t res = 0;
1111 #define SAT8(n) \
1112     s = x >> n; \
1113     if (s != (int8_t)s) { \
1114         d = (s >> 15) ^ 0x7f; \
1115         SET_QC(); \
1116     } else  { \
1117         d = s; \
1118     } \
1119     res |= (uint32_t)d << (n / 2);
1120
1121     SAT8(0);
1122     SAT8(16);
1123     SAT8(32);
1124     SAT8(48);
1125 #undef SAT8
1126     return res;
1127 }
1128
1129 uint32_t HELPER(neon_unarrow_sat16)(CPUState *env, uint64_t x)
1130 {
1131     uint32_t high;
1132     uint32_t low;
1133     low = x;
1134     if (low & 0x80000000) {
1135         low = 0;
1136         SET_QC();
1137     } else if (low > 0xffff) {
1138         low = 0xffff;
1139         SET_QC();
1140     }
1141     high = x >> 32;
1142     if (high & 0x80000000) {
1143         high = 0;
1144         SET_QC();
1145     } else if (high > 0xffff) {
1146         high = 0xffff;
1147         SET_QC();
1148     }
1149     return low | (high << 16);
1150 }
1151
1152 uint32_t HELPER(neon_narrow_sat_u16)(CPUState *env, uint64_t x)
1153 {
1154     uint32_t high;
1155     uint32_t low;
1156     low = x;
1157     if (low > 0xffff) {
1158         low = 0xffff;
1159         SET_QC();
1160     }
1161     high = x >> 32;
1162     if (high > 0xffff) {
1163         high = 0xffff;
1164         SET_QC();
1165     }
1166     return low | (high << 16);
1167 }
1168
1169 uint32_t HELPER(neon_narrow_sat_s16)(CPUState *env, uint64_t x)
1170 {
1171     int32_t low;
1172     int32_t high;
1173     low = x;
1174     if (low != (int16_t)low) {
1175         low = (low >> 31) ^ 0x7fff;
1176         SET_QC();
1177     }
1178     high = x >> 32;
1179     if (high != (int16_t)high) {
1180         high = (high >> 31) ^ 0x7fff;
1181         SET_QC();
1182     }
1183     return (uint16_t)low | (high << 16);
1184 }
1185
1186 uint32_t HELPER(neon_unarrow_sat32)(CPUState *env, uint64_t x)
1187 {
1188     if (x & 0x8000000000000000ull) {
1189         SET_QC();
1190         return 0;
1191     }
1192     if (x > 0xffffffffu) {
1193         SET_QC();
1194         return 0xffffffffu;
1195     }
1196     return x;
1197 }
1198
1199 uint32_t HELPER(neon_narrow_sat_u32)(CPUState *env, uint64_t x)
1200 {
1201     if (x > 0xffffffffu) {
1202         SET_QC();
1203         return 0xffffffffu;
1204     }
1205     return x;
1206 }
1207
1208 uint32_t HELPER(neon_narrow_sat_s32)(CPUState *env, uint64_t x)
1209 {
1210     if ((int64_t)x != (int32_t)x) {
1211         SET_QC();
1212         return ((int64_t)x >> 63) ^ 0x7fffffff;
1213     }
1214     return x;
1215 }
1216
1217 uint64_t HELPER(neon_widen_u8)(uint32_t x)
1218 {
1219     uint64_t tmp;
1220     uint64_t ret;
1221     ret = (uint8_t)x;
1222     tmp = (uint8_t)(x >> 8);
1223     ret |= tmp << 16;
1224     tmp = (uint8_t)(x >> 16);
1225     ret |= tmp << 32;
1226     tmp = (uint8_t)(x >> 24);
1227     ret |= tmp << 48;
1228     return ret;
1229 }
1230
1231 uint64_t HELPER(neon_widen_s8)(uint32_t x)
1232 {
1233     uint64_t tmp;
1234     uint64_t ret;
1235     ret = (uint16_t)(int8_t)x;
1236     tmp = (uint16_t)(int8_t)(x >> 8);
1237     ret |= tmp << 16;
1238     tmp = (uint16_t)(int8_t)(x >> 16);
1239     ret |= tmp << 32;
1240     tmp = (uint16_t)(int8_t)(x >> 24);
1241     ret |= tmp << 48;
1242     return ret;
1243 }
1244
1245 uint64_t HELPER(neon_widen_u16)(uint32_t x)
1246 {
1247     uint64_t high = (uint16_t)(x >> 16);
1248     return ((uint16_t)x) | (high << 32);
1249 }
1250
1251 uint64_t HELPER(neon_widen_s16)(uint32_t x)
1252 {
1253     uint64_t high = (int16_t)(x >> 16);
1254     return ((uint32_t)(int16_t)x) | (high << 32);
1255 }
1256
1257 uint64_t HELPER(neon_addl_u16)(uint64_t a, uint64_t b)
1258 {
1259     uint64_t mask;
1260     mask = (a ^ b) & 0x8000800080008000ull;
1261     a &= ~0x8000800080008000ull;
1262     b &= ~0x8000800080008000ull;
1263     return (a + b) ^ mask;
1264 }
1265
1266 uint64_t HELPER(neon_addl_u32)(uint64_t a, uint64_t b)
1267 {
1268     uint64_t mask;
1269     mask = (a ^ b) & 0x8000000080000000ull;
1270     a &= ~0x8000000080000000ull;
1271     b &= ~0x8000000080000000ull;
1272     return (a + b) ^ mask;
1273 }
1274
1275 uint64_t HELPER(neon_paddl_u16)(uint64_t a, uint64_t b)
1276 {
1277     uint64_t tmp;
1278     uint64_t tmp2;
1279
1280     tmp = a & 0x0000ffff0000ffffull;
1281     tmp += (a >> 16) & 0x0000ffff0000ffffull;
1282     tmp2 = b & 0xffff0000ffff0000ull;
1283     tmp2 += (b << 16) & 0xffff0000ffff0000ull;
1284     return    ( tmp         & 0xffff)
1285             | ((tmp  >> 16) & 0xffff0000ull)
1286             | ((tmp2 << 16) & 0xffff00000000ull)
1287             | ( tmp2        & 0xffff000000000000ull);
1288 }
1289
1290 uint64_t HELPER(neon_paddl_u32)(uint64_t a, uint64_t b)
1291 {
1292     uint32_t low = a + (a >> 32);
1293     uint32_t high = b + (b >> 32);
1294     return low + ((uint64_t)high << 32);
1295 }
1296
1297 uint64_t HELPER(neon_subl_u16)(uint64_t a, uint64_t b)
1298 {
1299     uint64_t mask;
1300     mask = (a ^ ~b) & 0x8000800080008000ull;
1301     a |= 0x8000800080008000ull;
1302     b &= ~0x8000800080008000ull;
1303     return (a - b) ^ mask;
1304 }
1305
1306 uint64_t HELPER(neon_subl_u32)(uint64_t a, uint64_t b)
1307 {
1308     uint64_t mask;
1309     mask = (a ^ ~b) & 0x8000000080000000ull;
1310     a |= 0x8000000080000000ull;
1311     b &= ~0x8000000080000000ull;
1312     return (a - b) ^ mask;
1313 }
1314
1315 uint64_t HELPER(neon_addl_saturate_s32)(CPUState *env, uint64_t a, uint64_t b)
1316 {
1317     uint32_t x, y;
1318     uint32_t low, high;
1319
1320     x = a;
1321     y = b;
1322     low = x + y;
1323     if (((low ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) {
1324         SET_QC();
1325         low = ((int32_t)x >> 31) ^ ~SIGNBIT;
1326     }
1327     x = a >> 32;
1328     y = b >> 32;
1329     high = x + y;
1330     if (((high ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) {
1331         SET_QC();
1332         high = ((int32_t)x >> 31) ^ ~SIGNBIT;
1333     }
1334     return low | ((uint64_t)high << 32);
1335 }
1336
1337 uint64_t HELPER(neon_addl_saturate_s64)(CPUState *env, uint64_t a, uint64_t b)
1338 {
1339     uint64_t result;
1340
1341     result = a + b;
1342     if (((result ^ a) & SIGNBIT64) && !((a ^ b) & SIGNBIT64)) {
1343         SET_QC();
1344         result = ((int64_t)a >> 63) ^ ~SIGNBIT64;
1345     }
1346     return result;
1347 }
1348
1349 #define DO_ABD(dest, x, y, type) do { \
1350     type tmp_x = x; \
1351     type tmp_y = y; \
1352     dest = ((tmp_x > tmp_y) ? tmp_x - tmp_y : tmp_y - tmp_x); \
1353     } while(0)
1354
1355 uint64_t HELPER(neon_abdl_u16)(uint32_t a, uint32_t b)
1356 {
1357     uint64_t tmp;
1358     uint64_t result;
1359     DO_ABD(result, a, b, uint8_t);
1360     DO_ABD(tmp, a >> 8, b >> 8, uint8_t);
1361     result |= tmp << 16;
1362     DO_ABD(tmp, a >> 16, b >> 16, uint8_t);
1363     result |= tmp << 32;
1364     DO_ABD(tmp, a >> 24, b >> 24, uint8_t);
1365     result |= tmp << 48;
1366     return result;
1367 }
1368
1369 uint64_t HELPER(neon_abdl_s16)(uint32_t a, uint32_t b)
1370 {
1371     uint64_t tmp;
1372     uint64_t result;
1373     DO_ABD(result, a, b, int8_t);
1374     DO_ABD(tmp, a >> 8, b >> 8, int8_t);
1375     result |= tmp << 16;
1376     DO_ABD(tmp, a >> 16, b >> 16, int8_t);
1377     result |= tmp << 32;
1378     DO_ABD(tmp, a >> 24, b >> 24, int8_t);
1379     result |= tmp << 48;
1380     return result;
1381 }
1382
1383 uint64_t HELPER(neon_abdl_u32)(uint32_t a, uint32_t b)
1384 {
1385     uint64_t tmp;
1386     uint64_t result;
1387     DO_ABD(result, a, b, uint16_t);
1388     DO_ABD(tmp, a >> 16, b >> 16, uint16_t);
1389     return result | (tmp << 32);
1390 }
1391
1392 uint64_t HELPER(neon_abdl_s32)(uint32_t a, uint32_t b)
1393 {
1394     uint64_t tmp;
1395     uint64_t result;
1396     DO_ABD(result, a, b, int16_t);
1397     DO_ABD(tmp, a >> 16, b >> 16, int16_t);
1398     return result | (tmp << 32);
1399 }
1400
1401 uint64_t HELPER(neon_abdl_u64)(uint32_t a, uint32_t b)
1402 {
1403     uint64_t result;
1404     DO_ABD(result, a, b, uint32_t);
1405     return result;
1406 }
1407
1408 uint64_t HELPER(neon_abdl_s64)(uint32_t a, uint32_t b)
1409 {
1410     uint64_t result;
1411     DO_ABD(result, a, b, int32_t);
1412     return result;
1413 }
1414 #undef DO_ABD
1415
1416 /* Widening multiply. Named type is the source type.  */
1417 #define DO_MULL(dest, x, y, type1, type2) do { \
1418     type1 tmp_x = x; \
1419     type1 tmp_y = y; \
1420     dest = (type2)((type2)tmp_x * (type2)tmp_y); \
1421     } while(0)
1422
1423 uint64_t HELPER(neon_mull_u8)(uint32_t a, uint32_t b)
1424 {
1425     uint64_t tmp;
1426     uint64_t result;
1427
1428     DO_MULL(result, a, b, uint8_t, uint16_t);
1429     DO_MULL(tmp, a >> 8, b >> 8, uint8_t, uint16_t);
1430     result |= tmp << 16;
1431     DO_MULL(tmp, a >> 16, b >> 16, uint8_t, uint16_t);
1432     result |= tmp << 32;
1433     DO_MULL(tmp, a >> 24, b >> 24, uint8_t, uint16_t);
1434     result |= tmp << 48;
1435     return result;
1436 }
1437
1438 uint64_t HELPER(neon_mull_s8)(uint32_t a, uint32_t b)
1439 {
1440     uint64_t tmp;
1441     uint64_t result;
1442
1443     DO_MULL(result, a, b, int8_t, uint16_t);
1444     DO_MULL(tmp, a >> 8, b >> 8, int8_t, uint16_t);
1445     result |= tmp << 16;
1446     DO_MULL(tmp, a >> 16, b >> 16, int8_t, uint16_t);
1447     result |= tmp << 32;
1448     DO_MULL(tmp, a >> 24, b >> 24, int8_t, uint16_t);
1449     result |= tmp << 48;
1450     return result;
1451 }
1452
1453 uint64_t HELPER(neon_mull_u16)(uint32_t a, uint32_t b)
1454 {
1455     uint64_t tmp;
1456     uint64_t result;
1457
1458     DO_MULL(result, a, b, uint16_t, uint32_t);
1459     DO_MULL(tmp, a >> 16, b >> 16, uint16_t, uint32_t);
1460     return result | (tmp << 32);
1461 }
1462
1463 uint64_t HELPER(neon_mull_s16)(uint32_t a, uint32_t b)
1464 {
1465     uint64_t tmp;
1466     uint64_t result;
1467
1468     DO_MULL(result, a, b, int16_t, uint32_t);
1469     DO_MULL(tmp, a >> 16, b >> 16, int16_t, uint32_t);
1470     return result | (tmp << 32);
1471 }
1472
1473 uint64_t HELPER(neon_negl_u16)(uint64_t x)
1474 {
1475     uint16_t tmp;
1476     uint64_t result;
1477     result = (uint16_t)-x;
1478     tmp = -(x >> 16);
1479     result |= (uint64_t)tmp << 16;
1480     tmp = -(x >> 32);
1481     result |= (uint64_t)tmp << 32;
1482     tmp = -(x >> 48);
1483     result |= (uint64_t)tmp << 48;
1484     return result;
1485 }
1486
1487 uint64_t HELPER(neon_negl_u32)(uint64_t x)
1488 {
1489     uint32_t low = -x;
1490     uint32_t high = -(x >> 32);
1491     return low | ((uint64_t)high << 32);
1492 }
1493
1494 /* FIXME:  There should be a native op for this.  */
1495 uint64_t HELPER(neon_negl_u64)(uint64_t x)
1496 {
1497     return -x;
1498 }
1499
1500 /* Saturnating sign manuipulation.  */
1501 /* ??? Make these use NEON_VOP1 */
1502 #define DO_QABS8(x) do { \
1503     if (x == (int8_t)0x80) { \
1504         x = 0x7f; \
1505         SET_QC(); \
1506     } else if (x < 0) { \
1507         x = -x; \
1508     }} while (0)
1509 uint32_t HELPER(neon_qabs_s8)(CPUState *env, uint32_t x)
1510 {
1511     neon_s8 vec;
1512     NEON_UNPACK(neon_s8, vec, x);
1513     DO_QABS8(vec.v1);
1514     DO_QABS8(vec.v2);
1515     DO_QABS8(vec.v3);
1516     DO_QABS8(vec.v4);
1517     NEON_PACK(neon_s8, x, vec);
1518     return x;
1519 }
1520 #undef DO_QABS8
1521
1522 #define DO_QNEG8(x) do { \
1523     if (x == (int8_t)0x80) { \
1524         x = 0x7f; \
1525         SET_QC(); \
1526     } else { \
1527         x = -x; \
1528     }} while (0)
1529 uint32_t HELPER(neon_qneg_s8)(CPUState *env, uint32_t x)
1530 {
1531     neon_s8 vec;
1532     NEON_UNPACK(neon_s8, vec, x);
1533     DO_QNEG8(vec.v1);
1534     DO_QNEG8(vec.v2);
1535     DO_QNEG8(vec.v3);
1536     DO_QNEG8(vec.v4);
1537     NEON_PACK(neon_s8, x, vec);
1538     return x;
1539 }
1540 #undef DO_QNEG8
1541
1542 #define DO_QABS16(x) do { \
1543     if (x == (int16_t)0x8000) { \
1544         x = 0x7fff; \
1545         SET_QC(); \
1546     } else if (x < 0) { \
1547         x = -x; \
1548     }} while (0)
1549 uint32_t HELPER(neon_qabs_s16)(CPUState *env, uint32_t x)
1550 {
1551     neon_s16 vec;
1552     NEON_UNPACK(neon_s16, vec, x);
1553     DO_QABS16(vec.v1);
1554     DO_QABS16(vec.v2);
1555     NEON_PACK(neon_s16, x, vec);
1556     return x;
1557 }
1558 #undef DO_QABS16
1559
1560 #define DO_QNEG16(x) do { \
1561     if (x == (int16_t)0x8000) { \
1562         x = 0x7fff; \
1563         SET_QC(); \
1564     } else { \
1565         x = -x; \
1566     }} while (0)
1567 uint32_t HELPER(neon_qneg_s16)(CPUState *env, uint32_t x)
1568 {
1569     neon_s16 vec;
1570     NEON_UNPACK(neon_s16, vec, x);
1571     DO_QNEG16(vec.v1);
1572     DO_QNEG16(vec.v2);
1573     NEON_PACK(neon_s16, x, vec);
1574     return x;
1575 }
1576 #undef DO_QNEG16
1577
1578 uint32_t HELPER(neon_qabs_s32)(CPUState *env, uint32_t x)
1579 {
1580     if (x == SIGNBIT) {
1581         SET_QC();
1582         x = ~SIGNBIT;
1583     } else if ((int32_t)x < 0) {
1584         x = -x;
1585     }
1586     return x;
1587 }
1588
1589 uint32_t HELPER(neon_qneg_s32)(CPUState *env, uint32_t x)
1590 {
1591     if (x == SIGNBIT) {
1592         SET_QC();
1593         x = ~SIGNBIT;
1594     } else {
1595         x = -x;
1596     }
1597     return x;
1598 }
1599
1600 /* NEON Float helpers.  */
1601 uint32_t HELPER(neon_min_f32)(uint32_t a, uint32_t b)
1602 {
1603     float32 f0 = vfp_itos(a);
1604     float32 f1 = vfp_itos(b);
1605     return (float32_compare_quiet(f0, f1, NFS) == -1) ? a : b;
1606 }
1607
1608 uint32_t HELPER(neon_max_f32)(uint32_t a, uint32_t b)
1609 {
1610     float32 f0 = vfp_itos(a);
1611     float32 f1 = vfp_itos(b);
1612     return (float32_compare_quiet(f0, f1, NFS) == 1) ? a : b;
1613 }
1614
1615 uint32_t HELPER(neon_abd_f32)(uint32_t a, uint32_t b)
1616 {
1617     float32 f0 = vfp_itos(a);
1618     float32 f1 = vfp_itos(b);
1619     return vfp_stoi((float32_compare_quiet(f0, f1, NFS) == 1)
1620                     ? float32_sub(f0, f1, NFS)
1621                     : float32_sub(f1, f0, NFS));
1622 }
1623
1624 uint32_t HELPER(neon_add_f32)(uint32_t a, uint32_t b)
1625 {
1626     return vfp_stoi(float32_add(vfp_itos(a), vfp_itos(b), NFS));
1627 }
1628
1629 uint32_t HELPER(neon_sub_f32)(uint32_t a, uint32_t b)
1630 {
1631     return vfp_stoi(float32_sub(vfp_itos(a), vfp_itos(b), NFS));
1632 }
1633
1634 uint32_t HELPER(neon_mul_f32)(uint32_t a, uint32_t b)
1635 {
1636     return vfp_stoi(float32_mul(vfp_itos(a), vfp_itos(b), NFS));
1637 }
1638
1639 /* Floating point comparisons produce an integer result.  */
1640 #define NEON_VOP_FCMP(name, cmp) \
1641 uint32_t HELPER(neon_##name)(uint32_t a, uint32_t b) \
1642 { \
1643     if (float32_compare_quiet(vfp_itos(a), vfp_itos(b), NFS) cmp 0) \
1644         return ~0; \
1645     else \
1646         return 0; \
1647 }
1648
1649 NEON_VOP_FCMP(ceq_f32, ==)
1650 NEON_VOP_FCMP(cge_f32, >=)
1651 NEON_VOP_FCMP(cgt_f32, >)
1652
1653 uint32_t HELPER(neon_acge_f32)(uint32_t a, uint32_t b)
1654 {
1655     float32 f0 = float32_abs(vfp_itos(a));
1656     float32 f1 = float32_abs(vfp_itos(b));
1657     return (float32_compare_quiet(f0, f1,NFS) >= 0) ? ~0 : 0;
1658 }
1659
1660 uint32_t HELPER(neon_acgt_f32)(uint32_t a, uint32_t b)
1661 {
1662     float32 f0 = float32_abs(vfp_itos(a));
1663     float32 f1 = float32_abs(vfp_itos(b));
1664     return (float32_compare_quiet(f0, f1, NFS) > 0) ? ~0 : 0;
1665 }