qemu

FORK: QEMU emulator
git clone https://git.neptards.moe/neptards/qemu.git
Log | Files | Refs | Submodules | LICENSE

emit.c.inc (80977B)


      1 /*
      2  * New-style TCG opcode generator for i386 instructions
      3  *
      4  *  Copyright (c) 2022 Red Hat, Inc.
      5  *
      6  * Author: Paolo Bonzini <pbonzini@redhat.com>
      7  *
      8  * This library is free software; you can redistribute it and/or
      9  * modify it under the terms of the GNU Lesser General Public
     10  * License as published by the Free Software Foundation; either
     11  * version 2.1 of the License, or (at your option) any later version.
     12  *
     13  * This library is distributed in the hope that it will be useful,
     14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
     15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
     16  * Lesser General Public License for more details.
     17  *
     18  * You should have received a copy of the GNU Lesser General Public
     19  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
     20  */
     21 
     22 #define ZMM_OFFSET(reg) offsetof(CPUX86State, xmm_regs[reg])
     23 
     24 typedef void (*SSEFunc_i_ep)(TCGv_i32 val, TCGv_ptr env, TCGv_ptr reg);
     25 typedef void (*SSEFunc_l_ep)(TCGv_i64 val, TCGv_ptr env, TCGv_ptr reg);
     26 typedef void (*SSEFunc_0_epp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b);
     27 typedef void (*SSEFunc_0_eppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
     28                                TCGv_ptr reg_c);
     29 typedef void (*SSEFunc_0_epppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
     30                                 TCGv_ptr reg_c, TCGv_ptr reg_d);
     31 typedef void (*SSEFunc_0_eppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
     32                                TCGv_i32 val);
     33 typedef void (*SSEFunc_0_epppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
     34                                 TCGv_ptr reg_c, TCGv_i32 val);
     35 typedef void (*SSEFunc_0_ppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_i32 val);
     36 typedef void (*SSEFunc_0_pppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_ptr reg_c,
     37                                TCGv_i32 val);
     38 typedef void (*SSEFunc_0_eppt)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
     39                                TCGv val);
     40 typedef void (*SSEFunc_0_epppti)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
     41                                  TCGv_ptr reg_c, TCGv a0, TCGv_i32 scale);
     42 typedef void (*SSEFunc_0_eppppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
     43                                   TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 flags);
     44 typedef void (*SSEFunc_0_eppppii)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
     45                                   TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 even,
     46                                   TCGv_i32 odd);
     47 
     48 static inline TCGv_i32 tcg_constant8u_i32(uint8_t val)
     49 {
     50     return tcg_constant_i32(val);
     51 }
     52 
     53 static void gen_NM_exception(DisasContext *s)
     54 {
     55     gen_exception(s, EXCP07_PREX);
     56 }
     57 
     58 static void gen_illegal(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
     59 {
     60     gen_illegal_opcode(s);
     61 }
     62 
     63 static void gen_load_ea(DisasContext *s, AddressParts *mem, bool is_vsib)
     64 {
     65     TCGv ea = gen_lea_modrm_1(s, *mem, is_vsib);
     66     gen_lea_v_seg(s, s->aflag, ea, mem->def_seg, s->override);
     67 }
     68 
     69 static inline int mmx_offset(MemOp ot)
     70 {
     71     switch (ot) {
     72     case MO_8:
     73         return offsetof(MMXReg, MMX_B(0));
     74     case MO_16:
     75         return offsetof(MMXReg, MMX_W(0));
     76     case MO_32:
     77         return offsetof(MMXReg, MMX_L(0));
     78     case MO_64:
     79         return offsetof(MMXReg, MMX_Q(0));
     80     default:
     81         g_assert_not_reached();
     82     }
     83 }
     84 
     85 static inline int xmm_offset(MemOp ot)
     86 {
     87     switch (ot) {
     88     case MO_8:
     89         return offsetof(ZMMReg, ZMM_B(0));
     90     case MO_16:
     91         return offsetof(ZMMReg, ZMM_W(0));
     92     case MO_32:
     93         return offsetof(ZMMReg, ZMM_L(0));
     94     case MO_64:
     95         return offsetof(ZMMReg, ZMM_Q(0));
     96     case MO_128:
     97         return offsetof(ZMMReg, ZMM_X(0));
     98     case MO_256:
     99         return offsetof(ZMMReg, ZMM_Y(0));
    100     default:
    101         g_assert_not_reached();
    102     }
    103 }
    104 
    105 static int vector_reg_offset(X86DecodedOp *op)
    106 {
    107     assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
    108 
    109     if (op->unit == X86_OP_MMX) {
    110         return op->offset - mmx_offset(op->ot);
    111     } else {
    112         return op->offset - xmm_offset(op->ot);
    113     }
    114 }
    115 
    116 static int vector_elem_offset(X86DecodedOp *op, MemOp ot, int n)
    117 {
    118     int base_ofs = vector_reg_offset(op);
    119     switch(ot) {
    120     case MO_8:
    121         if (op->unit == X86_OP_MMX) {
    122             return base_ofs + offsetof(MMXReg, MMX_B(n));
    123         } else {
    124             return base_ofs + offsetof(ZMMReg, ZMM_B(n));
    125         }
    126     case MO_16:
    127         if (op->unit == X86_OP_MMX) {
    128             return base_ofs + offsetof(MMXReg, MMX_W(n));
    129         } else {
    130             return base_ofs + offsetof(ZMMReg, ZMM_W(n));
    131         }
    132     case MO_32:
    133         if (op->unit == X86_OP_MMX) {
    134             return base_ofs + offsetof(MMXReg, MMX_L(n));
    135         } else {
    136             return base_ofs + offsetof(ZMMReg, ZMM_L(n));
    137         }
    138     case MO_64:
    139         if (op->unit == X86_OP_MMX) {
    140             return base_ofs;
    141         } else {
    142             return base_ofs + offsetof(ZMMReg, ZMM_Q(n));
    143         }
    144     case MO_128:
    145         assert(op->unit == X86_OP_SSE);
    146         return base_ofs + offsetof(ZMMReg, ZMM_X(n));
    147     case MO_256:
    148         assert(op->unit == X86_OP_SSE);
    149         return base_ofs + offsetof(ZMMReg, ZMM_Y(n));
    150     default:
    151         g_assert_not_reached();
    152     }
    153 }
    154 
    155 static void compute_mmx_offset(X86DecodedOp *op)
    156 {
    157     if (!op->has_ea) {
    158         op->offset = offsetof(CPUX86State, fpregs[op->n].mmx) + mmx_offset(op->ot);
    159     } else {
    160         op->offset = offsetof(CPUX86State, mmx_t0) + mmx_offset(op->ot);
    161     }
    162 }
    163 
    164 static void compute_xmm_offset(X86DecodedOp *op)
    165 {
    166     if (!op->has_ea) {
    167         op->offset = ZMM_OFFSET(op->n) + xmm_offset(op->ot);
    168     } else {
    169         op->offset = offsetof(CPUX86State, xmm_t0) + xmm_offset(op->ot);
    170     }
    171 }
    172 
    173 static void gen_load_sse(DisasContext *s, TCGv temp, MemOp ot, int dest_ofs, bool aligned)
    174 {
    175     switch(ot) {
    176     case MO_8:
    177         gen_op_ld_v(s, MO_8, temp, s->A0);
    178         tcg_gen_st8_tl(temp, cpu_env, dest_ofs);
    179         break;
    180     case MO_16:
    181         gen_op_ld_v(s, MO_16, temp, s->A0);
    182         tcg_gen_st16_tl(temp, cpu_env, dest_ofs);
    183         break;
    184     case MO_32:
    185         gen_op_ld_v(s, MO_32, temp, s->A0);
    186         tcg_gen_st32_tl(temp, cpu_env, dest_ofs);
    187         break;
    188     case MO_64:
    189         gen_ldq_env_A0(s, dest_ofs);
    190         break;
    191     case MO_128:
    192         gen_ldo_env_A0(s, dest_ofs, aligned);
    193         break;
    194     case MO_256:
    195         gen_ldy_env_A0(s, dest_ofs, aligned);
    196         break;
    197     default:
    198         g_assert_not_reached();
    199     }
    200 }
    201 
    202 static bool sse_needs_alignment(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
    203 {
    204     switch (decode->e.vex_class) {
    205     case 2:
    206     case 4:
    207         if ((s->prefix & PREFIX_VEX) ||
    208             decode->e.vex_special == X86_VEX_SSEUnaligned) {
    209             /* MOST legacy SSE instructions require aligned memory operands, but not all.  */
    210             return false;
    211         }
    212         /* fall through */
    213     case 1:
    214         return ot >= MO_128;
    215 
    216     default:
    217         return false;
    218     }
    219 }
    220 
    221 static void gen_load(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
    222 {
    223     X86DecodedOp *op = &decode->op[opn];
    224 
    225     switch (op->unit) {
    226     case X86_OP_SKIP:
    227         return;
    228     case X86_OP_SEG:
    229         tcg_gen_ld32u_tl(v, cpu_env,
    230                          offsetof(CPUX86State,segs[op->n].selector));
    231         break;
    232     case X86_OP_CR:
    233         tcg_gen_ld_tl(v, cpu_env, offsetof(CPUX86State, cr[op->n]));
    234         break;
    235     case X86_OP_DR:
    236         tcg_gen_ld_tl(v, cpu_env, offsetof(CPUX86State, dr[op->n]));
    237         break;
    238     case X86_OP_INT:
    239         if (op->has_ea) {
    240             gen_op_ld_v(s, op->ot, v, s->A0);
    241         } else {
    242             gen_op_mov_v_reg(s, op->ot, v, op->n);
    243         }
    244         break;
    245     case X86_OP_IMM:
    246         tcg_gen_movi_tl(v, decode->immediate);
    247         break;
    248 
    249     case X86_OP_MMX:
    250         compute_mmx_offset(op);
    251         goto load_vector;
    252 
    253     case X86_OP_SSE:
    254         compute_xmm_offset(op);
    255     load_vector:
    256         if (op->has_ea) {
    257             bool aligned = sse_needs_alignment(s, decode, op->ot);
    258             gen_load_sse(s, v, op->ot, op->offset, aligned);
    259         }
    260         break;
    261 
    262     default:
    263         g_assert_not_reached();
    264     }
    265 }
    266 
    267 static TCGv_ptr op_ptr(X86DecodedInsn *decode, int opn)
    268 {
    269     X86DecodedOp *op = &decode->op[opn];
    270     if (op->v_ptr) {
    271         return op->v_ptr;
    272     }
    273     op->v_ptr = tcg_temp_new_ptr();
    274 
    275     /* The temporary points to the MMXReg or ZMMReg.  */
    276     tcg_gen_addi_ptr(op->v_ptr, cpu_env, vector_reg_offset(op));
    277     return op->v_ptr;
    278 }
    279 
    280 #define OP_PTR0 op_ptr(decode, 0)
    281 #define OP_PTR1 op_ptr(decode, 1)
    282 #define OP_PTR2 op_ptr(decode, 2)
    283 
    284 static void gen_writeback(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
    285 {
    286     X86DecodedOp *op = &decode->op[opn];
    287     switch (op->unit) {
    288     case X86_OP_SKIP:
    289         break;
    290     case X86_OP_SEG:
    291         /* Note that gen_movl_seg_T0 takes care of interrupt shadow and TF.  */
    292         gen_movl_seg_T0(s, op->n);
    293         break;
    294     case X86_OP_INT:
    295         if (op->has_ea) {
    296             gen_op_st_v(s, op->ot, v, s->A0);
    297         } else {
    298             gen_op_mov_reg_v(s, op->ot, op->n, v);
    299         }
    300         break;
    301     case X86_OP_MMX:
    302         break;
    303     case X86_OP_SSE:
    304         if (!op->has_ea && (s->prefix & PREFIX_VEX) && op->ot <= MO_128) {
    305             tcg_gen_gvec_dup_imm(MO_64,
    306                                  offsetof(CPUX86State, xmm_regs[op->n].ZMM_X(1)),
    307                                  16, 16, 0);
    308         }
    309         break;
    310     case X86_OP_CR:
    311     case X86_OP_DR:
    312     default:
    313         g_assert_not_reached();
    314     }
    315 }
    316 
    317 static inline int vector_len(DisasContext *s, X86DecodedInsn *decode)
    318 {
    319     if (decode->e.special == X86_SPECIAL_MMX &&
    320         !(s->prefix & (PREFIX_DATA | PREFIX_REPZ | PREFIX_REPNZ))) {
    321         return 8;
    322     }
    323     return s->vex_l ? 32 : 16;
    324 }
    325 
    326 static void gen_store_sse(DisasContext *s, X86DecodedInsn *decode, int src_ofs)
    327 {
    328     MemOp ot = decode->op[0].ot;
    329     int vec_len = vector_len(s, decode);
    330     bool aligned = sse_needs_alignment(s, decode, ot);
    331 
    332     if (!decode->op[0].has_ea) {
    333         tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, vec_len, vec_len);
    334         return;
    335     }
    336 
    337     switch (ot) {
    338     case MO_64:
    339         gen_stq_env_A0(s, src_ofs);
    340         break;
    341     case MO_128:
    342         gen_sto_env_A0(s, src_ofs, aligned);
    343         break;
    344     case MO_256:
    345         gen_sty_env_A0(s, src_ofs, aligned);
    346         break;
    347     default:
    348         g_assert_not_reached();
    349     }
    350 }
    351 
    352 static void gen_helper_pavgusb(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b)
    353 {
    354     gen_helper_pavgb_mmx(env, reg_a, reg_a, reg_b);
    355 }
    356 
    357 #define FN_3DNOW_MOVE ((SSEFunc_0_epp) (uintptr_t) 1)
    358 static const SSEFunc_0_epp fns_3dnow[] = {
    359     [0x0c] = gen_helper_pi2fw,
    360     [0x0d] = gen_helper_pi2fd,
    361     [0x1c] = gen_helper_pf2iw,
    362     [0x1d] = gen_helper_pf2id,
    363     [0x8a] = gen_helper_pfnacc,
    364     [0x8e] = gen_helper_pfpnacc,
    365     [0x90] = gen_helper_pfcmpge,
    366     [0x94] = gen_helper_pfmin,
    367     [0x96] = gen_helper_pfrcp,
    368     [0x97] = gen_helper_pfrsqrt,
    369     [0x9a] = gen_helper_pfsub,
    370     [0x9e] = gen_helper_pfadd,
    371     [0xa0] = gen_helper_pfcmpgt,
    372     [0xa4] = gen_helper_pfmax,
    373     [0xa6] = FN_3DNOW_MOVE, /* PFRCPIT1; no need to actually increase precision */
    374     [0xa7] = FN_3DNOW_MOVE, /* PFRSQIT1 */
    375     [0xb6] = FN_3DNOW_MOVE, /* PFRCPIT2 */
    376     [0xaa] = gen_helper_pfsubr,
    377     [0xae] = gen_helper_pfacc,
    378     [0xb0] = gen_helper_pfcmpeq,
    379     [0xb4] = gen_helper_pfmul,
    380     [0xb7] = gen_helper_pmulhrw_mmx,
    381     [0xbb] = gen_helper_pswapd,
    382     [0xbf] = gen_helper_pavgusb,
    383 };
    384 
    385 static void gen_3dnow(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
    386 {
    387     uint8_t b = decode->immediate;
    388     SSEFunc_0_epp fn = b < ARRAY_SIZE(fns_3dnow) ? fns_3dnow[b] : NULL;
    389 
    390     if (!fn) {
    391         gen_illegal_opcode(s);
    392         return;
    393     }
    394     if (s->flags & HF_TS_MASK) {
    395         gen_NM_exception(s);
    396         return;
    397     }
    398     if (s->flags & HF_EM_MASK) {
    399         gen_illegal_opcode(s);
    400         return;
    401     }
    402 
    403     gen_helper_enter_mmx(cpu_env);
    404     if (fn == FN_3DNOW_MOVE) {
    405        tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[1].offset);
    406        tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset);
    407     } else {
    408        fn(cpu_env, OP_PTR0, OP_PTR1);
    409     }
    410 }
    411 
    412 /*
    413  * 00 = v*ps Vps, Hps, Wpd
    414  * 66 = v*pd Vpd, Hpd, Wps
    415  * f3 = v*ss Vss, Hss, Wps
    416  * f2 = v*sd Vsd, Hsd, Wps
    417  */
    418 static inline void gen_unary_fp_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    419                               SSEFunc_0_epp pd_xmm, SSEFunc_0_epp ps_xmm,
    420                               SSEFunc_0_epp pd_ymm, SSEFunc_0_epp ps_ymm,
    421                               SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
    422 {
    423     if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
    424         SSEFunc_0_eppp fn = s->prefix & PREFIX_REPZ ? ss : sd;
    425         if (!fn) {
    426             gen_illegal_opcode(s);
    427             return;
    428         }
    429         fn(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
    430     } else {
    431         SSEFunc_0_epp ps, pd, fn;
    432         ps = s->vex_l ? ps_ymm : ps_xmm;
    433         pd = s->vex_l ? pd_ymm : pd_xmm;
    434         fn = s->prefix & PREFIX_DATA ? pd : ps;
    435         if (!fn) {
    436             gen_illegal_opcode(s);
    437             return;
    438         }
    439         fn(cpu_env, OP_PTR0, OP_PTR2);
    440     }
    441 }
    442 #define UNARY_FP_SSE(uname, lname)                                                 \
    443 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    444 {                                                                                  \
    445     gen_unary_fp_sse(s, env, decode,                                               \
    446                      gen_helper_##lname##pd_xmm,                                   \
    447                      gen_helper_##lname##ps_xmm,                                   \
    448                      gen_helper_##lname##pd_ymm,                                   \
    449                      gen_helper_##lname##ps_ymm,                                   \
    450                      gen_helper_##lname##sd,                                       \
    451                      gen_helper_##lname##ss);                                      \
    452 }
    453 UNARY_FP_SSE(VSQRT, sqrt)
    454 
    455 /*
    456  * 00 = v*ps Vps, Hps, Wpd
    457  * 66 = v*pd Vpd, Hpd, Wps
    458  * f3 = v*ss Vss, Hss, Wps
    459  * f2 = v*sd Vsd, Hsd, Wps
    460  */
    461 static inline void gen_fp_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    462                               SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
    463                               SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm,
    464                               SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
    465 {
    466     SSEFunc_0_eppp ps, pd, fn;
    467     if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
    468         fn = s->prefix & PREFIX_REPZ ? ss : sd;
    469     } else {
    470         ps = s->vex_l ? ps_ymm : ps_xmm;
    471         pd = s->vex_l ? pd_ymm : pd_xmm;
    472         fn = s->prefix & PREFIX_DATA ? pd : ps;
    473     }
    474     if (fn) {
    475         fn(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
    476     } else {
    477         gen_illegal_opcode(s);
    478     }
    479 }
    480 
    481 #define FP_SSE(uname, lname)                                                       \
    482 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    483 {                                                                                  \
    484     gen_fp_sse(s, env, decode,                                                     \
    485                gen_helper_##lname##pd_xmm,                                         \
    486                gen_helper_##lname##ps_xmm,                                         \
    487                gen_helper_##lname##pd_ymm,                                         \
    488                gen_helper_##lname##ps_ymm,                                         \
    489                gen_helper_##lname##sd,                                             \
    490                gen_helper_##lname##ss);                                            \
    491 }
    492 FP_SSE(VADD, add)
    493 FP_SSE(VMUL, mul)
    494 FP_SSE(VSUB, sub)
    495 FP_SSE(VMIN, min)
    496 FP_SSE(VDIV, div)
    497 FP_SSE(VMAX, max)
    498 
    499 #define FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, even, odd)                         \
    500 static void gen_##uname##Px(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    501 {                                                                                  \
    502     SSEFunc_0_eppppii xmm = s->vex_w ? gen_helper_fma4pd_xmm : gen_helper_fma4ps_xmm; \
    503     SSEFunc_0_eppppii ymm = s->vex_w ? gen_helper_fma4pd_ymm : gen_helper_fma4ps_ymm; \
    504     SSEFunc_0_eppppii fn = s->vex_l ? ymm : xmm;                                   \
    505                                                                                    \
    506     fn(cpu_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
    507        tcg_constant_i32(even),                                                     \
    508        tcg_constant_i32((even) ^ (odd)));                                          \
    509 }
    510 
    511 #define FMA_SSE(uname, ptr0, ptr1, ptr2, flags)                                    \
    512 FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, flags, flags)                              \
    513 static void gen_##uname##Sx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    514 {                                                                                  \
    515     SSEFunc_0_eppppi fn = s->vex_w ? gen_helper_fma4sd : gen_helper_fma4ss;        \
    516                                                                                    \
    517     fn(cpu_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
    518        tcg_constant_i32(flags));                                                   \
    519 }                                                                                  \
    520 
    521 FMA_SSE(VFMADD231,  OP_PTR1, OP_PTR2, OP_PTR0, 0)
    522 FMA_SSE(VFMADD213,  OP_PTR1, OP_PTR0, OP_PTR2, 0)
    523 FMA_SSE(VFMADD132,  OP_PTR0, OP_PTR2, OP_PTR1, 0)
    524 
    525 FMA_SSE(VFNMADD231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_product)
    526 FMA_SSE(VFNMADD213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_product)
    527 FMA_SSE(VFNMADD132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_product)
    528 
    529 FMA_SSE(VFMSUB231,  OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c)
    530 FMA_SSE(VFMSUB213,  OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c)
    531 FMA_SSE(VFMSUB132,  OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c)
    532 
    533 FMA_SSE(VFNMSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c|float_muladd_negate_product)
    534 FMA_SSE(VFNMSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c|float_muladd_negate_product)
    535 FMA_SSE(VFNMSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c|float_muladd_negate_product)
    536 
    537 FMA_SSE_PACKED(VFMADDSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c, 0)
    538 FMA_SSE_PACKED(VFMADDSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c, 0)
    539 FMA_SSE_PACKED(VFMADDSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c, 0)
    540 
    541 FMA_SSE_PACKED(VFMSUBADD231, OP_PTR1, OP_PTR2, OP_PTR0, 0, float_muladd_negate_c)
    542 FMA_SSE_PACKED(VFMSUBADD213, OP_PTR1, OP_PTR0, OP_PTR2, 0, float_muladd_negate_c)
    543 FMA_SSE_PACKED(VFMSUBADD132, OP_PTR0, OP_PTR2, OP_PTR1, 0, float_muladd_negate_c)
    544 
    545 #define FP_UNPACK_SSE(uname, lname)                                                \
    546 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    547 {                                                                                  \
    548     /* PS maps to the DQ integer instruction, PD maps to QDQ.  */                  \
    549     gen_fp_sse(s, env, decode,                                                     \
    550                gen_helper_##lname##qdq_xmm,                                        \
    551                gen_helper_##lname##dq_xmm,                                         \
    552                gen_helper_##lname##qdq_ymm,                                        \
    553                gen_helper_##lname##dq_ymm,                                         \
    554                NULL, NULL);                                                        \
    555 }
    556 FP_UNPACK_SSE(VUNPCKLPx, punpckl)
    557 FP_UNPACK_SSE(VUNPCKHPx, punpckh)
    558 
    559 /*
    560  * 00 = v*ps Vps, Wpd
    561  * f3 = v*ss Vss, Wps
    562  */
    563 static inline void gen_unary_fp32_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    564                                       SSEFunc_0_epp ps_xmm,
    565                                       SSEFunc_0_epp ps_ymm,
    566                                       SSEFunc_0_eppp ss)
    567 {
    568     if ((s->prefix & (PREFIX_DATA | PREFIX_REPNZ)) != 0) {
    569         goto illegal_op;
    570     } else if (s->prefix & PREFIX_REPZ) {
    571         if (!ss) {
    572             goto illegal_op;
    573         }
    574         ss(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
    575     } else {
    576         SSEFunc_0_epp fn = s->vex_l ? ps_ymm : ps_xmm;
    577         if (!fn) {
    578             goto illegal_op;
    579         }
    580         fn(cpu_env, OP_PTR0, OP_PTR2);
    581     }
    582     return;
    583 
    584 illegal_op:
    585     gen_illegal_opcode(s);
    586 }
    587 #define UNARY_FP32_SSE(uname, lname)                                               \
    588 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    589 {                                                                                  \
    590     gen_unary_fp32_sse(s, env, decode,                                             \
    591                        gen_helper_##lname##ps_xmm,                                 \
    592                        gen_helper_##lname##ps_ymm,                                 \
    593                        gen_helper_##lname##ss);                                    \
    594 }
    595 UNARY_FP32_SSE(VRSQRT, rsqrt)
    596 UNARY_FP32_SSE(VRCP, rcp)
    597 
    598 /*
    599  * 66 = v*pd Vpd, Hpd, Wpd
    600  * f2 = v*ps Vps, Hps, Wps
    601  */
    602 static inline void gen_horizontal_fp_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    603                                          SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
    604                                          SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm)
    605 {
    606     SSEFunc_0_eppp ps, pd, fn;
    607     ps = s->vex_l ? ps_ymm : ps_xmm;
    608     pd = s->vex_l ? pd_ymm : pd_xmm;
    609     fn = s->prefix & PREFIX_DATA ? pd : ps;
    610     fn(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
    611 }
    612 #define HORIZONTAL_FP_SSE(uname, lname)                                            \
    613 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    614 {                                                                                  \
    615     gen_horizontal_fp_sse(s, env, decode,                                          \
    616                           gen_helper_##lname##pd_xmm, gen_helper_##lname##ps_xmm,  \
    617                           gen_helper_##lname##pd_ymm, gen_helper_##lname##ps_ymm); \
    618 }
    619 HORIZONTAL_FP_SSE(VHADD, hadd)
    620 HORIZONTAL_FP_SSE(VHSUB, hsub)
    621 HORIZONTAL_FP_SSE(VADDSUB, addsub)
    622 
    623 static inline void gen_ternary_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    624                                    int op3, SSEFunc_0_epppp xmm, SSEFunc_0_epppp ymm)
    625 {
    626     SSEFunc_0_epppp fn = s->vex_l ? ymm : xmm;
    627     TCGv_ptr ptr3 = tcg_temp_new_ptr();
    628 
    629     /* The format of the fourth input is Lx */
    630     tcg_gen_addi_ptr(ptr3, cpu_env, ZMM_OFFSET(op3));
    631     fn(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2, ptr3);
    632     tcg_temp_free_ptr(ptr3);
    633 }
    634 #define TERNARY_SSE(uname, uvname, lname)                                          \
    635 static void gen_##uvname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    636 {                                                                                  \
    637     gen_ternary_sse(s, env, decode, (uint8_t)decode->immediate >> 4,               \
    638                     gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);           \
    639 }                                                                                  \
    640 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    641 {                                                                                  \
    642     gen_ternary_sse(s, env, decode, 0,                                             \
    643                   gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);             \
    644 }
    645 TERNARY_SSE(BLENDVPS, VBLENDVPS, blendvps)
    646 TERNARY_SSE(BLENDVPD, VBLENDVPD, blendvpd)
    647 TERNARY_SSE(PBLENDVB, VPBLENDVB, pblendvb)
    648 
    649 static inline void gen_binary_imm_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    650                                       SSEFunc_0_epppi xmm, SSEFunc_0_epppi ymm)
    651 {
    652     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
    653     if (!s->vex_l) {
    654         xmm(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
    655     } else {
    656         ymm(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
    657     }
    658 }
    659 
    660 #define BINARY_IMM_SSE(uname, lname)                                               \
    661 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    662 {                                                                                  \
    663     gen_binary_imm_sse(s, env, decode,                                             \
    664                        gen_helper_##lname##_xmm,                                   \
    665                        gen_helper_##lname##_ymm);                                  \
    666 }
    667 
    668 BINARY_IMM_SSE(VBLENDPD,   blendpd)
    669 BINARY_IMM_SSE(VBLENDPS,   blendps)
    670 BINARY_IMM_SSE(VPBLENDW,   pblendw)
    671 BINARY_IMM_SSE(VDDPS,      dpps)
    672 #define gen_helper_dppd_ymm NULL
    673 BINARY_IMM_SSE(VDDPD,      dppd)
    674 BINARY_IMM_SSE(VMPSADBW,   mpsadbw)
    675 BINARY_IMM_SSE(PCLMULQDQ,  pclmulqdq)
    676 
    677 
    678 #define UNARY_INT_GVEC(uname, func, ...)                                           \
    679 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    680 {                                                                                  \
    681     int vec_len = vector_len(s, decode);                                          \
    682                                                                                    \
    683     func(__VA_ARGS__, decode->op[0].offset,                                        \
    684          decode->op[2].offset, vec_len, vec_len);                                  \
    685 }
    686 UNARY_INT_GVEC(PABSB,          tcg_gen_gvec_abs, MO_8)
    687 UNARY_INT_GVEC(PABSW,          tcg_gen_gvec_abs, MO_16)
    688 UNARY_INT_GVEC(PABSD,          tcg_gen_gvec_abs, MO_32)
    689 UNARY_INT_GVEC(VBROADCASTx128, tcg_gen_gvec_dup_mem, MO_128)
    690 UNARY_INT_GVEC(VPBROADCASTB,   tcg_gen_gvec_dup_mem, MO_8)
    691 UNARY_INT_GVEC(VPBROADCASTW,   tcg_gen_gvec_dup_mem, MO_16)
    692 UNARY_INT_GVEC(VPBROADCASTD,   tcg_gen_gvec_dup_mem, MO_32)
    693 UNARY_INT_GVEC(VPBROADCASTQ,   tcg_gen_gvec_dup_mem, MO_64)
    694 
    695 
    696 #define BINARY_INT_GVEC(uname, func, ...)                                          \
    697 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    698 {                                                                                  \
    699     int vec_len = vector_len(s, decode);                                          \
    700                                                                                    \
    701     func(__VA_ARGS__,                                                              \
    702          decode->op[0].offset, decode->op[1].offset,                               \
    703          decode->op[2].offset, vec_len, vec_len);                                  \
    704 }
    705 
    706 BINARY_INT_GVEC(PADDB,   tcg_gen_gvec_add, MO_8)
    707 BINARY_INT_GVEC(PADDW,   tcg_gen_gvec_add, MO_16)
    708 BINARY_INT_GVEC(PADDD,   tcg_gen_gvec_add, MO_32)
    709 BINARY_INT_GVEC(PADDQ,   tcg_gen_gvec_add, MO_64)
    710 BINARY_INT_GVEC(PADDSB,  tcg_gen_gvec_ssadd, MO_8)
    711 BINARY_INT_GVEC(PADDSW,  tcg_gen_gvec_ssadd, MO_16)
    712 BINARY_INT_GVEC(PADDUSB, tcg_gen_gvec_usadd, MO_8)
    713 BINARY_INT_GVEC(PADDUSW, tcg_gen_gvec_usadd, MO_16)
    714 BINARY_INT_GVEC(PAND,    tcg_gen_gvec_and, MO_64)
    715 BINARY_INT_GVEC(PCMPEQB, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_8)
    716 BINARY_INT_GVEC(PCMPEQD, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_32)
    717 BINARY_INT_GVEC(PCMPEQW, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_16)
    718 BINARY_INT_GVEC(PCMPEQQ, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_64)
    719 BINARY_INT_GVEC(PCMPGTB, tcg_gen_gvec_cmp, TCG_COND_GT, MO_8)
    720 BINARY_INT_GVEC(PCMPGTW, tcg_gen_gvec_cmp, TCG_COND_GT, MO_16)
    721 BINARY_INT_GVEC(PCMPGTD, tcg_gen_gvec_cmp, TCG_COND_GT, MO_32)
    722 BINARY_INT_GVEC(PCMPGTQ, tcg_gen_gvec_cmp, TCG_COND_GT, MO_64)
    723 BINARY_INT_GVEC(PMAXSB,  tcg_gen_gvec_smax, MO_8)
    724 BINARY_INT_GVEC(PMAXSW,  tcg_gen_gvec_smax, MO_16)
    725 BINARY_INT_GVEC(PMAXSD,  tcg_gen_gvec_smax, MO_32)
    726 BINARY_INT_GVEC(PMAXUB,  tcg_gen_gvec_umax, MO_8)
    727 BINARY_INT_GVEC(PMAXUW,  tcg_gen_gvec_umax, MO_16)
    728 BINARY_INT_GVEC(PMAXUD,  tcg_gen_gvec_umax, MO_32)
    729 BINARY_INT_GVEC(PMINSB,  tcg_gen_gvec_smin, MO_8)
    730 BINARY_INT_GVEC(PMINSW,  tcg_gen_gvec_smin, MO_16)
    731 BINARY_INT_GVEC(PMINSD,  tcg_gen_gvec_smin, MO_32)
    732 BINARY_INT_GVEC(PMINUB,  tcg_gen_gvec_umin, MO_8)
    733 BINARY_INT_GVEC(PMINUW,  tcg_gen_gvec_umin, MO_16)
    734 BINARY_INT_GVEC(PMINUD,  tcg_gen_gvec_umin, MO_32)
    735 BINARY_INT_GVEC(PMULLW,  tcg_gen_gvec_mul, MO_16)
    736 BINARY_INT_GVEC(PMULLD,  tcg_gen_gvec_mul, MO_32)
    737 BINARY_INT_GVEC(POR,     tcg_gen_gvec_or, MO_64)
    738 BINARY_INT_GVEC(PSUBB,   tcg_gen_gvec_sub, MO_8)
    739 BINARY_INT_GVEC(PSUBW,   tcg_gen_gvec_sub, MO_16)
    740 BINARY_INT_GVEC(PSUBD,   tcg_gen_gvec_sub, MO_32)
    741 BINARY_INT_GVEC(PSUBQ,   tcg_gen_gvec_sub, MO_64)
    742 BINARY_INT_GVEC(PSUBSB,  tcg_gen_gvec_sssub, MO_8)
    743 BINARY_INT_GVEC(PSUBSW,  tcg_gen_gvec_sssub, MO_16)
    744 BINARY_INT_GVEC(PSUBUSB, tcg_gen_gvec_ussub, MO_8)
    745 BINARY_INT_GVEC(PSUBUSW, tcg_gen_gvec_ussub, MO_16)
    746 BINARY_INT_GVEC(PXOR,    tcg_gen_gvec_xor, MO_64)
    747 
    748 
    749 /*
    750  * 00 = p*  Pq, Qq (if mmx not NULL; no VEX)
    751  * 66 = vp* Vx, Hx, Wx
    752  *
    753  * These are really the same encoding, because 1) V is the same as P when VEX.V
    754  * is not present 2) P and Q are the same as H and W apart from MM/XMM
    755  */
    756 static inline void gen_binary_int_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    757                                       SSEFunc_0_eppp mmx, SSEFunc_0_eppp xmm, SSEFunc_0_eppp ymm)
    758 {
    759     assert(!!mmx == !!(decode->e.special == X86_SPECIAL_MMX));
    760 
    761     if (mmx && (s->prefix & PREFIX_VEX) && !(s->prefix & PREFIX_DATA)) {
    762         /* VEX encoding is not applicable to MMX instructions.  */
    763         gen_illegal_opcode(s);
    764         return;
    765     }
    766     if (!(s->prefix & PREFIX_DATA)) {
    767         mmx(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
    768     } else if (!s->vex_l) {
    769         xmm(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
    770     } else {
    771         ymm(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
    772     }
    773 }
    774 
    775 
    776 #define BINARY_INT_MMX(uname, lname)                                               \
    777 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    778 {                                                                                  \
    779     gen_binary_int_sse(s, env, decode,                                             \
    780                           gen_helper_##lname##_mmx,                                \
    781                           gen_helper_##lname##_xmm,                                \
    782                           gen_helper_##lname##_ymm);                               \
    783 }
    784 BINARY_INT_MMX(PUNPCKLBW,  punpcklbw)
    785 BINARY_INT_MMX(PUNPCKLWD,  punpcklwd)
    786 BINARY_INT_MMX(PUNPCKLDQ,  punpckldq)
    787 BINARY_INT_MMX(PACKSSWB,   packsswb)
    788 BINARY_INT_MMX(PACKUSWB,   packuswb)
    789 BINARY_INT_MMX(PUNPCKHBW,  punpckhbw)
    790 BINARY_INT_MMX(PUNPCKHWD,  punpckhwd)
    791 BINARY_INT_MMX(PUNPCKHDQ,  punpckhdq)
    792 BINARY_INT_MMX(PACKSSDW,   packssdw)
    793 
    794 BINARY_INT_MMX(PAVGB,   pavgb)
    795 BINARY_INT_MMX(PAVGW,   pavgw)
    796 BINARY_INT_MMX(PMADDWD, pmaddwd)
    797 BINARY_INT_MMX(PMULHUW, pmulhuw)
    798 BINARY_INT_MMX(PMULHW,  pmulhw)
    799 BINARY_INT_MMX(PMULUDQ, pmuludq)
    800 BINARY_INT_MMX(PSADBW,  psadbw)
    801 
    802 BINARY_INT_MMX(PSLLW_r, psllw)
    803 BINARY_INT_MMX(PSLLD_r, pslld)
    804 BINARY_INT_MMX(PSLLQ_r, psllq)
    805 BINARY_INT_MMX(PSRLW_r, psrlw)
    806 BINARY_INT_MMX(PSRLD_r, psrld)
    807 BINARY_INT_MMX(PSRLQ_r, psrlq)
    808 BINARY_INT_MMX(PSRAW_r, psraw)
    809 BINARY_INT_MMX(PSRAD_r, psrad)
    810 
    811 BINARY_INT_MMX(PHADDW,    phaddw)
    812 BINARY_INT_MMX(PHADDSW,   phaddsw)
    813 BINARY_INT_MMX(PHADDD,    phaddd)
    814 BINARY_INT_MMX(PHSUBW,    phsubw)
    815 BINARY_INT_MMX(PHSUBSW,   phsubsw)
    816 BINARY_INT_MMX(PHSUBD,    phsubd)
    817 BINARY_INT_MMX(PMADDUBSW, pmaddubsw)
    818 BINARY_INT_MMX(PSHUFB,    pshufb)
    819 BINARY_INT_MMX(PSIGNB,    psignb)
    820 BINARY_INT_MMX(PSIGNW,    psignw)
    821 BINARY_INT_MMX(PSIGND,    psignd)
    822 BINARY_INT_MMX(PMULHRSW,  pmulhrsw)
    823 
    824 /* Instructions with no MMX equivalent.  */
    825 #define BINARY_INT_SSE(uname, lname)                                               \
    826 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    827 {                                                                                  \
    828     gen_binary_int_sse(s, env, decode,                                             \
    829                           NULL,                                                    \
    830                           gen_helper_##lname##_xmm,                                \
    831                           gen_helper_##lname##_ymm);                               \
    832 }
    833 
    834 /* Instructions with no MMX equivalent.  */
    835 BINARY_INT_SSE(PUNPCKLQDQ, punpcklqdq)
    836 BINARY_INT_SSE(PUNPCKHQDQ, punpckhqdq)
    837 BINARY_INT_SSE(VPACKUSDW,  packusdw)
    838 BINARY_INT_SSE(VPERMILPS,  vpermilps)
    839 BINARY_INT_SSE(VPERMILPD,  vpermilpd)
    840 BINARY_INT_SSE(VMASKMOVPS, vpmaskmovd)
    841 BINARY_INT_SSE(VMASKMOVPD, vpmaskmovq)
    842 
    843 BINARY_INT_SSE(PMULDQ,    pmuldq)
    844 
    845 BINARY_INT_SSE(VAESDEC, aesdec)
    846 BINARY_INT_SSE(VAESDECLAST, aesdeclast)
    847 BINARY_INT_SSE(VAESENC, aesenc)
    848 BINARY_INT_SSE(VAESENCLAST, aesenclast)
    849 
    850 #define UNARY_CMP_SSE(uname, lname)                                                \
    851 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    852 {                                                                                  \
    853     if (!s->vex_l) {                                                               \
    854         gen_helper_##lname##_xmm(cpu_env, OP_PTR1, OP_PTR2);                       \
    855     } else {                                                                       \
    856         gen_helper_##lname##_ymm(cpu_env, OP_PTR1, OP_PTR2);                       \
    857     }                                                                              \
    858     set_cc_op(s, CC_OP_EFLAGS);                                                    \
    859 }
    860 UNARY_CMP_SSE(VPTEST,     ptest)
    861 UNARY_CMP_SSE(VTESTPS,    vtestps)
    862 UNARY_CMP_SSE(VTESTPD,    vtestpd)
    863 
    864 static inline void gen_unary_int_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    865                                      SSEFunc_0_epp xmm, SSEFunc_0_epp ymm)
    866 {
    867     if (!s->vex_l) {
    868         xmm(cpu_env, OP_PTR0, OP_PTR2);
    869     } else {
    870         ymm(cpu_env, OP_PTR0, OP_PTR2);
    871     }
    872 }
    873 
    874 #define UNARY_INT_SSE(uname, lname)                                                \
    875 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    876 {                                                                                  \
    877     gen_unary_int_sse(s, env, decode,                                              \
    878                       gen_helper_##lname##_xmm,                                    \
    879                       gen_helper_##lname##_ymm);                                   \
    880 }
    881 
    882 UNARY_INT_SSE(VPMOVSXBW,    pmovsxbw)
    883 UNARY_INT_SSE(VPMOVSXBD,    pmovsxbd)
    884 UNARY_INT_SSE(VPMOVSXBQ,    pmovsxbq)
    885 UNARY_INT_SSE(VPMOVSXWD,    pmovsxwd)
    886 UNARY_INT_SSE(VPMOVSXWQ,    pmovsxwq)
    887 UNARY_INT_SSE(VPMOVSXDQ,    pmovsxdq)
    888 
    889 UNARY_INT_SSE(VPMOVZXBW,    pmovzxbw)
    890 UNARY_INT_SSE(VPMOVZXBD,    pmovzxbd)
    891 UNARY_INT_SSE(VPMOVZXBQ,    pmovzxbq)
    892 UNARY_INT_SSE(VPMOVZXWD,    pmovzxwd)
    893 UNARY_INT_SSE(VPMOVZXWQ,    pmovzxwq)
    894 UNARY_INT_SSE(VPMOVZXDQ,    pmovzxdq)
    895 
    896 UNARY_INT_SSE(VMOVSLDUP,    pmovsldup)
    897 UNARY_INT_SSE(VMOVSHDUP,    pmovshdup)
    898 UNARY_INT_SSE(VMOVDDUP,     pmovdldup)
    899 
    900 UNARY_INT_SSE(VCVTDQ2PD, cvtdq2pd)
    901 UNARY_INT_SSE(VCVTPD2DQ, cvtpd2dq)
    902 UNARY_INT_SSE(VCVTTPD2DQ, cvttpd2dq)
    903 UNARY_INT_SSE(VCVTDQ2PS, cvtdq2ps)
    904 UNARY_INT_SSE(VCVTPS2DQ, cvtps2dq)
    905 UNARY_INT_SSE(VCVTTPS2DQ, cvttps2dq)
    906 UNARY_INT_SSE(VCVTPH2PS, cvtph2ps)
    907 
    908 
    909 static inline void gen_unary_imm_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    910                                      SSEFunc_0_ppi xmm, SSEFunc_0_ppi ymm)
    911 {
    912     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
    913     if (!s->vex_l) {
    914         xmm(OP_PTR0, OP_PTR1, imm);
    915     } else {
    916         ymm(OP_PTR0, OP_PTR1, imm);
    917     }
    918 }
    919 
    920 #define UNARY_IMM_SSE(uname, lname)                                                \
    921 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    922 {                                                                                  \
    923     gen_unary_imm_sse(s, env, decode,                                              \
    924                       gen_helper_##lname##_xmm,                                    \
    925                       gen_helper_##lname##_ymm);                                   \
    926 }
    927 
    928 UNARY_IMM_SSE(PSHUFD,     pshufd)
    929 UNARY_IMM_SSE(PSHUFHW,    pshufhw)
    930 UNARY_IMM_SSE(PSHUFLW,    pshuflw)
    931 #define gen_helper_vpermq_xmm NULL
    932 UNARY_IMM_SSE(VPERMQ,      vpermq)
    933 UNARY_IMM_SSE(VPERMILPS_i, vpermilps_imm)
    934 UNARY_IMM_SSE(VPERMILPD_i, vpermilpd_imm)
    935 
    936 static inline void gen_unary_imm_fp_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    937                                         SSEFunc_0_eppi xmm, SSEFunc_0_eppi ymm)
    938 {
    939     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
    940     if (!s->vex_l) {
    941         xmm(cpu_env, OP_PTR0, OP_PTR1, imm);
    942     } else {
    943         ymm(cpu_env, OP_PTR0, OP_PTR1, imm);
    944     }
    945 }
    946 
    947 #define UNARY_IMM_FP_SSE(uname, lname)                                             \
    948 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    949 {                                                                                  \
    950     gen_unary_imm_fp_sse(s, env, decode,                                           \
    951                       gen_helper_##lname##_xmm,                                    \
    952                       gen_helper_##lname##_ymm);                                   \
    953 }
    954 
    955 UNARY_IMM_FP_SSE(VROUNDPS,    roundps)
    956 UNARY_IMM_FP_SSE(VROUNDPD,    roundpd)
    957 
    958 static inline void gen_vexw_avx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    959                                 SSEFunc_0_eppp d_xmm, SSEFunc_0_eppp q_xmm,
    960                                 SSEFunc_0_eppp d_ymm, SSEFunc_0_eppp q_ymm)
    961 {
    962     SSEFunc_0_eppp d = s->vex_l ? d_ymm : d_xmm;
    963     SSEFunc_0_eppp q = s->vex_l ? q_ymm : q_xmm;
    964     SSEFunc_0_eppp fn = s->vex_w ? q : d;
    965     fn(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
    966 }
    967 
    968 /* VEX.W affects whether to operate on 32- or 64-bit elements.  */
    969 #define VEXW_AVX(uname, lname)                                                     \
    970 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
    971 {                                                                                  \
    972     gen_vexw_avx(s, env, decode,                                                   \
    973                  gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
    974                  gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
    975 }
    976 VEXW_AVX(VPSLLV,    vpsllv)
    977 VEXW_AVX(VPSRLV,    vpsrlv)
    978 VEXW_AVX(VPSRAV,    vpsrav)
    979 VEXW_AVX(VPMASKMOV, vpmaskmov)
    980 
    981 /* Same as above, but with extra arguments to the helper.  */
    982 static inline void gen_vsib_avx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
    983                                 SSEFunc_0_epppti d_xmm, SSEFunc_0_epppti q_xmm,
    984                                 SSEFunc_0_epppti d_ymm, SSEFunc_0_epppti q_ymm)
    985 {
    986     SSEFunc_0_epppti d = s->vex_l ? d_ymm : d_xmm;
    987     SSEFunc_0_epppti q = s->vex_l ? q_ymm : q_xmm;
    988     SSEFunc_0_epppti fn = s->vex_w ? q : d;
    989     TCGv_i32 scale = tcg_constant_i32(decode->mem.scale);
    990     TCGv_ptr index = tcg_temp_new_ptr();
    991 
    992     /* Pass third input as (index, base, scale) */
    993     tcg_gen_addi_ptr(index, cpu_env, ZMM_OFFSET(decode->mem.index));
    994     fn(cpu_env, OP_PTR0, OP_PTR1, index, s->A0, scale);
    995 
    996     /*
    997      * There are two output operands, so zero OP1's high 128 bits
    998      * in the VEX.128 case.
    999      */
   1000     if (!s->vex_l) {
   1001         int ymmh_ofs = vector_elem_offset(&decode->op[1], MO_128, 1);
   1002         tcg_gen_gvec_dup_imm(MO_64, ymmh_ofs, 16, 16, 0);
   1003     }
   1004     tcg_temp_free_ptr(index);
   1005 }
   1006 #define VSIB_AVX(uname, lname)                                                     \
   1007 static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
   1008 {                                                                                  \
   1009     gen_vsib_avx(s, env, decode,                                                   \
   1010                  gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
   1011                  gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
   1012 }
   1013 VSIB_AVX(VPGATHERD, vpgatherd)
   1014 VSIB_AVX(VPGATHERQ, vpgatherq)
   1015 
   1016 static void gen_ADCOX(DisasContext *s, CPUX86State *env, MemOp ot, int cc_op)
   1017 {
   1018     TCGv carry_in = NULL;
   1019     TCGv carry_out = (cc_op == CC_OP_ADCX ? cpu_cc_dst : cpu_cc_src2);
   1020     TCGv zero;
   1021 
   1022     if (cc_op == s->cc_op || s->cc_op == CC_OP_ADCOX) {
   1023         /* Re-use the carry-out from a previous round.  */
   1024         carry_in = carry_out;
   1025         cc_op = s->cc_op;
   1026     } else if (s->cc_op == CC_OP_ADCX || s->cc_op == CC_OP_ADOX) {
   1027         /* Merge with the carry-out from the opposite instruction.  */
   1028         cc_op = CC_OP_ADCOX;
   1029     }
   1030 
   1031     /* If we don't have a carry-in, get it out of EFLAGS.  */
   1032     if (!carry_in) {
   1033         if (s->cc_op != CC_OP_ADCX && s->cc_op != CC_OP_ADOX) {
   1034             gen_compute_eflags(s);
   1035         }
   1036         carry_in = s->tmp0;
   1037         tcg_gen_extract_tl(carry_in, cpu_cc_src,
   1038             ctz32(cc_op == CC_OP_ADCX ? CC_C : CC_O), 1);
   1039     }
   1040 
   1041     switch (ot) {
   1042 #ifdef TARGET_X86_64
   1043     case MO_32:
   1044         /* If TL is 64-bit just do everything in 64-bit arithmetic.  */
   1045         tcg_gen_add_i64(s->T0, s->T0, s->T1);
   1046         tcg_gen_add_i64(s->T0, s->T0, carry_in);
   1047         tcg_gen_shri_i64(carry_out, s->T0, 32);
   1048         break;
   1049 #endif
   1050     default:
   1051         zero = tcg_constant_tl(0);
   1052         tcg_gen_add2_tl(s->T0, carry_out, s->T0, zero, carry_in, zero);
   1053         tcg_gen_add2_tl(s->T0, carry_out, s->T0, carry_out, s->T1, zero);
   1054         break;
   1055     }
   1056     set_cc_op(s, cc_op);
   1057 }
   1058 
   1059 static void gen_ADCX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1060 {
   1061     gen_ADCOX(s, env, decode->op[0].ot, CC_OP_ADCX);
   1062 }
   1063 
   1064 static void gen_ADOX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1065 {
   1066     gen_ADCOX(s, env, decode->op[0].ot, CC_OP_ADOX);
   1067 }
   1068 
   1069 static void gen_ANDN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1070 {
   1071     MemOp ot = decode->op[0].ot;
   1072 
   1073     tcg_gen_andc_tl(s->T0, s->T1, s->T0);
   1074     gen_op_update1_cc(s);
   1075     set_cc_op(s, CC_OP_LOGICB + ot);
   1076 }
   1077 
   1078 static void gen_BEXTR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1079 {
   1080     MemOp ot = decode->op[0].ot;
   1081     TCGv bound, zero;
   1082 
   1083     /*
   1084      * Extract START, and shift the operand.
   1085      * Shifts larger than operand size get zeros.
   1086      */
   1087     tcg_gen_ext8u_tl(s->A0, s->T1);
   1088     tcg_gen_shr_tl(s->T0, s->T0, s->A0);
   1089 
   1090     bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
   1091     zero = tcg_constant_tl(0);
   1092     tcg_gen_movcond_tl(TCG_COND_LEU, s->T0, s->A0, bound, s->T0, zero);
   1093 
   1094     /*
   1095      * Extract the LEN into a mask.  Lengths larger than
   1096      * operand size get all ones.
   1097      */
   1098     tcg_gen_extract_tl(s->A0, s->T1, 8, 8);
   1099     tcg_gen_movcond_tl(TCG_COND_LEU, s->A0, s->A0, bound, s->A0, bound);
   1100 
   1101     tcg_gen_movi_tl(s->T1, 1);
   1102     tcg_gen_shl_tl(s->T1, s->T1, s->A0);
   1103     tcg_gen_subi_tl(s->T1, s->T1, 1);
   1104     tcg_gen_and_tl(s->T0, s->T0, s->T1);
   1105 
   1106     gen_op_update1_cc(s);
   1107     set_cc_op(s, CC_OP_LOGICB + ot);
   1108 }
   1109 
   1110 static void gen_BLSI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1111 {
   1112     MemOp ot = decode->op[0].ot;
   1113 
   1114     tcg_gen_neg_tl(s->T1, s->T0);
   1115     tcg_gen_and_tl(s->T0, s->T0, s->T1);
   1116     tcg_gen_mov_tl(cpu_cc_dst, s->T0);
   1117     set_cc_op(s, CC_OP_BMILGB + ot);
   1118 }
   1119 
   1120 static void gen_BLSMSK(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1121 {
   1122     MemOp ot = decode->op[0].ot;
   1123 
   1124     tcg_gen_subi_tl(s->T1, s->T0, 1);
   1125     tcg_gen_xor_tl(s->T0, s->T0, s->T1);
   1126     tcg_gen_mov_tl(cpu_cc_dst, s->T0);
   1127     set_cc_op(s, CC_OP_BMILGB + ot);
   1128 }
   1129 
   1130 static void gen_BLSR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1131 {
   1132     MemOp ot = decode->op[0].ot;
   1133 
   1134     tcg_gen_subi_tl(s->T1, s->T0, 1);
   1135     tcg_gen_and_tl(s->T0, s->T0, s->T1);
   1136     tcg_gen_mov_tl(cpu_cc_dst, s->T0);
   1137     set_cc_op(s, CC_OP_BMILGB + ot);
   1138 }
   1139 
   1140 static void gen_BZHI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1141 {
   1142     MemOp ot = decode->op[0].ot;
   1143     TCGv bound;
   1144 
   1145     tcg_gen_ext8u_tl(s->T1, cpu_regs[s->vex_v]);
   1146     bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
   1147 
   1148     /*
   1149      * Note that since we're using BMILG (in order to get O
   1150      * cleared) we need to store the inverse into C.
   1151      */
   1152     tcg_gen_setcond_tl(TCG_COND_LT, cpu_cc_src, s->T1, bound);
   1153     tcg_gen_movcond_tl(TCG_COND_GT, s->T1, s->T1, bound, bound, s->T1);
   1154 
   1155     tcg_gen_movi_tl(s->A0, -1);
   1156     tcg_gen_shl_tl(s->A0, s->A0, s->T1);
   1157     tcg_gen_andc_tl(s->T0, s->T0, s->A0);
   1158 
   1159     gen_op_update1_cc(s);
   1160     set_cc_op(s, CC_OP_BMILGB + ot);
   1161 }
   1162 
   1163 static void gen_CRC32(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1164 {
   1165     MemOp ot = decode->op[2].ot;
   1166 
   1167     tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
   1168     gen_helper_crc32(s->T0, s->tmp2_i32, s->T1, tcg_constant_i32(8 << ot));
   1169 }
   1170 
   1171 static void gen_CVTPI2Px(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1172 {
   1173     gen_helper_enter_mmx(cpu_env);
   1174     if (s->prefix & PREFIX_DATA) {
   1175         gen_helper_cvtpi2pd(cpu_env, OP_PTR0, OP_PTR2);
   1176     } else {
   1177         gen_helper_cvtpi2ps(cpu_env, OP_PTR0, OP_PTR2);
   1178     }
   1179 }
   1180 
   1181 static void gen_CVTPx2PI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1182 {
   1183     gen_helper_enter_mmx(cpu_env);
   1184     if (s->prefix & PREFIX_DATA) {
   1185         gen_helper_cvtpd2pi(cpu_env, OP_PTR0, OP_PTR2);
   1186     } else {
   1187         gen_helper_cvtps2pi(cpu_env, OP_PTR0, OP_PTR2);
   1188     }
   1189 }
   1190 
   1191 static void gen_CVTTPx2PI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1192 {
   1193     gen_helper_enter_mmx(cpu_env);
   1194     if (s->prefix & PREFIX_DATA) {
   1195         gen_helper_cvttpd2pi(cpu_env, OP_PTR0, OP_PTR2);
   1196     } else {
   1197         gen_helper_cvttps2pi(cpu_env, OP_PTR0, OP_PTR2);
   1198     }
   1199 }
   1200 
   1201 static void gen_EMMS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1202 {
   1203     gen_helper_emms(cpu_env);
   1204 }
   1205 
   1206 static void gen_EXTRQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1207 {
   1208     TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
   1209     TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
   1210 
   1211     gen_helper_extrq_i(cpu_env, OP_PTR0, index, length);
   1212 }
   1213 
   1214 static void gen_EXTRQ_r(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1215 {
   1216     gen_helper_extrq_r(cpu_env, OP_PTR0, OP_PTR2);
   1217 }
   1218 
   1219 static void gen_INSERTQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1220 {
   1221     TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
   1222     TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
   1223 
   1224     gen_helper_insertq_i(cpu_env, OP_PTR0, OP_PTR1, index, length);
   1225 }
   1226 
   1227 static void gen_INSERTQ_r(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1228 {
   1229     gen_helper_insertq_r(cpu_env, OP_PTR0, OP_PTR2);
   1230 }
   1231 
   1232 static void gen_LDMXCSR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1233 {
   1234     if (s->vex_l) {
   1235         gen_illegal_opcode(s);
   1236         return;
   1237     }
   1238     tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T1);
   1239     gen_helper_ldmxcsr(cpu_env, s->tmp2_i32);
   1240 }
   1241 
   1242 static void gen_MASKMOV(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1243 {
   1244     tcg_gen_mov_tl(s->A0, cpu_regs[R_EDI]);
   1245     gen_extu(s->aflag, s->A0);
   1246     gen_add_A0_ds_seg(s);
   1247 
   1248     if (s->prefix & PREFIX_DATA) {
   1249         gen_helper_maskmov_xmm(cpu_env, OP_PTR1, OP_PTR2, s->A0);
   1250     } else {
   1251         gen_helper_maskmov_mmx(cpu_env, OP_PTR1, OP_PTR2, s->A0);
   1252     }
   1253 }
   1254 
   1255 static void gen_MOVBE(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1256 {
   1257     MemOp ot = decode->op[0].ot;
   1258 
   1259     /* M operand type does not load/store */
   1260     if (decode->e.op0 == X86_TYPE_M) {
   1261         tcg_gen_qemu_st_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
   1262     } else {
   1263         tcg_gen_qemu_ld_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
   1264     }
   1265 }
   1266 
   1267 static void gen_MOVD_from(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1268 {
   1269     MemOp ot = decode->op[2].ot;
   1270 
   1271     switch (ot) {
   1272     case MO_32:
   1273 #ifdef TARGET_X86_64
   1274         tcg_gen_ld32u_tl(s->T0, cpu_env, decode->op[2].offset);
   1275         break;
   1276     case MO_64:
   1277 #endif
   1278         tcg_gen_ld_tl(s->T0, cpu_env, decode->op[2].offset);
   1279         break;
   1280     default:
   1281         abort();
   1282     }
   1283 }
   1284 
   1285 static void gen_MOVD_to(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1286 {
   1287     MemOp ot = decode->op[2].ot;
   1288     int vec_len = vector_len(s, decode);
   1289     int lo_ofs = vector_elem_offset(&decode->op[0], ot, 0);
   1290 
   1291     tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   1292 
   1293     switch (ot) {
   1294     case MO_32:
   1295 #ifdef TARGET_X86_64
   1296         tcg_gen_st32_tl(s->T1, cpu_env, lo_ofs);
   1297         break;
   1298     case MO_64:
   1299 #endif
   1300         tcg_gen_st_tl(s->T1, cpu_env, lo_ofs);
   1301         break;
   1302     default:
   1303         g_assert_not_reached();
   1304     }
   1305 }
   1306 
   1307 static void gen_MOVDQ(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1308 {
   1309     gen_store_sse(s, decode, decode->op[2].offset);
   1310 }
   1311 
   1312 static void gen_MOVMSK(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1313 {
   1314     typeof(gen_helper_movmskps_ymm) *ps, *pd, *fn;
   1315     ps = s->vex_l ? gen_helper_movmskps_ymm : gen_helper_movmskps_xmm;
   1316     pd = s->vex_l ? gen_helper_movmskpd_ymm : gen_helper_movmskpd_xmm;
   1317     fn = s->prefix & PREFIX_DATA ? pd : ps;
   1318     fn(s->tmp2_i32, cpu_env, OP_PTR2);
   1319     tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
   1320 }
   1321 
   1322 static void gen_MOVQ(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1323 {
   1324     int vec_len = vector_len(s, decode);
   1325     int lo_ofs = vector_elem_offset(&decode->op[0], MO_64, 0);
   1326 
   1327     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[2].offset);
   1328     if (decode->op[0].has_ea) {
   1329         tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
   1330     } else {
   1331         /*
   1332          * tcg_gen_gvec_dup_i64(MO_64, op0.offset, 8, vec_len, s->tmp1_64) would
   1333          * seem to work, but it does not on big-endian platforms; the cleared parts
   1334          * are always at higher addresses, but cross-endian emulation inverts the
   1335          * byte order so that the cleared parts need to be at *lower* addresses.
   1336          * Because oprsz is 8, we see this here even for SSE; but more in general,
   1337          * it disqualifies using oprsz < maxsz to emulate VEX128.
   1338          */
   1339         tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   1340         tcg_gen_st_i64(s->tmp1_i64, cpu_env, lo_ofs);
   1341     }
   1342 }
   1343 
   1344 static void gen_MOVq_dq(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1345 {
   1346     gen_helper_enter_mmx(cpu_env);
   1347     /* Otherwise the same as any other movq.  */
   1348     return gen_MOVQ(s, env, decode);
   1349 }
   1350 
   1351 static void gen_MULX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1352 {
   1353     MemOp ot = decode->op[0].ot;
   1354 
   1355     /* low part of result in VEX.vvvv, high in MODRM */
   1356     switch (ot) {
   1357     default:
   1358         tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
   1359         tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
   1360         tcg_gen_mulu2_i32(s->tmp2_i32, s->tmp3_i32,
   1361                           s->tmp2_i32, s->tmp3_i32);
   1362         tcg_gen_extu_i32_tl(cpu_regs[s->vex_v], s->tmp2_i32);
   1363         tcg_gen_extu_i32_tl(s->T0, s->tmp3_i32);
   1364         break;
   1365 #ifdef TARGET_X86_64
   1366     case MO_64:
   1367         tcg_gen_mulu2_i64(cpu_regs[s->vex_v], s->T0, s->T0, s->T1);
   1368         break;
   1369 #endif
   1370     }
   1371 
   1372 }
   1373 
   1374 static void gen_PALIGNR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1375 {
   1376     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   1377     if (!(s->prefix & PREFIX_DATA)) {
   1378         gen_helper_palignr_mmx(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
   1379     } else if (!s->vex_l) {
   1380         gen_helper_palignr_xmm(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
   1381     } else {
   1382         gen_helper_palignr_ymm(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
   1383     }
   1384 }
   1385 
   1386 static void gen_PANDN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1387 {
   1388     int vec_len = vector_len(s, decode);
   1389 
   1390     /* Careful, operand order is reversed!  */
   1391     tcg_gen_gvec_andc(MO_64,
   1392                       decode->op[0].offset, decode->op[2].offset,
   1393                       decode->op[1].offset, vec_len, vec_len);
   1394 }
   1395 
   1396 static void gen_PCMPESTRI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1397 {
   1398     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   1399     gen_helper_pcmpestri_xmm(cpu_env, OP_PTR1, OP_PTR2, imm);
   1400     set_cc_op(s, CC_OP_EFLAGS);
   1401 }
   1402 
   1403 static void gen_PCMPESTRM(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1404 {
   1405     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   1406     gen_helper_pcmpestrm_xmm(cpu_env, OP_PTR1, OP_PTR2, imm);
   1407     set_cc_op(s, CC_OP_EFLAGS);
   1408     if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
   1409         tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
   1410                              16, 16, 0);
   1411     }
   1412 }
   1413 
   1414 static void gen_PCMPISTRI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1415 {
   1416     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   1417     gen_helper_pcmpistri_xmm(cpu_env, OP_PTR1, OP_PTR2, imm);
   1418     set_cc_op(s, CC_OP_EFLAGS);
   1419 }
   1420 
   1421 static void gen_PCMPISTRM(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1422 {
   1423     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   1424     gen_helper_pcmpistrm_xmm(cpu_env, OP_PTR1, OP_PTR2, imm);
   1425     set_cc_op(s, CC_OP_EFLAGS);
   1426     if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
   1427         tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
   1428                              16, 16, 0);
   1429     }
   1430 }
   1431 
   1432 static void gen_PDEP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1433 {
   1434     MemOp ot = decode->op[1].ot;
   1435     if (ot < MO_64) {
   1436         tcg_gen_ext32u_tl(s->T0, s->T0);
   1437     }
   1438     gen_helper_pdep(s->T0, s->T0, s->T1);
   1439 }
   1440 
   1441 static void gen_PEXT(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1442 {
   1443     MemOp ot = decode->op[1].ot;
   1444     if (ot < MO_64) {
   1445         tcg_gen_ext32u_tl(s->T0, s->T0);
   1446     }
   1447     gen_helper_pext(s->T0, s->T0, s->T1);
   1448 }
   1449 
   1450 static inline void gen_pextr(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode, MemOp ot)
   1451 {
   1452     int vec_len = vector_len(s, decode);
   1453     int mask = (vec_len >> ot) - 1;
   1454     int val = decode->immediate & mask;
   1455 
   1456     switch (ot) {
   1457     case MO_8:
   1458         tcg_gen_ld8u_tl(s->T0, cpu_env, vector_elem_offset(&decode->op[1], ot, val));
   1459         break;
   1460     case MO_16:
   1461         tcg_gen_ld16u_tl(s->T0, cpu_env, vector_elem_offset(&decode->op[1], ot, val));
   1462         break;
   1463     case MO_32:
   1464 #ifdef TARGET_X86_64
   1465         tcg_gen_ld32u_tl(s->T0, cpu_env, vector_elem_offset(&decode->op[1], ot, val));
   1466         break;
   1467     case MO_64:
   1468 #endif
   1469         tcg_gen_ld_tl(s->T0, cpu_env, vector_elem_offset(&decode->op[1], ot, val));
   1470         break;
   1471     default:
   1472         abort();
   1473     }
   1474 }
   1475 
   1476 static void gen_PEXTRB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1477 {
   1478     gen_pextr(s, env, decode, MO_8);
   1479 }
   1480 
   1481 static void gen_PEXTRW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1482 {
   1483     gen_pextr(s, env, decode, MO_16);
   1484 }
   1485 
   1486 static void gen_PEXTR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1487 {
   1488     MemOp ot = decode->op[0].ot;
   1489     gen_pextr(s, env, decode, ot);
   1490 }
   1491 
   1492 static inline void gen_pinsr(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode, MemOp ot)
   1493 {
   1494     int vec_len = vector_len(s, decode);
   1495     int mask = (vec_len >> ot) - 1;
   1496     int val = decode->immediate & mask;
   1497 
   1498     if (decode->op[1].offset != decode->op[0].offset) {
   1499         assert(vec_len == 16);
   1500         gen_store_sse(s, decode, decode->op[1].offset);
   1501     }
   1502 
   1503     switch (ot) {
   1504     case MO_8:
   1505         tcg_gen_st8_tl(s->T1, cpu_env, vector_elem_offset(&decode->op[0], ot, val));
   1506         break;
   1507     case MO_16:
   1508         tcg_gen_st16_tl(s->T1, cpu_env, vector_elem_offset(&decode->op[0], ot, val));
   1509         break;
   1510     case MO_32:
   1511 #ifdef TARGET_X86_64
   1512         tcg_gen_st32_tl(s->T1, cpu_env, vector_elem_offset(&decode->op[0], ot, val));
   1513         break;
   1514     case MO_64:
   1515 #endif
   1516         tcg_gen_st_tl(s->T1, cpu_env, vector_elem_offset(&decode->op[0], ot, val));
   1517         break;
   1518     default:
   1519         abort();
   1520     }
   1521 }
   1522 
   1523 static void gen_PINSRB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1524 {
   1525     gen_pinsr(s, env, decode, MO_8);
   1526 }
   1527 
   1528 static void gen_PINSRW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1529 {
   1530     gen_pinsr(s, env, decode, MO_16);
   1531 }
   1532 
   1533 static void gen_PINSR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1534 {
   1535     gen_pinsr(s, env, decode, decode->op[2].ot);
   1536 }
   1537 
   1538 static void gen_pmovmskb_i64(TCGv_i64 d, TCGv_i64 s)
   1539 {
   1540     TCGv_i64 t = tcg_temp_new_i64();
   1541 
   1542     tcg_gen_andi_i64(d, s, 0x8080808080808080ull);
   1543 
   1544     /*
   1545      * After each shift+or pair:
   1546      * 0:  a.......b.......c.......d.......e.......f.......g.......h.......
   1547      * 7:  ab......bc......cd......de......ef......fg......gh......h.......
   1548      * 14: abcd....bcde....cdef....defg....efgh....fgh.....gh......h.......
   1549      * 28: abcdefghbcdefgh.cdefgh..defgh...efgh....fgh.....gh......h.......
   1550      * The result is left in the high bits of the word.
   1551      */
   1552     tcg_gen_shli_i64(t, d, 7);
   1553     tcg_gen_or_i64(d, d, t);
   1554     tcg_gen_shli_i64(t, d, 14);
   1555     tcg_gen_or_i64(d, d, t);
   1556     tcg_gen_shli_i64(t, d, 28);
   1557     tcg_gen_or_i64(d, d, t);
   1558 }
   1559 
   1560 static void gen_pmovmskb_vec(unsigned vece, TCGv_vec d, TCGv_vec s)
   1561 {
   1562     TCGv_vec t = tcg_temp_new_vec_matching(d);
   1563     TCGv_vec m = tcg_constant_vec_matching(d, MO_8, 0x80);
   1564 
   1565     /* See above */
   1566     tcg_gen_and_vec(vece, d, s, m);
   1567     tcg_gen_shli_vec(vece, t, d, 7);
   1568     tcg_gen_or_vec(vece, d, d, t);
   1569     tcg_gen_shli_vec(vece, t, d, 14);
   1570     tcg_gen_or_vec(vece, d, d, t);
   1571     tcg_gen_shli_vec(vece, t, d, 28);
   1572     tcg_gen_or_vec(vece, d, d, t);
   1573 }
   1574 
   1575 #ifdef TARGET_X86_64
   1576 #define TCG_TARGET_HAS_extract2_tl TCG_TARGET_HAS_extract2_i64
   1577 #else
   1578 #define TCG_TARGET_HAS_extract2_tl TCG_TARGET_HAS_extract2_i32
   1579 #endif
   1580 
   1581 static void gen_PMOVMSKB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1582 {
   1583     static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
   1584     static const GVecGen2 g = {
   1585         .fni8 = gen_pmovmskb_i64,
   1586         .fniv = gen_pmovmskb_vec,
   1587         .opt_opc = vecop_list,
   1588         .vece = MO_64,
   1589         .prefer_i64 = TCG_TARGET_REG_BITS == 64
   1590     };
   1591     MemOp ot = decode->op[2].ot;
   1592     int vec_len = vector_len(s, decode);
   1593     TCGv t = tcg_temp_new();
   1594 
   1595     tcg_gen_gvec_2(offsetof(CPUX86State, xmm_t0) + xmm_offset(ot), decode->op[2].offset,
   1596                    vec_len, vec_len, &g);
   1597     tcg_gen_ld8u_tl(s->T0, cpu_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
   1598     while (vec_len > 8) {
   1599         vec_len -= 8;
   1600         if (TCG_TARGET_HAS_extract2_tl) {
   1601             /*
   1602              * Load the next byte of the result into the high byte of T.
   1603              * TCG does a similar expansion of deposit to shl+extract2; by
   1604              * loading the whole word, the shift left is avoided.
   1605              */
   1606 #ifdef TARGET_X86_64
   1607             tcg_gen_ld_tl(t, cpu_env, offsetof(CPUX86State, xmm_t0.ZMM_Q((vec_len - 1) / 8)));
   1608 #else
   1609             tcg_gen_ld_tl(t, cpu_env, offsetof(CPUX86State, xmm_t0.ZMM_L((vec_len - 1) / 4)));
   1610 #endif
   1611 
   1612             tcg_gen_extract2_tl(s->T0, t, s->T0, TARGET_LONG_BITS - 8);
   1613         } else {
   1614             /*
   1615              * The _previous_ value is deposited into bits 8 and higher of t.  Because
   1616              * those bits are known to be zero after ld8u, this becomes a shift+or
   1617              * if deposit is not available.
   1618              */
   1619             tcg_gen_ld8u_tl(t, cpu_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
   1620             tcg_gen_deposit_tl(s->T0, t, s->T0, 8, TARGET_LONG_BITS - 8);
   1621         }
   1622     }
   1623     tcg_temp_free(t);
   1624 }
   1625 
   1626 static void gen_PSHUFW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1627 {
   1628     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   1629     gen_helper_pshufw_mmx(OP_PTR0, OP_PTR1, imm);
   1630 }
   1631 
   1632 static void gen_PSRLW_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1633 {
   1634     int vec_len = vector_len(s, decode);
   1635 
   1636     if (decode->immediate >= 16) {
   1637         tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   1638     } else {
   1639         tcg_gen_gvec_shri(MO_16,
   1640                           decode->op[0].offset, decode->op[1].offset,
   1641                           decode->immediate, vec_len, vec_len);
   1642     }
   1643 }
   1644 
   1645 static void gen_PSLLW_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1646 {
   1647     int vec_len = vector_len(s, decode);
   1648 
   1649     if (decode->immediate >= 16) {
   1650         tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   1651     } else {
   1652         tcg_gen_gvec_shli(MO_16,
   1653                           decode->op[0].offset, decode->op[1].offset,
   1654                           decode->immediate, vec_len, vec_len);
   1655     }
   1656 }
   1657 
   1658 static void gen_PSRAW_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1659 {
   1660     int vec_len = vector_len(s, decode);
   1661 
   1662     if (decode->immediate >= 16) {
   1663         decode->immediate = 15;
   1664     }
   1665     tcg_gen_gvec_sari(MO_16,
   1666                       decode->op[0].offset, decode->op[1].offset,
   1667                       decode->immediate, vec_len, vec_len);
   1668 }
   1669 
   1670 static void gen_PSRLD_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1671 {
   1672     int vec_len = vector_len(s, decode);
   1673 
   1674     if (decode->immediate >= 32) {
   1675         tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   1676     } else {
   1677         tcg_gen_gvec_shri(MO_32,
   1678                           decode->op[0].offset, decode->op[1].offset,
   1679                           decode->immediate, vec_len, vec_len);
   1680     }
   1681 }
   1682 
   1683 static void gen_PSLLD_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1684 {
   1685     int vec_len = vector_len(s, decode);
   1686 
   1687     if (decode->immediate >= 32) {
   1688         tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   1689     } else {
   1690         tcg_gen_gvec_shli(MO_32,
   1691                           decode->op[0].offset, decode->op[1].offset,
   1692                           decode->immediate, vec_len, vec_len);
   1693     }
   1694 }
   1695 
   1696 static void gen_PSRAD_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1697 {
   1698     int vec_len = vector_len(s, decode);
   1699 
   1700     if (decode->immediate >= 32) {
   1701         decode->immediate = 31;
   1702     }
   1703     tcg_gen_gvec_sari(MO_32,
   1704                       decode->op[0].offset, decode->op[1].offset,
   1705                       decode->immediate, vec_len, vec_len);
   1706 }
   1707 
   1708 static void gen_PSRLQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1709 {
   1710     int vec_len = vector_len(s, decode);
   1711 
   1712     if (decode->immediate >= 64) {
   1713         tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   1714     } else {
   1715         tcg_gen_gvec_shri(MO_64,
   1716                           decode->op[0].offset, decode->op[1].offset,
   1717                           decode->immediate, vec_len, vec_len);
   1718     }
   1719 }
   1720 
   1721 static void gen_PSLLQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1722 {
   1723     int vec_len = vector_len(s, decode);
   1724 
   1725     if (decode->immediate >= 64) {
   1726         tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   1727     } else {
   1728         tcg_gen_gvec_shli(MO_64,
   1729                           decode->op[0].offset, decode->op[1].offset,
   1730                           decode->immediate, vec_len, vec_len);
   1731     }
   1732 }
   1733 
   1734 static TCGv_ptr make_imm8u_xmm_vec(uint8_t imm, int vec_len)
   1735 {
   1736     MemOp ot = vec_len == 16 ? MO_128 : MO_256;
   1737     TCGv_i32 imm_v = tcg_constant8u_i32(imm);
   1738     TCGv_ptr ptr = tcg_temp_new_ptr();
   1739 
   1740     tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_t0) + xmm_offset(ot),
   1741                          vec_len, vec_len, 0);
   1742 
   1743     tcg_gen_addi_ptr(ptr, cpu_env, offsetof(CPUX86State, xmm_t0));
   1744     tcg_gen_st_i32(imm_v, cpu_env, offsetof(CPUX86State, xmm_t0.ZMM_L(0)));
   1745     return ptr;
   1746 }
   1747 
   1748 static void gen_PSRLDQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1749 {
   1750     int vec_len = vector_len(s, decode);
   1751     TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
   1752 
   1753     if (s->vex_l) {
   1754         gen_helper_psrldq_ymm(cpu_env, OP_PTR0, OP_PTR1, imm_vec);
   1755     } else {
   1756         gen_helper_psrldq_xmm(cpu_env, OP_PTR0, OP_PTR1, imm_vec);
   1757     }
   1758     tcg_temp_free_ptr(imm_vec);
   1759 }
   1760 
   1761 static void gen_PSLLDQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1762 {
   1763     int vec_len = vector_len(s, decode);
   1764     TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
   1765 
   1766     if (s->vex_l) {
   1767         gen_helper_pslldq_ymm(cpu_env, OP_PTR0, OP_PTR1, imm_vec);
   1768     } else {
   1769         gen_helper_pslldq_xmm(cpu_env, OP_PTR0, OP_PTR1, imm_vec);
   1770     }
   1771     tcg_temp_free_ptr(imm_vec);
   1772 }
   1773 
   1774 static void gen_RORX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1775 {
   1776     MemOp ot = decode->op[0].ot;
   1777     int b = decode->immediate;
   1778 
   1779     if (ot == MO_64) {
   1780         tcg_gen_rotri_tl(s->T0, s->T0, b & 63);
   1781     } else {
   1782         tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
   1783         tcg_gen_rotri_i32(s->tmp2_i32, s->tmp2_i32, b & 31);
   1784         tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
   1785     }
   1786 }
   1787 
   1788 static void gen_SARX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1789 {
   1790     MemOp ot = decode->op[0].ot;
   1791     int mask;
   1792 
   1793     mask = ot == MO_64 ? 63 : 31;
   1794     tcg_gen_andi_tl(s->T1, s->T1, mask);
   1795     if (ot != MO_64) {
   1796         tcg_gen_ext32s_tl(s->T0, s->T0);
   1797     }
   1798     tcg_gen_sar_tl(s->T0, s->T0, s->T1);
   1799 }
   1800 
   1801 static void gen_SHLX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1802 {
   1803     MemOp ot = decode->op[0].ot;
   1804     int mask;
   1805 
   1806     mask = ot == MO_64 ? 63 : 31;
   1807     tcg_gen_andi_tl(s->T1, s->T1, mask);
   1808     tcg_gen_shl_tl(s->T0, s->T0, s->T1);
   1809 }
   1810 
   1811 static void gen_SHRX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1812 {
   1813     MemOp ot = decode->op[0].ot;
   1814     int mask;
   1815 
   1816     mask = ot == MO_64 ? 63 : 31;
   1817     tcg_gen_andi_tl(s->T1, s->T1, mask);
   1818     if (ot != MO_64) {
   1819         tcg_gen_ext32u_tl(s->T0, s->T0);
   1820     }
   1821     tcg_gen_shr_tl(s->T0, s->T0, s->T1);
   1822 }
   1823 
   1824 static void gen_VAESKEYGEN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1825 {
   1826     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   1827     assert(!s->vex_l);
   1828     gen_helper_aeskeygenassist_xmm(cpu_env, OP_PTR0, OP_PTR1, imm);
   1829 }
   1830 
   1831 static void gen_STMXCSR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1832 {
   1833     if (s->vex_l) {
   1834         gen_illegal_opcode(s);
   1835         return;
   1836     }
   1837     gen_helper_update_mxcsr(cpu_env);
   1838     tcg_gen_ld32u_tl(s->T0, cpu_env, offsetof(CPUX86State, mxcsr));
   1839 }
   1840 
   1841 static void gen_VAESIMC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1842 {
   1843     assert(!s->vex_l);
   1844     gen_helper_aesimc_xmm(cpu_env, OP_PTR0, OP_PTR2);
   1845 }
   1846 
   1847 /*
   1848  * 00 = v*ps Vps, Hps, Wpd
   1849  * 66 = v*pd Vpd, Hpd, Wps
   1850  * f3 = v*ss Vss, Hss, Wps
   1851  * f2 = v*sd Vsd, Hsd, Wps
   1852  */
   1853 #define SSE_CMP(x) { \
   1854     gen_helper_ ## x ## ps ## _xmm, gen_helper_ ## x ## pd ## _xmm, \
   1855     gen_helper_ ## x ## ss, gen_helper_ ## x ## sd, \
   1856     gen_helper_ ## x ## ps ## _ymm, gen_helper_ ## x ## pd ## _ymm}
   1857 static const SSEFunc_0_eppp gen_helper_cmp_funcs[32][6] = {
   1858     SSE_CMP(cmpeq),
   1859     SSE_CMP(cmplt),
   1860     SSE_CMP(cmple),
   1861     SSE_CMP(cmpunord),
   1862     SSE_CMP(cmpneq),
   1863     SSE_CMP(cmpnlt),
   1864     SSE_CMP(cmpnle),
   1865     SSE_CMP(cmpord),
   1866 
   1867     SSE_CMP(cmpequ),
   1868     SSE_CMP(cmpnge),
   1869     SSE_CMP(cmpngt),
   1870     SSE_CMP(cmpfalse),
   1871     SSE_CMP(cmpnequ),
   1872     SSE_CMP(cmpge),
   1873     SSE_CMP(cmpgt),
   1874     SSE_CMP(cmptrue),
   1875 
   1876     SSE_CMP(cmpeqs),
   1877     SSE_CMP(cmpltq),
   1878     SSE_CMP(cmpleq),
   1879     SSE_CMP(cmpunords),
   1880     SSE_CMP(cmpneqq),
   1881     SSE_CMP(cmpnltq),
   1882     SSE_CMP(cmpnleq),
   1883     SSE_CMP(cmpords),
   1884 
   1885     SSE_CMP(cmpequs),
   1886     SSE_CMP(cmpngeq),
   1887     SSE_CMP(cmpngtq),
   1888     SSE_CMP(cmpfalses),
   1889     SSE_CMP(cmpnequs),
   1890     SSE_CMP(cmpgeq),
   1891     SSE_CMP(cmpgtq),
   1892     SSE_CMP(cmptrues),
   1893 };
   1894 #undef SSE_CMP
   1895 
   1896 static void gen_VCMP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1897 {
   1898     int index = decode->immediate & (s->prefix & PREFIX_VEX ? 31 : 7);
   1899     int b =
   1900         s->prefix & PREFIX_REPZ  ? 2 /* ss */ :
   1901         s->prefix & PREFIX_REPNZ ? 3 /* sd */ :
   1902         !!(s->prefix & PREFIX_DATA) /* pd */ + (s->vex_l << 2);
   1903 
   1904     gen_helper_cmp_funcs[index][b](cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
   1905 }
   1906 
   1907 static void gen_VCOMI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1908 {
   1909     SSEFunc_0_epp fn;
   1910     fn = s->prefix & PREFIX_DATA ? gen_helper_comisd : gen_helper_comiss;
   1911     fn(cpu_env, OP_PTR1, OP_PTR2);
   1912     set_cc_op(s, CC_OP_EFLAGS);
   1913 }
   1914 
   1915 static void gen_VCVTfp2fp(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1916 {
   1917     gen_unary_fp_sse(s, env, decode,
   1918                      gen_helper_cvtpd2ps_xmm, gen_helper_cvtps2pd_xmm,
   1919                      gen_helper_cvtpd2ps_ymm, gen_helper_cvtps2pd_ymm,
   1920                      gen_helper_cvtsd2ss, gen_helper_cvtss2sd);
   1921 }
   1922 
   1923 static void gen_VCVTPS2PH(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1924 {
   1925     gen_unary_imm_fp_sse(s, env, decode,
   1926                       gen_helper_cvtps2ph_xmm,
   1927                       gen_helper_cvtps2ph_ymm);
   1928     /*
   1929      * VCVTPS2PH is the only instruction that performs an operation on a
   1930      * register source and then *stores* into memory.
   1931      */
   1932     if (decode->op[0].has_ea) {
   1933         gen_store_sse(s, decode, decode->op[0].offset);
   1934     }
   1935 }
   1936 
   1937 static void gen_VCVTSI2Sx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   1938 {
   1939     int vec_len = vector_len(s, decode);
   1940     TCGv_i32 in;
   1941 
   1942     tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
   1943 
   1944 #ifdef TARGET_X86_64
   1945     MemOp ot = decode->op[2].ot;
   1946     if (ot == MO_64) {
   1947         if (s->prefix & PREFIX_REPNZ) {
   1948             gen_helper_cvtsq2sd(cpu_env, OP_PTR0, s->T1);
   1949         } else {
   1950             gen_helper_cvtsq2ss(cpu_env, OP_PTR0, s->T1);
   1951         }
   1952         return;
   1953     }
   1954     in = s->tmp2_i32;
   1955     tcg_gen_trunc_tl_i32(in, s->T1);
   1956 #else
   1957     in = s->T1;
   1958 #endif
   1959 
   1960     if (s->prefix & PREFIX_REPNZ) {
   1961         gen_helper_cvtsi2sd(cpu_env, OP_PTR0, in);
   1962     } else {
   1963         gen_helper_cvtsi2ss(cpu_env, OP_PTR0, in);
   1964     }
   1965 }
   1966 
   1967 static inline void gen_VCVTtSx2SI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
   1968                                   SSEFunc_i_ep ss2si, SSEFunc_l_ep ss2sq,
   1969                                   SSEFunc_i_ep sd2si, SSEFunc_l_ep sd2sq)
   1970 {
   1971     TCGv_i32 out;
   1972 
   1973 #ifdef TARGET_X86_64
   1974     MemOp ot = decode->op[0].ot;
   1975     if (ot == MO_64) {
   1976         if (s->prefix & PREFIX_REPNZ) {
   1977             sd2sq(s->T0, cpu_env, OP_PTR2);
   1978         } else {
   1979             ss2sq(s->T0, cpu_env, OP_PTR2);
   1980         }
   1981         return;
   1982     }
   1983 
   1984     out = s->tmp2_i32;
   1985 #else
   1986     out = s->T0;
   1987 #endif
   1988     if (s->prefix & PREFIX_REPNZ) {
   1989         sd2si(out, cpu_env, OP_PTR2);
   1990     } else {
   1991         ss2si(out, cpu_env, OP_PTR2);
   1992     }
   1993 #ifdef TARGET_X86_64
   1994     tcg_gen_extu_i32_tl(s->T0, out);
   1995 #endif
   1996 }
   1997 
   1998 #ifndef TARGET_X86_64
   1999 #define gen_helper_cvtss2sq NULL
   2000 #define gen_helper_cvtsd2sq NULL
   2001 #define gen_helper_cvttss2sq NULL
   2002 #define gen_helper_cvttsd2sq NULL
   2003 #endif
   2004 
   2005 static void gen_VCVTSx2SI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2006 {
   2007     gen_VCVTtSx2SI(s, env, decode,
   2008                    gen_helper_cvtss2si, gen_helper_cvtss2sq,
   2009                    gen_helper_cvtsd2si, gen_helper_cvtsd2sq);
   2010 }
   2011 
   2012 static void gen_VCVTTSx2SI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2013 {
   2014     gen_VCVTtSx2SI(s, env, decode,
   2015                    gen_helper_cvttss2si, gen_helper_cvttss2sq,
   2016                    gen_helper_cvttsd2si, gen_helper_cvttsd2sq);
   2017 }
   2018 
   2019 static void gen_VEXTRACTx128(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2020 {
   2021     int mask = decode->immediate & 1;
   2022     int src_ofs = vector_elem_offset(&decode->op[1], MO_128, mask);
   2023     if (decode->op[0].has_ea) {
   2024         /* VEX-only instruction, no alignment requirements.  */
   2025         gen_sto_env_A0(s, src_ofs, false);
   2026     } else {
   2027         tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, 16, 16);
   2028     }
   2029 }
   2030 
   2031 static void gen_VEXTRACTPS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2032 {
   2033     gen_pextr(s, env, decode, MO_32);
   2034 }
   2035 
   2036 static void gen_vinsertps(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2037 {
   2038     int val = decode->immediate;
   2039     int dest_word = (val >> 4) & 3;
   2040     int new_mask = (val & 15) | (1 << dest_word);
   2041     int vec_len = 16;
   2042 
   2043     assert(!s->vex_l);
   2044 
   2045     if (new_mask == 15) {
   2046         /* All zeroes except possibly for the inserted element */
   2047         tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   2048     } else if (decode->op[1].offset != decode->op[0].offset) {
   2049         gen_store_sse(s, decode, decode->op[1].offset);
   2050     }
   2051 
   2052     if (new_mask != (val & 15)) {
   2053         tcg_gen_st_i32(s->tmp2_i32, cpu_env,
   2054                        vector_elem_offset(&decode->op[0], MO_32, dest_word));
   2055     }
   2056 
   2057     if (new_mask != 15) {
   2058         TCGv_i32 zero = tcg_constant_i32(0); /* float32_zero */
   2059         int i;
   2060         for (i = 0; i < 4; i++) {
   2061             if ((val >> i) & 1) {
   2062                 tcg_gen_st_i32(zero, cpu_env,
   2063                                vector_elem_offset(&decode->op[0], MO_32, i));
   2064             }
   2065         }
   2066     }
   2067 }
   2068 
   2069 static void gen_VINSERTPS_r(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2070 {
   2071     int val = decode->immediate;
   2072     tcg_gen_ld_i32(s->tmp2_i32, cpu_env,
   2073                    vector_elem_offset(&decode->op[2], MO_32, (val >> 6) & 3));
   2074     gen_vinsertps(s, env, decode);
   2075 }
   2076 
   2077 static void gen_VINSERTPS_m(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2078 {
   2079     tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
   2080     gen_vinsertps(s, env, decode);
   2081 }
   2082 
   2083 static void gen_VINSERTx128(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2084 {
   2085     int mask = decode->immediate & 1;
   2086     tcg_gen_gvec_mov(MO_64,
   2087                      decode->op[0].offset + offsetof(YMMReg, YMM_X(mask)),
   2088                      decode->op[2].offset + offsetof(YMMReg, YMM_X(0)), 16, 16);
   2089     tcg_gen_gvec_mov(MO_64,
   2090                      decode->op[0].offset + offsetof(YMMReg, YMM_X(!mask)),
   2091                      decode->op[1].offset + offsetof(YMMReg, YMM_X(!mask)), 16, 16);
   2092 }
   2093 
   2094 static inline void gen_maskmov(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
   2095                                SSEFunc_0_eppt xmm, SSEFunc_0_eppt ymm)
   2096 {
   2097     if (!s->vex_l) {
   2098         xmm(cpu_env, OP_PTR2, OP_PTR1, s->A0);
   2099     } else {
   2100         ymm(cpu_env, OP_PTR2, OP_PTR1, s->A0);
   2101     }
   2102 }
   2103 
   2104 static void gen_VMASKMOVPD_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2105 {
   2106     gen_maskmov(s, env, decode, gen_helper_vpmaskmovq_st_xmm, gen_helper_vpmaskmovq_st_ymm);
   2107 }
   2108 
   2109 static void gen_VMASKMOVPS_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2110 {
   2111     gen_maskmov(s, env, decode, gen_helper_vpmaskmovd_st_xmm, gen_helper_vpmaskmovd_st_ymm);
   2112 }
   2113 
   2114 static void gen_VMOVHPx_ld(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2115 {
   2116     gen_ldq_env_A0(s, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
   2117     if (decode->op[0].offset != decode->op[1].offset) {
   2118         tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
   2119         tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
   2120     }
   2121 }
   2122 
   2123 static void gen_VMOVHPx_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2124 {
   2125     gen_stq_env_A0(s, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
   2126 }
   2127 
   2128 static void gen_VMOVHPx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2129 {
   2130     if (decode->op[0].offset != decode->op[2].offset) {
   2131         tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
   2132         tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
   2133     }
   2134     if (decode->op[0].offset != decode->op[1].offset) {
   2135         tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
   2136         tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
   2137     }
   2138 }
   2139 
   2140 static void gen_VMOVHLPS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2141 {
   2142     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
   2143     tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
   2144     if (decode->op[0].offset != decode->op[1].offset) {
   2145         tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(1)));
   2146         tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
   2147     }
   2148 }
   2149 
   2150 static void gen_VMOVLHPS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2151 {
   2152     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[2].offset);
   2153     tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
   2154     if (decode->op[0].offset != decode->op[1].offset) {
   2155         tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
   2156         tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
   2157     }
   2158 }
   2159 
   2160 /*
   2161  * Note that MOVLPx supports 256-bit operation unlike MOVHLPx, MOVLHPx, MOXHPx.
   2162  * Use a gvec move to move everything above the bottom 64 bits.
   2163  */
   2164 
   2165 static void gen_VMOVLPx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2166 {
   2167     int vec_len = vector_len(s, decode);
   2168 
   2169     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(0)));
   2170     tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
   2171     tcg_gen_st_i64(s->tmp1_i64, cpu_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
   2172 }
   2173 
   2174 static void gen_VMOVLPx_ld(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2175 {
   2176     int vec_len = vector_len(s, decode);
   2177 
   2178     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
   2179     tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
   2180     tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
   2181 }
   2182 
   2183 static void gen_VMOVLPx_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2184 {
   2185     tcg_gen_ld_i64(s->tmp1_i64, OP_PTR2, offsetof(ZMMReg, ZMM_Q(0)));
   2186     tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
   2187 }
   2188 
   2189 static void gen_VMOVSD_ld(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2190 {
   2191     TCGv_i64 zero = tcg_constant_i64(0);
   2192 
   2193     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
   2194     tcg_gen_st_i64(zero, OP_PTR0, offsetof(ZMMReg, ZMM_Q(1)));
   2195     tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
   2196 }
   2197 
   2198 static void gen_VMOVSS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2199 {
   2200     int vec_len = vector_len(s, decode);
   2201 
   2202     tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
   2203     tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
   2204     tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
   2205 }
   2206 
   2207 static void gen_VMOVSS_ld(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2208 {
   2209     int vec_len = vector_len(s, decode);
   2210 
   2211     tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
   2212     tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
   2213     tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
   2214 }
   2215 
   2216 static void gen_VMOVSS_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2217 {
   2218     tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
   2219     tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
   2220 }
   2221 
   2222 static void gen_VPMASKMOV_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2223 {
   2224     if (s->vex_w) {
   2225         gen_VMASKMOVPD_st(s, env, decode);
   2226     } else {
   2227         gen_VMASKMOVPS_st(s, env, decode);
   2228     }
   2229 }
   2230 
   2231 static void gen_VPERMD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2232 {
   2233     assert(s->vex_l);
   2234     gen_helper_vpermd_ymm(OP_PTR0, OP_PTR1, OP_PTR2);
   2235 }
   2236 
   2237 static void gen_VPERM2x128(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2238 {
   2239     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   2240     assert(s->vex_l);
   2241     gen_helper_vpermdq_ymm(OP_PTR0, OP_PTR1, OP_PTR2, imm);
   2242 }
   2243 
   2244 static void gen_VPHMINPOSUW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2245 {
   2246     assert(!s->vex_l);
   2247     gen_helper_phminposuw_xmm(cpu_env, OP_PTR0, OP_PTR2);
   2248 }
   2249 
   2250 static void gen_VROUNDSD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2251 {
   2252     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   2253     assert(!s->vex_l);
   2254     gen_helper_roundsd_xmm(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
   2255 }
   2256 
   2257 static void gen_VROUNDSS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2258 {
   2259     TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
   2260     assert(!s->vex_l);
   2261     gen_helper_roundss_xmm(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
   2262 }
   2263 
   2264 static void gen_VSHUF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2265 {
   2266     TCGv_i32 imm = tcg_constant_i32(decode->immediate);
   2267     SSEFunc_0_pppi ps, pd, fn;
   2268     ps = s->vex_l ? gen_helper_shufps_ymm : gen_helper_shufps_xmm;
   2269     pd = s->vex_l ? gen_helper_shufpd_ymm : gen_helper_shufpd_xmm;
   2270     fn = s->prefix & PREFIX_DATA ? pd : ps;
   2271     fn(OP_PTR0, OP_PTR1, OP_PTR2, imm);
   2272 }
   2273 
   2274 static void gen_VUCOMI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2275 {
   2276     SSEFunc_0_epp fn;
   2277     fn = s->prefix & PREFIX_DATA ? gen_helper_ucomisd : gen_helper_ucomiss;
   2278     fn(cpu_env, OP_PTR1, OP_PTR2);
   2279     set_cc_op(s, CC_OP_EFLAGS);
   2280 }
   2281 
   2282 static void gen_VZEROALL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2283 {
   2284     TCGv_ptr ptr = tcg_temp_new_ptr();
   2285 
   2286     tcg_gen_addi_ptr(ptr, cpu_env, offsetof(CPUX86State, xmm_t0));
   2287     gen_helper_memset(ptr, ptr, tcg_constant_i32(0),
   2288                       tcg_constant_ptr(CPU_NB_REGS * sizeof(ZMMReg)));
   2289     tcg_temp_free_ptr(ptr);
   2290 }
   2291 
   2292 static void gen_VZEROUPPER(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
   2293 {
   2294     int i;
   2295 
   2296     for (i = 0; i < CPU_NB_REGS; i++) {
   2297         int offset = offsetof(CPUX86State, xmm_regs[i].ZMM_X(1));
   2298         tcg_gen_gvec_dup_imm(MO_64, offset, 16, 16, 0);
   2299     }
   2300 }