xref: /openbmc/qemu/target/i386/tcg/emit.c.inc (revision 2e1cacfb)
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/*
23 * Sometimes, knowing what the backend has can produce better code.
24 * The exact opcode to check depends on 32- vs. 64-bit.
25 */
26#ifdef TARGET_X86_64
27#define TCG_TARGET_HAS_extract2_tl      TCG_TARGET_HAS_extract2_i64
28#define TCG_TARGET_deposit_tl_valid     TCG_TARGET_deposit_i64_valid
29#define TCG_TARGET_extract_tl_valid     TCG_TARGET_extract_i64_valid
30#else
31#define TCG_TARGET_HAS_extract2_tl      TCG_TARGET_HAS_extract2_i32
32#define TCG_TARGET_deposit_tl_valid     TCG_TARGET_deposit_i32_valid
33#define TCG_TARGET_extract_tl_valid     TCG_TARGET_extract_i32_valid
34#endif
35
36#define MMX_OFFSET(reg)                        \
37  ({ assert((reg) >= 0 && (reg) <= 7);         \
38     offsetof(CPUX86State, fpregs[reg].mmx); })
39
40#define ZMM_OFFSET(reg)                        \
41  ({ assert((reg) >= 0 && (reg) <= 15);        \
42     offsetof(CPUX86State, xmm_regs[reg]); })
43
44typedef void (*SSEFunc_i_ep)(TCGv_i32 val, TCGv_ptr env, TCGv_ptr reg);
45typedef void (*SSEFunc_l_ep)(TCGv_i64 val, TCGv_ptr env, TCGv_ptr reg);
46typedef void (*SSEFunc_0_epp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b);
47typedef void (*SSEFunc_0_eppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
48                               TCGv_ptr reg_c);
49typedef void (*SSEFunc_0_epppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
50                                TCGv_ptr reg_c, TCGv_ptr reg_d);
51typedef void (*SSEFunc_0_eppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
52                               TCGv_i32 val);
53typedef void (*SSEFunc_0_epppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
54                                TCGv_ptr reg_c, TCGv_i32 val);
55typedef void (*SSEFunc_0_ppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_i32 val);
56typedef void (*SSEFunc_0_pppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_ptr reg_c,
57                               TCGv_i32 val);
58typedef void (*SSEFunc_0_eppt)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
59                               TCGv val);
60typedef void (*SSEFunc_0_epppti)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
61                                 TCGv_ptr reg_c, TCGv a0, TCGv_i32 scale);
62typedef void (*SSEFunc_0_eppppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
63                                  TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 flags);
64typedef void (*SSEFunc_0_eppppii)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
65                                  TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 even,
66                                  TCGv_i32 odd);
67
68static void gen_JMP_m(DisasContext *s, X86DecodedInsn *decode);
69static void gen_JMP(DisasContext *s, X86DecodedInsn *decode);
70
71static inline TCGv_i32 tcg_constant8u_i32(uint8_t val)
72{
73    return tcg_constant_i32(val);
74}
75
76static void gen_NM_exception(DisasContext *s)
77{
78    gen_exception(s, EXCP07_PREX);
79}
80
81static void gen_lea_modrm(DisasContext *s, X86DecodedInsn *decode)
82{
83    AddressParts *mem = &decode->mem;
84    TCGv ea;
85
86    ea = gen_lea_modrm_1(s, *mem, decode->e.vex_class == 12);
87    if (decode->e.special == X86_SPECIAL_BitTest) {
88        MemOp ot = decode->op[1].ot;
89        int poslen = 8 << ot;
90        int opn = decode->op[2].n;
91        TCGv ofs = tcg_temp_new();
92
93        /* Extract memory displacement from the second operand.  */
94        assert(decode->op[2].unit == X86_OP_INT && decode->op[2].ot != MO_8);
95        tcg_gen_sextract_tl(ofs, cpu_regs[opn], 3, poslen - 3);
96        tcg_gen_andi_tl(ofs, ofs, -1 << ot);
97        tcg_gen_add_tl(s->A0, ea, ofs);
98        ea = s->A0;
99    }
100
101    gen_lea_v_seg(s, ea, mem->def_seg, s->override);
102}
103
104static inline int mmx_offset(MemOp ot)
105{
106    switch (ot) {
107    case MO_8:
108        return offsetof(MMXReg, MMX_B(0));
109    case MO_16:
110        return offsetof(MMXReg, MMX_W(0));
111    case MO_32:
112        return offsetof(MMXReg, MMX_L(0));
113    case MO_64:
114        return offsetof(MMXReg, MMX_Q(0));
115    default:
116        g_assert_not_reached();
117    }
118}
119
120static inline int xmm_offset(MemOp ot)
121{
122    switch (ot) {
123    case MO_8:
124        return offsetof(ZMMReg, ZMM_B(0));
125    case MO_16:
126        return offsetof(ZMMReg, ZMM_W(0));
127    case MO_32:
128        return offsetof(ZMMReg, ZMM_L(0));
129    case MO_64:
130        return offsetof(ZMMReg, ZMM_Q(0));
131    case MO_128:
132        return offsetof(ZMMReg, ZMM_X(0));
133    case MO_256:
134        return offsetof(ZMMReg, ZMM_Y(0));
135    default:
136        g_assert_not_reached();
137    }
138}
139
140static int vector_reg_offset(X86DecodedOp *op)
141{
142    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
143
144    if (op->unit == X86_OP_MMX) {
145        return op->offset - mmx_offset(op->ot);
146    } else {
147        return op->offset - xmm_offset(op->ot);
148    }
149}
150
151static int vector_elem_offset(X86DecodedOp *op, MemOp ot, int n)
152{
153    int base_ofs = vector_reg_offset(op);
154    switch(ot) {
155    case MO_8:
156        if (op->unit == X86_OP_MMX) {
157            return base_ofs + offsetof(MMXReg, MMX_B(n));
158        } else {
159            return base_ofs + offsetof(ZMMReg, ZMM_B(n));
160        }
161    case MO_16:
162        if (op->unit == X86_OP_MMX) {
163            return base_ofs + offsetof(MMXReg, MMX_W(n));
164        } else {
165            return base_ofs + offsetof(ZMMReg, ZMM_W(n));
166        }
167    case MO_32:
168        if (op->unit == X86_OP_MMX) {
169            return base_ofs + offsetof(MMXReg, MMX_L(n));
170        } else {
171            return base_ofs + offsetof(ZMMReg, ZMM_L(n));
172        }
173    case MO_64:
174        if (op->unit == X86_OP_MMX) {
175            return base_ofs;
176        } else {
177            return base_ofs + offsetof(ZMMReg, ZMM_Q(n));
178        }
179    case MO_128:
180        assert(op->unit == X86_OP_SSE);
181        return base_ofs + offsetof(ZMMReg, ZMM_X(n));
182    case MO_256:
183        assert(op->unit == X86_OP_SSE);
184        return base_ofs + offsetof(ZMMReg, ZMM_Y(n));
185    default:
186        g_assert_not_reached();
187    }
188}
189
190static void compute_mmx_offset(X86DecodedOp *op)
191{
192    if (!op->has_ea) {
193        op->offset = MMX_OFFSET(op->n) + mmx_offset(op->ot);
194    } else {
195        op->offset = offsetof(CPUX86State, mmx_t0) + mmx_offset(op->ot);
196    }
197}
198
199static void compute_xmm_offset(X86DecodedOp *op)
200{
201    if (!op->has_ea) {
202        op->offset = ZMM_OFFSET(op->n) + xmm_offset(op->ot);
203    } else {
204        op->offset = offsetof(CPUX86State, xmm_t0) + xmm_offset(op->ot);
205    }
206}
207
208static void gen_load_sse(DisasContext *s, TCGv temp, MemOp ot, int dest_ofs, bool aligned)
209{
210    switch(ot) {
211    case MO_8:
212        gen_op_ld_v(s, MO_8, temp, s->A0);
213        tcg_gen_st8_tl(temp, tcg_env, dest_ofs);
214        break;
215    case MO_16:
216        gen_op_ld_v(s, MO_16, temp, s->A0);
217        tcg_gen_st16_tl(temp, tcg_env, dest_ofs);
218        break;
219    case MO_32:
220        gen_op_ld_v(s, MO_32, temp, s->A0);
221        tcg_gen_st32_tl(temp, tcg_env, dest_ofs);
222        break;
223    case MO_64:
224        gen_ldq_env_A0(s, dest_ofs);
225        break;
226    case MO_128:
227        gen_ldo_env_A0(s, dest_ofs, aligned);
228        break;
229    case MO_256:
230        gen_ldy_env_A0(s, dest_ofs, aligned);
231        break;
232    default:
233        g_assert_not_reached();
234    }
235}
236
237static bool sse_needs_alignment(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
238{
239    switch (decode->e.vex_class) {
240    case 2:
241    case 4:
242        if ((s->prefix & PREFIX_VEX) ||
243            decode->e.vex_special == X86_VEX_SSEUnaligned) {
244            /* MOST legacy SSE instructions require aligned memory operands, but not all.  */
245            return false;
246        }
247        /* fall through */
248    case 1:
249        return ot >= MO_128;
250
251    default:
252        return false;
253    }
254}
255
256static void gen_load(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
257{
258    X86DecodedOp *op = &decode->op[opn];
259
260    switch (op->unit) {
261    case X86_OP_SKIP:
262        return;
263    case X86_OP_SEG:
264        tcg_gen_ld32u_tl(v, tcg_env,
265                         offsetof(CPUX86State,segs[op->n].selector));
266        break;
267#ifndef CONFIG_USER_ONLY
268    case X86_OP_CR:
269        if (op->n == 8) {
270            translator_io_start(&s->base);
271            gen_helper_read_cr8(v, tcg_env);
272        } else {
273            tcg_gen_ld_tl(v, tcg_env, offsetof(CPUX86State, cr[op->n]));
274        }
275        break;
276    case X86_OP_DR:
277        /* CR4.DE tested in the helper.  */
278        gen_helper_get_dr(v, tcg_env, tcg_constant_i32(op->n));
279        break;
280#endif
281    case X86_OP_INT:
282        if (op->has_ea) {
283            if (v == s->T0 && decode->e.special == X86_SPECIAL_SExtT0) {
284                gen_op_ld_v(s, op->ot | MO_SIGN, v, s->A0);
285            } else {
286                gen_op_ld_v(s, op->ot, v, s->A0);
287            }
288
289        } else if (op->ot == MO_8 && byte_reg_is_xH(s, op->n)) {
290            if (v == s->T0 && decode->e.special == X86_SPECIAL_SExtT0) {
291                tcg_gen_sextract_tl(v, cpu_regs[op->n - 4], 8, 8);
292            } else {
293                tcg_gen_extract_tl(v, cpu_regs[op->n - 4], 8, 8);
294            }
295
296        } else if (op->ot < MO_TL && v == s->T0 &&
297                   (decode->e.special == X86_SPECIAL_SExtT0 ||
298                    decode->e.special == X86_SPECIAL_ZExtT0)) {
299            if (decode->e.special == X86_SPECIAL_SExtT0) {
300                tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot | MO_SIGN);
301            } else {
302                tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot);
303            }
304
305        } else {
306            tcg_gen_mov_tl(v, cpu_regs[op->n]);
307        }
308        break;
309    case X86_OP_IMM:
310        tcg_gen_movi_tl(v, op->imm);
311        break;
312
313    case X86_OP_MMX:
314        compute_mmx_offset(op);
315        goto load_vector;
316
317    case X86_OP_SSE:
318        compute_xmm_offset(op);
319    load_vector:
320        if (op->has_ea) {
321            bool aligned = sse_needs_alignment(s, decode, op->ot);
322            gen_load_sse(s, v, op->ot, op->offset, aligned);
323        }
324        break;
325
326    default:
327        g_assert_not_reached();
328    }
329}
330
331static TCGv_ptr op_ptr(X86DecodedInsn *decode, int opn)
332{
333    X86DecodedOp *op = &decode->op[opn];
334
335    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
336    if (op->v_ptr) {
337        return op->v_ptr;
338    }
339    op->v_ptr = tcg_temp_new_ptr();
340
341    /* The temporary points to the MMXReg or ZMMReg.  */
342    tcg_gen_addi_ptr(op->v_ptr, tcg_env, vector_reg_offset(op));
343    return op->v_ptr;
344}
345
346#define OP_PTR0 op_ptr(decode, 0)
347#define OP_PTR1 op_ptr(decode, 1)
348#define OP_PTR2 op_ptr(decode, 2)
349
350static void gen_writeback(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
351{
352    X86DecodedOp *op = &decode->op[opn];
353    switch (op->unit) {
354    case X86_OP_SKIP:
355        break;
356    case X86_OP_SEG:
357        /* Note that gen_movl_seg takes care of interrupt shadow and TF.  */
358        gen_movl_seg(s, op->n, s->T0);
359        break;
360    case X86_OP_INT:
361        if (op->has_ea) {
362            gen_op_st_v(s, op->ot, v, s->A0);
363        } else {
364            gen_op_mov_reg_v(s, op->ot, op->n, v);
365        }
366        break;
367    case X86_OP_MMX:
368        break;
369    case X86_OP_SSE:
370        if (!op->has_ea && (s->prefix & PREFIX_VEX) && op->ot <= MO_128) {
371            tcg_gen_gvec_dup_imm(MO_64,
372                                 offsetof(CPUX86State, xmm_regs[op->n].ZMM_X(1)),
373                                 16, 16, 0);
374        }
375        break;
376#ifndef CONFIG_USER_ONLY
377    case X86_OP_CR:
378        if (op->n == 8) {
379            translator_io_start(&s->base);
380        }
381        gen_helper_write_crN(tcg_env, tcg_constant_i32(op->n), v);
382        s->base.is_jmp = DISAS_EOB_NEXT;
383        break;
384    case X86_OP_DR:
385        /* CR4.DE tested in the helper.  */
386        gen_helper_set_dr(tcg_env, tcg_constant_i32(op->n), v);
387        s->base.is_jmp = DISAS_EOB_NEXT;
388        break;
389#endif
390    default:
391        g_assert_not_reached();
392    }
393    op->unit = X86_OP_SKIP;
394}
395
396static inline int vector_len(DisasContext *s, X86DecodedInsn *decode)
397{
398    if (decode->e.special == X86_SPECIAL_MMX &&
399        !(s->prefix & (PREFIX_DATA | PREFIX_REPZ | PREFIX_REPNZ))) {
400        return 8;
401    }
402    return s->vex_l ? 32 : 16;
403}
404
405static void prepare_update1_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
406{
407    decode->cc_dst = s->T0;
408    decode->cc_op = op;
409}
410
411static void prepare_update2_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
412{
413    decode->cc_src = s->T1;
414    decode->cc_dst = s->T0;
415    decode->cc_op = op;
416}
417
418static void prepare_update_cc_incdec(X86DecodedInsn *decode, DisasContext *s, CCOp op)
419{
420    gen_compute_eflags_c(s, s->T1);
421    prepare_update2_cc(decode, s, op);
422}
423
424static void prepare_update3_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op, TCGv reg)
425{
426    decode->cc_src2 = reg;
427    decode->cc_src = s->T1;
428    decode->cc_dst = s->T0;
429    decode->cc_op = op;
430}
431
432/* Set up decode->cc_* to modify CF while keeping other flags unchanged.  */
433static void prepare_update_cf(X86DecodedInsn *decode, DisasContext *s, TCGv cf)
434{
435    switch (s->cc_op) {
436    case CC_OP_ADOX:
437    case CC_OP_ADCOX:
438        decode->cc_src2 = cpu_cc_src2;
439        decode->cc_src = cpu_cc_src;
440        decode->cc_op = CC_OP_ADCOX;
441        break;
442
443    case CC_OP_EFLAGS:
444    case CC_OP_ADCX:
445        decode->cc_src = cpu_cc_src;
446        decode->cc_op = CC_OP_ADCX;
447        break;
448
449    default:
450        decode->cc_src = tcg_temp_new();
451        gen_mov_eflags(s, decode->cc_src);
452        decode->cc_op = CC_OP_ADCX;
453        break;
454    }
455    decode->cc_dst = cf;
456}
457
458static void gen_store_sse(DisasContext *s, X86DecodedInsn *decode, int src_ofs)
459{
460    MemOp ot = decode->op[0].ot;
461    int vec_len = vector_len(s, decode);
462    bool aligned = sse_needs_alignment(s, decode, ot);
463
464    if (!decode->op[0].has_ea) {
465        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, vec_len, vec_len);
466        return;
467    }
468
469    switch (ot) {
470    case MO_64:
471        gen_stq_env_A0(s, src_ofs);
472        break;
473    case MO_128:
474        gen_sto_env_A0(s, src_ofs, aligned);
475        break;
476    case MO_256:
477        gen_sty_env_A0(s, src_ofs, aligned);
478        break;
479    default:
480        g_assert_not_reached();
481    }
482}
483
484static void gen_helper_pavgusb(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b)
485{
486    gen_helper_pavgb_mmx(env, reg_a, reg_a, reg_b);
487}
488
489#define FN_3DNOW_MOVE ((SSEFunc_0_epp) (uintptr_t) 1)
490static const SSEFunc_0_epp fns_3dnow[] = {
491    [0x0c] = gen_helper_pi2fw,
492    [0x0d] = gen_helper_pi2fd,
493    [0x1c] = gen_helper_pf2iw,
494    [0x1d] = gen_helper_pf2id,
495    [0x8a] = gen_helper_pfnacc,
496    [0x8e] = gen_helper_pfpnacc,
497    [0x90] = gen_helper_pfcmpge,
498    [0x94] = gen_helper_pfmin,
499    [0x96] = gen_helper_pfrcp,
500    [0x97] = gen_helper_pfrsqrt,
501    [0x9a] = gen_helper_pfsub,
502    [0x9e] = gen_helper_pfadd,
503    [0xa0] = gen_helper_pfcmpgt,
504    [0xa4] = gen_helper_pfmax,
505    [0xa6] = FN_3DNOW_MOVE, /* PFRCPIT1; no need to actually increase precision */
506    [0xa7] = FN_3DNOW_MOVE, /* PFRSQIT1 */
507    [0xb6] = FN_3DNOW_MOVE, /* PFRCPIT2 */
508    [0xaa] = gen_helper_pfsubr,
509    [0xae] = gen_helper_pfacc,
510    [0xb0] = gen_helper_pfcmpeq,
511    [0xb4] = gen_helper_pfmul,
512    [0xb7] = gen_helper_pmulhrw_mmx,
513    [0xbb] = gen_helper_pswapd,
514    [0xbf] = gen_helper_pavgusb,
515};
516
517static void gen_3dnow(DisasContext *s, X86DecodedInsn *decode)
518{
519    uint8_t b = decode->immediate;
520    SSEFunc_0_epp fn = b < ARRAY_SIZE(fns_3dnow) ? fns_3dnow[b] : NULL;
521
522    if (!fn) {
523        gen_illegal_opcode(s);
524        return;
525    }
526    if (s->flags & HF_TS_MASK) {
527        gen_NM_exception(s);
528        return;
529    }
530    if (s->flags & HF_EM_MASK) {
531        gen_illegal_opcode(s);
532        return;
533    }
534
535    gen_helper_enter_mmx(tcg_env);
536    if (fn == FN_3DNOW_MOVE) {
537       tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset);
538       tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset);
539    } else {
540       fn(tcg_env, OP_PTR0, OP_PTR1);
541    }
542}
543
544/*
545 * 00 = v*ps Vps, Hps, Wpd
546 * 66 = v*pd Vpd, Hpd, Wps
547 * f3 = v*ss Vss, Hss, Wps
548 * f2 = v*sd Vsd, Hsd, Wps
549 */
550static inline void gen_unary_fp_sse(DisasContext *s, X86DecodedInsn *decode,
551                              SSEFunc_0_epp pd_xmm, SSEFunc_0_epp ps_xmm,
552                              SSEFunc_0_epp pd_ymm, SSEFunc_0_epp ps_ymm,
553                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
554{
555    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
556        SSEFunc_0_eppp fn = s->prefix & PREFIX_REPZ ? ss : sd;
557        if (!fn) {
558            gen_illegal_opcode(s);
559            return;
560        }
561        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
562    } else {
563        SSEFunc_0_epp ps, pd, fn;
564        ps = s->vex_l ? ps_ymm : ps_xmm;
565        pd = s->vex_l ? pd_ymm : pd_xmm;
566        fn = s->prefix & PREFIX_DATA ? pd : ps;
567        if (!fn) {
568            gen_illegal_opcode(s);
569            return;
570        }
571        fn(tcg_env, OP_PTR0, OP_PTR2);
572    }
573}
574#define UNARY_FP_SSE(uname, lname)                                                 \
575static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
576{                                                                                  \
577    gen_unary_fp_sse(s, decode,                                                    \
578                     gen_helper_##lname##pd_xmm,                                   \
579                     gen_helper_##lname##ps_xmm,                                   \
580                     gen_helper_##lname##pd_ymm,                                   \
581                     gen_helper_##lname##ps_ymm,                                   \
582                     gen_helper_##lname##sd,                                       \
583                     gen_helper_##lname##ss);                                      \
584}
585UNARY_FP_SSE(VSQRT, sqrt)
586
587/*
588 * 00 = v*ps Vps, Hps, Wpd
589 * 66 = v*pd Vpd, Hpd, Wps
590 * f3 = v*ss Vss, Hss, Wps
591 * f2 = v*sd Vsd, Hsd, Wps
592 */
593static inline void gen_fp_sse(DisasContext *s, X86DecodedInsn *decode,
594                              SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
595                              SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm,
596                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
597{
598    SSEFunc_0_eppp ps, pd, fn;
599    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
600        fn = s->prefix & PREFIX_REPZ ? ss : sd;
601    } else {
602        ps = s->vex_l ? ps_ymm : ps_xmm;
603        pd = s->vex_l ? pd_ymm : pd_xmm;
604        fn = s->prefix & PREFIX_DATA ? pd : ps;
605    }
606    if (fn) {
607        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
608    } else {
609        gen_illegal_opcode(s);
610    }
611}
612
613#define FP_SSE(uname, lname)                                                       \
614static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
615{                                                                                  \
616    gen_fp_sse(s, decode,                                                          \
617               gen_helper_##lname##pd_xmm,                                         \
618               gen_helper_##lname##ps_xmm,                                         \
619               gen_helper_##lname##pd_ymm,                                         \
620               gen_helper_##lname##ps_ymm,                                         \
621               gen_helper_##lname##sd,                                             \
622               gen_helper_##lname##ss);                                            \
623}
624FP_SSE(VADD, add)
625FP_SSE(VMUL, mul)
626FP_SSE(VSUB, sub)
627FP_SSE(VMIN, min)
628FP_SSE(VDIV, div)
629FP_SSE(VMAX, max)
630
631#define FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, even, odd)                         \
632static void gen_##uname##Px(DisasContext *s, X86DecodedInsn *decode)               \
633{                                                                                  \
634    SSEFunc_0_eppppii xmm = s->vex_w ? gen_helper_fma4pd_xmm : gen_helper_fma4ps_xmm; \
635    SSEFunc_0_eppppii ymm = s->vex_w ? gen_helper_fma4pd_ymm : gen_helper_fma4ps_ymm; \
636    SSEFunc_0_eppppii fn = s->vex_l ? ymm : xmm;                                   \
637                                                                                   \
638    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
639       tcg_constant_i32(even),                                                     \
640       tcg_constant_i32((even) ^ (odd)));                                          \
641}
642
643#define FMA_SSE(uname, ptr0, ptr1, ptr2, flags)                                    \
644FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, flags, flags)                              \
645static void gen_##uname##Sx(DisasContext *s, X86DecodedInsn *decode)               \
646{                                                                                  \
647    SSEFunc_0_eppppi fn = s->vex_w ? gen_helper_fma4sd : gen_helper_fma4ss;        \
648                                                                                   \
649    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
650       tcg_constant_i32(flags));                                                   \
651}                                                                                  \
652
653FMA_SSE(VFMADD231,  OP_PTR1, OP_PTR2, OP_PTR0, 0)
654FMA_SSE(VFMADD213,  OP_PTR1, OP_PTR0, OP_PTR2, 0)
655FMA_SSE(VFMADD132,  OP_PTR0, OP_PTR2, OP_PTR1, 0)
656
657FMA_SSE(VFNMADD231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_product)
658FMA_SSE(VFNMADD213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_product)
659FMA_SSE(VFNMADD132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_product)
660
661FMA_SSE(VFMSUB231,  OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c)
662FMA_SSE(VFMSUB213,  OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c)
663FMA_SSE(VFMSUB132,  OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c)
664
665FMA_SSE(VFNMSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c|float_muladd_negate_product)
666FMA_SSE(VFNMSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c|float_muladd_negate_product)
667FMA_SSE(VFNMSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c|float_muladd_negate_product)
668
669FMA_SSE_PACKED(VFMADDSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c, 0)
670FMA_SSE_PACKED(VFMADDSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c, 0)
671FMA_SSE_PACKED(VFMADDSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c, 0)
672
673FMA_SSE_PACKED(VFMSUBADD231, OP_PTR1, OP_PTR2, OP_PTR0, 0, float_muladd_negate_c)
674FMA_SSE_PACKED(VFMSUBADD213, OP_PTR1, OP_PTR0, OP_PTR2, 0, float_muladd_negate_c)
675FMA_SSE_PACKED(VFMSUBADD132, OP_PTR0, OP_PTR2, OP_PTR1, 0, float_muladd_negate_c)
676
677#define FP_UNPACK_SSE(uname, lname)                                                \
678static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
679{                                                                                  \
680    /* PS maps to the DQ integer instruction, PD maps to QDQ.  */                  \
681    gen_fp_sse(s, decode,                                                          \
682               gen_helper_##lname##qdq_xmm,                                        \
683               gen_helper_##lname##dq_xmm,                                         \
684               gen_helper_##lname##qdq_ymm,                                        \
685               gen_helper_##lname##dq_ymm,                                         \
686               NULL, NULL);                                                        \
687}
688FP_UNPACK_SSE(VUNPCKLPx, punpckl)
689FP_UNPACK_SSE(VUNPCKHPx, punpckh)
690
691/*
692 * 00 = v*ps Vps, Wpd
693 * f3 = v*ss Vss, Wps
694 */
695static inline void gen_unary_fp32_sse(DisasContext *s, X86DecodedInsn *decode,
696                                      SSEFunc_0_epp ps_xmm,
697                                      SSEFunc_0_epp ps_ymm,
698                                      SSEFunc_0_eppp ss)
699{
700    if ((s->prefix & (PREFIX_DATA | PREFIX_REPNZ)) != 0) {
701        goto illegal_op;
702    } else if (s->prefix & PREFIX_REPZ) {
703        if (!ss) {
704            goto illegal_op;
705        }
706        ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
707    } else {
708        SSEFunc_0_epp fn = s->vex_l ? ps_ymm : ps_xmm;
709        if (!fn) {
710            goto illegal_op;
711        }
712        fn(tcg_env, OP_PTR0, OP_PTR2);
713    }
714    return;
715
716illegal_op:
717    gen_illegal_opcode(s);
718}
719#define UNARY_FP32_SSE(uname, lname)                                               \
720static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
721{                                                                                  \
722    gen_unary_fp32_sse(s, decode,                                                  \
723                       gen_helper_##lname##ps_xmm,                                 \
724                       gen_helper_##lname##ps_ymm,                                 \
725                       gen_helper_##lname##ss);                                    \
726}
727UNARY_FP32_SSE(VRSQRT, rsqrt)
728UNARY_FP32_SSE(VRCP, rcp)
729
730/*
731 * 66 = v*pd Vpd, Hpd, Wpd
732 * f2 = v*ps Vps, Hps, Wps
733 */
734static inline void gen_horizontal_fp_sse(DisasContext *s, X86DecodedInsn *decode,
735                                         SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
736                                         SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm)
737{
738    SSEFunc_0_eppp ps, pd, fn;
739    ps = s->vex_l ? ps_ymm : ps_xmm;
740    pd = s->vex_l ? pd_ymm : pd_xmm;
741    fn = s->prefix & PREFIX_DATA ? pd : ps;
742    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
743}
744#define HORIZONTAL_FP_SSE(uname, lname)                                            \
745static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
746{                                                                                  \
747    gen_horizontal_fp_sse(s, decode,                                               \
748                          gen_helper_##lname##pd_xmm, gen_helper_##lname##ps_xmm,  \
749                          gen_helper_##lname##pd_ymm, gen_helper_##lname##ps_ymm); \
750}
751HORIZONTAL_FP_SSE(VHADD, hadd)
752HORIZONTAL_FP_SSE(VHSUB, hsub)
753HORIZONTAL_FP_SSE(VADDSUB, addsub)
754
755static inline void gen_ternary_sse(DisasContext *s, X86DecodedInsn *decode,
756                                   int op3, SSEFunc_0_epppp xmm, SSEFunc_0_epppp ymm)
757{
758    SSEFunc_0_epppp fn = s->vex_l ? ymm : xmm;
759    TCGv_ptr ptr3 = tcg_temp_new_ptr();
760
761    /* The format of the fourth input is Lx */
762    tcg_gen_addi_ptr(ptr3, tcg_env, ZMM_OFFSET(op3));
763    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, ptr3);
764}
765#define TERNARY_SSE(uname, uvname, lname)                                          \
766static void gen_##uvname(DisasContext *s, X86DecodedInsn *decode)                  \
767{                                                                                  \
768    gen_ternary_sse(s, decode, (uint8_t)decode->immediate >> 4,                    \
769                    gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);           \
770}                                                                                  \
771static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
772{                                                                                  \
773    gen_ternary_sse(s, decode, 0,                                                  \
774                  gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);             \
775}
776TERNARY_SSE(BLENDVPS, VBLENDVPS, blendvps)
777TERNARY_SSE(BLENDVPD, VBLENDVPD, blendvpd)
778TERNARY_SSE(PBLENDVB, VPBLENDVB, pblendvb)
779
780static inline void gen_binary_imm_sse(DisasContext *s, X86DecodedInsn *decode,
781                                      SSEFunc_0_epppi xmm, SSEFunc_0_epppi ymm)
782{
783    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
784    if (!s->vex_l) {
785        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
786    } else {
787        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
788    }
789}
790
791#define BINARY_IMM_SSE(uname, lname)                                               \
792static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
793{                                                                                  \
794    gen_binary_imm_sse(s, decode,                                                  \
795                       gen_helper_##lname##_xmm,                                   \
796                       gen_helper_##lname##_ymm);                                  \
797}
798
799BINARY_IMM_SSE(VBLENDPD,   blendpd)
800BINARY_IMM_SSE(VBLENDPS,   blendps)
801BINARY_IMM_SSE(VPBLENDW,   pblendw)
802BINARY_IMM_SSE(VDDPS,      dpps)
803#define gen_helper_dppd_ymm NULL
804BINARY_IMM_SSE(VDDPD,      dppd)
805BINARY_IMM_SSE(VMPSADBW,   mpsadbw)
806BINARY_IMM_SSE(PCLMULQDQ,  pclmulqdq)
807
808
809#define UNARY_INT_GVEC(uname, func, ...)                                           \
810static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
811{                                                                                  \
812    int vec_len = vector_len(s, decode);                                          \
813                                                                                   \
814    func(__VA_ARGS__, decode->op[0].offset,                                        \
815         decode->op[2].offset, vec_len, vec_len);                                  \
816}
817UNARY_INT_GVEC(PABSB,          tcg_gen_gvec_abs, MO_8)
818UNARY_INT_GVEC(PABSW,          tcg_gen_gvec_abs, MO_16)
819UNARY_INT_GVEC(PABSD,          tcg_gen_gvec_abs, MO_32)
820UNARY_INT_GVEC(VBROADCASTx128, tcg_gen_gvec_dup_mem, MO_128)
821UNARY_INT_GVEC(VPBROADCASTB,   tcg_gen_gvec_dup_mem, MO_8)
822UNARY_INT_GVEC(VPBROADCASTW,   tcg_gen_gvec_dup_mem, MO_16)
823UNARY_INT_GVEC(VPBROADCASTD,   tcg_gen_gvec_dup_mem, MO_32)
824UNARY_INT_GVEC(VPBROADCASTQ,   tcg_gen_gvec_dup_mem, MO_64)
825
826
827#define BINARY_INT_GVEC(uname, func, ...)                                          \
828static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
829{                                                                                  \
830    int vec_len = vector_len(s, decode);                                          \
831                                                                                   \
832    func(__VA_ARGS__,                                                              \
833         decode->op[0].offset, decode->op[1].offset,                               \
834         decode->op[2].offset, vec_len, vec_len);                                  \
835}
836
837BINARY_INT_GVEC(PADDB,   tcg_gen_gvec_add, MO_8)
838BINARY_INT_GVEC(PADDW,   tcg_gen_gvec_add, MO_16)
839BINARY_INT_GVEC(PADDD,   tcg_gen_gvec_add, MO_32)
840BINARY_INT_GVEC(PADDQ,   tcg_gen_gvec_add, MO_64)
841BINARY_INT_GVEC(PADDSB,  tcg_gen_gvec_ssadd, MO_8)
842BINARY_INT_GVEC(PADDSW,  tcg_gen_gvec_ssadd, MO_16)
843BINARY_INT_GVEC(PADDUSB, tcg_gen_gvec_usadd, MO_8)
844BINARY_INT_GVEC(PADDUSW, tcg_gen_gvec_usadd, MO_16)
845BINARY_INT_GVEC(PAND,    tcg_gen_gvec_and, MO_64)
846BINARY_INT_GVEC(PCMPEQB, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_8)
847BINARY_INT_GVEC(PCMPEQD, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_32)
848BINARY_INT_GVEC(PCMPEQW, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_16)
849BINARY_INT_GVEC(PCMPEQQ, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_64)
850BINARY_INT_GVEC(PCMPGTB, tcg_gen_gvec_cmp, TCG_COND_GT, MO_8)
851BINARY_INT_GVEC(PCMPGTW, tcg_gen_gvec_cmp, TCG_COND_GT, MO_16)
852BINARY_INT_GVEC(PCMPGTD, tcg_gen_gvec_cmp, TCG_COND_GT, MO_32)
853BINARY_INT_GVEC(PCMPGTQ, tcg_gen_gvec_cmp, TCG_COND_GT, MO_64)
854BINARY_INT_GVEC(PMAXSB,  tcg_gen_gvec_smax, MO_8)
855BINARY_INT_GVEC(PMAXSW,  tcg_gen_gvec_smax, MO_16)
856BINARY_INT_GVEC(PMAXSD,  tcg_gen_gvec_smax, MO_32)
857BINARY_INT_GVEC(PMAXUB,  tcg_gen_gvec_umax, MO_8)
858BINARY_INT_GVEC(PMAXUW,  tcg_gen_gvec_umax, MO_16)
859BINARY_INT_GVEC(PMAXUD,  tcg_gen_gvec_umax, MO_32)
860BINARY_INT_GVEC(PMINSB,  tcg_gen_gvec_smin, MO_8)
861BINARY_INT_GVEC(PMINSW,  tcg_gen_gvec_smin, MO_16)
862BINARY_INT_GVEC(PMINSD,  tcg_gen_gvec_smin, MO_32)
863BINARY_INT_GVEC(PMINUB,  tcg_gen_gvec_umin, MO_8)
864BINARY_INT_GVEC(PMINUW,  tcg_gen_gvec_umin, MO_16)
865BINARY_INT_GVEC(PMINUD,  tcg_gen_gvec_umin, MO_32)
866BINARY_INT_GVEC(PMULLW,  tcg_gen_gvec_mul, MO_16)
867BINARY_INT_GVEC(PMULLD,  tcg_gen_gvec_mul, MO_32)
868BINARY_INT_GVEC(POR,     tcg_gen_gvec_or, MO_64)
869BINARY_INT_GVEC(PSUBB,   tcg_gen_gvec_sub, MO_8)
870BINARY_INT_GVEC(PSUBW,   tcg_gen_gvec_sub, MO_16)
871BINARY_INT_GVEC(PSUBD,   tcg_gen_gvec_sub, MO_32)
872BINARY_INT_GVEC(PSUBQ,   tcg_gen_gvec_sub, MO_64)
873BINARY_INT_GVEC(PSUBSB,  tcg_gen_gvec_sssub, MO_8)
874BINARY_INT_GVEC(PSUBSW,  tcg_gen_gvec_sssub, MO_16)
875BINARY_INT_GVEC(PSUBUSB, tcg_gen_gvec_ussub, MO_8)
876BINARY_INT_GVEC(PSUBUSW, tcg_gen_gvec_ussub, MO_16)
877BINARY_INT_GVEC(PXOR,    tcg_gen_gvec_xor, MO_64)
878
879
880/*
881 * 00 = p*  Pq, Qq (if mmx not NULL; no VEX)
882 * 66 = vp* Vx, Hx, Wx
883 *
884 * These are really the same encoding, because 1) V is the same as P when VEX.V
885 * is not present 2) P and Q are the same as H and W apart from MM/XMM
886 */
887static inline void gen_binary_int_sse(DisasContext *s, X86DecodedInsn *decode,
888                                      SSEFunc_0_eppp mmx, SSEFunc_0_eppp xmm, SSEFunc_0_eppp ymm)
889{
890    assert(!!mmx == !!(decode->e.special == X86_SPECIAL_MMX));
891
892    if (mmx && (s->prefix & PREFIX_VEX) && !(s->prefix & PREFIX_DATA)) {
893        /* VEX encoding is not applicable to MMX instructions.  */
894        gen_illegal_opcode(s);
895        return;
896    }
897    if (!(s->prefix & PREFIX_DATA)) {
898        mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
899    } else if (!s->vex_l) {
900        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
901    } else {
902        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
903    }
904}
905
906
907#define BINARY_INT_MMX(uname, lname)                                               \
908static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
909{                                                                                  \
910    gen_binary_int_sse(s, decode,                                                  \
911                          gen_helper_##lname##_mmx,                                \
912                          gen_helper_##lname##_xmm,                                \
913                          gen_helper_##lname##_ymm);                               \
914}
915BINARY_INT_MMX(PUNPCKLBW,  punpcklbw)
916BINARY_INT_MMX(PUNPCKLWD,  punpcklwd)
917BINARY_INT_MMX(PUNPCKLDQ,  punpckldq)
918BINARY_INT_MMX(PACKSSWB,   packsswb)
919BINARY_INT_MMX(PACKUSWB,   packuswb)
920BINARY_INT_MMX(PUNPCKHBW,  punpckhbw)
921BINARY_INT_MMX(PUNPCKHWD,  punpckhwd)
922BINARY_INT_MMX(PUNPCKHDQ,  punpckhdq)
923BINARY_INT_MMX(PACKSSDW,   packssdw)
924
925BINARY_INT_MMX(PAVGB,   pavgb)
926BINARY_INT_MMX(PAVGW,   pavgw)
927BINARY_INT_MMX(PMADDWD, pmaddwd)
928BINARY_INT_MMX(PMULHUW, pmulhuw)
929BINARY_INT_MMX(PMULHW,  pmulhw)
930BINARY_INT_MMX(PMULUDQ, pmuludq)
931BINARY_INT_MMX(PSADBW,  psadbw)
932
933BINARY_INT_MMX(PSLLW_r, psllw)
934BINARY_INT_MMX(PSLLD_r, pslld)
935BINARY_INT_MMX(PSLLQ_r, psllq)
936BINARY_INT_MMX(PSRLW_r, psrlw)
937BINARY_INT_MMX(PSRLD_r, psrld)
938BINARY_INT_MMX(PSRLQ_r, psrlq)
939BINARY_INT_MMX(PSRAW_r, psraw)
940BINARY_INT_MMX(PSRAD_r, psrad)
941
942BINARY_INT_MMX(PHADDW,    phaddw)
943BINARY_INT_MMX(PHADDSW,   phaddsw)
944BINARY_INT_MMX(PHADDD,    phaddd)
945BINARY_INT_MMX(PHSUBW,    phsubw)
946BINARY_INT_MMX(PHSUBSW,   phsubsw)
947BINARY_INT_MMX(PHSUBD,    phsubd)
948BINARY_INT_MMX(PMADDUBSW, pmaddubsw)
949BINARY_INT_MMX(PSHUFB,    pshufb)
950BINARY_INT_MMX(PSIGNB,    psignb)
951BINARY_INT_MMX(PSIGNW,    psignw)
952BINARY_INT_MMX(PSIGND,    psignd)
953BINARY_INT_MMX(PMULHRSW,  pmulhrsw)
954
955/* Instructions with no MMX equivalent.  */
956#define BINARY_INT_SSE(uname, lname)                                               \
957static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
958{                                                                                  \
959    gen_binary_int_sse(s, decode,                                                  \
960                          NULL,                                                    \
961                          gen_helper_##lname##_xmm,                                \
962                          gen_helper_##lname##_ymm);                               \
963}
964
965/* Instructions with no MMX equivalent.  */
966BINARY_INT_SSE(PUNPCKLQDQ, punpcklqdq)
967BINARY_INT_SSE(PUNPCKHQDQ, punpckhqdq)
968BINARY_INT_SSE(VPACKUSDW,  packusdw)
969BINARY_INT_SSE(VPERMILPS,  vpermilps)
970BINARY_INT_SSE(VPERMILPD,  vpermilpd)
971BINARY_INT_SSE(VMASKMOVPS, vpmaskmovd)
972BINARY_INT_SSE(VMASKMOVPD, vpmaskmovq)
973
974BINARY_INT_SSE(PMULDQ,    pmuldq)
975
976BINARY_INT_SSE(VAESDEC, aesdec)
977BINARY_INT_SSE(VAESDECLAST, aesdeclast)
978BINARY_INT_SSE(VAESENC, aesenc)
979BINARY_INT_SSE(VAESENCLAST, aesenclast)
980
981#define UNARY_CMP_SSE(uname, lname)                                                \
982static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
983{                                                                                  \
984    if (!s->vex_l) {                                                               \
985        gen_helper_##lname##_xmm(tcg_env, OP_PTR1, OP_PTR2);                       \
986    } else {                                                                       \
987        gen_helper_##lname##_ymm(tcg_env, OP_PTR1, OP_PTR2);                       \
988    }                                                                              \
989    assume_cc_op(s, CC_OP_EFLAGS);                                                  \
990}
991UNARY_CMP_SSE(VPTEST,     ptest)
992UNARY_CMP_SSE(VTESTPS,    vtestps)
993UNARY_CMP_SSE(VTESTPD,    vtestpd)
994
995static inline void gen_unary_int_sse(DisasContext *s, X86DecodedInsn *decode,
996                                     SSEFunc_0_epp xmm, SSEFunc_0_epp ymm)
997{
998    if (!s->vex_l) {
999        xmm(tcg_env, OP_PTR0, OP_PTR2);
1000    } else {
1001        ymm(tcg_env, OP_PTR0, OP_PTR2);
1002    }
1003}
1004
1005#define UNARY_INT_SSE(uname, lname)                                                \
1006static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1007{                                                                                  \
1008    gen_unary_int_sse(s, decode,                                                   \
1009                      gen_helper_##lname##_xmm,                                    \
1010                      gen_helper_##lname##_ymm);                                   \
1011}
1012
1013UNARY_INT_SSE(VPMOVSXBW,    pmovsxbw)
1014UNARY_INT_SSE(VPMOVSXBD,    pmovsxbd)
1015UNARY_INT_SSE(VPMOVSXBQ,    pmovsxbq)
1016UNARY_INT_SSE(VPMOVSXWD,    pmovsxwd)
1017UNARY_INT_SSE(VPMOVSXWQ,    pmovsxwq)
1018UNARY_INT_SSE(VPMOVSXDQ,    pmovsxdq)
1019
1020UNARY_INT_SSE(VPMOVZXBW,    pmovzxbw)
1021UNARY_INT_SSE(VPMOVZXBD,    pmovzxbd)
1022UNARY_INT_SSE(VPMOVZXBQ,    pmovzxbq)
1023UNARY_INT_SSE(VPMOVZXWD,    pmovzxwd)
1024UNARY_INT_SSE(VPMOVZXWQ,    pmovzxwq)
1025UNARY_INT_SSE(VPMOVZXDQ,    pmovzxdq)
1026
1027UNARY_INT_SSE(VMOVSLDUP,    pmovsldup)
1028UNARY_INT_SSE(VMOVSHDUP,    pmovshdup)
1029UNARY_INT_SSE(VMOVDDUP,     pmovdldup)
1030
1031UNARY_INT_SSE(VCVTDQ2PD, cvtdq2pd)
1032UNARY_INT_SSE(VCVTPD2DQ, cvtpd2dq)
1033UNARY_INT_SSE(VCVTTPD2DQ, cvttpd2dq)
1034UNARY_INT_SSE(VCVTDQ2PS, cvtdq2ps)
1035UNARY_INT_SSE(VCVTPS2DQ, cvtps2dq)
1036UNARY_INT_SSE(VCVTTPS2DQ, cvttps2dq)
1037UNARY_INT_SSE(VCVTPH2PS, cvtph2ps)
1038
1039
1040static inline void gen_unary_imm_sse(DisasContext *s, X86DecodedInsn *decode,
1041                                     SSEFunc_0_ppi xmm, SSEFunc_0_ppi ymm)
1042{
1043    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
1044    if (!s->vex_l) {
1045        xmm(OP_PTR0, OP_PTR1, imm);
1046    } else {
1047        ymm(OP_PTR0, OP_PTR1, imm);
1048    }
1049}
1050
1051#define UNARY_IMM_SSE(uname, lname)                                                \
1052static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1053{                                                                                  \
1054    gen_unary_imm_sse(s, decode,                                                   \
1055                      gen_helper_##lname##_xmm,                                    \
1056                      gen_helper_##lname##_ymm);                                   \
1057}
1058
1059UNARY_IMM_SSE(PSHUFD,     pshufd)
1060UNARY_IMM_SSE(PSHUFHW,    pshufhw)
1061UNARY_IMM_SSE(PSHUFLW,    pshuflw)
1062#define gen_helper_vpermq_xmm NULL
1063UNARY_IMM_SSE(VPERMQ,      vpermq)
1064UNARY_IMM_SSE(VPERMILPS_i, vpermilps_imm)
1065UNARY_IMM_SSE(VPERMILPD_i, vpermilpd_imm)
1066
1067static inline void gen_unary_imm_fp_sse(DisasContext *s, X86DecodedInsn *decode,
1068                                        SSEFunc_0_eppi xmm, SSEFunc_0_eppi ymm)
1069{
1070    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
1071    if (!s->vex_l) {
1072        xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
1073    } else {
1074        ymm(tcg_env, OP_PTR0, OP_PTR1, imm);
1075    }
1076}
1077
1078#define UNARY_IMM_FP_SSE(uname, lname)                                             \
1079static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1080{                                                                                  \
1081    gen_unary_imm_fp_sse(s, decode,                                                \
1082                      gen_helper_##lname##_xmm,                                    \
1083                      gen_helper_##lname##_ymm);                                   \
1084}
1085
1086UNARY_IMM_FP_SSE(VROUNDPS,    roundps)
1087UNARY_IMM_FP_SSE(VROUNDPD,    roundpd)
1088
1089static inline void gen_vexw_avx(DisasContext *s, X86DecodedInsn *decode,
1090                                SSEFunc_0_eppp d_xmm, SSEFunc_0_eppp q_xmm,
1091                                SSEFunc_0_eppp d_ymm, SSEFunc_0_eppp q_ymm)
1092{
1093    SSEFunc_0_eppp d = s->vex_l ? d_ymm : d_xmm;
1094    SSEFunc_0_eppp q = s->vex_l ? q_ymm : q_xmm;
1095    SSEFunc_0_eppp fn = s->vex_w ? q : d;
1096    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
1097}
1098
1099/* VEX.W affects whether to operate on 32- or 64-bit elements.  */
1100#define VEXW_AVX(uname, lname)                                                     \
1101static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1102{                                                                                  \
1103    gen_vexw_avx(s, decode,                                                        \
1104                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1105                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1106}
1107VEXW_AVX(VPSLLV,    vpsllv)
1108VEXW_AVX(VPSRLV,    vpsrlv)
1109VEXW_AVX(VPSRAV,    vpsrav)
1110VEXW_AVX(VPMASKMOV, vpmaskmov)
1111
1112/* Same as above, but with extra arguments to the helper.  */
1113static inline void gen_vsib_avx(DisasContext *s, X86DecodedInsn *decode,
1114                                SSEFunc_0_epppti d_xmm, SSEFunc_0_epppti q_xmm,
1115                                SSEFunc_0_epppti d_ymm, SSEFunc_0_epppti q_ymm)
1116{
1117    SSEFunc_0_epppti d = s->vex_l ? d_ymm : d_xmm;
1118    SSEFunc_0_epppti q = s->vex_l ? q_ymm : q_xmm;
1119    SSEFunc_0_epppti fn = s->vex_w ? q : d;
1120    TCGv_i32 scale = tcg_constant_i32(decode->mem.scale);
1121    TCGv_ptr index = tcg_temp_new_ptr();
1122
1123    /* Pass third input as (index, base, scale) */
1124    tcg_gen_addi_ptr(index, tcg_env, ZMM_OFFSET(decode->mem.index));
1125    fn(tcg_env, OP_PTR0, OP_PTR1, index, s->A0, scale);
1126
1127    /*
1128     * There are two output operands, so zero OP1's high 128 bits
1129     * in the VEX.128 case.
1130     */
1131    if (!s->vex_l) {
1132        int ymmh_ofs = vector_elem_offset(&decode->op[1], MO_128, 1);
1133        tcg_gen_gvec_dup_imm(MO_64, ymmh_ofs, 16, 16, 0);
1134    }
1135}
1136#define VSIB_AVX(uname, lname)                                                     \
1137static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1138{                                                                                  \
1139    gen_vsib_avx(s, decode,                                                        \
1140                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1141                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1142}
1143VSIB_AVX(VPGATHERD, vpgatherd)
1144VSIB_AVX(VPGATHERQ, vpgatherq)
1145
1146static void gen_AAA(DisasContext *s, X86DecodedInsn *decode)
1147{
1148    gen_update_cc_op(s);
1149    gen_helper_aaa(tcg_env);
1150    assume_cc_op(s, CC_OP_EFLAGS);
1151}
1152
1153static void gen_AAD(DisasContext *s, X86DecodedInsn *decode)
1154{
1155    gen_helper_aad(s->T0, s->T0, s->T1);
1156    prepare_update1_cc(decode, s, CC_OP_LOGICB);
1157}
1158
1159static void gen_AAM(DisasContext *s, X86DecodedInsn *decode)
1160{
1161    if (decode->immediate == 0) {
1162        gen_exception(s, EXCP00_DIVZ);
1163    } else {
1164        gen_helper_aam(s->T0, s->T0, s->T1);
1165        prepare_update1_cc(decode, s, CC_OP_LOGICB);
1166    }
1167}
1168
1169static void gen_AAS(DisasContext *s, X86DecodedInsn *decode)
1170{
1171    gen_update_cc_op(s);
1172    gen_helper_aas(tcg_env);
1173    assume_cc_op(s, CC_OP_EFLAGS);
1174}
1175
1176static void gen_ADC(DisasContext *s, X86DecodedInsn *decode)
1177{
1178    MemOp ot = decode->op[1].ot;
1179    TCGv c_in = tcg_temp_new();
1180
1181    gen_compute_eflags_c(s, c_in);
1182    if (s->prefix & PREFIX_LOCK) {
1183        tcg_gen_add_tl(s->T0, c_in, s->T1);
1184        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
1185                                    s->mem_index, ot | MO_LE);
1186    } else {
1187        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1188        tcg_gen_add_tl(s->T0, s->T0, c_in);
1189    }
1190    prepare_update3_cc(decode, s, CC_OP_ADCB + ot, c_in);
1191}
1192
1193static void gen_ADCOX(DisasContext *s, X86DecodedInsn *decode, int cc_op)
1194{
1195    MemOp ot = decode->op[0].ot;
1196    TCGv carry_in = NULL;
1197    TCGv *carry_out = (cc_op == CC_OP_ADCX ? &decode->cc_dst : &decode->cc_src2);
1198    TCGv zero;
1199
1200    decode->cc_op = cc_op;
1201    *carry_out = tcg_temp_new();
1202    if (CC_OP_HAS_EFLAGS(s->cc_op)) {
1203        decode->cc_src = cpu_cc_src;
1204
1205        /* Re-use the carry-out from a previous round?  */
1206        if (s->cc_op == cc_op || s->cc_op == CC_OP_ADCOX) {
1207            carry_in = (cc_op == CC_OP_ADCX ? cpu_cc_dst : cpu_cc_src2);
1208        }
1209
1210        /* Preserve the opposite carry from previous rounds?  */
1211        if (s->cc_op != cc_op && s->cc_op != CC_OP_EFLAGS) {
1212            decode->cc_op = CC_OP_ADCOX;
1213            if (carry_out == &decode->cc_dst) {
1214                decode->cc_src2 = cpu_cc_src2;
1215            } else {
1216                decode->cc_dst = cpu_cc_dst;
1217            }
1218        }
1219    } else {
1220        decode->cc_src = tcg_temp_new();
1221        gen_mov_eflags(s, decode->cc_src);
1222    }
1223
1224    if (!carry_in) {
1225        /* Get carry_in out of EFLAGS.  */
1226        carry_in = tcg_temp_new();
1227        tcg_gen_extract_tl(carry_in, decode->cc_src,
1228            ctz32(cc_op == CC_OP_ADCX ? CC_C : CC_O), 1);
1229    }
1230
1231    switch (ot) {
1232#ifdef TARGET_X86_64
1233    case MO_32:
1234        /* If TL is 64-bit just do everything in 64-bit arithmetic.  */
1235        tcg_gen_ext32u_tl(s->T0, s->T0);
1236        tcg_gen_ext32u_tl(s->T1, s->T1);
1237        tcg_gen_add_i64(s->T0, s->T0, s->T1);
1238        tcg_gen_add_i64(s->T0, s->T0, carry_in);
1239        tcg_gen_shri_i64(*carry_out, s->T0, 32);
1240        break;
1241#endif
1242    default:
1243        zero = tcg_constant_tl(0);
1244        tcg_gen_add2_tl(s->T0, *carry_out, s->T0, zero, carry_in, zero);
1245        tcg_gen_add2_tl(s->T0, *carry_out, s->T0, *carry_out, s->T1, zero);
1246        break;
1247    }
1248}
1249
1250static void gen_ADCX(DisasContext *s, X86DecodedInsn *decode)
1251{
1252    gen_ADCOX(s, decode, CC_OP_ADCX);
1253}
1254
1255static void gen_ADD(DisasContext *s, X86DecodedInsn *decode)
1256{
1257    MemOp ot = decode->op[1].ot;
1258
1259    if (s->prefix & PREFIX_LOCK) {
1260        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1261                                    s->mem_index, ot | MO_LE);
1262    } else {
1263        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1264    }
1265    prepare_update2_cc(decode, s, CC_OP_ADDB + ot);
1266}
1267
1268static void gen_ADOX(DisasContext *s, X86DecodedInsn *decode)
1269{
1270    gen_ADCOX(s, decode, CC_OP_ADOX);
1271}
1272
1273static void gen_AND(DisasContext *s, X86DecodedInsn *decode)
1274{
1275    MemOp ot = decode->op[1].ot;
1276
1277    if (s->prefix & PREFIX_LOCK) {
1278        tcg_gen_atomic_and_fetch_tl(s->T0, s->A0, s->T1,
1279                                    s->mem_index, ot | MO_LE);
1280    } else {
1281        tcg_gen_and_tl(s->T0, s->T0, s->T1);
1282    }
1283    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1284}
1285
1286static void gen_ANDN(DisasContext *s, X86DecodedInsn *decode)
1287{
1288    MemOp ot = decode->op[0].ot;
1289
1290    tcg_gen_andc_tl(s->T0, s->T1, s->T0);
1291    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1292}
1293
1294static void gen_ARPL(DisasContext *s, X86DecodedInsn *decode)
1295{
1296    TCGv zf = tcg_temp_new();
1297    TCGv flags = tcg_temp_new();
1298
1299    gen_mov_eflags(s, flags);
1300
1301    /* Compute adjusted DST in T1, merging in SRC[RPL].  */
1302    tcg_gen_deposit_tl(s->T1, s->T0, s->T1, 0, 2);
1303
1304    /* Z flag set if DST[RPL] < SRC[RPL] */
1305    tcg_gen_setcond_tl(TCG_COND_LTU, zf, s->T0, s->T1);
1306    tcg_gen_deposit_tl(flags, flags, zf, ctz32(CC_Z), 1);
1307
1308    /* Place maximum RPL in DST */
1309    tcg_gen_umax_tl(s->T0, s->T0, s->T1);
1310
1311    decode->cc_src = flags;
1312    decode->cc_op = CC_OP_EFLAGS;
1313}
1314
1315static void gen_BEXTR(DisasContext *s, X86DecodedInsn *decode)
1316{
1317    MemOp ot = decode->op[0].ot;
1318    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1319    TCGv zero = tcg_constant_tl(0);
1320    TCGv mone = tcg_constant_tl(-1);
1321
1322    /*
1323     * Extract START, and shift the operand.
1324     * Shifts larger than operand size get zeros.
1325     */
1326    tcg_gen_ext8u_tl(s->A0, s->T1);
1327    tcg_gen_shr_tl(s->T0, s->T0, s->A0);
1328
1329    tcg_gen_movcond_tl(TCG_COND_LEU, s->T0, s->A0, bound, s->T0, zero);
1330
1331    /*
1332     * Extract the LEN into an inverse mask.  Lengths larger than
1333     * operand size get all zeros, length 0 gets all ones.
1334     */
1335    tcg_gen_extract_tl(s->A0, s->T1, 8, 8);
1336    tcg_gen_shl_tl(s->T1, mone, s->A0);
1337    tcg_gen_movcond_tl(TCG_COND_LEU, s->T1, s->A0, bound, s->T1, zero);
1338    tcg_gen_andc_tl(s->T0, s->T0, s->T1);
1339
1340    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1341}
1342
1343static void gen_BLSI(DisasContext *s, X86DecodedInsn *decode)
1344{
1345    MemOp ot = decode->op[0].ot;
1346
1347    /* input in T1, which is ready for prepare_update2_cc  */
1348    tcg_gen_neg_tl(s->T0, s->T1);
1349    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1350    prepare_update2_cc(decode, s, CC_OP_BLSIB + ot);
1351}
1352
1353static void gen_BLSMSK(DisasContext *s, X86DecodedInsn *decode)
1354{
1355    MemOp ot = decode->op[0].ot;
1356
1357    /* input in T1, which is ready for prepare_update2_cc  */
1358    tcg_gen_subi_tl(s->T0, s->T1, 1);
1359    tcg_gen_xor_tl(s->T0, s->T0, s->T1);
1360    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1361}
1362
1363static void gen_BLSR(DisasContext *s, X86DecodedInsn *decode)
1364{
1365    MemOp ot = decode->op[0].ot;
1366
1367    /* input in T1, which is ready for prepare_update2_cc  */
1368    tcg_gen_subi_tl(s->T0, s->T1, 1);
1369    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1370    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1371}
1372
1373static void gen_BOUND(DisasContext *s, X86DecodedInsn *decode)
1374{
1375    TCGv_i32 op = tcg_temp_new_i32();
1376    tcg_gen_trunc_tl_i32(op, s->T0);
1377    if (decode->op[1].ot == MO_16) {
1378        gen_helper_boundw(tcg_env, s->A0, op);
1379    } else {
1380        gen_helper_boundl(tcg_env, s->A0, op);
1381    }
1382}
1383
1384/* Non-standard convention - on entry T0 is zero-extended input, T1 is the output.  */
1385static void gen_BSF(DisasContext *s, X86DecodedInsn *decode)
1386{
1387    MemOp ot = decode->op[0].ot;
1388
1389    /* Only the Z bit is defined and it is related to the input.  */
1390    decode->cc_dst = tcg_temp_new();
1391    decode->cc_op = CC_OP_LOGICB + ot;
1392    tcg_gen_mov_tl(decode->cc_dst, s->T0);
1393
1394    /*
1395     * The manual says that the output is undefined when the
1396     * input is zero, but real hardware leaves it unchanged, and
1397     * real programs appear to depend on that.  Accomplish this
1398     * by passing the output as the value to return upon zero.
1399     */
1400    tcg_gen_ctz_tl(s->T0, s->T0, s->T1);
1401}
1402
1403/* Non-standard convention - on entry T0 is zero-extended input, T1 is the output.  */
1404static void gen_BSR(DisasContext *s, X86DecodedInsn *decode)
1405{
1406    MemOp ot = decode->op[0].ot;
1407
1408    /* Only the Z bit is defined and it is related to the input.  */
1409    decode->cc_dst = tcg_temp_new();
1410    decode->cc_op = CC_OP_LOGICB + ot;
1411    tcg_gen_mov_tl(decode->cc_dst, s->T0);
1412
1413    /*
1414     * The manual says that the output is undefined when the
1415     * input is zero, but real hardware leaves it unchanged, and
1416     * real programs appear to depend on that.  Accomplish this
1417     * by passing the output as the value to return upon zero.
1418     * Plus, return the bit index of the first 1 bit.
1419     */
1420    tcg_gen_xori_tl(s->T1, s->T1, TARGET_LONG_BITS - 1);
1421    tcg_gen_clz_tl(s->T0, s->T0, s->T1);
1422    tcg_gen_xori_tl(s->T0, s->T0, TARGET_LONG_BITS - 1);
1423}
1424
1425static void gen_BSWAP(DisasContext *s, X86DecodedInsn *decode)
1426{
1427#ifdef TARGET_X86_64
1428    if (s->dflag == MO_64) {
1429        tcg_gen_bswap64_i64(s->T0, s->T0);
1430        return;
1431    }
1432#endif
1433    tcg_gen_bswap32_tl(s->T0, s->T0, TCG_BSWAP_OZ);
1434}
1435
1436static TCGv gen_bt_mask(DisasContext *s, X86DecodedInsn *decode)
1437{
1438    MemOp ot = decode->op[1].ot;
1439    TCGv mask = tcg_temp_new();
1440
1441    tcg_gen_andi_tl(s->T1, s->T1, (8 << ot) - 1);
1442    tcg_gen_shl_tl(mask, tcg_constant_tl(1), s->T1);
1443    return mask;
1444}
1445
1446/* Expects truncated bit index in s->T1, 1 << s->T1 in MASK.  */
1447static void gen_bt_flags(DisasContext *s, X86DecodedInsn *decode, TCGv src, TCGv mask)
1448{
1449    TCGv cf;
1450
1451    /*
1452     * C is the result of the test, Z is unchanged, and the others
1453     * are all undefined.
1454     */
1455    switch (s->cc_op) {
1456    case CC_OP_DYNAMIC:
1457    case CC_OP_CLR:
1458    case CC_OP_EFLAGS:
1459    case CC_OP_ADCX:
1460    case CC_OP_ADOX:
1461    case CC_OP_ADCOX:
1462        /* Generate EFLAGS and replace the C bit.  */
1463        cf = tcg_temp_new();
1464        tcg_gen_setcond_tl(TCG_COND_TSTNE, cf, src, mask);
1465        prepare_update_cf(decode, s, cf);
1466        break;
1467    default:
1468        /*
1469         * Z was going to be computed from the non-zero status of CC_DST.
1470         * We can get that same Z value (and the new C value) by leaving
1471         * CC_DST alone, setting CC_SRC, and using a CC_OP_SAR of the
1472         * same width.
1473         */
1474        decode->cc_src = tcg_temp_new();
1475        decode->cc_dst = cpu_cc_dst;
1476        decode->cc_op = ((s->cc_op - CC_OP_MULB) & 3) + CC_OP_SARB;
1477        tcg_gen_shr_tl(decode->cc_src, src, s->T1);
1478        break;
1479    }
1480}
1481
1482static void gen_BT(DisasContext *s, X86DecodedInsn *decode)
1483{
1484    TCGv mask = gen_bt_mask(s, decode);
1485
1486    gen_bt_flags(s, decode, s->T0, mask);
1487}
1488
1489static void gen_BTC(DisasContext *s, X86DecodedInsn *decode)
1490{
1491    MemOp ot = decode->op[0].ot;
1492    TCGv old = tcg_temp_new();
1493    TCGv mask = gen_bt_mask(s, decode);
1494
1495    if (s->prefix & PREFIX_LOCK) {
1496        tcg_gen_atomic_fetch_xor_tl(old, s->A0, mask, s->mem_index, ot | MO_LE);
1497    } else {
1498        tcg_gen_mov_tl(old, s->T0);
1499        tcg_gen_xor_tl(s->T0, s->T0, mask);
1500    }
1501
1502    gen_bt_flags(s, decode, old, mask);
1503}
1504
1505static void gen_BTR(DisasContext *s, X86DecodedInsn *decode)
1506{
1507    MemOp ot = decode->op[0].ot;
1508    TCGv old = tcg_temp_new();
1509    TCGv mask = gen_bt_mask(s, decode);
1510
1511    if (s->prefix & PREFIX_LOCK) {
1512        TCGv maskc = tcg_temp_new();
1513        tcg_gen_not_tl(maskc, mask);
1514        tcg_gen_atomic_fetch_and_tl(old, s->A0, maskc, s->mem_index, ot | MO_LE);
1515    } else {
1516        tcg_gen_mov_tl(old, s->T0);
1517        tcg_gen_andc_tl(s->T0, s->T0, mask);
1518    }
1519
1520    gen_bt_flags(s, decode, old, mask);
1521}
1522
1523static void gen_BTS(DisasContext *s, X86DecodedInsn *decode)
1524{
1525    MemOp ot = decode->op[0].ot;
1526    TCGv old = tcg_temp_new();
1527    TCGv mask = gen_bt_mask(s, decode);
1528
1529    if (s->prefix & PREFIX_LOCK) {
1530        tcg_gen_atomic_fetch_or_tl(old, s->A0, mask, s->mem_index, ot | MO_LE);
1531    } else {
1532        tcg_gen_mov_tl(old, s->T0);
1533        tcg_gen_or_tl(s->T0, s->T0, mask);
1534    }
1535
1536    gen_bt_flags(s, decode, old, mask);
1537}
1538
1539static void gen_BZHI(DisasContext *s, X86DecodedInsn *decode)
1540{
1541    MemOp ot = decode->op[0].ot;
1542    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1543    TCGv zero = tcg_constant_tl(0);
1544    TCGv mone = tcg_constant_tl(-1);
1545
1546    tcg_gen_ext8u_tl(s->T1, s->T1);
1547
1548    tcg_gen_shl_tl(s->A0, mone, s->T1);
1549    tcg_gen_movcond_tl(TCG_COND_LEU, s->A0, s->T1, bound, s->A0, zero);
1550    tcg_gen_andc_tl(s->T0, s->T0, s->A0);
1551    /*
1552     * Note that since we're using BMILG (in order to get O
1553     * cleared) we need to store the inverse into C.
1554     */
1555    tcg_gen_setcond_tl(TCG_COND_LEU, s->T1, s->T1, bound);
1556    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1557}
1558
1559static void gen_CALL(DisasContext *s, X86DecodedInsn *decode)
1560{
1561    gen_push_v(s, eip_next_tl(s));
1562    gen_JMP(s, decode);
1563}
1564
1565static void gen_CALL_m(DisasContext *s, X86DecodedInsn *decode)
1566{
1567    gen_push_v(s, eip_next_tl(s));
1568    gen_JMP_m(s, decode);
1569}
1570
1571static void gen_CALLF(DisasContext *s, X86DecodedInsn *decode)
1572{
1573    gen_far_call(s);
1574}
1575
1576static void gen_CALLF_m(DisasContext *s, X86DecodedInsn *decode)
1577{
1578    MemOp ot = decode->op[1].ot;
1579
1580    gen_op_ld_v(s, ot, s->T0, s->A0);
1581    gen_add_A0_im(s, 1 << ot);
1582    gen_op_ld_v(s, MO_16, s->T1, s->A0);
1583    gen_far_call(s);
1584}
1585
1586static void gen_CBW(DisasContext *s, X86DecodedInsn *decode)
1587{
1588    MemOp src_ot = decode->op[0].ot - 1;
1589
1590    tcg_gen_ext_tl(s->T0, s->T0, src_ot | MO_SIGN);
1591}
1592
1593static void gen_CLC(DisasContext *s, X86DecodedInsn *decode)
1594{
1595    gen_compute_eflags(s);
1596    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, ~CC_C);
1597}
1598
1599static void gen_CLD(DisasContext *s, X86DecodedInsn *decode)
1600{
1601    tcg_gen_st_i32(tcg_constant_i32(1), tcg_env, offsetof(CPUX86State, df));
1602}
1603
1604static void gen_CLI(DisasContext *s, X86DecodedInsn *decode)
1605{
1606    gen_reset_eflags(s, IF_MASK);
1607}
1608
1609static void gen_CLTS(DisasContext *s, X86DecodedInsn *decode)
1610{
1611    gen_helper_clts(tcg_env);
1612    /* abort block because static cpu state changed */
1613    s->base.is_jmp = DISAS_EOB_NEXT;
1614}
1615
1616static void gen_CMC(DisasContext *s, X86DecodedInsn *decode)
1617{
1618    gen_compute_eflags(s);
1619    tcg_gen_xori_tl(cpu_cc_src, cpu_cc_src, CC_C);
1620}
1621
1622static void gen_CMOVcc(DisasContext *s, X86DecodedInsn *decode)
1623{
1624    gen_cmovcc1(s, decode->b & 0xf, s->T0, s->T1);
1625}
1626
1627static void gen_CMPccXADD(DisasContext *s, X86DecodedInsn *decode)
1628{
1629    TCGLabel *label_top = gen_new_label();
1630    TCGLabel *label_bottom = gen_new_label();
1631    TCGv oldv = tcg_temp_new();
1632    TCGv newv = tcg_temp_new();
1633    TCGv cmpv = tcg_temp_new();
1634    TCGCond cond;
1635
1636    TCGv cmp_lhs, cmp_rhs;
1637    MemOp ot, ot_full;
1638
1639    int jcc_op = (decode->b >> 1) & 7;
1640    static const TCGCond cond_table[8] = {
1641        [JCC_O] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1642        [JCC_B] = TCG_COND_LTU,
1643        [JCC_Z] = TCG_COND_EQ,
1644        [JCC_BE] = TCG_COND_LEU,
1645        [JCC_S] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1646        [JCC_P] = TCG_COND_TSTEQ,  /* even parity - tests low bit of popcount */
1647        [JCC_L] = TCG_COND_LT,
1648        [JCC_LE] = TCG_COND_LE,
1649    };
1650
1651    cond = cond_table[jcc_op];
1652    if (decode->b & 1) {
1653        cond = tcg_invert_cond(cond);
1654    }
1655
1656    ot = decode->op[0].ot;
1657    ot_full = ot | MO_LE;
1658    if (jcc_op >= JCC_S) {
1659        /*
1660         * Sign-extend values before subtracting for S, P (zero/sign extension
1661         * does not matter there) L, LE and their inverses.
1662         */
1663        ot_full |= MO_SIGN;
1664    }
1665
1666    /*
1667     * cmpv will be moved to cc_src *after* cpu_regs[] is written back, so use
1668     * tcg_gen_ext_tl instead of gen_ext_tl.
1669     */
1670    tcg_gen_ext_tl(cmpv, cpu_regs[decode->op[1].n], ot_full);
1671
1672    /*
1673     * Cmpxchg loop starts here.
1674     * - s->T1: addition operand (from decoder)
1675     * - s->A0: dest address (from decoder)
1676     * - s->cc_srcT: memory operand (lhs for comparison)
1677     * - cmpv: rhs for comparison
1678     */
1679    gen_set_label(label_top);
1680    gen_op_ld_v(s, ot_full, s->cc_srcT, s->A0);
1681    tcg_gen_sub_tl(s->T0, s->cc_srcT, cmpv);
1682
1683    /* Compute the comparison result by hand, to avoid clobbering cc_*.  */
1684    switch (jcc_op) {
1685    case JCC_O:
1686        /* (src1 ^ src2) & (src1 ^ dst). newv is only used here for a moment */
1687        tcg_gen_xor_tl(newv, s->cc_srcT, s->T0);
1688        tcg_gen_xor_tl(s->tmp0, s->cc_srcT, cmpv);
1689        tcg_gen_and_tl(s->tmp0, s->tmp0, newv);
1690        tcg_gen_sextract_tl(s->tmp0, s->tmp0, 0, 8 << ot);
1691        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(0);
1692        break;
1693
1694    case JCC_P:
1695        tcg_gen_ext8u_tl(s->tmp0, s->T0);
1696        tcg_gen_ctpop_tl(s->tmp0, s->tmp0);
1697        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(1);
1698        break;
1699
1700    case JCC_S:
1701        tcg_gen_sextract_tl(s->tmp0, s->T0, 0, 8 << ot);
1702        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(0);
1703        break;
1704
1705    default:
1706        cmp_lhs = s->cc_srcT, cmp_rhs = cmpv;
1707        break;
1708    }
1709
1710    /* Compute new value: if condition does not hold, just store back s->cc_srcT */
1711    tcg_gen_add_tl(newv, s->cc_srcT, s->T1);
1712    tcg_gen_movcond_tl(cond, newv, cmp_lhs, cmp_rhs, newv, s->cc_srcT);
1713    tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, s->cc_srcT, newv, s->mem_index, ot_full);
1714
1715    /* Exit unconditionally if cmpxchg succeeded.  */
1716    tcg_gen_brcond_tl(TCG_COND_EQ, oldv, s->cc_srcT, label_bottom);
1717
1718    /* Try again if there was actually a store to make.  */
1719    tcg_gen_brcond_tl(cond, cmp_lhs, cmp_rhs, label_top);
1720    gen_set_label(label_bottom);
1721
1722    /* Store old value to registers only after a successful store.  */
1723    gen_writeback(s, decode, 1, s->cc_srcT);
1724
1725    decode->cc_dst = s->T0;
1726    decode->cc_src = cmpv;
1727    decode->cc_op = CC_OP_SUBB + ot;
1728}
1729
1730static void gen_CMPS(DisasContext *s, X86DecodedInsn *decode)
1731{
1732    MemOp ot = decode->op[2].ot;
1733    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
1734        gen_repz_nz(s, ot, gen_cmps);
1735    } else {
1736        gen_cmps(s, ot);
1737    }
1738}
1739
1740static void gen_CMPXCHG(DisasContext *s, X86DecodedInsn *decode)
1741{
1742    MemOp ot = decode->op[2].ot;
1743    TCGv cmpv = tcg_temp_new();
1744    TCGv oldv = tcg_temp_new();
1745    TCGv newv = tcg_temp_new();
1746    TCGv dest;
1747
1748    tcg_gen_ext_tl(cmpv, cpu_regs[R_EAX], ot);
1749    tcg_gen_ext_tl(newv, s->T1, ot);
1750    if (s->prefix & PREFIX_LOCK) {
1751        tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, cmpv, newv,
1752                                  s->mem_index, ot | MO_LE);
1753    } else {
1754        tcg_gen_ext_tl(oldv, s->T0, ot);
1755        if (decode->op[0].has_ea) {
1756            /*
1757             * Perform an unconditional store cycle like physical cpu;
1758             * must be before changing accumulator to ensure
1759             * idempotency if the store faults and the instruction
1760             * is restarted
1761             */
1762            tcg_gen_movcond_tl(TCG_COND_EQ, newv, oldv, cmpv, newv, oldv);
1763            gen_op_st_v(s, ot, newv, s->A0);
1764        } else {
1765            /*
1766             * Unlike the memory case, where "the destination operand receives
1767             * a write cycle without regard to the result of the comparison",
1768             * rm must not be touched altogether if the write fails, including
1769             * not zero-extending it on 64-bit processors.  So, precompute
1770             * the result of a successful writeback and perform the movcond
1771             * directly on cpu_regs.  In case rm is part of RAX, note that this
1772             * movcond and the one below are mutually exclusive is executed.
1773             */
1774            dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, newv, newv);
1775            tcg_gen_movcond_tl(TCG_COND_EQ, dest, oldv, cmpv, newv, dest);
1776        }
1777        decode->op[0].unit = X86_OP_SKIP;
1778    }
1779
1780    /* Write RAX only if the cmpxchg fails.  */
1781    dest = gen_op_deposit_reg_v(s, ot, R_EAX, s->T0, oldv);
1782    tcg_gen_movcond_tl(TCG_COND_NE, dest, oldv, cmpv, s->T0, dest);
1783
1784    tcg_gen_mov_tl(s->cc_srcT, cmpv);
1785    tcg_gen_sub_tl(cmpv, cmpv, oldv);
1786    decode->cc_dst = cmpv;
1787    decode->cc_src = oldv;
1788    decode->cc_op = CC_OP_SUBB + ot;
1789}
1790
1791static void gen_CMPXCHG16B(DisasContext *s, X86DecodedInsn *decode)
1792{
1793#ifdef TARGET_X86_64
1794    MemOp mop = MO_TE | MO_128 | MO_ALIGN;
1795    TCGv_i64 t0, t1;
1796    TCGv_i128 cmp, val;
1797
1798    cmp = tcg_temp_new_i128();
1799    val = tcg_temp_new_i128();
1800    tcg_gen_concat_i64_i128(cmp, cpu_regs[R_EAX], cpu_regs[R_EDX]);
1801    tcg_gen_concat_i64_i128(val, cpu_regs[R_EBX], cpu_regs[R_ECX]);
1802
1803    /* Only require atomic with LOCK; non-parallel handled in generator. */
1804    if (s->prefix & PREFIX_LOCK) {
1805        tcg_gen_atomic_cmpxchg_i128(val, s->A0, cmp, val, s->mem_index, mop);
1806    } else {
1807        tcg_gen_nonatomic_cmpxchg_i128(val, s->A0, cmp, val, s->mem_index, mop);
1808    }
1809
1810    tcg_gen_extr_i128_i64(s->T0, s->T1, val);
1811
1812    /* Determine success after the fact. */
1813    t0 = tcg_temp_new_i64();
1814    t1 = tcg_temp_new_i64();
1815    tcg_gen_xor_i64(t0, s->T0, cpu_regs[R_EAX]);
1816    tcg_gen_xor_i64(t1, s->T1, cpu_regs[R_EDX]);
1817    tcg_gen_or_i64(t0, t0, t1);
1818
1819    /* Update Z. */
1820    gen_compute_eflags(s);
1821    tcg_gen_setcondi_i64(TCG_COND_EQ, t0, t0, 0);
1822    tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, t0, ctz32(CC_Z), 1);
1823
1824    /*
1825     * Extract the result values for the register pair.  We may do this
1826     * unconditionally, because on success (Z=1), the old value matches
1827     * the previous value in RDX:RAX.
1828     */
1829    tcg_gen_mov_i64(cpu_regs[R_EAX], s->T0);
1830    tcg_gen_mov_i64(cpu_regs[R_EDX], s->T1);
1831#else
1832    abort();
1833#endif
1834}
1835
1836static void gen_CMPXCHG8B(DisasContext *s, X86DecodedInsn *decode)
1837{
1838    TCGv_i64 cmp, val, old;
1839    TCGv Z;
1840
1841    cmp = tcg_temp_new_i64();
1842    val = tcg_temp_new_i64();
1843    old = tcg_temp_new_i64();
1844
1845    /* Construct the comparison values from the register pair. */
1846    tcg_gen_concat_tl_i64(cmp, cpu_regs[R_EAX], cpu_regs[R_EDX]);
1847    tcg_gen_concat_tl_i64(val, cpu_regs[R_EBX], cpu_regs[R_ECX]);
1848
1849    /* Only require atomic with LOCK; non-parallel handled in generator. */
1850    if (s->prefix & PREFIX_LOCK) {
1851        tcg_gen_atomic_cmpxchg_i64(old, s->A0, cmp, val, s->mem_index, MO_TEUQ);
1852    } else {
1853        tcg_gen_nonatomic_cmpxchg_i64(old, s->A0, cmp, val,
1854                                      s->mem_index, MO_TEUQ);
1855    }
1856
1857    /* Set tmp0 to match the required value of Z. */
1858    tcg_gen_setcond_i64(TCG_COND_EQ, cmp, old, cmp);
1859    Z = tcg_temp_new();
1860    tcg_gen_trunc_i64_tl(Z, cmp);
1861
1862    /*
1863     * Extract the result values for the register pair.
1864     * For 32-bit, we may do this unconditionally, because on success (Z=1),
1865     * the old value matches the previous value in EDX:EAX.  For x86_64,
1866     * the store must be conditional, because we must leave the source
1867     * registers unchanged on success, and zero-extend the writeback
1868     * on failure (Z=0).
1869     */
1870    if (TARGET_LONG_BITS == 32) {
1871        tcg_gen_extr_i64_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], old);
1872    } else {
1873        TCGv zero = tcg_constant_tl(0);
1874
1875        tcg_gen_extr_i64_tl(s->T0, s->T1, old);
1876        tcg_gen_movcond_tl(TCG_COND_EQ, cpu_regs[R_EAX], Z, zero,
1877                           s->T0, cpu_regs[R_EAX]);
1878        tcg_gen_movcond_tl(TCG_COND_EQ, cpu_regs[R_EDX], Z, zero,
1879                           s->T1, cpu_regs[R_EDX]);
1880    }
1881
1882    /* Update Z. */
1883    gen_compute_eflags(s);
1884    tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, Z, ctz32(CC_Z), 1);
1885}
1886
1887static void gen_CPUID(DisasContext *s, X86DecodedInsn *decode)
1888{
1889    gen_update_cc_op(s);
1890    gen_update_eip_cur(s);
1891    gen_helper_cpuid(tcg_env);
1892}
1893
1894static void gen_CRC32(DisasContext *s, X86DecodedInsn *decode)
1895{
1896    MemOp ot = decode->op[2].ot;
1897
1898    tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
1899    gen_helper_crc32(s->T0, s->tmp2_i32, s->T1, tcg_constant_i32(8 << ot));
1900}
1901
1902static void gen_CVTPI2Px(DisasContext *s, X86DecodedInsn *decode)
1903{
1904    gen_helper_enter_mmx(tcg_env);
1905    if (s->prefix & PREFIX_DATA) {
1906        gen_helper_cvtpi2pd(tcg_env, OP_PTR0, OP_PTR2);
1907    } else {
1908        gen_helper_cvtpi2ps(tcg_env, OP_PTR0, OP_PTR2);
1909    }
1910}
1911
1912static void gen_CVTPx2PI(DisasContext *s, X86DecodedInsn *decode)
1913{
1914    gen_helper_enter_mmx(tcg_env);
1915    if (s->prefix & PREFIX_DATA) {
1916        gen_helper_cvtpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1917    } else {
1918        gen_helper_cvtps2pi(tcg_env, OP_PTR0, OP_PTR2);
1919    }
1920}
1921
1922static void gen_CVTTPx2PI(DisasContext *s, X86DecodedInsn *decode)
1923{
1924    gen_helper_enter_mmx(tcg_env);
1925    if (s->prefix & PREFIX_DATA) {
1926        gen_helper_cvttpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1927    } else {
1928        gen_helper_cvttps2pi(tcg_env, OP_PTR0, OP_PTR2);
1929    }
1930}
1931
1932static void gen_CWD(DisasContext *s, X86DecodedInsn *decode)
1933{
1934    int shift = 8 << decode->op[0].ot;
1935
1936    tcg_gen_sextract_tl(s->T0, s->T0, shift - 1, 1);
1937}
1938
1939static void gen_DAA(DisasContext *s, X86DecodedInsn *decode)
1940{
1941    gen_update_cc_op(s);
1942    gen_helper_daa(tcg_env);
1943    assume_cc_op(s, CC_OP_EFLAGS);
1944}
1945
1946static void gen_DAS(DisasContext *s, X86DecodedInsn *decode)
1947{
1948    gen_update_cc_op(s);
1949    gen_helper_das(tcg_env);
1950    assume_cc_op(s, CC_OP_EFLAGS);
1951}
1952
1953static void gen_DEC(DisasContext *s, X86DecodedInsn *decode)
1954{
1955    MemOp ot = decode->op[1].ot;
1956
1957    tcg_gen_movi_tl(s->T1, -1);
1958    if (s->prefix & PREFIX_LOCK) {
1959        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1960                                    s->mem_index, ot | MO_LE);
1961    } else {
1962        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1963    }
1964    prepare_update_cc_incdec(decode, s, CC_OP_DECB + ot);
1965}
1966
1967static void gen_DIV(DisasContext *s, X86DecodedInsn *decode)
1968{
1969    MemOp ot = decode->op[1].ot;
1970
1971    switch(ot) {
1972    case MO_8:
1973        gen_helper_divb_AL(tcg_env, s->T0);
1974        break;
1975    case MO_16:
1976        gen_helper_divw_AX(tcg_env, s->T0);
1977        break;
1978    default:
1979    case MO_32:
1980        gen_helper_divl_EAX(tcg_env, s->T0);
1981        break;
1982#ifdef TARGET_X86_64
1983    case MO_64:
1984        gen_helper_divq_EAX(tcg_env, s->T0);
1985        break;
1986#endif
1987    }
1988}
1989
1990static void gen_EMMS(DisasContext *s, X86DecodedInsn *decode)
1991{
1992    gen_helper_emms(tcg_env);
1993}
1994
1995static void gen_ENTER(DisasContext *s, X86DecodedInsn *decode)
1996{
1997   gen_enter(s, decode->op[1].imm, decode->op[2].imm);
1998}
1999
2000static void gen_EXTRQ_i(DisasContext *s, X86DecodedInsn *decode)
2001{
2002    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
2003    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
2004
2005    gen_helper_extrq_i(tcg_env, OP_PTR0, index, length);
2006}
2007
2008static void gen_EXTRQ_r(DisasContext *s, X86DecodedInsn *decode)
2009{
2010    gen_helper_extrq_r(tcg_env, OP_PTR0, OP_PTR2);
2011}
2012
2013static void gen_FXRSTOR(DisasContext *s, X86DecodedInsn *decode)
2014{
2015    if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
2016        gen_NM_exception(s);
2017    } else {
2018        gen_helper_fxrstor(tcg_env, s->A0);
2019    }
2020}
2021
2022static void gen_FXSAVE(DisasContext *s, X86DecodedInsn *decode)
2023{
2024    if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
2025        gen_NM_exception(s);
2026    } else {
2027        gen_helper_fxsave(tcg_env, s->A0);
2028    }
2029}
2030
2031static void gen_HLT(DisasContext *s, X86DecodedInsn *decode)
2032{
2033#ifdef CONFIG_SYSTEM_ONLY
2034    gen_update_cc_op(s);
2035    gen_update_eip_next(s);
2036    gen_helper_hlt(tcg_env);
2037    s->base.is_jmp = DISAS_NORETURN;
2038#endif
2039}
2040
2041static void gen_IDIV(DisasContext *s, X86DecodedInsn *decode)
2042{
2043    MemOp ot = decode->op[1].ot;
2044
2045    switch(ot) {
2046    case MO_8:
2047        gen_helper_idivb_AL(tcg_env, s->T0);
2048        break;
2049    case MO_16:
2050        gen_helper_idivw_AX(tcg_env, s->T0);
2051        break;
2052    default:
2053    case MO_32:
2054        gen_helper_idivl_EAX(tcg_env, s->T0);
2055        break;
2056#ifdef TARGET_X86_64
2057    case MO_64:
2058        gen_helper_idivq_EAX(tcg_env, s->T0);
2059        break;
2060#endif
2061    }
2062}
2063
2064static void gen_IMUL3(DisasContext *s, X86DecodedInsn *decode)
2065{
2066    MemOp ot = decode->op[0].ot;
2067    TCGv cc_src_rhs;
2068
2069    switch (ot) {
2070    case MO_16:
2071        /* s->T0 already sign-extended */
2072        tcg_gen_ext16s_tl(s->T1, s->T1);
2073        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2074        /* Compare the full result to the extension of the truncated result.  */
2075        tcg_gen_ext16s_tl(s->T1, s->T0);
2076        cc_src_rhs = s->T0;
2077        break;
2078
2079    case MO_32:
2080#ifdef TARGET_X86_64
2081        if (TCG_TARGET_REG_BITS == 64) {
2082            /*
2083             * This produces fewer TCG ops, and better code if flags are needed,
2084             * but it requires a 64-bit multiply even if they are not.  Use it
2085             * only if the target has 64-bits registers.
2086             *
2087             * s->T0 is already sign-extended.
2088             */
2089            tcg_gen_ext32s_tl(s->T1, s->T1);
2090            tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2091            /* Compare the full result to the extension of the truncated result.  */
2092            tcg_gen_ext32s_tl(s->T1, s->T0);
2093            cc_src_rhs = s->T0;
2094        } else {
2095            /* Variant that only needs a 32-bit widening multiply.  */
2096            TCGv_i32 hi = tcg_temp_new_i32();
2097            TCGv_i32 lo = tcg_temp_new_i32();
2098            tcg_gen_trunc_tl_i32(lo, s->T0);
2099            tcg_gen_trunc_tl_i32(hi, s->T1);
2100            tcg_gen_muls2_i32(lo, hi, lo, hi);
2101            tcg_gen_extu_i32_tl(s->T0, lo);
2102
2103            cc_src_rhs = tcg_temp_new();
2104            tcg_gen_extu_i32_tl(cc_src_rhs, hi);
2105            /* Compare the high part to the sign bit of the truncated result */
2106            tcg_gen_sari_i32(lo, lo, 31);
2107            tcg_gen_extu_i32_tl(s->T1, lo);
2108        }
2109        break;
2110
2111    case MO_64:
2112#endif
2113        cc_src_rhs = tcg_temp_new();
2114        tcg_gen_muls2_tl(s->T0, cc_src_rhs, s->T0, s->T1);
2115        /* Compare the high part to the sign bit of the truncated result */
2116        tcg_gen_sari_tl(s->T1, s->T0, TARGET_LONG_BITS - 1);
2117        break;
2118
2119    default:
2120        g_assert_not_reached();
2121    }
2122
2123    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
2124    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
2125}
2126
2127static void gen_IMUL(DisasContext *s, X86DecodedInsn *decode)
2128{
2129    MemOp ot = decode->op[1].ot;
2130    TCGv cc_src_rhs;
2131
2132    switch (ot) {
2133    case MO_8:
2134        /* s->T0 already sign-extended */
2135        tcg_gen_ext8s_tl(s->T1, s->T1);
2136        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2137        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2138        /* Compare the full result to the extension of the truncated result.  */
2139        tcg_gen_ext8s_tl(s->T1, s->T0);
2140        cc_src_rhs = s->T0;
2141        break;
2142
2143    case MO_16:
2144        /* s->T0 already sign-extended */
2145        tcg_gen_ext16s_tl(s->T1, s->T1);
2146        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2147        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2148        tcg_gen_shri_tl(s->T1, s->T0, 16);
2149        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
2150        /* Compare the full result to the extension of the truncated result.  */
2151        tcg_gen_ext16s_tl(s->T1, s->T0);
2152        cc_src_rhs = s->T0;
2153        break;
2154
2155    case MO_32:
2156#ifdef TARGET_X86_64
2157        /* s->T0 already sign-extended */
2158        tcg_gen_ext32s_tl(s->T1, s->T1);
2159        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2160        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
2161        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
2162        /* Compare the full result to the extension of the truncated result.  */
2163        tcg_gen_ext32s_tl(s->T1, s->T0);
2164        cc_src_rhs = s->T0;
2165        break;
2166
2167    case MO_64:
2168#endif
2169        tcg_gen_muls2_tl(s->T0, cpu_regs[R_EDX], s->T0, s->T1);
2170        tcg_gen_mov_tl(cpu_regs[R_EAX], s->T0);
2171
2172        /* Compare the high part to the sign bit of the truncated result */
2173        tcg_gen_negsetcondi_tl(TCG_COND_LT, s->T1, s->T0, 0);
2174        cc_src_rhs = cpu_regs[R_EDX];
2175        break;
2176
2177    default:
2178        g_assert_not_reached();
2179    }
2180
2181    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
2182    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
2183}
2184
2185static void gen_IN(DisasContext *s, X86DecodedInsn *decode)
2186{
2187    MemOp ot = decode->op[0].ot;
2188    TCGv_i32 port = tcg_temp_new_i32();
2189
2190    tcg_gen_trunc_tl_i32(port, s->T0);
2191    tcg_gen_ext16u_i32(port, port);
2192    if (!gen_check_io(s, ot, port, SVM_IOIO_TYPE_MASK)) {
2193        return;
2194    }
2195    translator_io_start(&s->base);
2196    gen_helper_in_func(ot, s->T0, port);
2197    gen_writeback(s, decode, 0, s->T0);
2198    gen_bpt_io(s, port, ot);
2199}
2200
2201static void gen_INC(DisasContext *s, X86DecodedInsn *decode)
2202{
2203    MemOp ot = decode->op[1].ot;
2204
2205    tcg_gen_movi_tl(s->T1, 1);
2206    if (s->prefix & PREFIX_LOCK) {
2207        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
2208                                    s->mem_index, ot | MO_LE);
2209    } else {
2210        tcg_gen_add_tl(s->T0, s->T0, s->T1);
2211    }
2212    prepare_update_cc_incdec(decode, s, CC_OP_INCB + ot);
2213}
2214
2215static void gen_INS(DisasContext *s, X86DecodedInsn *decode)
2216{
2217    MemOp ot = decode->op[1].ot;
2218    TCGv_i32 port = tcg_temp_new_i32();
2219
2220    tcg_gen_trunc_tl_i32(port, s->T1);
2221    tcg_gen_ext16u_i32(port, port);
2222    if (!gen_check_io(s, ot, port,
2223                      SVM_IOIO_TYPE_MASK | SVM_IOIO_STR_MASK)) {
2224        return;
2225    }
2226
2227    translator_io_start(&s->base);
2228    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2229        gen_repz(s, ot, gen_ins);
2230    } else {
2231        gen_ins(s, ot);
2232    }
2233}
2234
2235static void gen_INSERTQ_i(DisasContext *s, X86DecodedInsn *decode)
2236{
2237    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
2238    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
2239
2240    gen_helper_insertq_i(tcg_env, OP_PTR0, OP_PTR1, index, length);
2241}
2242
2243static void gen_INSERTQ_r(DisasContext *s, X86DecodedInsn *decode)
2244{
2245    gen_helper_insertq_r(tcg_env, OP_PTR0, OP_PTR2);
2246}
2247
2248static void gen_INT(DisasContext *s, X86DecodedInsn *decode)
2249{
2250    gen_interrupt(s, decode->immediate);
2251}
2252
2253static void gen_INT1(DisasContext *s, X86DecodedInsn *decode)
2254{
2255    gen_update_cc_op(s);
2256    gen_update_eip_next(s);
2257    gen_helper_icebp(tcg_env);
2258    s->base.is_jmp = DISAS_NORETURN;
2259}
2260
2261static void gen_INT3(DisasContext *s, X86DecodedInsn *decode)
2262{
2263    gen_interrupt(s, EXCP03_INT3);
2264}
2265
2266static void gen_INTO(DisasContext *s, X86DecodedInsn *decode)
2267{
2268    gen_update_cc_op(s);
2269    gen_update_eip_cur(s);
2270    gen_helper_into(tcg_env, cur_insn_len_i32(s));
2271}
2272
2273static void gen_IRET(DisasContext *s, X86DecodedInsn *decode)
2274{
2275    if (!PE(s) || VM86(s)) {
2276        gen_helper_iret_real(tcg_env, tcg_constant_i32(s->dflag - 1));
2277    } else {
2278        gen_helper_iret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
2279                                  eip_next_i32(s));
2280    }
2281    assume_cc_op(s, CC_OP_EFLAGS);
2282    s->base.is_jmp = DISAS_EOB_ONLY;
2283}
2284
2285static void gen_Jcc(DisasContext *s, X86DecodedInsn *decode)
2286{
2287    gen_bnd_jmp(s);
2288    gen_jcc(s, decode->b & 0xf, decode->immediate);
2289}
2290
2291static void gen_JCXZ(DisasContext *s, X86DecodedInsn *decode)
2292{
2293    TCGLabel *taken = gen_new_label();
2294
2295    gen_update_cc_op(s);
2296    gen_op_jz_ecx(s, taken);
2297    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2298}
2299
2300static void gen_JMP(DisasContext *s, X86DecodedInsn *decode)
2301{
2302    gen_update_cc_op(s);
2303    gen_jmp_rel(s, s->dflag, decode->immediate, 0);
2304}
2305
2306static void gen_JMP_m(DisasContext *s, X86DecodedInsn *decode)
2307{
2308    gen_op_jmp_v(s, s->T0);
2309    gen_bnd_jmp(s);
2310    s->base.is_jmp = DISAS_JUMP;
2311}
2312
2313static void gen_JMPF(DisasContext *s, X86DecodedInsn *decode)
2314{
2315    gen_far_jmp(s);
2316}
2317
2318static void gen_JMPF_m(DisasContext *s, X86DecodedInsn *decode)
2319{
2320    MemOp ot = decode->op[1].ot;
2321
2322    gen_op_ld_v(s, ot, s->T0, s->A0);
2323    gen_add_A0_im(s, 1 << ot);
2324    gen_op_ld_v(s, MO_16, s->T1, s->A0);
2325    gen_far_jmp(s);
2326}
2327
2328static void gen_LAHF(DisasContext *s, X86DecodedInsn *decode)
2329{
2330    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
2331        return gen_illegal_opcode(s);
2332    }
2333    gen_compute_eflags(s);
2334    /* Note: gen_compute_eflags() only gives the condition codes */
2335    tcg_gen_ori_tl(s->T0, cpu_cc_src, 0x02);
2336    tcg_gen_deposit_tl(cpu_regs[R_EAX], cpu_regs[R_EAX], s->T0, 8, 8);
2337}
2338
2339static void gen_LAR(DisasContext *s, X86DecodedInsn *decode)
2340{
2341    MemOp ot = decode->op[0].ot;
2342    TCGv result = tcg_temp_new();
2343    TCGv dest;
2344
2345    gen_compute_eflags(s);
2346    gen_update_cc_op(s);
2347    gen_helper_lar(result, tcg_env, s->T0);
2348
2349    /* Perform writeback here to skip it if ZF=0.  */
2350    decode->op[0].unit = X86_OP_SKIP;
2351    dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, result, result);
2352    tcg_gen_movcond_tl(TCG_COND_TSTNE, dest, cpu_cc_src, tcg_constant_tl(CC_Z),
2353                       result, dest);
2354}
2355
2356static void gen_LDMXCSR(DisasContext *s, X86DecodedInsn *decode)
2357{
2358    tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
2359    gen_helper_ldmxcsr(tcg_env, s->tmp2_i32);
2360}
2361
2362static void gen_lxx_seg(DisasContext *s, X86DecodedInsn *decode, int seg)
2363{
2364    MemOp ot = decode->op[0].ot;
2365
2366    /* Offset already in s->T0.  */
2367    gen_add_A0_im(s, 1 << ot);
2368    gen_op_ld_v(s, MO_16, s->T1, s->A0);
2369
2370    /* load the segment here to handle exceptions properly */
2371    gen_movl_seg(s, seg, s->T1);
2372}
2373
2374static void gen_LDS(DisasContext *s, X86DecodedInsn *decode)
2375{
2376    gen_lxx_seg(s, decode, R_DS);
2377}
2378
2379static void gen_LEA(DisasContext *s, X86DecodedInsn *decode)
2380{
2381    TCGv ea = gen_lea_modrm_1(s, decode->mem, false);
2382    gen_lea_v_seg_dest(s, s->aflag, s->T0, ea, -1, -1);
2383}
2384
2385static void gen_LEAVE(DisasContext *s, X86DecodedInsn *decode)
2386{
2387    gen_leave(s);
2388}
2389
2390static void gen_LES(DisasContext *s, X86DecodedInsn *decode)
2391{
2392    gen_lxx_seg(s, decode, R_ES);
2393}
2394
2395static void gen_LFENCE(DisasContext *s, X86DecodedInsn *decode)
2396{
2397    tcg_gen_mb(TCG_MO_LD_LD | TCG_BAR_SC);
2398}
2399
2400static void gen_LFS(DisasContext *s, X86DecodedInsn *decode)
2401{
2402    gen_lxx_seg(s, decode, R_FS);
2403}
2404
2405static void gen_LGS(DisasContext *s, X86DecodedInsn *decode)
2406{
2407    gen_lxx_seg(s, decode, R_GS);
2408}
2409
2410static void gen_LODS(DisasContext *s, X86DecodedInsn *decode)
2411{
2412    MemOp ot = decode->op[1].ot;
2413    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2414        gen_repz(s, ot, gen_lods);
2415    } else {
2416        gen_lods(s, ot);
2417    }
2418}
2419
2420static void gen_LOOP(DisasContext *s, X86DecodedInsn *decode)
2421{
2422    TCGLabel *taken = gen_new_label();
2423
2424    gen_update_cc_op(s);
2425    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2426    gen_op_jnz_ecx(s, taken);
2427    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2428}
2429
2430static void gen_LOOPE(DisasContext *s, X86DecodedInsn *decode)
2431{
2432    TCGLabel *taken = gen_new_label();
2433    TCGLabel *not_taken = gen_new_label();
2434
2435    gen_update_cc_op(s);
2436    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2437    gen_op_jz_ecx(s, not_taken);
2438    gen_jcc1(s, (JCC_Z << 1), taken); /* jz taken */
2439    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2440}
2441
2442static void gen_LOOPNE(DisasContext *s, X86DecodedInsn *decode)
2443{
2444    TCGLabel *taken = gen_new_label();
2445    TCGLabel *not_taken = gen_new_label();
2446
2447    gen_update_cc_op(s);
2448    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2449    gen_op_jz_ecx(s, not_taken);
2450    gen_jcc1(s, (JCC_Z << 1) | 1, taken); /* jnz taken */
2451    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2452}
2453
2454static void gen_LSL(DisasContext *s, X86DecodedInsn *decode)
2455{
2456    MemOp ot = decode->op[0].ot;
2457    TCGv result = tcg_temp_new();
2458    TCGv dest;
2459
2460    gen_compute_eflags(s);
2461    gen_update_cc_op(s);
2462    gen_helper_lsl(result, tcg_env, s->T0);
2463
2464    /* Perform writeback here to skip it if ZF=0.  */
2465    decode->op[0].unit = X86_OP_SKIP;
2466    dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, result, result);
2467    tcg_gen_movcond_tl(TCG_COND_TSTNE, dest, cpu_cc_src, tcg_constant_tl(CC_Z),
2468                       result, dest);
2469}
2470
2471static void gen_LSS(DisasContext *s, X86DecodedInsn *decode)
2472{
2473    gen_lxx_seg(s, decode, R_SS);
2474}
2475
2476static void gen_LZCNT(DisasContext *s, X86DecodedInsn *decode)
2477{
2478    MemOp ot = decode->op[0].ot;
2479
2480    /* C bit (cc_src) is defined related to the input.  */
2481    decode->cc_src = tcg_temp_new();
2482    decode->cc_dst = s->T0;
2483    decode->cc_op = CC_OP_BMILGB + ot;
2484    tcg_gen_mov_tl(decode->cc_src, s->T0);
2485
2486    /*
2487     * Reduce the target_ulong result by the number of zeros that
2488     * we expect to find at the top.
2489     */
2490    tcg_gen_clzi_tl(s->T0, s->T0, TARGET_LONG_BITS);
2491    tcg_gen_subi_tl(s->T0, s->T0, TARGET_LONG_BITS - (8 << ot));
2492}
2493
2494static void gen_MFENCE(DisasContext *s, X86DecodedInsn *decode)
2495{
2496    tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
2497}
2498
2499static void gen_MOV(DisasContext *s, X86DecodedInsn *decode)
2500{
2501    /* nothing to do! */
2502}
2503#define gen_NOP gen_MOV
2504
2505static void gen_MASKMOV(DisasContext *s, X86DecodedInsn *decode)
2506{
2507    gen_lea_v_seg(s, cpu_regs[R_EDI], R_DS, s->override);
2508
2509    if (s->prefix & PREFIX_DATA) {
2510        gen_helper_maskmov_xmm(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2511    } else {
2512        gen_helper_maskmov_mmx(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2513    }
2514}
2515
2516static void gen_MOVBE(DisasContext *s, X86DecodedInsn *decode)
2517{
2518    MemOp ot = decode->op[0].ot;
2519
2520    /* M operand type does not load/store */
2521    if (decode->e.op0 == X86_TYPE_M) {
2522        tcg_gen_qemu_st_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2523    } else {
2524        tcg_gen_qemu_ld_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2525    }
2526}
2527
2528static void gen_MOVD_from(DisasContext *s, X86DecodedInsn *decode)
2529{
2530    MemOp ot = decode->op[2].ot;
2531
2532    switch (ot) {
2533    case MO_32:
2534#ifdef TARGET_X86_64
2535        tcg_gen_ld32u_tl(s->T0, tcg_env, decode->op[2].offset);
2536        break;
2537    case MO_64:
2538#endif
2539        tcg_gen_ld_tl(s->T0, tcg_env, decode->op[2].offset);
2540        break;
2541    default:
2542        abort();
2543    }
2544}
2545
2546static void gen_MOVD_to(DisasContext *s, X86DecodedInsn *decode)
2547{
2548    MemOp ot = decode->op[2].ot;
2549    int vec_len = vector_len(s, decode);
2550    int lo_ofs = vector_elem_offset(&decode->op[0], ot, 0);
2551
2552    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2553
2554    switch (ot) {
2555    case MO_32:
2556#ifdef TARGET_X86_64
2557        tcg_gen_st32_tl(s->T1, tcg_env, lo_ofs);
2558        break;
2559    case MO_64:
2560#endif
2561        tcg_gen_st_tl(s->T1, tcg_env, lo_ofs);
2562        break;
2563    default:
2564        g_assert_not_reached();
2565    }
2566}
2567
2568static void gen_MOVDQ(DisasContext *s, X86DecodedInsn *decode)
2569{
2570    gen_store_sse(s, decode, decode->op[2].offset);
2571}
2572
2573static void gen_MOVMSK(DisasContext *s, X86DecodedInsn *decode)
2574{
2575    typeof(gen_helper_movmskps_ymm) *ps, *pd, *fn;
2576    ps = s->vex_l ? gen_helper_movmskps_ymm : gen_helper_movmskps_xmm;
2577    pd = s->vex_l ? gen_helper_movmskpd_ymm : gen_helper_movmskpd_xmm;
2578    fn = s->prefix & PREFIX_DATA ? pd : ps;
2579    fn(s->tmp2_i32, tcg_env, OP_PTR2);
2580    tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
2581}
2582
2583static void gen_MOVQ(DisasContext *s, X86DecodedInsn *decode)
2584{
2585    int vec_len = vector_len(s, decode);
2586    int lo_ofs = vector_elem_offset(&decode->op[0], MO_64, 0);
2587
2588    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
2589    if (decode->op[0].has_ea) {
2590        tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
2591    } else {
2592        /*
2593         * tcg_gen_gvec_dup_i64(MO_64, op0.offset, 8, vec_len, s->tmp1_64) would
2594         * seem to work, but it does not on big-endian platforms; the cleared parts
2595         * are always at higher addresses, but cross-endian emulation inverts the
2596         * byte order so that the cleared parts need to be at *lower* addresses.
2597         * Because oprsz is 8, we see this here even for SSE; but more in general,
2598         * it disqualifies using oprsz < maxsz to emulate VEX128.
2599         */
2600        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2601        tcg_gen_st_i64(s->tmp1_i64, tcg_env, lo_ofs);
2602    }
2603}
2604
2605static void gen_MOVq_dq(DisasContext *s, X86DecodedInsn *decode)
2606{
2607    gen_helper_enter_mmx(tcg_env);
2608    /* Otherwise the same as any other movq.  */
2609    return gen_MOVQ(s, decode);
2610}
2611
2612static void gen_MOVS(DisasContext *s, X86DecodedInsn *decode)
2613{
2614    MemOp ot = decode->op[2].ot;
2615    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2616        gen_repz(s, ot, gen_movs);
2617    } else {
2618        gen_movs(s, ot);
2619    }
2620}
2621
2622static void gen_MUL(DisasContext *s, X86DecodedInsn *decode)
2623{
2624    MemOp ot = decode->op[1].ot;
2625
2626    switch (ot) {
2627    case MO_8:
2628        /* s->T0 already zero-extended */
2629        tcg_gen_ext8u_tl(s->T1, s->T1);
2630        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2631        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2632        tcg_gen_andi_tl(s->T1, s->T0, 0xff00);
2633        decode->cc_dst = s->T0;
2634        decode->cc_src = s->T1;
2635        break;
2636
2637    case MO_16:
2638        /* s->T0 already zero-extended */
2639        tcg_gen_ext16u_tl(s->T1, s->T1);
2640        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2641        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2642        tcg_gen_shri_tl(s->T1, s->T0, 16);
2643        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
2644        decode->cc_dst = s->T0;
2645        decode->cc_src = s->T1;
2646        break;
2647
2648    case MO_32:
2649#ifdef TARGET_X86_64
2650        /* s->T0 already zero-extended */
2651        tcg_gen_ext32u_tl(s->T1, s->T1);
2652        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2653        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
2654        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
2655        decode->cc_dst = cpu_regs[R_EAX];
2656        decode->cc_src = cpu_regs[R_EDX];
2657        break;
2658
2659    case MO_64:
2660#endif
2661        tcg_gen_mulu2_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], s->T0, s->T1);
2662        decode->cc_dst = cpu_regs[R_EAX];
2663        decode->cc_src = cpu_regs[R_EDX];
2664        break;
2665
2666    default:
2667        g_assert_not_reached();
2668    }
2669
2670    decode->cc_op = CC_OP_MULB + ot;
2671}
2672
2673static void gen_MULX(DisasContext *s, X86DecodedInsn *decode)
2674{
2675    MemOp ot = decode->op[0].ot;
2676
2677    /* low part of result in VEX.vvvv, high in MODRM */
2678    switch (ot) {
2679    case MO_32:
2680#ifdef TARGET_X86_64
2681        tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
2682        tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
2683        tcg_gen_mulu2_i32(s->tmp2_i32, s->tmp3_i32,
2684                          s->tmp2_i32, s->tmp3_i32);
2685        tcg_gen_extu_i32_tl(cpu_regs[s->vex_v], s->tmp2_i32);
2686        tcg_gen_extu_i32_tl(s->T0, s->tmp3_i32);
2687        break;
2688
2689    case MO_64:
2690#endif
2691        tcg_gen_mulu2_tl(cpu_regs[s->vex_v], s->T0, s->T0, s->T1);
2692        break;
2693
2694    default:
2695        g_assert_not_reached();
2696    }
2697}
2698
2699static void gen_NEG(DisasContext *s, X86DecodedInsn *decode)
2700{
2701    MemOp ot = decode->op[0].ot;
2702    TCGv oldv = tcg_temp_new();
2703
2704    if (s->prefix & PREFIX_LOCK) {
2705        TCGv newv = tcg_temp_new();
2706        TCGv cmpv = tcg_temp_new();
2707        TCGLabel *label1 = gen_new_label();
2708
2709        gen_set_label(label1);
2710        gen_op_ld_v(s, ot, oldv, s->A0);
2711        tcg_gen_neg_tl(newv, oldv);
2712        tcg_gen_atomic_cmpxchg_tl(cmpv, s->A0, oldv, newv,
2713                                  s->mem_index, ot | MO_LE);
2714        tcg_gen_brcond_tl(TCG_COND_NE, oldv, cmpv, label1);
2715    } else {
2716        tcg_gen_mov_tl(oldv, s->T0);
2717    }
2718    tcg_gen_neg_tl(s->T0, oldv);
2719
2720    decode->cc_dst = s->T0;
2721    decode->cc_src = oldv;
2722    tcg_gen_movi_tl(s->cc_srcT, 0);
2723    decode->cc_op = CC_OP_SUBB + ot;
2724}
2725
2726static void gen_NOT(DisasContext *s, X86DecodedInsn *decode)
2727{
2728    MemOp ot = decode->op[0].ot;
2729
2730    if (s->prefix & PREFIX_LOCK) {
2731        tcg_gen_movi_tl(s->T0, ~0);
2732        tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T0,
2733                                    s->mem_index, ot | MO_LE);
2734    } else {
2735        tcg_gen_not_tl(s->T0, s->T0);
2736    }
2737}
2738
2739static void gen_OR(DisasContext *s, X86DecodedInsn *decode)
2740{
2741    MemOp ot = decode->op[1].ot;
2742
2743    if (s->prefix & PREFIX_LOCK) {
2744        tcg_gen_atomic_or_fetch_tl(s->T0, s->A0, s->T1,
2745                                   s->mem_index, ot | MO_LE);
2746    } else {
2747        tcg_gen_or_tl(s->T0, s->T0, s->T1);
2748    }
2749    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
2750}
2751
2752static void gen_OUT(DisasContext *s, X86DecodedInsn *decode)
2753{
2754    MemOp ot = decode->op[1].ot;
2755    TCGv_i32 port = tcg_temp_new_i32();
2756    TCGv_i32 value = tcg_temp_new_i32();
2757
2758    tcg_gen_trunc_tl_i32(port, s->T1);
2759    tcg_gen_ext16u_i32(port, port);
2760    if (!gen_check_io(s, ot, port, 0)) {
2761        return;
2762    }
2763    tcg_gen_trunc_tl_i32(value, s->T0);
2764    translator_io_start(&s->base);
2765    gen_helper_out_func(ot, port, value);
2766    gen_bpt_io(s, port, ot);
2767}
2768
2769static void gen_OUTS(DisasContext *s, X86DecodedInsn *decode)
2770{
2771    MemOp ot = decode->op[1].ot;
2772    TCGv_i32 port = tcg_temp_new_i32();
2773
2774    tcg_gen_trunc_tl_i32(port, s->T1);
2775    tcg_gen_ext16u_i32(port, port);
2776    if (!gen_check_io(s, ot, port, SVM_IOIO_STR_MASK)) {
2777        return;
2778    }
2779
2780    translator_io_start(&s->base);
2781    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2782        gen_repz(s, ot, gen_outs);
2783    } else {
2784        gen_outs(s, ot);
2785    }
2786}
2787
2788static void gen_PALIGNR(DisasContext *s, X86DecodedInsn *decode)
2789{
2790    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2791    if (!(s->prefix & PREFIX_DATA)) {
2792        gen_helper_palignr_mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2793    } else if (!s->vex_l) {
2794        gen_helper_palignr_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2795    } else {
2796        gen_helper_palignr_ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2797    }
2798}
2799
2800static void gen_PANDN(DisasContext *s, X86DecodedInsn *decode)
2801{
2802    int vec_len = vector_len(s, decode);
2803
2804    /* Careful, operand order is reversed!  */
2805    tcg_gen_gvec_andc(MO_64,
2806                      decode->op[0].offset, decode->op[2].offset,
2807                      decode->op[1].offset, vec_len, vec_len);
2808}
2809
2810static void gen_PAUSE(DisasContext *s, X86DecodedInsn *decode)
2811{
2812    gen_update_cc_op(s);
2813    gen_update_eip_next(s);
2814    gen_helper_pause(tcg_env);
2815    s->base.is_jmp = DISAS_NORETURN;
2816}
2817
2818static void gen_PCMPESTRI(DisasContext *s, X86DecodedInsn *decode)
2819{
2820    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2821    gen_helper_pcmpestri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2822    assume_cc_op(s, CC_OP_EFLAGS);
2823}
2824
2825static void gen_PCMPESTRM(DisasContext *s, X86DecodedInsn *decode)
2826{
2827    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2828    gen_helper_pcmpestrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2829    assume_cc_op(s, CC_OP_EFLAGS);
2830    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2831        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2832                             16, 16, 0);
2833    }
2834}
2835
2836static void gen_PCMPISTRI(DisasContext *s, X86DecodedInsn *decode)
2837{
2838    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2839    gen_helper_pcmpistri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2840    assume_cc_op(s, CC_OP_EFLAGS);
2841}
2842
2843static void gen_PCMPISTRM(DisasContext *s, X86DecodedInsn *decode)
2844{
2845    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2846    gen_helper_pcmpistrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2847    assume_cc_op(s, CC_OP_EFLAGS);
2848    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2849        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2850                             16, 16, 0);
2851    }
2852}
2853
2854static void gen_PDEP(DisasContext *s, X86DecodedInsn *decode)
2855{
2856    gen_helper_pdep(s->T0, s->T0, s->T1);
2857}
2858
2859static void gen_PEXT(DisasContext *s, X86DecodedInsn *decode)
2860{
2861    gen_helper_pext(s->T0, s->T0, s->T1);
2862}
2863
2864static inline void gen_pextr(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
2865{
2866    int vec_len = vector_len(s, decode);
2867    int mask = (vec_len >> ot) - 1;
2868    int val = decode->immediate & mask;
2869
2870    switch (ot) {
2871    case MO_8:
2872        tcg_gen_ld8u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2873        break;
2874    case MO_16:
2875        tcg_gen_ld16u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2876        break;
2877    case MO_32:
2878#ifdef TARGET_X86_64
2879        tcg_gen_ld32u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2880        break;
2881    case MO_64:
2882#endif
2883        tcg_gen_ld_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2884        break;
2885    default:
2886        abort();
2887    }
2888}
2889
2890static void gen_PEXTRB(DisasContext *s, X86DecodedInsn *decode)
2891{
2892    gen_pextr(s, decode, MO_8);
2893}
2894
2895static void gen_PEXTRW(DisasContext *s, X86DecodedInsn *decode)
2896{
2897    gen_pextr(s, decode, MO_16);
2898}
2899
2900static void gen_PEXTR(DisasContext *s, X86DecodedInsn *decode)
2901{
2902    MemOp ot = decode->op[0].ot;
2903    gen_pextr(s, decode, ot);
2904}
2905
2906static inline void gen_pinsr(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
2907{
2908    int vec_len = vector_len(s, decode);
2909    int mask = (vec_len >> ot) - 1;
2910    int val = decode->immediate & mask;
2911
2912    if (decode->op[1].offset != decode->op[0].offset) {
2913        assert(vec_len == 16);
2914        gen_store_sse(s, decode, decode->op[1].offset);
2915    }
2916
2917    switch (ot) {
2918    case MO_8:
2919        tcg_gen_st8_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2920        break;
2921    case MO_16:
2922        tcg_gen_st16_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2923        break;
2924    case MO_32:
2925#ifdef TARGET_X86_64
2926        tcg_gen_st32_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2927        break;
2928    case MO_64:
2929#endif
2930        tcg_gen_st_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2931        break;
2932    default:
2933        abort();
2934    }
2935}
2936
2937static void gen_PINSRB(DisasContext *s, X86DecodedInsn *decode)
2938{
2939    gen_pinsr(s, decode, MO_8);
2940}
2941
2942static void gen_PINSRW(DisasContext *s, X86DecodedInsn *decode)
2943{
2944    gen_pinsr(s, decode, MO_16);
2945}
2946
2947static void gen_PINSR(DisasContext *s, X86DecodedInsn *decode)
2948{
2949    gen_pinsr(s, decode, decode->op[2].ot);
2950}
2951
2952static void gen_pmovmskb_i64(TCGv_i64 d, TCGv_i64 s)
2953{
2954    TCGv_i64 t = tcg_temp_new_i64();
2955
2956    tcg_gen_andi_i64(d, s, 0x8080808080808080ull);
2957
2958    /*
2959     * After each shift+or pair:
2960     * 0:  a.......b.......c.......d.......e.......f.......g.......h.......
2961     * 7:  ab......bc......cd......de......ef......fg......gh......h.......
2962     * 14: abcd....bcde....cdef....defg....efgh....fgh.....gh......h.......
2963     * 28: abcdefghbcdefgh.cdefgh..defgh...efgh....fgh.....gh......h.......
2964     * The result is left in the high bits of the word.
2965     */
2966    tcg_gen_shli_i64(t, d, 7);
2967    tcg_gen_or_i64(d, d, t);
2968    tcg_gen_shli_i64(t, d, 14);
2969    tcg_gen_or_i64(d, d, t);
2970    tcg_gen_shli_i64(t, d, 28);
2971    tcg_gen_or_i64(d, d, t);
2972}
2973
2974static void gen_pmovmskb_vec(unsigned vece, TCGv_vec d, TCGv_vec s)
2975{
2976    TCGv_vec t = tcg_temp_new_vec_matching(d);
2977    TCGv_vec m = tcg_constant_vec_matching(d, MO_8, 0x80);
2978
2979    /* See above */
2980    tcg_gen_and_vec(vece, d, s, m);
2981    tcg_gen_shli_vec(vece, t, d, 7);
2982    tcg_gen_or_vec(vece, d, d, t);
2983    tcg_gen_shli_vec(vece, t, d, 14);
2984    tcg_gen_or_vec(vece, d, d, t);
2985    tcg_gen_shli_vec(vece, t, d, 28);
2986    tcg_gen_or_vec(vece, d, d, t);
2987}
2988
2989static void gen_PMOVMSKB(DisasContext *s, X86DecodedInsn *decode)
2990{
2991    static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
2992    static const GVecGen2 g = {
2993        .fni8 = gen_pmovmskb_i64,
2994        .fniv = gen_pmovmskb_vec,
2995        .opt_opc = vecop_list,
2996        .vece = MO_64,
2997        .prefer_i64 = TCG_TARGET_REG_BITS == 64
2998    };
2999    MemOp ot = decode->op[2].ot;
3000    int vec_len = vector_len(s, decode);
3001    TCGv t = tcg_temp_new();
3002
3003    tcg_gen_gvec_2(offsetof(CPUX86State, xmm_t0) + xmm_offset(ot), decode->op[2].offset,
3004                   vec_len, vec_len, &g);
3005    tcg_gen_ld8u_tl(s->T0, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
3006    while (vec_len > 8) {
3007        vec_len -= 8;
3008        if (TCG_TARGET_HAS_extract2_tl) {
3009            /*
3010             * Load the next byte of the result into the high byte of T.
3011             * TCG does a similar expansion of deposit to shl+extract2; by
3012             * loading the whole word, the shift left is avoided.
3013             */
3014#ifdef TARGET_X86_64
3015            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_Q((vec_len - 1) / 8)));
3016#else
3017            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L((vec_len - 1) / 4)));
3018#endif
3019
3020            tcg_gen_extract2_tl(s->T0, t, s->T0, TARGET_LONG_BITS - 8);
3021        } else {
3022            /*
3023             * The _previous_ value is deposited into bits 8 and higher of t.  Because
3024             * those bits are known to be zero after ld8u, this becomes a shift+or
3025             * if deposit is not available.
3026             */
3027            tcg_gen_ld8u_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
3028            tcg_gen_deposit_tl(s->T0, t, s->T0, 8, TARGET_LONG_BITS - 8);
3029        }
3030    }
3031}
3032
3033static void gen_POP(DisasContext *s, X86DecodedInsn *decode)
3034{
3035    X86DecodedOp *op = &decode->op[0];
3036    MemOp ot = gen_pop_T0(s);
3037
3038    assert(ot >= op->ot);
3039    if (op->has_ea || op->unit == X86_OP_SEG) {
3040        /* NOTE: order is important for MMU exceptions */
3041        gen_writeback(s, decode, 0, s->T0);
3042    }
3043
3044    /* NOTE: writing back registers after update is important for pop %sp */
3045    gen_pop_update(s, ot);
3046}
3047
3048static void gen_POPA(DisasContext *s, X86DecodedInsn *decode)
3049{
3050    gen_popa(s);
3051}
3052
3053static void gen_POPCNT(DisasContext *s, X86DecodedInsn *decode)
3054{
3055    decode->cc_dst = tcg_temp_new();
3056    decode->cc_op = CC_OP_POPCNT;
3057
3058    tcg_gen_mov_tl(decode->cc_dst, s->T0);
3059    tcg_gen_ctpop_tl(s->T0, s->T0);
3060}
3061
3062static void gen_POPF(DisasContext *s, X86DecodedInsn *decode)
3063{
3064    MemOp ot;
3065    int mask = TF_MASK | AC_MASK | ID_MASK | NT_MASK;
3066
3067    if (CPL(s) == 0) {
3068        mask |= IF_MASK | IOPL_MASK;
3069    } else if (CPL(s) <= IOPL(s)) {
3070        mask |= IF_MASK;
3071    }
3072    if (s->dflag == MO_16) {
3073        mask &= 0xffff;
3074    }
3075
3076    ot = gen_pop_T0(s);
3077    gen_helper_write_eflags(tcg_env, s->T0, tcg_constant_i32(mask));
3078    gen_pop_update(s, ot);
3079    set_cc_op(s, CC_OP_EFLAGS);
3080    /* abort translation because TF/AC flag may change */
3081    s->base.is_jmp = DISAS_EOB_NEXT;
3082}
3083
3084static void gen_PSHUFW(DisasContext *s, X86DecodedInsn *decode)
3085{
3086    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
3087    gen_helper_pshufw_mmx(OP_PTR0, OP_PTR1, imm);
3088}
3089
3090static void gen_PSRLW_i(DisasContext *s, X86DecodedInsn *decode)
3091{
3092    int vec_len = vector_len(s, decode);
3093
3094    if (decode->immediate >= 16) {
3095        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3096    } else {
3097        tcg_gen_gvec_shri(MO_16,
3098                          decode->op[0].offset, decode->op[1].offset,
3099                          decode->immediate, vec_len, vec_len);
3100    }
3101}
3102
3103static void gen_PSLLW_i(DisasContext *s, X86DecodedInsn *decode)
3104{
3105    int vec_len = vector_len(s, decode);
3106
3107    if (decode->immediate >= 16) {
3108        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3109    } else {
3110        tcg_gen_gvec_shli(MO_16,
3111                          decode->op[0].offset, decode->op[1].offset,
3112                          decode->immediate, vec_len, vec_len);
3113    }
3114}
3115
3116static void gen_PSRAW_i(DisasContext *s, X86DecodedInsn *decode)
3117{
3118    int vec_len = vector_len(s, decode);
3119
3120    if (decode->immediate >= 16) {
3121        decode->immediate = 15;
3122    }
3123    tcg_gen_gvec_sari(MO_16,
3124                      decode->op[0].offset, decode->op[1].offset,
3125                      decode->immediate, vec_len, vec_len);
3126}
3127
3128static void gen_PSRLD_i(DisasContext *s, X86DecodedInsn *decode)
3129{
3130    int vec_len = vector_len(s, decode);
3131
3132    if (decode->immediate >= 32) {
3133        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3134    } else {
3135        tcg_gen_gvec_shri(MO_32,
3136                          decode->op[0].offset, decode->op[1].offset,
3137                          decode->immediate, vec_len, vec_len);
3138    }
3139}
3140
3141static void gen_PSLLD_i(DisasContext *s, X86DecodedInsn *decode)
3142{
3143    int vec_len = vector_len(s, decode);
3144
3145    if (decode->immediate >= 32) {
3146        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3147    } else {
3148        tcg_gen_gvec_shli(MO_32,
3149                          decode->op[0].offset, decode->op[1].offset,
3150                          decode->immediate, vec_len, vec_len);
3151    }
3152}
3153
3154static void gen_PSRAD_i(DisasContext *s, X86DecodedInsn *decode)
3155{
3156    int vec_len = vector_len(s, decode);
3157
3158    if (decode->immediate >= 32) {
3159        decode->immediate = 31;
3160    }
3161    tcg_gen_gvec_sari(MO_32,
3162                      decode->op[0].offset, decode->op[1].offset,
3163                      decode->immediate, vec_len, vec_len);
3164}
3165
3166static void gen_PSRLQ_i(DisasContext *s, X86DecodedInsn *decode)
3167{
3168    int vec_len = vector_len(s, decode);
3169
3170    if (decode->immediate >= 64) {
3171        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3172    } else {
3173        tcg_gen_gvec_shri(MO_64,
3174                          decode->op[0].offset, decode->op[1].offset,
3175                          decode->immediate, vec_len, vec_len);
3176    }
3177}
3178
3179static void gen_PSLLQ_i(DisasContext *s, X86DecodedInsn *decode)
3180{
3181    int vec_len = vector_len(s, decode);
3182
3183    if (decode->immediate >= 64) {
3184        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3185    } else {
3186        tcg_gen_gvec_shli(MO_64,
3187                          decode->op[0].offset, decode->op[1].offset,
3188                          decode->immediate, vec_len, vec_len);
3189    }
3190}
3191
3192static TCGv_ptr make_imm8u_xmm_vec(uint8_t imm, int vec_len)
3193{
3194    MemOp ot = vec_len == 16 ? MO_128 : MO_256;
3195    TCGv_i32 imm_v = tcg_constant8u_i32(imm);
3196    TCGv_ptr ptr = tcg_temp_new_ptr();
3197
3198    tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_t0) + xmm_offset(ot),
3199                         vec_len, vec_len, 0);
3200
3201    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_t0));
3202    tcg_gen_st_i32(imm_v, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L(0)));
3203    return ptr;
3204}
3205
3206static void gen_PSRLDQ_i(DisasContext *s, X86DecodedInsn *decode)
3207{
3208    int vec_len = vector_len(s, decode);
3209    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
3210
3211    if (s->vex_l) {
3212        gen_helper_psrldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3213    } else {
3214        gen_helper_psrldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3215    }
3216}
3217
3218static void gen_PSLLDQ_i(DisasContext *s, X86DecodedInsn *decode)
3219{
3220    int vec_len = vector_len(s, decode);
3221    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
3222
3223    if (s->vex_l) {
3224        gen_helper_pslldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3225    } else {
3226        gen_helper_pslldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3227    }
3228}
3229
3230static void gen_PUSH(DisasContext *s, X86DecodedInsn *decode)
3231{
3232    gen_push_v(s, s->T0);
3233}
3234
3235static void gen_PUSHA(DisasContext *s, X86DecodedInsn *decode)
3236{
3237    gen_pusha(s);
3238}
3239
3240static void gen_PUSHF(DisasContext *s, X86DecodedInsn *decode)
3241{
3242    gen_update_cc_op(s);
3243    gen_helper_read_eflags(s->T0, tcg_env);
3244    gen_push_v(s, s->T0);
3245}
3246
3247static MemOp gen_shift_count(DisasContext *s, X86DecodedInsn *decode,
3248                             bool *can_be_zero, TCGv *count, int unit)
3249{
3250    MemOp ot = decode->op[0].ot;
3251    int mask = (ot <= MO_32 ? 0x1f : 0x3f);
3252
3253    *can_be_zero = false;
3254    switch (unit) {
3255    case X86_OP_INT:
3256        *count = tcg_temp_new();
3257        tcg_gen_andi_tl(*count, cpu_regs[R_ECX], mask);
3258        *can_be_zero = true;
3259        break;
3260
3261    case X86_OP_IMM:
3262        if ((decode->immediate & mask) == 0) {
3263            *count = NULL;
3264            break;
3265        }
3266        *count = tcg_temp_new();
3267        tcg_gen_movi_tl(*count, decode->immediate & mask);
3268        break;
3269
3270    case X86_OP_SKIP:
3271        *count = tcg_temp_new();
3272        tcg_gen_movi_tl(*count, 1);
3273        break;
3274
3275    default:
3276        g_assert_not_reached();
3277    }
3278
3279    return ot;
3280}
3281
3282/*
3283 * Compute existing flags in decode->cc_src, for gen_* functions that wants
3284 * to set the cc_op set to CC_OP_ADCOX.  In particular, this allows rotate
3285 * operations to compute the carry in decode->cc_dst and the overflow in
3286 * decode->cc_src2.
3287 *
3288 * If need_flags is true, decode->cc_dst and decode->cc_src2 are preloaded
3289 * with the value of CF and OF before the instruction, so that it is possible
3290 * to keep the flags unmodified.
3291 *
3292 * Return true if carry could be made available cheaply as a 1-bit value in
3293 * decode->cc_dst (trying a bit harder if want_carry is true).  If false is
3294 * returned, decode->cc_dst is uninitialized and the carry is only available
3295 * as bit 0 of decode->cc_src.
3296 */
3297static bool gen_eflags_adcox(DisasContext *s, X86DecodedInsn *decode, bool want_carry, bool need_flags)
3298{
3299    bool got_cf = false;
3300    bool got_of = false;
3301
3302    decode->cc_dst = tcg_temp_new();
3303    decode->cc_src = tcg_temp_new();
3304    decode->cc_src2 = tcg_temp_new();
3305    decode->cc_op = CC_OP_ADCOX;
3306
3307    /* A lot more cc_ops could be "optimized" to avoid the extracts at
3308     * the end (INC/DEC, BMILG, MUL), but they are all really unlikely
3309     * to be followed by rotations within the same basic block.
3310     */
3311    switch (s->cc_op) {
3312    case CC_OP_ADCOX:
3313        /* No need to compute the full EFLAGS, CF/OF are already isolated.  */
3314        tcg_gen_mov_tl(decode->cc_src, cpu_cc_src);
3315        if (need_flags) {
3316            tcg_gen_mov_tl(decode->cc_src2, cpu_cc_src2);
3317            got_of = true;
3318        }
3319        if (want_carry || need_flags) {
3320            tcg_gen_mov_tl(decode->cc_dst, cpu_cc_dst);
3321            got_cf = true;
3322        }
3323        break;
3324
3325    case CC_OP_LOGICB ... CC_OP_LOGICQ:
3326        /* CF and OF are zero, do it just because it's easy.  */
3327        gen_mov_eflags(s, decode->cc_src);
3328        if (need_flags) {
3329            tcg_gen_movi_tl(decode->cc_src2, 0);
3330            got_of = true;
3331        }
3332        if (want_carry || need_flags) {
3333            tcg_gen_movi_tl(decode->cc_dst, 0);
3334            got_cf = true;
3335        }
3336        break;
3337
3338    case CC_OP_SARB ... CC_OP_SARQ:
3339        /*
3340         * SHR/RCR/SHR/RCR/... is a relatively common occurrence of RCR.
3341         * By computing CF without using eflags, the calls to cc_compute_all
3342         * can be eliminated as dead code (except for the last RCR).
3343         */
3344        if (want_carry || need_flags) {
3345            tcg_gen_andi_tl(decode->cc_dst, cpu_cc_src, 1);
3346            got_cf = true;
3347        }
3348        gen_mov_eflags(s, decode->cc_src);
3349        break;
3350
3351    case CC_OP_SHLB ... CC_OP_SHLQ:
3352        /*
3353         * Likewise for SHL/RCL/SHL/RCL/... but, if CF is not in the sign
3354         * bit, we might as well fish CF out of EFLAGS and save a shift.
3355         */
3356        if (want_carry && (!need_flags || s->cc_op == CC_OP_SHLB + MO_TL)) {
3357            tcg_gen_shri_tl(decode->cc_dst, cpu_cc_src, (8 << (s->cc_op - CC_OP_SHLB)) - 1);
3358            got_cf = true;
3359        }
3360        gen_mov_eflags(s, decode->cc_src);
3361        break;
3362
3363    default:
3364        gen_mov_eflags(s, decode->cc_src);
3365        break;
3366    }
3367
3368    if (need_flags) {
3369        /* If the flags could be left unmodified, always load them.  */
3370        if (!got_of) {
3371            tcg_gen_extract_tl(decode->cc_src2, decode->cc_src, ctz32(CC_O), 1);
3372            got_of = true;
3373        }
3374        if (!got_cf) {
3375            tcg_gen_extract_tl(decode->cc_dst, decode->cc_src, ctz32(CC_C), 1);
3376            got_cf = true;
3377        }
3378    }
3379    return got_cf;
3380}
3381
3382static void gen_rot_overflow(X86DecodedInsn *decode, TCGv result, TCGv old,
3383                             bool can_be_zero, TCGv count)
3384{
3385    MemOp ot = decode->op[0].ot;
3386    TCGv temp = can_be_zero ? tcg_temp_new() : decode->cc_src2;
3387
3388    tcg_gen_xor_tl(temp, old, result);
3389    tcg_gen_extract_tl(temp, temp, (8 << ot) - 1, 1);
3390    if (can_be_zero) {
3391        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src2, count, tcg_constant_tl(0),
3392                           decode->cc_src2, temp);
3393    }
3394}
3395
3396/*
3397 * RCx operations are invariant modulo 8*operand_size+1.  For 8 and 16-bit operands,
3398 * this is less than 0x1f (the mask applied by gen_shift_count) so reduce further.
3399 */
3400static void gen_rotc_mod(MemOp ot, TCGv count)
3401{
3402    TCGv temp;
3403
3404    switch (ot) {
3405    case MO_8:
3406        temp = tcg_temp_new();
3407        tcg_gen_subi_tl(temp, count, 18);
3408        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3409        tcg_gen_subi_tl(temp, count, 9);
3410        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3411        break;
3412
3413    case MO_16:
3414        temp = tcg_temp_new();
3415        tcg_gen_subi_tl(temp, count, 17);
3416        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3417        break;
3418
3419    default:
3420        break;
3421    }
3422}
3423
3424/*
3425 * The idea here is that the bit to the right of the new bit 0 is the
3426 * new carry, and the bit to the right of the old bit 0 is the old carry.
3427 * Just like a regular rotation, the result of the rotation is composed
3428 * from a right shifted part and a left shifted part of s->T0.  The new carry
3429 * is extracted from the right-shifted portion, and the old carry is
3430 * inserted at the end of the left-shifted portion.
3431 *
3432 * Because of the separate shifts involving the carry, gen_RCL and gen_RCR
3433 * mostly operate on count-1.  This also comes in handy when computing
3434 * length - count, because (length-1) - (count-1) can be computed with
3435 * a XOR, and that is commutative unlike subtraction.
3436 */
3437static void gen_RCL(DisasContext *s, X86DecodedInsn *decode)
3438{
3439    bool have_1bit_cin, can_be_zero;
3440    TCGv count;
3441    TCGLabel *zero_label = NULL;
3442    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3443    TCGv low, high, low_count;
3444
3445    if (!count) {
3446        return;
3447    }
3448
3449    low = tcg_temp_new();
3450    high = tcg_temp_new();
3451    low_count = tcg_temp_new();
3452
3453    gen_rotc_mod(ot, count);
3454    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3455    if (can_be_zero) {
3456        zero_label = gen_new_label();
3457        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3458    }
3459
3460    /* Compute high part, including incoming carry.  */
3461    if (!have_1bit_cin || TCG_TARGET_deposit_tl_valid(1, TARGET_LONG_BITS - 1)) {
3462        /* high = (T0 << 1) | cin */
3463        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3464        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3465    } else {
3466        /* Same as above but without deposit; cin in cc_dst.  */
3467        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3468        tcg_gen_add_tl(high, high, s->T0);
3469    }
3470    tcg_gen_subi_tl(count, count, 1);
3471    tcg_gen_shl_tl(high, high, count);
3472
3473    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3474    tcg_gen_xori_tl(low_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3475    tcg_gen_shr_tl(low, s->T0, low_count);
3476    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3477    tcg_gen_shri_tl(low, low, 1);
3478
3479    /* Compute result and outgoing overflow */
3480    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3481    tcg_gen_or_tl(s->T0, low, high);
3482    gen_rot_overflow(decode, s->T0, decode->cc_src2, false, NULL);
3483
3484    if (zero_label) {
3485        gen_set_label(zero_label);
3486    }
3487}
3488
3489static void gen_RCR(DisasContext *s, X86DecodedInsn *decode)
3490{
3491    bool have_1bit_cin, can_be_zero;
3492    TCGv count;
3493    TCGLabel *zero_label = NULL;
3494    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3495    TCGv low, high, high_count;
3496
3497    if (!count) {
3498        return;
3499    }
3500
3501    low = tcg_temp_new();
3502    high = tcg_temp_new();
3503    high_count = tcg_temp_new();
3504
3505    gen_rotc_mod(ot, count);
3506    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3507    if (can_be_zero) {
3508        zero_label = gen_new_label();
3509        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3510    }
3511
3512    /* Save incoming carry into high, it will be shifted later.  */
3513    if (!have_1bit_cin || TCG_TARGET_deposit_tl_valid(1, TARGET_LONG_BITS - 1)) {
3514        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3515        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3516    } else {
3517        /* Same as above but without deposit; cin in cc_dst.  */
3518        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3519        tcg_gen_add_tl(high, high, s->T0);
3520    }
3521
3522    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3523    tcg_gen_subi_tl(count, count, 1);
3524    tcg_gen_shr_tl(low, s->T0, count);
3525    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3526    tcg_gen_shri_tl(low, low, 1);
3527
3528    /* Move high part to the right position */
3529    tcg_gen_xori_tl(high_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3530    tcg_gen_shl_tl(high, high, high_count);
3531
3532    /* Compute result and outgoing overflow */
3533    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3534    tcg_gen_or_tl(s->T0, low, high);
3535    gen_rot_overflow(decode, s->T0, decode->cc_src2, false, NULL);
3536
3537    if (zero_label) {
3538        gen_set_label(zero_label);
3539    }
3540}
3541
3542#ifdef CONFIG_USER_ONLY
3543static void gen_unreachable(DisasContext *s, X86DecodedInsn *decode)
3544{
3545    g_assert_not_reached();
3546}
3547#endif
3548
3549#ifndef CONFIG_USER_ONLY
3550static void gen_RDMSR(DisasContext *s, X86DecodedInsn *decode)
3551{
3552    gen_update_cc_op(s);
3553    gen_update_eip_cur(s);
3554    gen_helper_rdmsr(tcg_env);
3555}
3556#else
3557#define gen_RDMSR gen_unreachable
3558#endif
3559
3560static void gen_RDPMC(DisasContext *s, X86DecodedInsn *decode)
3561{
3562    gen_update_cc_op(s);
3563    gen_update_eip_cur(s);
3564    translator_io_start(&s->base);
3565    gen_helper_rdpmc(tcg_env);
3566    s->base.is_jmp = DISAS_NORETURN;
3567}
3568
3569static void gen_RDTSC(DisasContext *s, X86DecodedInsn *decode)
3570{
3571    gen_update_cc_op(s);
3572    gen_update_eip_cur(s);
3573    translator_io_start(&s->base);
3574    gen_helper_rdtsc(tcg_env);
3575}
3576
3577static void gen_RDxxBASE(DisasContext *s, X86DecodedInsn *decode)
3578{
3579    TCGv base = cpu_seg_base[s->modrm & 8 ? R_GS : R_FS];
3580
3581    /* Preserve hflags bits by testing CR4 at runtime.  */
3582    gen_helper_cr4_testbit(tcg_env, tcg_constant_i32(CR4_FSGSBASE_MASK));
3583    tcg_gen_mov_tl(s->T0, base);
3584}
3585
3586static void gen_RET(DisasContext *s, X86DecodedInsn *decode)
3587{
3588    int16_t adjust = decode->e.op1 == X86_TYPE_I ? decode->immediate : 0;
3589
3590    MemOp ot = gen_pop_T0(s);
3591    gen_stack_update(s, adjust + (1 << ot));
3592    gen_op_jmp_v(s, s->T0);
3593    gen_bnd_jmp(s);
3594    s->base.is_jmp = DISAS_JUMP;
3595}
3596
3597static void gen_RETF(DisasContext *s, X86DecodedInsn *decode)
3598{
3599    int16_t adjust = decode->e.op1 == X86_TYPE_I ? decode->immediate : 0;
3600
3601    if (!PE(s) || VM86(s)) {
3602        gen_lea_ss_ofs(s, s->A0, cpu_regs[R_ESP], 0);
3603        /* pop offset */
3604        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3605        /* NOTE: keeping EIP updated is not a problem in case of
3606           exception */
3607        gen_op_jmp_v(s, s->T0);
3608        /* pop selector */
3609        gen_add_A0_im(s, 1 << s->dflag);
3610        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3611        gen_op_movl_seg_real(s, R_CS, s->T0);
3612        /* add stack offset */
3613        gen_stack_update(s, adjust + (2 << s->dflag));
3614    } else {
3615        gen_update_cc_op(s);
3616        gen_update_eip_cur(s);
3617        gen_helper_lret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
3618                                  tcg_constant_i32(adjust));
3619    }
3620    s->base.is_jmp = DISAS_EOB_ONLY;
3621}
3622
3623/*
3624 * Return non-NULL if a 32-bit rotate works, after possibly replicating the input.
3625 * The input has already been zero-extended upon operand decode.
3626 */
3627static TCGv_i32 gen_rot_replicate(MemOp ot, TCGv in)
3628{
3629    TCGv_i32 temp;
3630    switch (ot) {
3631    case MO_8:
3632        temp = tcg_temp_new_i32();
3633        tcg_gen_trunc_tl_i32(temp, in);
3634        tcg_gen_muli_i32(temp, temp, 0x01010101);
3635        return temp;
3636
3637    case MO_16:
3638        temp = tcg_temp_new_i32();
3639        tcg_gen_trunc_tl_i32(temp, in);
3640        tcg_gen_deposit_i32(temp, temp, temp, 16, 16);
3641        return temp;
3642
3643#ifdef TARGET_X86_64
3644    case MO_32:
3645        temp = tcg_temp_new_i32();
3646        tcg_gen_trunc_tl_i32(temp, in);
3647        return temp;
3648#endif
3649
3650    default:
3651        return NULL;
3652    }
3653}
3654
3655static void gen_rot_carry(X86DecodedInsn *decode, TCGv result,
3656                          bool can_be_zero, TCGv count, int bit)
3657{
3658    if (!can_be_zero) {
3659        tcg_gen_extract_tl(decode->cc_dst, result, bit, 1);
3660    } else {
3661        TCGv temp = tcg_temp_new();
3662        tcg_gen_extract_tl(temp, result, bit, 1);
3663        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3664                           decode->cc_dst, temp);
3665    }
3666}
3667
3668static void gen_ROL(DisasContext *s, X86DecodedInsn *decode)
3669{
3670    bool can_be_zero;
3671    TCGv count;
3672    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3673    TCGv_i32 temp32, count32;
3674    TCGv old = tcg_temp_new();
3675
3676    if (!count) {
3677        return;
3678    }
3679
3680    gen_eflags_adcox(s, decode, false, can_be_zero);
3681    tcg_gen_mov_tl(old, s->T0);
3682    temp32 = gen_rot_replicate(ot, s->T0);
3683    if (temp32) {
3684        count32 = tcg_temp_new_i32();
3685        tcg_gen_trunc_tl_i32(count32, count);
3686        tcg_gen_rotl_i32(temp32, temp32, count32);
3687        /* Zero extend to facilitate later optimization.  */
3688        tcg_gen_extu_i32_tl(s->T0, temp32);
3689    } else {
3690        tcg_gen_rotl_tl(s->T0, s->T0, count);
3691    }
3692    gen_rot_carry(decode, s->T0, can_be_zero, count, 0);
3693    gen_rot_overflow(decode, s->T0, old, can_be_zero, count);
3694}
3695
3696static void gen_ROR(DisasContext *s, X86DecodedInsn *decode)
3697{
3698    bool can_be_zero;
3699    TCGv count;
3700    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3701    TCGv_i32 temp32, count32;
3702    TCGv old = tcg_temp_new();
3703
3704    if (!count) {
3705        return;
3706    }
3707
3708    gen_eflags_adcox(s, decode, false, can_be_zero);
3709    tcg_gen_mov_tl(old, s->T0);
3710    temp32 = gen_rot_replicate(ot, s->T0);
3711    if (temp32) {
3712        count32 = tcg_temp_new_i32();
3713        tcg_gen_trunc_tl_i32(count32, count);
3714        tcg_gen_rotr_i32(temp32, temp32, count32);
3715        /* Zero extend to facilitate later optimization.  */
3716        tcg_gen_extu_i32_tl(s->T0, temp32);
3717        gen_rot_carry(decode, s->T0, can_be_zero, count, 31);
3718    } else {
3719        tcg_gen_rotr_tl(s->T0, s->T0, count);
3720        gen_rot_carry(decode, s->T0, can_be_zero, count, TARGET_LONG_BITS - 1);
3721    }
3722    gen_rot_overflow(decode, s->T0, old, can_be_zero, count);
3723}
3724
3725static void gen_RORX(DisasContext *s, X86DecodedInsn *decode)
3726{
3727    MemOp ot = decode->op[0].ot;
3728    int mask = ot == MO_64 ? 63 : 31;
3729    int b = decode->immediate & mask;
3730
3731    switch (ot) {
3732    case MO_32:
3733#ifdef TARGET_X86_64
3734        tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3735        tcg_gen_rotri_i32(s->tmp2_i32, s->tmp2_i32, b);
3736        tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
3737        break;
3738
3739    case MO_64:
3740#endif
3741        tcg_gen_rotri_tl(s->T0, s->T0, b);
3742        break;
3743
3744    default:
3745        g_assert_not_reached();
3746    }
3747}
3748
3749#ifndef CONFIG_USER_ONLY
3750static void gen_RSM(DisasContext *s, X86DecodedInsn *decode)
3751{
3752    gen_helper_rsm(tcg_env);
3753    assume_cc_op(s, CC_OP_EFLAGS);
3754    s->base.is_jmp = DISAS_EOB_ONLY;
3755}
3756#else
3757#define gen_RSM gen_UD
3758#endif
3759
3760static void gen_SAHF(DisasContext *s, X86DecodedInsn *decode)
3761{
3762    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
3763        return gen_illegal_opcode(s);
3764    }
3765    tcg_gen_shri_tl(s->T0, cpu_regs[R_EAX], 8);
3766    gen_compute_eflags(s);
3767    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, CC_O);
3768    tcg_gen_andi_tl(s->T0, s->T0, CC_S | CC_Z | CC_A | CC_P | CC_C);
3769    tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, s->T0);
3770}
3771
3772static void gen_SALC(DisasContext *s, X86DecodedInsn *decode)
3773{
3774    gen_compute_eflags_c(s, s->T0);
3775    tcg_gen_neg_tl(s->T0, s->T0);
3776}
3777
3778static void gen_shift_dynamic_flags(DisasContext *s, X86DecodedInsn *decode, TCGv count, CCOp cc_op)
3779{
3780    TCGv_i32 count32 = tcg_temp_new_i32();
3781    TCGv_i32 old_cc_op;
3782
3783    decode->cc_op = CC_OP_DYNAMIC;
3784    decode->cc_op_dynamic = tcg_temp_new_i32();
3785
3786    assert(decode->cc_dst == s->T0);
3787    if (cc_op_live[s->cc_op] & USES_CC_DST) {
3788        decode->cc_dst = tcg_temp_new();
3789        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3790                           cpu_cc_dst, s->T0);
3791    }
3792
3793    if (cc_op_live[s->cc_op] & USES_CC_SRC) {
3794        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src, count, tcg_constant_tl(0),
3795                           cpu_cc_src, decode->cc_src);
3796    }
3797
3798    tcg_gen_trunc_tl_i32(count32, count);
3799    if (s->cc_op == CC_OP_DYNAMIC) {
3800        old_cc_op = cpu_cc_op;
3801    } else {
3802        old_cc_op = tcg_constant_i32(s->cc_op);
3803    }
3804    tcg_gen_movcond_i32(TCG_COND_EQ, decode->cc_op_dynamic, count32, tcg_constant_i32(0),
3805                        old_cc_op, tcg_constant_i32(cc_op));
3806}
3807
3808static void gen_SAR(DisasContext *s, X86DecodedInsn *decode)
3809{
3810    bool can_be_zero;
3811    TCGv count;
3812    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3813
3814    if (!count) {
3815        return;
3816    }
3817
3818    decode->cc_dst = s->T0;
3819    decode->cc_src = tcg_temp_new();
3820    tcg_gen_subi_tl(decode->cc_src, count, 1);
3821    tcg_gen_sar_tl(decode->cc_src, s->T0, decode->cc_src);
3822    tcg_gen_sar_tl(s->T0, s->T0, count);
3823    if (can_be_zero) {
3824        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3825    } else {
3826        decode->cc_op = CC_OP_SARB + ot;
3827    }
3828}
3829
3830static void gen_SARX(DisasContext *s, X86DecodedInsn *decode)
3831{
3832    MemOp ot = decode->op[0].ot;
3833    int mask;
3834
3835    mask = ot == MO_64 ? 63 : 31;
3836    tcg_gen_andi_tl(s->T1, s->T1, mask);
3837    tcg_gen_sar_tl(s->T0, s->T0, s->T1);
3838}
3839
3840static void gen_SBB(DisasContext *s, X86DecodedInsn *decode)
3841{
3842    MemOp ot = decode->op[0].ot;
3843    TCGv c_in = tcg_temp_new();
3844
3845    gen_compute_eflags_c(s, c_in);
3846    if (s->prefix & PREFIX_LOCK) {
3847        tcg_gen_add_tl(s->T0, s->T1, c_in);
3848        tcg_gen_neg_tl(s->T0, s->T0);
3849        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
3850                                    s->mem_index, ot | MO_LE);
3851    } else {
3852        /*
3853         * TODO: SBB reg, reg could use gen_prepare_eflags_c followed by
3854         * negsetcond, and CC_OP_SUBB as the cc_op.
3855         */
3856        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
3857        tcg_gen_sub_tl(s->T0, s->T0, c_in);
3858    }
3859    prepare_update3_cc(decode, s, CC_OP_SBBB + ot, c_in);
3860}
3861
3862static void gen_SCAS(DisasContext *s, X86DecodedInsn *decode)
3863{
3864    MemOp ot = decode->op[2].ot;
3865    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
3866        gen_repz_nz(s, ot, gen_scas);
3867    } else {
3868        gen_scas(s, ot);
3869    }
3870}
3871
3872static void gen_SETcc(DisasContext *s, X86DecodedInsn *decode)
3873{
3874    gen_setcc1(s, decode->b & 0xf, s->T0);
3875}
3876
3877static void gen_SFENCE(DisasContext *s, X86DecodedInsn *decode)
3878{
3879    tcg_gen_mb(TCG_MO_ST_ST | TCG_BAR_SC);
3880}
3881
3882static void gen_SHA1NEXTE(DisasContext *s, X86DecodedInsn *decode)
3883{
3884    gen_helper_sha1nexte(OP_PTR0, OP_PTR1, OP_PTR2);
3885}
3886
3887static void gen_SHA1MSG1(DisasContext *s, X86DecodedInsn *decode)
3888{
3889    gen_helper_sha1msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3890}
3891
3892static void gen_SHA1MSG2(DisasContext *s, X86DecodedInsn *decode)
3893{
3894    gen_helper_sha1msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3895}
3896
3897static void gen_SHA1RNDS4(DisasContext *s, X86DecodedInsn *decode)
3898{
3899    switch(decode->immediate & 3) {
3900    case 0:
3901        gen_helper_sha1rnds4_f0(OP_PTR0, OP_PTR0, OP_PTR1);
3902        break;
3903    case 1:
3904        gen_helper_sha1rnds4_f1(OP_PTR0, OP_PTR0, OP_PTR1);
3905        break;
3906    case 2:
3907        gen_helper_sha1rnds4_f2(OP_PTR0, OP_PTR0, OP_PTR1);
3908        break;
3909    case 3:
3910        gen_helper_sha1rnds4_f3(OP_PTR0, OP_PTR0, OP_PTR1);
3911        break;
3912    }
3913}
3914
3915static void gen_SHA256MSG1(DisasContext *s, X86DecodedInsn *decode)
3916{
3917    gen_helper_sha256msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3918}
3919
3920static void gen_SHA256MSG2(DisasContext *s, X86DecodedInsn *decode)
3921{
3922    gen_helper_sha256msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3923}
3924
3925static void gen_SHA256RNDS2(DisasContext *s, X86DecodedInsn *decode)
3926{
3927    TCGv_i32 wk0 = tcg_temp_new_i32();
3928    TCGv_i32 wk1 = tcg_temp_new_i32();
3929
3930    tcg_gen_ld_i32(wk0, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(0)));
3931    tcg_gen_ld_i32(wk1, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(1)));
3932
3933    gen_helper_sha256rnds2(OP_PTR0, OP_PTR1, OP_PTR2, wk0, wk1);
3934}
3935
3936static void gen_SHL(DisasContext *s, X86DecodedInsn *decode)
3937{
3938    bool can_be_zero;
3939    TCGv count;
3940    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3941
3942    if (!count) {
3943        return;
3944    }
3945
3946    decode->cc_dst = s->T0;
3947    decode->cc_src = tcg_temp_new();
3948    tcg_gen_subi_tl(decode->cc_src, count, 1);
3949    tcg_gen_shl_tl(decode->cc_src, s->T0, decode->cc_src);
3950    tcg_gen_shl_tl(s->T0, s->T0, count);
3951    if (can_be_zero) {
3952        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
3953    } else {
3954        decode->cc_op = CC_OP_SHLB + ot;
3955    }
3956}
3957
3958static void gen_SHLD(DisasContext *s, X86DecodedInsn *decode)
3959{
3960    bool can_be_zero;
3961    TCGv count;
3962    int unit = decode->e.op3 == X86_TYPE_I ? X86_OP_IMM : X86_OP_INT;
3963    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, unit);
3964
3965    if (!count) {
3966        return;
3967    }
3968
3969    decode->cc_dst = s->T0;
3970    decode->cc_src = s->tmp0;
3971    gen_shiftd_rm_T1(s, ot, false, count);
3972    if (can_be_zero) {
3973        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
3974    } else {
3975        decode->cc_op = CC_OP_SHLB + ot;
3976    }
3977}
3978
3979static void gen_SHLX(DisasContext *s, X86DecodedInsn *decode)
3980{
3981    MemOp ot = decode->op[0].ot;
3982    int mask;
3983
3984    mask = ot == MO_64 ? 63 : 31;
3985    tcg_gen_andi_tl(s->T1, s->T1, mask);
3986    tcg_gen_shl_tl(s->T0, s->T0, s->T1);
3987}
3988
3989static void gen_SHR(DisasContext *s, X86DecodedInsn *decode)
3990{
3991    bool can_be_zero;
3992    TCGv count;
3993    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3994
3995    if (!count) {
3996        return;
3997    }
3998
3999    decode->cc_dst = s->T0;
4000    decode->cc_src = tcg_temp_new();
4001    tcg_gen_subi_tl(decode->cc_src, count, 1);
4002    tcg_gen_shr_tl(decode->cc_src, s->T0, decode->cc_src);
4003    tcg_gen_shr_tl(s->T0, s->T0, count);
4004    if (can_be_zero) {
4005        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
4006    } else {
4007        decode->cc_op = CC_OP_SARB + ot;
4008    }
4009}
4010
4011static void gen_SHRD(DisasContext *s, X86DecodedInsn *decode)
4012{
4013    bool can_be_zero;
4014    TCGv count;
4015    int unit = decode->e.op3 == X86_TYPE_I ? X86_OP_IMM : X86_OP_INT;
4016    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, unit);
4017
4018    if (!count) {
4019        return;
4020    }
4021
4022    decode->cc_dst = s->T0;
4023    decode->cc_src = s->tmp0;
4024    gen_shiftd_rm_T1(s, ot, true, count);
4025    if (can_be_zero) {
4026        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
4027    } else {
4028        decode->cc_op = CC_OP_SARB + ot;
4029    }
4030}
4031
4032static void gen_SHRX(DisasContext *s, X86DecodedInsn *decode)
4033{
4034    MemOp ot = decode->op[0].ot;
4035    int mask;
4036
4037    mask = ot == MO_64 ? 63 : 31;
4038    tcg_gen_andi_tl(s->T1, s->T1, mask);
4039    tcg_gen_shr_tl(s->T0, s->T0, s->T1);
4040}
4041
4042static void gen_STC(DisasContext *s, X86DecodedInsn *decode)
4043{
4044    gen_compute_eflags(s);
4045    tcg_gen_ori_tl(cpu_cc_src, cpu_cc_src, CC_C);
4046}
4047
4048static void gen_STD(DisasContext *s, X86DecodedInsn *decode)
4049{
4050    tcg_gen_st_i32(tcg_constant_i32(-1), tcg_env, offsetof(CPUX86State, df));
4051}
4052
4053static void gen_STI(DisasContext *s, X86DecodedInsn *decode)
4054{
4055    gen_set_eflags(s, IF_MASK);
4056    s->base.is_jmp = DISAS_EOB_INHIBIT_IRQ;
4057}
4058
4059static void gen_VAESKEYGEN(DisasContext *s, X86DecodedInsn *decode)
4060{
4061    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4062    assert(!s->vex_l);
4063    gen_helper_aeskeygenassist_xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
4064}
4065
4066static void gen_STMXCSR(DisasContext *s, X86DecodedInsn *decode)
4067{
4068    gen_helper_update_mxcsr(tcg_env);
4069    tcg_gen_ld32u_tl(s->T0, tcg_env, offsetof(CPUX86State, mxcsr));
4070}
4071
4072static void gen_STOS(DisasContext *s, X86DecodedInsn *decode)
4073{
4074    MemOp ot = decode->op[1].ot;
4075    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
4076        gen_repz(s, ot, gen_stos);
4077    } else {
4078        gen_stos(s, ot);
4079    }
4080}
4081
4082static void gen_SUB(DisasContext *s, X86DecodedInsn *decode)
4083{
4084    MemOp ot = decode->op[1].ot;
4085
4086    if (s->prefix & PREFIX_LOCK) {
4087        tcg_gen_neg_tl(s->T0, s->T1);
4088        tcg_gen_atomic_fetch_add_tl(s->cc_srcT, s->A0, s->T0,
4089                                    s->mem_index, ot | MO_LE);
4090        tcg_gen_sub_tl(s->T0, s->cc_srcT, s->T1);
4091    } else {
4092        tcg_gen_mov_tl(s->cc_srcT, s->T0);
4093        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
4094    }
4095    prepare_update2_cc(decode, s, CC_OP_SUBB + ot);
4096}
4097
4098static void gen_SYSCALL(DisasContext *s, X86DecodedInsn *decode)
4099{
4100    gen_update_cc_op(s);
4101    gen_update_eip_cur(s);
4102    gen_helper_syscall(tcg_env, cur_insn_len_i32(s));
4103    if (LMA(s)) {
4104        assume_cc_op(s, CC_OP_EFLAGS);
4105    }
4106
4107    /*
4108     * TF handling for the syscall insn is different. The TF bit is checked
4109     * after the syscall insn completes. This allows #DB to not be
4110     * generated after one has entered CPL0 if TF is set in FMASK.
4111     */
4112    s->base.is_jmp = DISAS_EOB_RECHECK_TF;
4113}
4114
4115static void gen_SYSENTER(DisasContext *s, X86DecodedInsn *decode)
4116{
4117    gen_helper_sysenter(tcg_env);
4118    s->base.is_jmp = DISAS_EOB_ONLY;
4119}
4120
4121static void gen_SYSEXIT(DisasContext *s, X86DecodedInsn *decode)
4122{
4123    gen_helper_sysexit(tcg_env, tcg_constant_i32(s->dflag - 1));
4124    s->base.is_jmp = DISAS_EOB_ONLY;
4125}
4126
4127static void gen_SYSRET(DisasContext *s, X86DecodedInsn *decode)
4128{
4129    gen_helper_sysret(tcg_env, tcg_constant_i32(s->dflag - 1));
4130    if (LMA(s)) {
4131        assume_cc_op(s, CC_OP_EFLAGS);
4132    }
4133
4134    /*
4135     * TF handling for the sysret insn is different. The TF bit is checked
4136     * after the sysret insn completes. This allows #DB to be
4137     * generated "as if" the syscall insn in userspace has just
4138     * completed.
4139     */
4140    s->base.is_jmp = DISAS_EOB_RECHECK_TF;
4141}
4142
4143static void gen_TZCNT(DisasContext *s, X86DecodedInsn *decode)
4144{
4145    MemOp ot = decode->op[0].ot;
4146
4147    /* C bit (cc_src) is defined related to the input.  */
4148    decode->cc_src = tcg_temp_new();
4149    decode->cc_dst = s->T0;
4150    decode->cc_op = CC_OP_BMILGB + ot;
4151    tcg_gen_mov_tl(decode->cc_src, s->T0);
4152
4153    /* A zero input returns the operand size.  */
4154    tcg_gen_ctzi_tl(s->T0, s->T0, 8 << ot);
4155}
4156
4157static void gen_UD(DisasContext *s, X86DecodedInsn *decode)
4158{
4159    gen_illegal_opcode(s);
4160}
4161
4162static void gen_VAESIMC(DisasContext *s, X86DecodedInsn *decode)
4163{
4164    assert(!s->vex_l);
4165    gen_helper_aesimc_xmm(tcg_env, OP_PTR0, OP_PTR2);
4166}
4167
4168/*
4169 * 00 = v*ps Vps, Hps, Wpd
4170 * 66 = v*pd Vpd, Hpd, Wps
4171 * f3 = v*ss Vss, Hss, Wps
4172 * f2 = v*sd Vsd, Hsd, Wps
4173 */
4174#define SSE_CMP(x) { \
4175    gen_helper_ ## x ## ps ## _xmm, gen_helper_ ## x ## pd ## _xmm, \
4176    gen_helper_ ## x ## ss, gen_helper_ ## x ## sd, \
4177    gen_helper_ ## x ## ps ## _ymm, gen_helper_ ## x ## pd ## _ymm}
4178static const SSEFunc_0_eppp gen_helper_cmp_funcs[32][6] = {
4179    SSE_CMP(cmpeq),
4180    SSE_CMP(cmplt),
4181    SSE_CMP(cmple),
4182    SSE_CMP(cmpunord),
4183    SSE_CMP(cmpneq),
4184    SSE_CMP(cmpnlt),
4185    SSE_CMP(cmpnle),
4186    SSE_CMP(cmpord),
4187
4188    SSE_CMP(cmpequ),
4189    SSE_CMP(cmpnge),
4190    SSE_CMP(cmpngt),
4191    SSE_CMP(cmpfalse),
4192    SSE_CMP(cmpnequ),
4193    SSE_CMP(cmpge),
4194    SSE_CMP(cmpgt),
4195    SSE_CMP(cmptrue),
4196
4197    SSE_CMP(cmpeqs),
4198    SSE_CMP(cmpltq),
4199    SSE_CMP(cmpleq),
4200    SSE_CMP(cmpunords),
4201    SSE_CMP(cmpneqq),
4202    SSE_CMP(cmpnltq),
4203    SSE_CMP(cmpnleq),
4204    SSE_CMP(cmpords),
4205
4206    SSE_CMP(cmpequs),
4207    SSE_CMP(cmpngeq),
4208    SSE_CMP(cmpngtq),
4209    SSE_CMP(cmpfalses),
4210    SSE_CMP(cmpnequs),
4211    SSE_CMP(cmpgeq),
4212    SSE_CMP(cmpgtq),
4213    SSE_CMP(cmptrues),
4214};
4215#undef SSE_CMP
4216
4217static void gen_VCMP(DisasContext *s, X86DecodedInsn *decode)
4218{
4219    int index = decode->immediate & (s->prefix & PREFIX_VEX ? 31 : 7);
4220    int b =
4221        s->prefix & PREFIX_REPZ  ? 2 /* ss */ :
4222        s->prefix & PREFIX_REPNZ ? 3 /* sd */ :
4223        !!(s->prefix & PREFIX_DATA) /* pd */ + (s->vex_l << 2);
4224
4225    gen_helper_cmp_funcs[index][b](tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4226}
4227
4228static void gen_VCOMI(DisasContext *s, X86DecodedInsn *decode)
4229{
4230    SSEFunc_0_epp fn;
4231    fn = s->prefix & PREFIX_DATA ? gen_helper_comisd : gen_helper_comiss;
4232    fn(tcg_env, OP_PTR1, OP_PTR2);
4233    assume_cc_op(s, CC_OP_EFLAGS);
4234}
4235
4236static void gen_VCVTPD2PS(DisasContext *s, X86DecodedInsn *decode)
4237{
4238    if (s->vex_l) {
4239        gen_helper_cvtpd2ps_ymm(tcg_env, OP_PTR0, OP_PTR2);
4240    } else {
4241        gen_helper_cvtpd2ps_xmm(tcg_env, OP_PTR0, OP_PTR2);
4242    }
4243}
4244
4245static void gen_VCVTPS2PD(DisasContext *s, X86DecodedInsn *decode)
4246{
4247    if (s->vex_l) {
4248        gen_helper_cvtps2pd_ymm(tcg_env, OP_PTR0, OP_PTR2);
4249    } else {
4250        gen_helper_cvtps2pd_xmm(tcg_env, OP_PTR0, OP_PTR2);
4251    }
4252}
4253
4254static void gen_VCVTPS2PH(DisasContext *s, X86DecodedInsn *decode)
4255{
4256    gen_unary_imm_fp_sse(s, decode,
4257                      gen_helper_cvtps2ph_xmm,
4258                      gen_helper_cvtps2ph_ymm);
4259    /*
4260     * VCVTPS2PH is the only instruction that performs an operation on a
4261     * register source and then *stores* into memory.
4262     */
4263    if (decode->op[0].has_ea) {
4264        gen_store_sse(s, decode, decode->op[0].offset);
4265    }
4266}
4267
4268static void gen_VCVTSD2SS(DisasContext *s, X86DecodedInsn *decode)
4269{
4270    gen_helper_cvtsd2ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4271}
4272
4273static void gen_VCVTSS2SD(DisasContext *s, X86DecodedInsn *decode)
4274{
4275    gen_helper_cvtss2sd(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4276}
4277
4278static void gen_VCVTSI2Sx(DisasContext *s, X86DecodedInsn *decode)
4279{
4280    int vec_len = vector_len(s, decode);
4281    TCGv_i32 in;
4282
4283    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4284
4285#ifdef TARGET_X86_64
4286    MemOp ot = decode->op[2].ot;
4287    if (ot == MO_64) {
4288        if (s->prefix & PREFIX_REPNZ) {
4289            gen_helper_cvtsq2sd(tcg_env, OP_PTR0, s->T1);
4290        } else {
4291            gen_helper_cvtsq2ss(tcg_env, OP_PTR0, s->T1);
4292        }
4293        return;
4294    }
4295    in = s->tmp2_i32;
4296    tcg_gen_trunc_tl_i32(in, s->T1);
4297#else
4298    in = s->T1;
4299#endif
4300
4301    if (s->prefix & PREFIX_REPNZ) {
4302        gen_helper_cvtsi2sd(tcg_env, OP_PTR0, in);
4303    } else {
4304        gen_helper_cvtsi2ss(tcg_env, OP_PTR0, in);
4305    }
4306}
4307
4308static inline void gen_VCVTtSx2SI(DisasContext *s, X86DecodedInsn *decode,
4309                                  SSEFunc_i_ep ss2si, SSEFunc_l_ep ss2sq,
4310                                  SSEFunc_i_ep sd2si, SSEFunc_l_ep sd2sq)
4311{
4312    TCGv_i32 out;
4313
4314#ifdef TARGET_X86_64
4315    MemOp ot = decode->op[0].ot;
4316    if (ot == MO_64) {
4317        if (s->prefix & PREFIX_REPNZ) {
4318            sd2sq(s->T0, tcg_env, OP_PTR2);
4319        } else {
4320            ss2sq(s->T0, tcg_env, OP_PTR2);
4321        }
4322        return;
4323    }
4324
4325    out = s->tmp2_i32;
4326#else
4327    out = s->T0;
4328#endif
4329    if (s->prefix & PREFIX_REPNZ) {
4330        sd2si(out, tcg_env, OP_PTR2);
4331    } else {
4332        ss2si(out, tcg_env, OP_PTR2);
4333    }
4334#ifdef TARGET_X86_64
4335    tcg_gen_extu_i32_tl(s->T0, out);
4336#endif
4337}
4338
4339#ifndef TARGET_X86_64
4340#define gen_helper_cvtss2sq NULL
4341#define gen_helper_cvtsd2sq NULL
4342#define gen_helper_cvttss2sq NULL
4343#define gen_helper_cvttsd2sq NULL
4344#endif
4345
4346static void gen_VCVTSx2SI(DisasContext *s, X86DecodedInsn *decode)
4347{
4348    gen_VCVTtSx2SI(s, decode,
4349                   gen_helper_cvtss2si, gen_helper_cvtss2sq,
4350                   gen_helper_cvtsd2si, gen_helper_cvtsd2sq);
4351}
4352
4353static void gen_VCVTTSx2SI(DisasContext *s, X86DecodedInsn *decode)
4354{
4355    gen_VCVTtSx2SI(s, decode,
4356                   gen_helper_cvttss2si, gen_helper_cvttss2sq,
4357                   gen_helper_cvttsd2si, gen_helper_cvttsd2sq);
4358}
4359
4360static void gen_VEXTRACTx128(DisasContext *s, X86DecodedInsn *decode)
4361{
4362    int mask = decode->immediate & 1;
4363    int src_ofs = vector_elem_offset(&decode->op[1], MO_128, mask);
4364    if (decode->op[0].has_ea) {
4365        /* VEX-only instruction, no alignment requirements.  */
4366        gen_sto_env_A0(s, src_ofs, false);
4367    } else {
4368        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, 16, 16);
4369    }
4370}
4371
4372static void gen_VEXTRACTPS(DisasContext *s, X86DecodedInsn *decode)
4373{
4374    gen_pextr(s, decode, MO_32);
4375}
4376
4377static void gen_vinsertps(DisasContext *s, X86DecodedInsn *decode)
4378{
4379    int val = decode->immediate;
4380    int dest_word = (val >> 4) & 3;
4381    int new_mask = (val & 15) | (1 << dest_word);
4382    int vec_len = 16;
4383
4384    assert(!s->vex_l);
4385
4386    if (new_mask == 15) {
4387        /* All zeroes except possibly for the inserted element */
4388        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
4389    } else if (decode->op[1].offset != decode->op[0].offset) {
4390        gen_store_sse(s, decode, decode->op[1].offset);
4391    }
4392
4393    if (new_mask != (val & 15)) {
4394        tcg_gen_st_i32(s->tmp2_i32, tcg_env,
4395                       vector_elem_offset(&decode->op[0], MO_32, dest_word));
4396    }
4397
4398    if (new_mask != 15) {
4399        TCGv_i32 zero = tcg_constant_i32(0); /* float32_zero */
4400        int i;
4401        for (i = 0; i < 4; i++) {
4402            if ((val >> i) & 1) {
4403                tcg_gen_st_i32(zero, tcg_env,
4404                               vector_elem_offset(&decode->op[0], MO_32, i));
4405            }
4406        }
4407    }
4408}
4409
4410static void gen_VINSERTPS_r(DisasContext *s, X86DecodedInsn *decode)
4411{
4412    int val = decode->immediate;
4413    tcg_gen_ld_i32(s->tmp2_i32, tcg_env,
4414                   vector_elem_offset(&decode->op[2], MO_32, (val >> 6) & 3));
4415    gen_vinsertps(s, decode);
4416}
4417
4418static void gen_VINSERTPS_m(DisasContext *s, X86DecodedInsn *decode)
4419{
4420    tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4421    gen_vinsertps(s, decode);
4422}
4423
4424static void gen_VINSERTx128(DisasContext *s, X86DecodedInsn *decode)
4425{
4426    int mask = decode->immediate & 1;
4427    tcg_gen_gvec_mov(MO_64,
4428                     decode->op[0].offset + offsetof(YMMReg, YMM_X(mask)),
4429                     decode->op[2].offset + offsetof(YMMReg, YMM_X(0)), 16, 16);
4430    tcg_gen_gvec_mov(MO_64,
4431                     decode->op[0].offset + offsetof(YMMReg, YMM_X(!mask)),
4432                     decode->op[1].offset + offsetof(YMMReg, YMM_X(!mask)), 16, 16);
4433}
4434
4435static inline void gen_maskmov(DisasContext *s, X86DecodedInsn *decode,
4436                               SSEFunc_0_eppt xmm, SSEFunc_0_eppt ymm)
4437{
4438    if (!s->vex_l) {
4439        xmm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
4440    } else {
4441        ymm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
4442    }
4443}
4444
4445static void gen_VMASKMOVPD_st(DisasContext *s, X86DecodedInsn *decode)
4446{
4447    gen_maskmov(s, decode, gen_helper_vpmaskmovq_st_xmm, gen_helper_vpmaskmovq_st_ymm);
4448}
4449
4450static void gen_VMASKMOVPS_st(DisasContext *s, X86DecodedInsn *decode)
4451{
4452    gen_maskmov(s, decode, gen_helper_vpmaskmovd_st_xmm, gen_helper_vpmaskmovd_st_ymm);
4453}
4454
4455static void gen_VMOVHPx_ld(DisasContext *s, X86DecodedInsn *decode)
4456{
4457    gen_ldq_env_A0(s, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4458    if (decode->op[0].offset != decode->op[1].offset) {
4459        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4460        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4461    }
4462}
4463
4464static void gen_VMOVHPx_st(DisasContext *s, X86DecodedInsn *decode)
4465{
4466    gen_stq_env_A0(s, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4467}
4468
4469static void gen_VMOVHPx(DisasContext *s, X86DecodedInsn *decode)
4470{
4471    if (decode->op[0].offset != decode->op[2].offset) {
4472        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4473        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4474    }
4475    if (decode->op[0].offset != decode->op[1].offset) {
4476        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4477        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4478    }
4479}
4480
4481static void gen_VMOVHLPS(DisasContext *s, X86DecodedInsn *decode)
4482{
4483    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4484    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4485    if (decode->op[0].offset != decode->op[1].offset) {
4486        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(1)));
4487        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4488    }
4489}
4490
4491static void gen_VMOVLHPS(DisasContext *s, X86DecodedInsn *decode)
4492{
4493    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
4494    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4495    if (decode->op[0].offset != decode->op[1].offset) {
4496        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4497        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4498    }
4499}
4500
4501/*
4502 * Note that MOVLPx supports 256-bit operation unlike MOVHLPx, MOVLHPx, MOXHPx.
4503 * Use a gvec move to move everything above the bottom 64 bits.
4504 */
4505
4506static void gen_VMOVLPx(DisasContext *s, X86DecodedInsn *decode)
4507{
4508    int vec_len = vector_len(s, decode);
4509
4510    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(0)));
4511    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4512    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4513}
4514
4515static void gen_VMOVLPx_ld(DisasContext *s, X86DecodedInsn *decode)
4516{
4517    int vec_len = vector_len(s, decode);
4518
4519    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4520    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4521    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
4522}
4523
4524static void gen_VMOVLPx_st(DisasContext *s, X86DecodedInsn *decode)
4525{
4526    tcg_gen_ld_i64(s->tmp1_i64, OP_PTR2, offsetof(ZMMReg, ZMM_Q(0)));
4527    tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4528}
4529
4530static void gen_VMOVSD_ld(DisasContext *s, X86DecodedInsn *decode)
4531{
4532    TCGv_i64 zero = tcg_constant_i64(0);
4533
4534    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4535    tcg_gen_st_i64(zero, OP_PTR0, offsetof(ZMMReg, ZMM_Q(1)));
4536    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
4537}
4538
4539static void gen_VMOVSS(DisasContext *s, X86DecodedInsn *decode)
4540{
4541    int vec_len = vector_len(s, decode);
4542
4543    tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
4544    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4545    tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
4546}
4547
4548static void gen_VMOVSS_ld(DisasContext *s, X86DecodedInsn *decode)
4549{
4550    int vec_len = vector_len(s, decode);
4551
4552    tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4553    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
4554    tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
4555}
4556
4557static void gen_VMOVSS_st(DisasContext *s, X86DecodedInsn *decode)
4558{
4559    tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
4560    tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4561}
4562
4563static void gen_VPMASKMOV_st(DisasContext *s, X86DecodedInsn *decode)
4564{
4565    if (s->vex_w) {
4566        gen_VMASKMOVPD_st(s, decode);
4567    } else {
4568        gen_VMASKMOVPS_st(s, decode);
4569    }
4570}
4571
4572static void gen_VPERMD(DisasContext *s, X86DecodedInsn *decode)
4573{
4574    assert(s->vex_l);
4575    gen_helper_vpermd_ymm(OP_PTR0, OP_PTR1, OP_PTR2);
4576}
4577
4578static void gen_VPERM2x128(DisasContext *s, X86DecodedInsn *decode)
4579{
4580    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4581    assert(s->vex_l);
4582    gen_helper_vpermdq_ymm(OP_PTR0, OP_PTR1, OP_PTR2, imm);
4583}
4584
4585static void gen_VPHMINPOSUW(DisasContext *s, X86DecodedInsn *decode)
4586{
4587    assert(!s->vex_l);
4588    gen_helper_phminposuw_xmm(tcg_env, OP_PTR0, OP_PTR2);
4589}
4590
4591static void gen_VROUNDSD(DisasContext *s, X86DecodedInsn *decode)
4592{
4593    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4594    assert(!s->vex_l);
4595    gen_helper_roundsd_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
4596}
4597
4598static void gen_VROUNDSS(DisasContext *s, X86DecodedInsn *decode)
4599{
4600    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4601    assert(!s->vex_l);
4602    gen_helper_roundss_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
4603}
4604
4605static void gen_VSHUF(DisasContext *s, X86DecodedInsn *decode)
4606{
4607    TCGv_i32 imm = tcg_constant_i32(decode->immediate);
4608    SSEFunc_0_pppi ps, pd, fn;
4609    ps = s->vex_l ? gen_helper_shufps_ymm : gen_helper_shufps_xmm;
4610    pd = s->vex_l ? gen_helper_shufpd_ymm : gen_helper_shufpd_xmm;
4611    fn = s->prefix & PREFIX_DATA ? pd : ps;
4612    fn(OP_PTR0, OP_PTR1, OP_PTR2, imm);
4613}
4614
4615static void gen_VUCOMI(DisasContext *s, X86DecodedInsn *decode)
4616{
4617    SSEFunc_0_epp fn;
4618    fn = s->prefix & PREFIX_DATA ? gen_helper_ucomisd : gen_helper_ucomiss;
4619    fn(tcg_env, OP_PTR1, OP_PTR2);
4620    assume_cc_op(s, CC_OP_EFLAGS);
4621}
4622
4623static void gen_VZEROALL(DisasContext *s, X86DecodedInsn *decode)
4624{
4625    TCGv_ptr ptr = tcg_temp_new_ptr();
4626
4627    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_regs));
4628    gen_helper_memset(ptr, ptr, tcg_constant_i32(0),
4629                      tcg_constant_ptr(CPU_NB_REGS * sizeof(ZMMReg)));
4630}
4631
4632static void gen_VZEROUPPER(DisasContext *s, X86DecodedInsn *decode)
4633{
4634    int i;
4635
4636    for (i = 0; i < CPU_NB_REGS; i++) {
4637        int offset = offsetof(CPUX86State, xmm_regs[i].ZMM_X(1));
4638        tcg_gen_gvec_dup_imm(MO_64, offset, 16, 16, 0);
4639    }
4640}
4641
4642static void gen_WAIT(DisasContext *s, X86DecodedInsn *decode)
4643{
4644    if ((s->flags & (HF_MP_MASK | HF_TS_MASK)) == (HF_MP_MASK | HF_TS_MASK)) {
4645        gen_NM_exception(s);
4646    } else {
4647        /* needs to be treated as I/O because of ferr_irq */
4648        translator_io_start(&s->base);
4649        gen_helper_fwait(tcg_env);
4650    }
4651}
4652
4653#ifndef CONFIG_USER_ONLY
4654static void gen_WRMSR(DisasContext *s, X86DecodedInsn *decode)
4655{
4656    gen_update_cc_op(s);
4657    gen_update_eip_cur(s);
4658    gen_helper_wrmsr(tcg_env);
4659    s->base.is_jmp = DISAS_EOB_NEXT;
4660}
4661#else
4662#define gen_WRMSR gen_unreachable
4663#endif
4664
4665static void gen_WRxxBASE(DisasContext *s, X86DecodedInsn *decode)
4666{
4667    TCGv base = cpu_seg_base[s->modrm & 8 ? R_GS : R_FS];
4668
4669    /* Preserve hflags bits by testing CR4 at runtime.  */
4670    gen_helper_cr4_testbit(tcg_env, tcg_constant_i32(CR4_FSGSBASE_MASK));
4671    tcg_gen_mov_tl(base, s->T0);
4672}
4673
4674static void gen_XADD(DisasContext *s, X86DecodedInsn *decode)
4675{
4676    MemOp ot = decode->op[1].ot;
4677
4678    decode->cc_dst = tcg_temp_new();
4679    decode->cc_src = s->T1;
4680    decode->cc_op = CC_OP_ADDB + ot;
4681
4682    if (s->prefix & PREFIX_LOCK) {
4683        tcg_gen_atomic_fetch_add_tl(s->T0, s->A0, s->T1, s->mem_index, ot | MO_LE);
4684        tcg_gen_add_tl(decode->cc_dst, s->T0, s->T1);
4685    } else {
4686        tcg_gen_add_tl(decode->cc_dst, s->T0, s->T1);
4687        /*
4688         * NOTE: writing memory first is important for MMU exceptions,
4689         * but "new result" wins for XADD AX, AX.
4690         */
4691        gen_writeback(s, decode, 0, decode->cc_dst);
4692    }
4693    if (decode->op[0].has_ea || decode->op[2].n != decode->op[0].n) {
4694        gen_writeback(s, decode, 2, s->T0);
4695    }
4696}
4697
4698static void gen_XCHG(DisasContext *s, X86DecodedInsn *decode)
4699{
4700    if (s->prefix & PREFIX_LOCK) {
4701        tcg_gen_atomic_xchg_tl(s->T0, s->A0, s->T1,
4702                               s->mem_index, decode->op[0].ot | MO_LE);
4703        /* now store old value into register operand */
4704        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4705    } else {
4706        /* move destination value into source operand, source preserved in T1 */
4707        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4708        tcg_gen_mov_tl(s->T0, s->T1);
4709    }
4710}
4711
4712static void gen_XLAT(DisasContext *s, X86DecodedInsn *decode)
4713{
4714    /* AL is already zero-extended into s->T0.  */
4715    tcg_gen_add_tl(s->A0, cpu_regs[R_EBX], s->T0);
4716    gen_lea_v_seg(s, s->A0, R_DS, s->override);
4717    gen_op_ld_v(s, MO_8, s->T0, s->A0);
4718}
4719
4720static void gen_XOR(DisasContext *s, X86DecodedInsn *decode)
4721{
4722    /* special case XOR reg, reg */
4723    if (decode->op[1].unit == X86_OP_INT &&
4724        decode->op[2].unit == X86_OP_INT &&
4725        decode->op[1].n == decode->op[2].n) {
4726        tcg_gen_movi_tl(s->T0, 0);
4727        decode->cc_op = CC_OP_CLR;
4728    } else {
4729        MemOp ot = decode->op[1].ot;
4730
4731        if (s->prefix & PREFIX_LOCK) {
4732            tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T1,
4733                                        s->mem_index, ot | MO_LE);
4734        } else {
4735            tcg_gen_xor_tl(s->T0, s->T0, s->T1);
4736        }
4737        prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
4738    }
4739}
4740
4741static void gen_XRSTOR(DisasContext *s, X86DecodedInsn *decode)
4742{
4743    TCGv_i64 features = tcg_temp_new_i64();
4744
4745    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4746    gen_helper_xrstor(tcg_env, s->A0, features);
4747    if (s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_MPX) {
4748        /*
4749         * XRSTOR is how MPX is enabled, which changes how
4750         * we translate.  Thus we need to end the TB.
4751         */
4752        s->base.is_jmp = DISAS_EOB_NEXT;
4753    }
4754}
4755
4756static void gen_XSAVE(DisasContext *s, X86DecodedInsn *decode)
4757{
4758    TCGv_i64 features = tcg_temp_new_i64();
4759
4760    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4761    gen_helper_xsave(tcg_env, s->A0, features);
4762}
4763
4764static void gen_XSAVEOPT(DisasContext *s, X86DecodedInsn *decode)
4765{
4766    TCGv_i64 features = tcg_temp_new_i64();
4767
4768    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4769    gen_helper_xsave(tcg_env, s->A0, features);
4770}
4771