xref: /openbmc/qemu/target/i386/tcg/emit.c.inc (revision dc1ed8f256c446cbf33e090f0e214d0311a771a7)
1/*
2 * New-style TCG opcode generator for i386 instructions
3 *
4 *  Copyright (c) 2022 Red Hat, Inc.
5 *
6 * Author: Paolo Bonzini <pbonzini@redhat.com>
7 *
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 */
21
22#define MMX_OFFSET(reg)                        \
23  ({ assert((reg) >= 0 && (reg) <= 7);         \
24     offsetof(CPUX86State, fpregs[reg].mmx); })
25
26#define ZMM_OFFSET(reg)                        \
27  ({ assert((reg) >= 0 && (reg) <= 15);        \
28     offsetof(CPUX86State, xmm_regs[reg]); })
29
30typedef void (*SSEFunc_i_ep)(TCGv_i32 val, TCGv_ptr env, TCGv_ptr reg);
31typedef void (*SSEFunc_l_ep)(TCGv_i64 val, TCGv_ptr env, TCGv_ptr reg);
32typedef void (*SSEFunc_0_epp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b);
33typedef void (*SSEFunc_0_eppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
34                               TCGv_ptr reg_c);
35typedef void (*SSEFunc_0_epppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
36                                TCGv_ptr reg_c, TCGv_ptr reg_d);
37typedef void (*SSEFunc_0_eppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
38                               TCGv_i32 val);
39typedef void (*SSEFunc_0_epppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
40                                TCGv_ptr reg_c, TCGv_i32 val);
41typedef void (*SSEFunc_0_ppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_i32 val);
42typedef void (*SSEFunc_0_pppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_ptr reg_c,
43                               TCGv_i32 val);
44typedef void (*SSEFunc_0_eppt)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
45                               TCGv val);
46typedef void (*SSEFunc_0_epppti)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
47                                 TCGv_ptr reg_c, TCGv a0, TCGv_i32 scale);
48typedef void (*SSEFunc_0_eppppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
49                                  TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 flags);
50typedef void (*SSEFunc_0_eppppii)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
51                                  TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 even,
52                                  TCGv_i32 odd);
53
54static void gen_JMP_m(DisasContext *s, X86DecodedInsn *decode);
55static void gen_JMP(DisasContext *s, X86DecodedInsn *decode);
56
57static inline TCGv_i32 tcg_constant8u_i32(uint8_t val)
58{
59    return tcg_constant_i32(val);
60}
61
62static void gen_NM_exception(DisasContext *s)
63{
64    gen_exception(s, EXCP07_PREX);
65}
66
67static void gen_lea_modrm(DisasContext *s, X86DecodedInsn *decode)
68{
69    AddressParts *mem = &decode->mem;
70    TCGv ea;
71
72    ea = gen_lea_modrm_1(s, *mem, decode->e.vex_class == 12);
73    if (decode->e.special == X86_SPECIAL_BitTest) {
74        MemOp ot = decode->op[1].ot;
75        int poslen = 8 << ot;
76        int opn = decode->op[2].n;
77        TCGv ofs = tcg_temp_new();
78
79        /* Extract memory displacement from the second operand.  */
80        assert(decode->op[2].unit == X86_OP_INT && decode->op[2].ot != MO_8);
81        tcg_gen_sextract_tl(ofs, cpu_regs[opn], 3, poslen - 3);
82        tcg_gen_andi_tl(ofs, ofs, -1 << ot);
83        tcg_gen_add_tl(s->A0, ea, ofs);
84        ea = s->A0;
85    }
86
87    gen_lea_v_seg(s, ea, mem->def_seg, s->override);
88}
89
90static inline int mmx_offset(MemOp ot)
91{
92    switch (ot) {
93    case MO_8:
94        return offsetof(MMXReg, MMX_B(0));
95    case MO_16:
96        return offsetof(MMXReg, MMX_W(0));
97    case MO_32:
98        return offsetof(MMXReg, MMX_L(0));
99    case MO_64:
100        return offsetof(MMXReg, MMX_Q(0));
101    default:
102        g_assert_not_reached();
103    }
104}
105
106static inline int xmm_offset(MemOp ot)
107{
108    switch (ot) {
109    case MO_8:
110        return offsetof(ZMMReg, ZMM_B(0));
111    case MO_16:
112        return offsetof(ZMMReg, ZMM_W(0));
113    case MO_32:
114        return offsetof(ZMMReg, ZMM_L(0));
115    case MO_64:
116        return offsetof(ZMMReg, ZMM_Q(0));
117    case MO_128:
118        return offsetof(ZMMReg, ZMM_X(0));
119    case MO_256:
120        return offsetof(ZMMReg, ZMM_Y(0));
121    default:
122        g_assert_not_reached();
123    }
124}
125
126static int vector_reg_offset(X86DecodedOp *op)
127{
128    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
129
130    if (op->unit == X86_OP_MMX) {
131        return op->offset - mmx_offset(op->ot);
132    } else {
133        return op->offset - xmm_offset(op->ot);
134    }
135}
136
137static int vector_elem_offset(X86DecodedOp *op, MemOp ot, int n)
138{
139    int base_ofs = vector_reg_offset(op);
140    switch(ot) {
141    case MO_8:
142        if (op->unit == X86_OP_MMX) {
143            return base_ofs + offsetof(MMXReg, MMX_B(n));
144        } else {
145            return base_ofs + offsetof(ZMMReg, ZMM_B(n));
146        }
147    case MO_16:
148        if (op->unit == X86_OP_MMX) {
149            return base_ofs + offsetof(MMXReg, MMX_W(n));
150        } else {
151            return base_ofs + offsetof(ZMMReg, ZMM_W(n));
152        }
153    case MO_32:
154        if (op->unit == X86_OP_MMX) {
155            return base_ofs + offsetof(MMXReg, MMX_L(n));
156        } else {
157            return base_ofs + offsetof(ZMMReg, ZMM_L(n));
158        }
159    case MO_64:
160        if (op->unit == X86_OP_MMX) {
161            return base_ofs;
162        } else {
163            return base_ofs + offsetof(ZMMReg, ZMM_Q(n));
164        }
165    case MO_128:
166        assert(op->unit == X86_OP_SSE);
167        return base_ofs + offsetof(ZMMReg, ZMM_X(n));
168    case MO_256:
169        assert(op->unit == X86_OP_SSE);
170        return base_ofs + offsetof(ZMMReg, ZMM_Y(n));
171    default:
172        g_assert_not_reached();
173    }
174}
175
176static void compute_mmx_offset(X86DecodedOp *op)
177{
178    if (!op->has_ea) {
179        op->offset = MMX_OFFSET(op->n) + mmx_offset(op->ot);
180    } else {
181        op->offset = offsetof(CPUX86State, mmx_t0) + mmx_offset(op->ot);
182    }
183}
184
185static void compute_xmm_offset(X86DecodedOp *op)
186{
187    if (!op->has_ea) {
188        op->offset = ZMM_OFFSET(op->n) + xmm_offset(op->ot);
189    } else {
190        op->offset = offsetof(CPUX86State, xmm_t0) + xmm_offset(op->ot);
191    }
192}
193
194static void gen_load_sse(DisasContext *s, TCGv temp, MemOp ot, int dest_ofs, bool aligned)
195{
196    switch(ot) {
197    case MO_8:
198        gen_op_ld_v(s, MO_8, temp, s->A0);
199        tcg_gen_st8_tl(temp, tcg_env, dest_ofs);
200        break;
201    case MO_16:
202        gen_op_ld_v(s, MO_16, temp, s->A0);
203        tcg_gen_st16_tl(temp, tcg_env, dest_ofs);
204        break;
205    case MO_32:
206        gen_op_ld_v(s, MO_32, temp, s->A0);
207        tcg_gen_st32_tl(temp, tcg_env, dest_ofs);
208        break;
209    case MO_64:
210        gen_ldq_env_A0(s, dest_ofs);
211        break;
212    case MO_128:
213        gen_ldo_env_A0(s, dest_ofs, aligned);
214        break;
215    case MO_256:
216        gen_ldy_env_A0(s, dest_ofs, aligned);
217        break;
218    default:
219        g_assert_not_reached();
220    }
221}
222
223static bool sse_needs_alignment(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
224{
225    switch (decode->e.vex_class) {
226    case 2:
227    case 4:
228        if ((s->prefix & PREFIX_VEX) ||
229            decode->e.vex_special == X86_VEX_SSEUnaligned) {
230            /* MOST legacy SSE instructions require aligned memory operands, but not all.  */
231            return false;
232        }
233        /* fall through */
234    case 1:
235        return ot >= MO_128;
236
237    default:
238        return false;
239    }
240}
241
242static void gen_load(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
243{
244    X86DecodedOp *op = &decode->op[opn];
245
246    switch (op->unit) {
247    case X86_OP_SKIP:
248        return;
249    case X86_OP_SEG:
250        tcg_gen_ld32u_tl(v, tcg_env,
251                         offsetof(CPUX86State,segs[op->n].selector));
252        break;
253#ifndef CONFIG_USER_ONLY
254    case X86_OP_CR:
255        if (op->n == 8) {
256            translator_io_start(&s->base);
257            gen_helper_read_cr8(v, tcg_env);
258        } else {
259            tcg_gen_ld_tl(v, tcg_env, offsetof(CPUX86State, cr[op->n]));
260        }
261        break;
262    case X86_OP_DR:
263        /* CR4.DE tested in the helper.  */
264        gen_helper_get_dr(v, tcg_env, tcg_constant_i32(op->n));
265        break;
266#endif
267    case X86_OP_INT:
268        if (op->has_ea) {
269            if (v == s->T0 && decode->e.special == X86_SPECIAL_SExtT0) {
270                gen_op_ld_v(s, op->ot | MO_SIGN, v, s->A0);
271            } else {
272                gen_op_ld_v(s, op->ot, v, s->A0);
273            }
274
275        } else if (op->ot < MO_TL && v == s->T0 &&
276                   (decode->e.special == X86_SPECIAL_SExtT0 ||
277                    decode->e.special == X86_SPECIAL_ZExtT0)) {
278            if (op->ot == MO_8 && byte_reg_is_xH(s, op->n)) {
279                if (decode->e.special == X86_SPECIAL_SExtT0) {
280                    tcg_gen_sextract_tl(v, cpu_regs[op->n - 4], 8, 8);
281                } else {
282                    tcg_gen_extract_tl(v, cpu_regs[op->n - 4], 8, 8);
283                }
284            } else {
285                if (decode->e.special == X86_SPECIAL_SExtT0) {
286                    tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot | MO_SIGN);
287                } else {
288                    tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot);
289                }
290            }
291
292        } else {
293            gen_op_mov_v_reg(s, op->ot, v, op->n);
294        }
295        break;
296    case X86_OP_IMM:
297        tcg_gen_movi_tl(v, op->imm);
298        break;
299
300    case X86_OP_MMX:
301        compute_mmx_offset(op);
302        goto load_vector;
303
304    case X86_OP_SSE:
305        compute_xmm_offset(op);
306    load_vector:
307        if (op->has_ea) {
308            bool aligned = sse_needs_alignment(s, decode, op->ot);
309            gen_load_sse(s, v, op->ot, op->offset, aligned);
310        }
311        break;
312
313    default:
314        g_assert_not_reached();
315    }
316}
317
318static TCGv_ptr op_ptr(X86DecodedInsn *decode, int opn)
319{
320    X86DecodedOp *op = &decode->op[opn];
321
322    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
323    if (op->v_ptr) {
324        return op->v_ptr;
325    }
326    op->v_ptr = tcg_temp_new_ptr();
327
328    /* The temporary points to the MMXReg or ZMMReg.  */
329    tcg_gen_addi_ptr(op->v_ptr, tcg_env, vector_reg_offset(op));
330    return op->v_ptr;
331}
332
333#define OP_PTR0 op_ptr(decode, 0)
334#define OP_PTR1 op_ptr(decode, 1)
335#define OP_PTR2 op_ptr(decode, 2)
336
337static void gen_writeback(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
338{
339    X86DecodedOp *op = &decode->op[opn];
340    switch (op->unit) {
341    case X86_OP_SKIP:
342        break;
343    case X86_OP_SEG:
344        /* Note that gen_movl_seg takes care of interrupt shadow and TF.  */
345        gen_movl_seg(s, op->n, v, op->n == R_SS);
346        break;
347    case X86_OP_INT:
348        if (op->has_ea) {
349            gen_op_st_v(s, op->ot, v, s->A0);
350        } else {
351            gen_op_mov_reg_v(s, op->ot, op->n, v);
352        }
353        break;
354    case X86_OP_MMX:
355        break;
356    case X86_OP_SSE:
357        if (!op->has_ea && (s->prefix & PREFIX_VEX) && op->ot <= MO_128) {
358            tcg_gen_gvec_dup_imm(MO_64,
359                                 offsetof(CPUX86State, xmm_regs[op->n].ZMM_X(1)),
360                                 16, 16, 0);
361        }
362        break;
363#ifndef CONFIG_USER_ONLY
364    case X86_OP_CR:
365        if (op->n == 8) {
366            translator_io_start(&s->base);
367        }
368        gen_helper_write_crN(tcg_env, tcg_constant_i32(op->n), v);
369        s->base.is_jmp = DISAS_EOB_NEXT;
370        break;
371    case X86_OP_DR:
372        /* CR4.DE tested in the helper.  */
373        gen_helper_set_dr(tcg_env, tcg_constant_i32(op->n), v);
374        s->base.is_jmp = DISAS_EOB_NEXT;
375        break;
376#endif
377    default:
378        g_assert_not_reached();
379    }
380    op->unit = X86_OP_SKIP;
381}
382
383static inline int vector_len(DisasContext *s, X86DecodedInsn *decode)
384{
385    if (decode->e.special == X86_SPECIAL_MMX &&
386        !(s->prefix & (PREFIX_DATA | PREFIX_REPZ | PREFIX_REPNZ))) {
387        return 8;
388    }
389    return s->vex_l ? 32 : 16;
390}
391
392static void prepare_update1_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
393{
394    decode->cc_dst = s->T0;
395    decode->cc_op = op;
396}
397
398static void prepare_update2_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
399{
400    decode->cc_src = s->T1;
401    decode->cc_dst = s->T0;
402    decode->cc_op = op;
403}
404
405static void prepare_update_cc_incdec(X86DecodedInsn *decode, DisasContext *s, CCOp op)
406{
407    gen_compute_eflags_c(s, s->T1);
408    prepare_update2_cc(decode, s, op);
409}
410
411static void prepare_update3_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op, TCGv reg)
412{
413    decode->cc_src2 = reg;
414    decode->cc_src = s->T1;
415    decode->cc_dst = s->T0;
416    decode->cc_op = op;
417}
418
419/* Set up decode->cc_* to modify CF while keeping other flags unchanged.  */
420static void prepare_update_cf(X86DecodedInsn *decode, DisasContext *s, TCGv cf)
421{
422    switch (s->cc_op) {
423    case CC_OP_ADOX:
424    case CC_OP_ADCOX:
425        decode->cc_src2 = cpu_cc_src2;
426        decode->cc_src = cpu_cc_src;
427        decode->cc_op = CC_OP_ADCOX;
428        break;
429
430    case CC_OP_EFLAGS:
431    case CC_OP_ADCX:
432        decode->cc_src = cpu_cc_src;
433        decode->cc_op = CC_OP_ADCX;
434        break;
435
436    default:
437        decode->cc_src = tcg_temp_new();
438        gen_mov_eflags(s, decode->cc_src);
439        decode->cc_op = CC_OP_ADCX;
440        break;
441    }
442    decode->cc_dst = cf;
443}
444
445static void gen_store_sse(DisasContext *s, X86DecodedInsn *decode, int src_ofs)
446{
447    MemOp ot = decode->op[0].ot;
448    int vec_len = vector_len(s, decode);
449    bool aligned = sse_needs_alignment(s, decode, ot);
450
451    if (!decode->op[0].has_ea) {
452        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, vec_len, vec_len);
453        return;
454    }
455
456    switch (ot) {
457    case MO_64:
458        gen_stq_env_A0(s, src_ofs);
459        break;
460    case MO_128:
461        gen_sto_env_A0(s, src_ofs, aligned);
462        break;
463    case MO_256:
464        gen_sty_env_A0(s, src_ofs, aligned);
465        break;
466    default:
467        g_assert_not_reached();
468    }
469}
470
471static void gen_helper_pavgusb(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b)
472{
473    gen_helper_pavgb_mmx(env, reg_a, reg_a, reg_b);
474}
475
476#define FN_3DNOW_MOVE ((SSEFunc_0_epp) (uintptr_t) 1)
477static const SSEFunc_0_epp fns_3dnow[] = {
478    [0x0c] = gen_helper_pi2fw,
479    [0x0d] = gen_helper_pi2fd,
480    [0x1c] = gen_helper_pf2iw,
481    [0x1d] = gen_helper_pf2id,
482    [0x8a] = gen_helper_pfnacc,
483    [0x8e] = gen_helper_pfpnacc,
484    [0x90] = gen_helper_pfcmpge,
485    [0x94] = gen_helper_pfmin,
486    [0x96] = gen_helper_pfrcp,
487    [0x97] = gen_helper_pfrsqrt,
488    [0x9a] = gen_helper_pfsub,
489    [0x9e] = gen_helper_pfadd,
490    [0xa0] = gen_helper_pfcmpgt,
491    [0xa4] = gen_helper_pfmax,
492    [0xa6] = FN_3DNOW_MOVE, /* PFRCPIT1; no need to actually increase precision */
493    [0xa7] = FN_3DNOW_MOVE, /* PFRSQIT1 */
494    [0xb6] = FN_3DNOW_MOVE, /* PFRCPIT2 */
495    [0xaa] = gen_helper_pfsubr,
496    [0xae] = gen_helper_pfacc,
497    [0xb0] = gen_helper_pfcmpeq,
498    [0xb4] = gen_helper_pfmul,
499    [0xb7] = gen_helper_pmulhrw_mmx,
500    [0xbb] = gen_helper_pswapd,
501    [0xbf] = gen_helper_pavgusb,
502};
503
504static void gen_3dnow(DisasContext *s, X86DecodedInsn *decode)
505{
506    uint8_t b = decode->immediate;
507    SSEFunc_0_epp fn = b < ARRAY_SIZE(fns_3dnow) ? fns_3dnow[b] : NULL;
508
509    if (!fn) {
510        gen_illegal_opcode(s);
511        return;
512    }
513    if (s->flags & HF_TS_MASK) {
514        gen_NM_exception(s);
515        return;
516    }
517    if (s->flags & HF_EM_MASK) {
518        gen_illegal_opcode(s);
519        return;
520    }
521
522    gen_helper_enter_mmx(tcg_env);
523    if (fn == FN_3DNOW_MOVE) {
524       tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset);
525       tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset);
526    } else {
527       fn(tcg_env, OP_PTR0, OP_PTR1);
528    }
529}
530
531/*
532 * 00 = v*ps Vps, Hps, Wpd
533 * 66 = v*pd Vpd, Hpd, Wps
534 * f3 = v*ss Vss, Hss, Wps
535 * f2 = v*sd Vsd, Hsd, Wps
536 */
537static inline void gen_unary_fp_sse(DisasContext *s, X86DecodedInsn *decode,
538                              SSEFunc_0_epp pd_xmm, SSEFunc_0_epp ps_xmm,
539                              SSEFunc_0_epp pd_ymm, SSEFunc_0_epp ps_ymm,
540                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
541{
542    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
543        SSEFunc_0_eppp fn = s->prefix & PREFIX_REPZ ? ss : sd;
544        if (!fn) {
545            gen_illegal_opcode(s);
546            return;
547        }
548        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
549    } else {
550        SSEFunc_0_epp ps, pd, fn;
551        ps = s->vex_l ? ps_ymm : ps_xmm;
552        pd = s->vex_l ? pd_ymm : pd_xmm;
553        fn = s->prefix & PREFIX_DATA ? pd : ps;
554        if (!fn) {
555            gen_illegal_opcode(s);
556            return;
557        }
558        fn(tcg_env, OP_PTR0, OP_PTR2);
559    }
560}
561#define UNARY_FP_SSE(uname, lname)                                                 \
562static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
563{                                                                                  \
564    gen_unary_fp_sse(s, decode,                                                    \
565                     gen_helper_##lname##pd_xmm,                                   \
566                     gen_helper_##lname##ps_xmm,                                   \
567                     gen_helper_##lname##pd_ymm,                                   \
568                     gen_helper_##lname##ps_ymm,                                   \
569                     gen_helper_##lname##sd,                                       \
570                     gen_helper_##lname##ss);                                      \
571}
572UNARY_FP_SSE(VSQRT, sqrt)
573
574/*
575 * 00 = v*ps Vps, Hps, Wpd
576 * 66 = v*pd Vpd, Hpd, Wps
577 * f3 = v*ss Vss, Hss, Wps
578 * f2 = v*sd Vsd, Hsd, Wps
579 */
580static inline void gen_fp_sse(DisasContext *s, X86DecodedInsn *decode,
581                              SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
582                              SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm,
583                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
584{
585    SSEFunc_0_eppp ps, pd, fn;
586    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
587        fn = s->prefix & PREFIX_REPZ ? ss : sd;
588    } else {
589        ps = s->vex_l ? ps_ymm : ps_xmm;
590        pd = s->vex_l ? pd_ymm : pd_xmm;
591        fn = s->prefix & PREFIX_DATA ? pd : ps;
592    }
593    if (fn) {
594        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
595    } else {
596        gen_illegal_opcode(s);
597    }
598}
599
600#define FP_SSE(uname, lname)                                                       \
601static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
602{                                                                                  \
603    gen_fp_sse(s, decode,                                                          \
604               gen_helper_##lname##pd_xmm,                                         \
605               gen_helper_##lname##ps_xmm,                                         \
606               gen_helper_##lname##pd_ymm,                                         \
607               gen_helper_##lname##ps_ymm,                                         \
608               gen_helper_##lname##sd,                                             \
609               gen_helper_##lname##ss);                                            \
610}
611FP_SSE(VADD, add)
612FP_SSE(VMUL, mul)
613FP_SSE(VSUB, sub)
614FP_SSE(VMIN, min)
615FP_SSE(VDIV, div)
616FP_SSE(VMAX, max)
617
618#define FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, even, odd)                         \
619static void gen_##uname##Px(DisasContext *s, X86DecodedInsn *decode)               \
620{                                                                                  \
621    SSEFunc_0_eppppii xmm = s->vex_w ? gen_helper_fma4pd_xmm : gen_helper_fma4ps_xmm; \
622    SSEFunc_0_eppppii ymm = s->vex_w ? gen_helper_fma4pd_ymm : gen_helper_fma4ps_ymm; \
623    SSEFunc_0_eppppii fn = s->vex_l ? ymm : xmm;                                   \
624                                                                                   \
625    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
626       tcg_constant_i32(even),                                                     \
627       tcg_constant_i32((even) ^ (odd)));                                          \
628}
629
630#define FMA_SSE(uname, ptr0, ptr1, ptr2, flags)                                    \
631FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, flags, flags)                              \
632static void gen_##uname##Sx(DisasContext *s, X86DecodedInsn *decode)               \
633{                                                                                  \
634    SSEFunc_0_eppppi fn = s->vex_w ? gen_helper_fma4sd : gen_helper_fma4ss;        \
635                                                                                   \
636    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
637       tcg_constant_i32(flags));                                                   \
638}                                                                                  \
639
640FMA_SSE(VFMADD231,  OP_PTR1, OP_PTR2, OP_PTR0, 0)
641FMA_SSE(VFMADD213,  OP_PTR1, OP_PTR0, OP_PTR2, 0)
642FMA_SSE(VFMADD132,  OP_PTR0, OP_PTR2, OP_PTR1, 0)
643
644FMA_SSE(VFNMADD231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_product)
645FMA_SSE(VFNMADD213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_product)
646FMA_SSE(VFNMADD132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_product)
647
648FMA_SSE(VFMSUB231,  OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c)
649FMA_SSE(VFMSUB213,  OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c)
650FMA_SSE(VFMSUB132,  OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c)
651
652FMA_SSE(VFNMSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c|float_muladd_negate_product)
653FMA_SSE(VFNMSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c|float_muladd_negate_product)
654FMA_SSE(VFNMSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c|float_muladd_negate_product)
655
656FMA_SSE_PACKED(VFMADDSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c, 0)
657FMA_SSE_PACKED(VFMADDSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c, 0)
658FMA_SSE_PACKED(VFMADDSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c, 0)
659
660FMA_SSE_PACKED(VFMSUBADD231, OP_PTR1, OP_PTR2, OP_PTR0, 0, float_muladd_negate_c)
661FMA_SSE_PACKED(VFMSUBADD213, OP_PTR1, OP_PTR0, OP_PTR2, 0, float_muladd_negate_c)
662FMA_SSE_PACKED(VFMSUBADD132, OP_PTR0, OP_PTR2, OP_PTR1, 0, float_muladd_negate_c)
663
664#define FP_UNPACK_SSE(uname, lname)                                                \
665static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
666{                                                                                  \
667    /* PS maps to the DQ integer instruction, PD maps to QDQ.  */                  \
668    gen_fp_sse(s, decode,                                                          \
669               gen_helper_##lname##qdq_xmm,                                        \
670               gen_helper_##lname##dq_xmm,                                         \
671               gen_helper_##lname##qdq_ymm,                                        \
672               gen_helper_##lname##dq_ymm,                                         \
673               NULL, NULL);                                                        \
674}
675FP_UNPACK_SSE(VUNPCKLPx, punpckl)
676FP_UNPACK_SSE(VUNPCKHPx, punpckh)
677
678/*
679 * 00 = v*ps Vps, Wpd
680 * f3 = v*ss Vss, Wps
681 */
682static inline void gen_unary_fp32_sse(DisasContext *s, X86DecodedInsn *decode,
683                                      SSEFunc_0_epp ps_xmm,
684                                      SSEFunc_0_epp ps_ymm,
685                                      SSEFunc_0_eppp ss)
686{
687    if ((s->prefix & (PREFIX_DATA | PREFIX_REPNZ)) != 0) {
688        goto illegal_op;
689    } else if (s->prefix & PREFIX_REPZ) {
690        if (!ss) {
691            goto illegal_op;
692        }
693        ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
694    } else {
695        SSEFunc_0_epp fn = s->vex_l ? ps_ymm : ps_xmm;
696        if (!fn) {
697            goto illegal_op;
698        }
699        fn(tcg_env, OP_PTR0, OP_PTR2);
700    }
701    return;
702
703illegal_op:
704    gen_illegal_opcode(s);
705}
706#define UNARY_FP32_SSE(uname, lname)                                               \
707static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
708{                                                                                  \
709    gen_unary_fp32_sse(s, decode,                                                  \
710                       gen_helper_##lname##ps_xmm,                                 \
711                       gen_helper_##lname##ps_ymm,                                 \
712                       gen_helper_##lname##ss);                                    \
713}
714UNARY_FP32_SSE(VRSQRT, rsqrt)
715UNARY_FP32_SSE(VRCP, rcp)
716
717/*
718 * 66 = v*pd Vpd, Hpd, Wpd
719 * f2 = v*ps Vps, Hps, Wps
720 */
721static inline void gen_horizontal_fp_sse(DisasContext *s, X86DecodedInsn *decode,
722                                         SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
723                                         SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm)
724{
725    SSEFunc_0_eppp ps, pd, fn;
726    ps = s->vex_l ? ps_ymm : ps_xmm;
727    pd = s->vex_l ? pd_ymm : pd_xmm;
728    fn = s->prefix & PREFIX_DATA ? pd : ps;
729    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
730}
731#define HORIZONTAL_FP_SSE(uname, lname)                                            \
732static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
733{                                                                                  \
734    gen_horizontal_fp_sse(s, decode,                                               \
735                          gen_helper_##lname##pd_xmm, gen_helper_##lname##ps_xmm,  \
736                          gen_helper_##lname##pd_ymm, gen_helper_##lname##ps_ymm); \
737}
738HORIZONTAL_FP_SSE(VHADD, hadd)
739HORIZONTAL_FP_SSE(VHSUB, hsub)
740HORIZONTAL_FP_SSE(VADDSUB, addsub)
741
742static inline void gen_ternary_sse(DisasContext *s, X86DecodedInsn *decode,
743                                   int op3, SSEFunc_0_epppp xmm, SSEFunc_0_epppp ymm)
744{
745    SSEFunc_0_epppp fn = s->vex_l ? ymm : xmm;
746    TCGv_ptr ptr3 = tcg_temp_new_ptr();
747
748    /* The format of the fourth input is Lx */
749    tcg_gen_addi_ptr(ptr3, tcg_env, ZMM_OFFSET(op3));
750    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, ptr3);
751}
752#define TERNARY_SSE(uname, uvname, lname)                                          \
753static void gen_##uvname(DisasContext *s, X86DecodedInsn *decode)                  \
754{                                                                                  \
755    gen_ternary_sse(s, decode, (uint8_t)decode->immediate >> 4,                    \
756                    gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);           \
757}                                                                                  \
758static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
759{                                                                                  \
760    gen_ternary_sse(s, decode, 0,                                                  \
761                  gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);             \
762}
763TERNARY_SSE(BLENDVPS, VBLENDVPS, blendvps)
764TERNARY_SSE(BLENDVPD, VBLENDVPD, blendvpd)
765TERNARY_SSE(PBLENDVB, VPBLENDVB, pblendvb)
766
767static inline void gen_binary_imm_sse(DisasContext *s, X86DecodedInsn *decode,
768                                      SSEFunc_0_epppi xmm, SSEFunc_0_epppi ymm)
769{
770    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
771    if (!s->vex_l) {
772        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
773    } else {
774        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
775    }
776}
777
778#define BINARY_IMM_SSE(uname, lname)                                               \
779static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
780{                                                                                  \
781    gen_binary_imm_sse(s, decode,                                                  \
782                       gen_helper_##lname##_xmm,                                   \
783                       gen_helper_##lname##_ymm);                                  \
784}
785
786BINARY_IMM_SSE(VBLENDPD,   blendpd)
787BINARY_IMM_SSE(VBLENDPS,   blendps)
788BINARY_IMM_SSE(VPBLENDW,   pblendw)
789BINARY_IMM_SSE(VDDPS,      dpps)
790#define gen_helper_dppd_ymm NULL
791BINARY_IMM_SSE(VDDPD,      dppd)
792BINARY_IMM_SSE(VMPSADBW,   mpsadbw)
793BINARY_IMM_SSE(PCLMULQDQ,  pclmulqdq)
794
795
796#define UNARY_INT_GVEC(uname, func, ...)                                           \
797static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
798{                                                                                  \
799    int vec_len = vector_len(s, decode);                                          \
800                                                                                   \
801    func(__VA_ARGS__, decode->op[0].offset,                                        \
802         decode->op[2].offset, vec_len, vec_len);                                  \
803}
804UNARY_INT_GVEC(PABSB,          tcg_gen_gvec_abs, MO_8)
805UNARY_INT_GVEC(PABSW,          tcg_gen_gvec_abs, MO_16)
806UNARY_INT_GVEC(PABSD,          tcg_gen_gvec_abs, MO_32)
807UNARY_INT_GVEC(VBROADCASTx128, tcg_gen_gvec_dup_mem, MO_128)
808UNARY_INT_GVEC(VPBROADCASTB,   tcg_gen_gvec_dup_mem, MO_8)
809UNARY_INT_GVEC(VPBROADCASTW,   tcg_gen_gvec_dup_mem, MO_16)
810UNARY_INT_GVEC(VPBROADCASTD,   tcg_gen_gvec_dup_mem, MO_32)
811UNARY_INT_GVEC(VPBROADCASTQ,   tcg_gen_gvec_dup_mem, MO_64)
812
813
814#define BINARY_INT_GVEC(uname, func, ...)                                          \
815static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
816{                                                                                  \
817    int vec_len = vector_len(s, decode);                                          \
818                                                                                   \
819    func(__VA_ARGS__,                                                              \
820         decode->op[0].offset, decode->op[1].offset,                               \
821         decode->op[2].offset, vec_len, vec_len);                                  \
822}
823
824BINARY_INT_GVEC(PADDB,   tcg_gen_gvec_add, MO_8)
825BINARY_INT_GVEC(PADDW,   tcg_gen_gvec_add, MO_16)
826BINARY_INT_GVEC(PADDD,   tcg_gen_gvec_add, MO_32)
827BINARY_INT_GVEC(PADDQ,   tcg_gen_gvec_add, MO_64)
828BINARY_INT_GVEC(PADDSB,  tcg_gen_gvec_ssadd, MO_8)
829BINARY_INT_GVEC(PADDSW,  tcg_gen_gvec_ssadd, MO_16)
830BINARY_INT_GVEC(PADDUSB, tcg_gen_gvec_usadd, MO_8)
831BINARY_INT_GVEC(PADDUSW, tcg_gen_gvec_usadd, MO_16)
832BINARY_INT_GVEC(PAND,    tcg_gen_gvec_and, MO_64)
833BINARY_INT_GVEC(PCMPEQB, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_8)
834BINARY_INT_GVEC(PCMPEQD, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_32)
835BINARY_INT_GVEC(PCMPEQW, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_16)
836BINARY_INT_GVEC(PCMPEQQ, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_64)
837BINARY_INT_GVEC(PCMPGTB, tcg_gen_gvec_cmp, TCG_COND_GT, MO_8)
838BINARY_INT_GVEC(PCMPGTW, tcg_gen_gvec_cmp, TCG_COND_GT, MO_16)
839BINARY_INT_GVEC(PCMPGTD, tcg_gen_gvec_cmp, TCG_COND_GT, MO_32)
840BINARY_INT_GVEC(PCMPGTQ, tcg_gen_gvec_cmp, TCG_COND_GT, MO_64)
841BINARY_INT_GVEC(PMAXSB,  tcg_gen_gvec_smax, MO_8)
842BINARY_INT_GVEC(PMAXSW,  tcg_gen_gvec_smax, MO_16)
843BINARY_INT_GVEC(PMAXSD,  tcg_gen_gvec_smax, MO_32)
844BINARY_INT_GVEC(PMAXUB,  tcg_gen_gvec_umax, MO_8)
845BINARY_INT_GVEC(PMAXUW,  tcg_gen_gvec_umax, MO_16)
846BINARY_INT_GVEC(PMAXUD,  tcg_gen_gvec_umax, MO_32)
847BINARY_INT_GVEC(PMINSB,  tcg_gen_gvec_smin, MO_8)
848BINARY_INT_GVEC(PMINSW,  tcg_gen_gvec_smin, MO_16)
849BINARY_INT_GVEC(PMINSD,  tcg_gen_gvec_smin, MO_32)
850BINARY_INT_GVEC(PMINUB,  tcg_gen_gvec_umin, MO_8)
851BINARY_INT_GVEC(PMINUW,  tcg_gen_gvec_umin, MO_16)
852BINARY_INT_GVEC(PMINUD,  tcg_gen_gvec_umin, MO_32)
853BINARY_INT_GVEC(PMULLW,  tcg_gen_gvec_mul, MO_16)
854BINARY_INT_GVEC(PMULLD,  tcg_gen_gvec_mul, MO_32)
855BINARY_INT_GVEC(POR,     tcg_gen_gvec_or, MO_64)
856BINARY_INT_GVEC(PSUBB,   tcg_gen_gvec_sub, MO_8)
857BINARY_INT_GVEC(PSUBW,   tcg_gen_gvec_sub, MO_16)
858BINARY_INT_GVEC(PSUBD,   tcg_gen_gvec_sub, MO_32)
859BINARY_INT_GVEC(PSUBQ,   tcg_gen_gvec_sub, MO_64)
860BINARY_INT_GVEC(PSUBSB,  tcg_gen_gvec_sssub, MO_8)
861BINARY_INT_GVEC(PSUBSW,  tcg_gen_gvec_sssub, MO_16)
862BINARY_INT_GVEC(PSUBUSB, tcg_gen_gvec_ussub, MO_8)
863BINARY_INT_GVEC(PSUBUSW, tcg_gen_gvec_ussub, MO_16)
864BINARY_INT_GVEC(PXOR,    tcg_gen_gvec_xor, MO_64)
865
866
867/*
868 * 00 = p*  Pq, Qq (if mmx not NULL; no VEX)
869 * 66 = vp* Vx, Hx, Wx
870 *
871 * These are really the same encoding, because 1) V is the same as P when VEX.V
872 * is not present 2) P and Q are the same as H and W apart from MM/XMM
873 */
874static inline void gen_binary_int_sse(DisasContext *s, X86DecodedInsn *decode,
875                                      SSEFunc_0_eppp mmx, SSEFunc_0_eppp xmm, SSEFunc_0_eppp ymm)
876{
877    assert(!!mmx == !!(decode->e.special == X86_SPECIAL_MMX));
878
879    if (mmx && (s->prefix & PREFIX_VEX) && !(s->prefix & PREFIX_DATA)) {
880        /* VEX encoding is not applicable to MMX instructions.  */
881        gen_illegal_opcode(s);
882        return;
883    }
884    if (!(s->prefix & PREFIX_DATA)) {
885        mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
886    } else if (!s->vex_l) {
887        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
888    } else {
889        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
890    }
891}
892
893
894#define BINARY_INT_MMX(uname, lname)                                               \
895static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
896{                                                                                  \
897    gen_binary_int_sse(s, decode,                                                  \
898                          gen_helper_##lname##_mmx,                                \
899                          gen_helper_##lname##_xmm,                                \
900                          gen_helper_##lname##_ymm);                               \
901}
902BINARY_INT_MMX(PUNPCKLBW,  punpcklbw)
903BINARY_INT_MMX(PUNPCKLWD,  punpcklwd)
904BINARY_INT_MMX(PUNPCKLDQ,  punpckldq)
905BINARY_INT_MMX(PACKSSWB,   packsswb)
906BINARY_INT_MMX(PACKUSWB,   packuswb)
907BINARY_INT_MMX(PUNPCKHBW,  punpckhbw)
908BINARY_INT_MMX(PUNPCKHWD,  punpckhwd)
909BINARY_INT_MMX(PUNPCKHDQ,  punpckhdq)
910BINARY_INT_MMX(PACKSSDW,   packssdw)
911
912BINARY_INT_MMX(PAVGB,   pavgb)
913BINARY_INT_MMX(PAVGW,   pavgw)
914BINARY_INT_MMX(PMADDWD, pmaddwd)
915BINARY_INT_MMX(PMULHUW, pmulhuw)
916BINARY_INT_MMX(PMULHW,  pmulhw)
917BINARY_INT_MMX(PMULUDQ, pmuludq)
918BINARY_INT_MMX(PSADBW,  psadbw)
919
920BINARY_INT_MMX(PSLLW_r, psllw)
921BINARY_INT_MMX(PSLLD_r, pslld)
922BINARY_INT_MMX(PSLLQ_r, psllq)
923BINARY_INT_MMX(PSRLW_r, psrlw)
924BINARY_INT_MMX(PSRLD_r, psrld)
925BINARY_INT_MMX(PSRLQ_r, psrlq)
926BINARY_INT_MMX(PSRAW_r, psraw)
927BINARY_INT_MMX(PSRAD_r, psrad)
928
929BINARY_INT_MMX(PHADDW,    phaddw)
930BINARY_INT_MMX(PHADDSW,   phaddsw)
931BINARY_INT_MMX(PHADDD,    phaddd)
932BINARY_INT_MMX(PHSUBW,    phsubw)
933BINARY_INT_MMX(PHSUBSW,   phsubsw)
934BINARY_INT_MMX(PHSUBD,    phsubd)
935BINARY_INT_MMX(PMADDUBSW, pmaddubsw)
936BINARY_INT_MMX(PSHUFB,    pshufb)
937BINARY_INT_MMX(PSIGNB,    psignb)
938BINARY_INT_MMX(PSIGNW,    psignw)
939BINARY_INT_MMX(PSIGND,    psignd)
940BINARY_INT_MMX(PMULHRSW,  pmulhrsw)
941
942/* Instructions with no MMX equivalent.  */
943#define BINARY_INT_SSE(uname, lname)                                               \
944static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
945{                                                                                  \
946    gen_binary_int_sse(s, decode,                                                  \
947                          NULL,                                                    \
948                          gen_helper_##lname##_xmm,                                \
949                          gen_helper_##lname##_ymm);                               \
950}
951
952/* Instructions with no MMX equivalent.  */
953BINARY_INT_SSE(PUNPCKLQDQ, punpcklqdq)
954BINARY_INT_SSE(PUNPCKHQDQ, punpckhqdq)
955BINARY_INT_SSE(VPACKUSDW,  packusdw)
956BINARY_INT_SSE(VPERMILPS,  vpermilps)
957BINARY_INT_SSE(VPERMILPD,  vpermilpd)
958BINARY_INT_SSE(VMASKMOVPS, vpmaskmovd)
959BINARY_INT_SSE(VMASKMOVPD, vpmaskmovq)
960
961BINARY_INT_SSE(PMULDQ,    pmuldq)
962
963BINARY_INT_SSE(VAESDEC, aesdec)
964BINARY_INT_SSE(VAESDECLAST, aesdeclast)
965BINARY_INT_SSE(VAESENC, aesenc)
966BINARY_INT_SSE(VAESENCLAST, aesenclast)
967
968#define UNARY_CMP_SSE(uname, lname)                                                \
969static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
970{                                                                                  \
971    if (!s->vex_l) {                                                               \
972        gen_helper_##lname##_xmm(tcg_env, OP_PTR1, OP_PTR2);                       \
973    } else {                                                                       \
974        gen_helper_##lname##_ymm(tcg_env, OP_PTR1, OP_PTR2);                       \
975    }                                                                              \
976    assume_cc_op(s, CC_OP_EFLAGS);                                                  \
977}
978UNARY_CMP_SSE(VPTEST,     ptest)
979UNARY_CMP_SSE(VTESTPS,    vtestps)
980UNARY_CMP_SSE(VTESTPD,    vtestpd)
981
982static inline void gen_unary_int_sse(DisasContext *s, X86DecodedInsn *decode,
983                                     SSEFunc_0_epp xmm, SSEFunc_0_epp ymm)
984{
985    if (!s->vex_l) {
986        xmm(tcg_env, OP_PTR0, OP_PTR2);
987    } else {
988        ymm(tcg_env, OP_PTR0, OP_PTR2);
989    }
990}
991
992#define UNARY_INT_SSE(uname, lname)                                                \
993static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
994{                                                                                  \
995    gen_unary_int_sse(s, decode,                                                   \
996                      gen_helper_##lname##_xmm,                                    \
997                      gen_helper_##lname##_ymm);                                   \
998}
999
1000UNARY_INT_SSE(VPMOVSXBW,    pmovsxbw)
1001UNARY_INT_SSE(VPMOVSXBD,    pmovsxbd)
1002UNARY_INT_SSE(VPMOVSXBQ,    pmovsxbq)
1003UNARY_INT_SSE(VPMOVSXWD,    pmovsxwd)
1004UNARY_INT_SSE(VPMOVSXWQ,    pmovsxwq)
1005UNARY_INT_SSE(VPMOVSXDQ,    pmovsxdq)
1006
1007UNARY_INT_SSE(VPMOVZXBW,    pmovzxbw)
1008UNARY_INT_SSE(VPMOVZXBD,    pmovzxbd)
1009UNARY_INT_SSE(VPMOVZXBQ,    pmovzxbq)
1010UNARY_INT_SSE(VPMOVZXWD,    pmovzxwd)
1011UNARY_INT_SSE(VPMOVZXWQ,    pmovzxwq)
1012UNARY_INT_SSE(VPMOVZXDQ,    pmovzxdq)
1013
1014UNARY_INT_SSE(VMOVSLDUP,    pmovsldup)
1015UNARY_INT_SSE(VMOVSHDUP,    pmovshdup)
1016UNARY_INT_SSE(VMOVDDUP,     pmovdldup)
1017
1018UNARY_INT_SSE(VCVTDQ2PD, cvtdq2pd)
1019UNARY_INT_SSE(VCVTPD2DQ, cvtpd2dq)
1020UNARY_INT_SSE(VCVTTPD2DQ, cvttpd2dq)
1021UNARY_INT_SSE(VCVTDQ2PS, cvtdq2ps)
1022UNARY_INT_SSE(VCVTPS2DQ, cvtps2dq)
1023UNARY_INT_SSE(VCVTTPS2DQ, cvttps2dq)
1024UNARY_INT_SSE(VCVTPH2PS, cvtph2ps)
1025
1026
1027static inline void gen_unary_imm_sse(DisasContext *s, X86DecodedInsn *decode,
1028                                     SSEFunc_0_ppi xmm, SSEFunc_0_ppi ymm)
1029{
1030    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
1031    if (!s->vex_l) {
1032        xmm(OP_PTR0, OP_PTR1, imm);
1033    } else {
1034        ymm(OP_PTR0, OP_PTR1, imm);
1035    }
1036}
1037
1038#define UNARY_IMM_SSE(uname, lname)                                                \
1039static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1040{                                                                                  \
1041    gen_unary_imm_sse(s, decode,                                                   \
1042                      gen_helper_##lname##_xmm,                                    \
1043                      gen_helper_##lname##_ymm);                                   \
1044}
1045
1046UNARY_IMM_SSE(PSHUFD,     pshufd)
1047UNARY_IMM_SSE(PSHUFHW,    pshufhw)
1048UNARY_IMM_SSE(PSHUFLW,    pshuflw)
1049#define gen_helper_vpermq_xmm NULL
1050UNARY_IMM_SSE(VPERMQ,      vpermq)
1051UNARY_IMM_SSE(VPERMILPS_i, vpermilps_imm)
1052UNARY_IMM_SSE(VPERMILPD_i, vpermilpd_imm)
1053
1054static inline void gen_unary_imm_fp_sse(DisasContext *s, X86DecodedInsn *decode,
1055                                        SSEFunc_0_eppi xmm, SSEFunc_0_eppi ymm)
1056{
1057    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
1058    if (!s->vex_l) {
1059        xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
1060    } else {
1061        ymm(tcg_env, OP_PTR0, OP_PTR1, imm);
1062    }
1063}
1064
1065#define UNARY_IMM_FP_SSE(uname, lname)                                             \
1066static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1067{                                                                                  \
1068    gen_unary_imm_fp_sse(s, decode,                                                \
1069                      gen_helper_##lname##_xmm,                                    \
1070                      gen_helper_##lname##_ymm);                                   \
1071}
1072
1073UNARY_IMM_FP_SSE(VROUNDPS,    roundps)
1074UNARY_IMM_FP_SSE(VROUNDPD,    roundpd)
1075
1076static inline void gen_vexw_avx(DisasContext *s, X86DecodedInsn *decode,
1077                                SSEFunc_0_eppp d_xmm, SSEFunc_0_eppp q_xmm,
1078                                SSEFunc_0_eppp d_ymm, SSEFunc_0_eppp q_ymm)
1079{
1080    SSEFunc_0_eppp d = s->vex_l ? d_ymm : d_xmm;
1081    SSEFunc_0_eppp q = s->vex_l ? q_ymm : q_xmm;
1082    SSEFunc_0_eppp fn = s->vex_w ? q : d;
1083    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
1084}
1085
1086/* VEX.W affects whether to operate on 32- or 64-bit elements.  */
1087#define VEXW_AVX(uname, lname)                                                     \
1088static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1089{                                                                                  \
1090    gen_vexw_avx(s, decode,                                                        \
1091                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1092                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1093}
1094VEXW_AVX(VPSLLV,    vpsllv)
1095VEXW_AVX(VPSRLV,    vpsrlv)
1096VEXW_AVX(VPSRAV,    vpsrav)
1097VEXW_AVX(VPMASKMOV, vpmaskmov)
1098
1099/* Same as above, but with extra arguments to the helper.  */
1100static inline void gen_vsib_avx(DisasContext *s, X86DecodedInsn *decode,
1101                                SSEFunc_0_epppti d_xmm, SSEFunc_0_epppti q_xmm,
1102                                SSEFunc_0_epppti d_ymm, SSEFunc_0_epppti q_ymm)
1103{
1104    SSEFunc_0_epppti d = s->vex_l ? d_ymm : d_xmm;
1105    SSEFunc_0_epppti q = s->vex_l ? q_ymm : q_xmm;
1106    SSEFunc_0_epppti fn = s->vex_w ? q : d;
1107    TCGv_i32 scale = tcg_constant_i32(decode->mem.scale);
1108    TCGv_ptr index = tcg_temp_new_ptr();
1109
1110    /* Pass third input as (index, base, scale) */
1111    tcg_gen_addi_ptr(index, tcg_env, ZMM_OFFSET(decode->mem.index));
1112    fn(tcg_env, OP_PTR0, OP_PTR1, index, s->A0, scale);
1113
1114    /*
1115     * There are two output operands, so zero OP1's high 128 bits
1116     * in the VEX.128 case.
1117     */
1118    if (!s->vex_l) {
1119        int ymmh_ofs = vector_elem_offset(&decode->op[1], MO_128, 1);
1120        tcg_gen_gvec_dup_imm(MO_64, ymmh_ofs, 16, 16, 0);
1121    }
1122}
1123#define VSIB_AVX(uname, lname)                                                     \
1124static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1125{                                                                                  \
1126    gen_vsib_avx(s, decode,                                                        \
1127                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1128                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1129}
1130VSIB_AVX(VPGATHERD, vpgatherd)
1131VSIB_AVX(VPGATHERQ, vpgatherq)
1132
1133static void gen_AAA(DisasContext *s, X86DecodedInsn *decode)
1134{
1135    gen_update_cc_op(s);
1136    gen_helper_aaa(tcg_env);
1137    assume_cc_op(s, CC_OP_EFLAGS);
1138}
1139
1140static void gen_AAD(DisasContext *s, X86DecodedInsn *decode)
1141{
1142    gen_helper_aad(s->T0, s->T0, s->T1);
1143    prepare_update1_cc(decode, s, CC_OP_LOGICB);
1144}
1145
1146static void gen_AAM(DisasContext *s, X86DecodedInsn *decode)
1147{
1148    if (decode->immediate == 0) {
1149        gen_exception(s, EXCP00_DIVZ);
1150    } else {
1151        gen_helper_aam(s->T0, s->T0, s->T1);
1152        prepare_update1_cc(decode, s, CC_OP_LOGICB);
1153    }
1154}
1155
1156static void gen_AAS(DisasContext *s, X86DecodedInsn *decode)
1157{
1158    gen_update_cc_op(s);
1159    gen_helper_aas(tcg_env);
1160    assume_cc_op(s, CC_OP_EFLAGS);
1161}
1162
1163static void gen_ADD(DisasContext *s, X86DecodedInsn *decode);
1164static void gen_ADC(DisasContext *s, X86DecodedInsn *decode)
1165{
1166    MemOp ot = decode->op[1].ot;
1167    TCGv c_in;
1168
1169    /*
1170     * Try to avoid CC_OP_ADC by transforming as follows:
1171     * CC_ADC: src1 = dst + c_in, src2 = 0, src3 = c_in
1172     * CC_ADD: src1 = dst + c_in, src2 = c_in (no src3)
1173     *
1174     * In general src2 vs. src3 matters when computing AF and OF, but not here:
1175     * - AF is bit 4 of dst^src1^src2, which is bit 4 of dst^src1 in both cases
1176     * - OF is a function of the two MSBs, and in both cases they are zero for src2
1177     */
1178    if (decode->e.op2 == X86_TYPE_I && decode->immediate == 0) {
1179        gen_compute_eflags_c(s, s->T1);
1180        gen_ADD(s, decode);
1181        return;
1182    }
1183
1184    c_in = tcg_temp_new();
1185    gen_compute_eflags_c(s, c_in);
1186    if (s->prefix & PREFIX_LOCK) {
1187        tcg_gen_add_tl(s->T0, c_in, s->T1);
1188        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
1189                                    s->mem_index, ot | MO_LE);
1190    } else {
1191        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1192        tcg_gen_add_tl(s->T0, s->T0, c_in);
1193    }
1194    prepare_update3_cc(decode, s, CC_OP_ADCB + ot, c_in);
1195}
1196
1197static void gen_ADCOX(DisasContext *s, X86DecodedInsn *decode, int cc_op)
1198{
1199    MemOp ot = decode->op[0].ot;
1200    TCGv carry_in = NULL;
1201    TCGv *carry_out = (cc_op == CC_OP_ADCX ? &decode->cc_dst : &decode->cc_src2);
1202    TCGv zero;
1203
1204    decode->cc_op = cc_op;
1205    *carry_out = tcg_temp_new();
1206    if (CC_OP_HAS_EFLAGS(s->cc_op)) {
1207        decode->cc_src = cpu_cc_src;
1208
1209        /* Re-use the carry-out from a previous round?  */
1210        if (s->cc_op == cc_op || s->cc_op == CC_OP_ADCOX) {
1211            carry_in = (cc_op == CC_OP_ADCX ? cpu_cc_dst : cpu_cc_src2);
1212        }
1213
1214        /* Preserve the opposite carry from previous rounds?  */
1215        if (s->cc_op != cc_op && s->cc_op != CC_OP_EFLAGS) {
1216            decode->cc_op = CC_OP_ADCOX;
1217            if (carry_out == &decode->cc_dst) {
1218                decode->cc_src2 = cpu_cc_src2;
1219            } else {
1220                decode->cc_dst = cpu_cc_dst;
1221            }
1222        }
1223    } else {
1224        decode->cc_src = tcg_temp_new();
1225        gen_mov_eflags(s, decode->cc_src);
1226    }
1227
1228    if (!carry_in) {
1229        /* Get carry_in out of EFLAGS.  */
1230        carry_in = tcg_temp_new();
1231        tcg_gen_extract_tl(carry_in, decode->cc_src,
1232            ctz32(cc_op == CC_OP_ADCX ? CC_C : CC_O), 1);
1233    }
1234
1235    switch (ot) {
1236#ifdef TARGET_X86_64
1237    case MO_32:
1238        /* If TL is 64-bit just do everything in 64-bit arithmetic.  */
1239        tcg_gen_ext32u_tl(s->T0, s->T0);
1240        tcg_gen_ext32u_tl(s->T1, s->T1);
1241        tcg_gen_add_i64(s->T0, s->T0, s->T1);
1242        tcg_gen_add_i64(s->T0, s->T0, carry_in);
1243        tcg_gen_shri_i64(*carry_out, s->T0, 32);
1244        break;
1245#endif
1246    default:
1247        zero = tcg_constant_tl(0);
1248        tcg_gen_add2_tl(s->T0, *carry_out, s->T0, zero, carry_in, zero);
1249        tcg_gen_add2_tl(s->T0, *carry_out, s->T0, *carry_out, s->T1, zero);
1250        break;
1251    }
1252}
1253
1254static void gen_ADCX(DisasContext *s, X86DecodedInsn *decode)
1255{
1256    gen_ADCOX(s, decode, CC_OP_ADCX);
1257}
1258
1259static void gen_ADD(DisasContext *s, X86DecodedInsn *decode)
1260{
1261    MemOp ot = decode->op[1].ot;
1262
1263    if (s->prefix & PREFIX_LOCK) {
1264        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1265                                    s->mem_index, ot | MO_LE);
1266    } else {
1267        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1268    }
1269    prepare_update2_cc(decode, s, CC_OP_ADDB + ot);
1270}
1271
1272static void gen_ADOX(DisasContext *s, X86DecodedInsn *decode)
1273{
1274    gen_ADCOX(s, decode, CC_OP_ADOX);
1275}
1276
1277static void gen_AND(DisasContext *s, X86DecodedInsn *decode)
1278{
1279    MemOp ot = decode->op[1].ot;
1280
1281    if (s->prefix & PREFIX_LOCK) {
1282        tcg_gen_atomic_and_fetch_tl(s->T0, s->A0, s->T1,
1283                                    s->mem_index, ot | MO_LE);
1284    } else {
1285        tcg_gen_and_tl(s->T0, s->T0, s->T1);
1286    }
1287    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1288}
1289
1290static void gen_ANDN(DisasContext *s, X86DecodedInsn *decode)
1291{
1292    MemOp ot = decode->op[0].ot;
1293
1294    tcg_gen_andc_tl(s->T0, s->T1, s->T0);
1295    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1296}
1297
1298static void gen_ARPL(DisasContext *s, X86DecodedInsn *decode)
1299{
1300    TCGv zf = tcg_temp_new();
1301    TCGv flags = tcg_temp_new();
1302
1303    gen_mov_eflags(s, flags);
1304
1305    /* Compute adjusted DST in T1, merging in SRC[RPL].  */
1306    tcg_gen_deposit_tl(s->T1, s->T0, s->T1, 0, 2);
1307
1308    /* Z flag set if DST[RPL] < SRC[RPL] */
1309    tcg_gen_setcond_tl(TCG_COND_LTU, zf, s->T0, s->T1);
1310    tcg_gen_deposit_tl(flags, flags, zf, ctz32(CC_Z), 1);
1311
1312    /* Place maximum RPL in DST */
1313    tcg_gen_umax_tl(s->T0, s->T0, s->T1);
1314
1315    decode->cc_src = flags;
1316    decode->cc_op = CC_OP_EFLAGS;
1317}
1318
1319static void gen_BEXTR(DisasContext *s, X86DecodedInsn *decode)
1320{
1321    MemOp ot = decode->op[0].ot;
1322    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1323    TCGv zero = tcg_constant_tl(0);
1324    TCGv mone = tcg_constant_tl(-1);
1325
1326    /*
1327     * Extract START, and shift the operand.
1328     * Shifts larger than operand size get zeros.
1329     */
1330    tcg_gen_ext8u_tl(s->A0, s->T1);
1331    tcg_gen_shr_tl(s->T0, s->T0, s->A0);
1332
1333    tcg_gen_movcond_tl(TCG_COND_LEU, s->T0, s->A0, bound, s->T0, zero);
1334
1335    /*
1336     * Extract the LEN into an inverse mask.  Lengths larger than
1337     * operand size get all zeros, length 0 gets all ones.
1338     */
1339    tcg_gen_extract_tl(s->A0, s->T1, 8, 8);
1340    tcg_gen_shl_tl(s->T1, mone, s->A0);
1341    tcg_gen_movcond_tl(TCG_COND_LEU, s->T1, s->A0, bound, s->T1, zero);
1342    tcg_gen_andc_tl(s->T0, s->T0, s->T1);
1343
1344    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1345}
1346
1347static void gen_BLSI(DisasContext *s, X86DecodedInsn *decode)
1348{
1349    MemOp ot = decode->op[0].ot;
1350
1351    /* input in T1, which is ready for prepare_update2_cc  */
1352    tcg_gen_neg_tl(s->T0, s->T1);
1353    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1354    prepare_update2_cc(decode, s, CC_OP_BLSIB + ot);
1355}
1356
1357static void gen_BLSMSK(DisasContext *s, X86DecodedInsn *decode)
1358{
1359    MemOp ot = decode->op[0].ot;
1360
1361    /* input in T1, which is ready for prepare_update2_cc  */
1362    tcg_gen_subi_tl(s->T0, s->T1, 1);
1363    tcg_gen_xor_tl(s->T0, s->T0, s->T1);
1364    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1365}
1366
1367static void gen_BLSR(DisasContext *s, X86DecodedInsn *decode)
1368{
1369    MemOp ot = decode->op[0].ot;
1370
1371    /* input in T1, which is ready for prepare_update2_cc  */
1372    tcg_gen_subi_tl(s->T0, s->T1, 1);
1373    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1374    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1375}
1376
1377static void gen_BOUND(DisasContext *s, X86DecodedInsn *decode)
1378{
1379    TCGv_i32 op = tcg_temp_new_i32();
1380    tcg_gen_trunc_tl_i32(op, s->T0);
1381    if (decode->op[1].ot == MO_16) {
1382        gen_helper_boundw(tcg_env, s->A0, op);
1383    } else {
1384        gen_helper_boundl(tcg_env, s->A0, op);
1385    }
1386}
1387
1388/* Non-standard convention - on entry T0 is zero-extended input, T1 is the output.  */
1389static void gen_BSF(DisasContext *s, X86DecodedInsn *decode)
1390{
1391    MemOp ot = decode->op[0].ot;
1392
1393    /* Only the Z bit is defined and it is related to the input.  */
1394    decode->cc_dst = tcg_temp_new();
1395    decode->cc_op = CC_OP_LOGICB + ot;
1396    tcg_gen_mov_tl(decode->cc_dst, s->T0);
1397
1398    /*
1399     * The manual says that the output is undefined when the
1400     * input is zero, but real hardware leaves it unchanged, and
1401     * real programs appear to depend on that.  Accomplish this
1402     * by passing the output as the value to return upon zero.
1403     */
1404    tcg_gen_ctz_tl(s->T0, s->T0, s->T1);
1405}
1406
1407/* Non-standard convention - on entry T0 is zero-extended input, T1 is the output.  */
1408static void gen_BSR(DisasContext *s, X86DecodedInsn *decode)
1409{
1410    MemOp ot = decode->op[0].ot;
1411
1412    /* Only the Z bit is defined and it is related to the input.  */
1413    decode->cc_dst = tcg_temp_new();
1414    decode->cc_op = CC_OP_LOGICB + ot;
1415    tcg_gen_mov_tl(decode->cc_dst, s->T0);
1416
1417    /*
1418     * The manual says that the output is undefined when the
1419     * input is zero, but real hardware leaves it unchanged, and
1420     * real programs appear to depend on that.  Accomplish this
1421     * by passing the output as the value to return upon zero.
1422     * Plus, return the bit index of the first 1 bit.
1423     */
1424    tcg_gen_xori_tl(s->T1, s->T1, TARGET_LONG_BITS - 1);
1425    tcg_gen_clz_tl(s->T0, s->T0, s->T1);
1426    tcg_gen_xori_tl(s->T0, s->T0, TARGET_LONG_BITS - 1);
1427}
1428
1429static void gen_BSWAP(DisasContext *s, X86DecodedInsn *decode)
1430{
1431#ifdef TARGET_X86_64
1432    if (s->dflag == MO_64) {
1433        tcg_gen_bswap64_i64(s->T0, s->T0);
1434        return;
1435    }
1436#endif
1437    tcg_gen_bswap32_tl(s->T0, s->T0, TCG_BSWAP_OZ);
1438}
1439
1440static TCGv gen_bt_mask(DisasContext *s, X86DecodedInsn *decode)
1441{
1442    MemOp ot = decode->op[1].ot;
1443    TCGv mask = tcg_temp_new();
1444
1445    tcg_gen_andi_tl(s->T1, s->T1, (8 << ot) - 1);
1446    tcg_gen_shl_tl(mask, tcg_constant_tl(1), s->T1);
1447    return mask;
1448}
1449
1450/* Expects truncated bit index in COUNT, 1 << COUNT in MASK.  */
1451static void gen_bt_flags(DisasContext *s, X86DecodedInsn *decode, TCGv src,
1452                         TCGv count, TCGv mask)
1453{
1454    TCGv cf;
1455
1456    /*
1457     * C is the result of the test, Z is unchanged, and the others
1458     * are all undefined.
1459     */
1460    if (s->cc_op == CC_OP_DYNAMIC || CC_OP_HAS_EFLAGS(s->cc_op)) {
1461        /* Generate EFLAGS and replace the C bit.  */
1462        cf = tcg_temp_new();
1463        tcg_gen_setcond_tl(TCG_COND_TSTNE, cf, src, mask);
1464        prepare_update_cf(decode, s, cf);
1465    } else {
1466        /*
1467         * Z was going to be computed from the non-zero status of CC_DST.
1468         * We can get that same Z value (and the new C value) by leaving
1469         * CC_DST alone, setting CC_SRC, and using a CC_OP_SAR of the
1470         * same width.
1471         */
1472        decode->cc_src = tcg_temp_new();
1473        decode->cc_dst = cpu_cc_dst;
1474        decode->cc_op = CC_OP_SARB + cc_op_size(s->cc_op);
1475        tcg_gen_shr_tl(decode->cc_src, src, count);
1476    }
1477}
1478
1479static void gen_BT(DisasContext *s, X86DecodedInsn *decode)
1480{
1481    TCGv count = s->T1;
1482    TCGv mask;
1483
1484    /*
1485     * Try to ensure that the rhs of the TSTNE condition is a constant (and a
1486     * power of two), as that is more readily available on most TCG backends.
1487     *
1488     * For immediate bit number gen_bt_mask()'s output is already a constant;
1489     * for register bit number, shift the source right and check bit 0.
1490     */
1491    if (decode->e.op2 == X86_TYPE_I) {
1492        mask = gen_bt_mask(s, decode);
1493    } else {
1494        MemOp ot = decode->op[1].ot;
1495
1496        tcg_gen_andi_tl(s->T1, s->T1, (8 << ot) - 1);
1497        tcg_gen_shr_tl(s->T0, s->T0, s->T1);
1498
1499        count = tcg_constant_tl(0);
1500        mask = tcg_constant_tl(1);
1501    }
1502    gen_bt_flags(s, decode, s->T0, count, mask);
1503}
1504
1505static void gen_BTC(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        tcg_gen_atomic_fetch_xor_tl(old, s->A0, mask, s->mem_index, ot | MO_LE);
1513    } else {
1514        tcg_gen_mov_tl(old, s->T0);
1515        tcg_gen_xor_tl(s->T0, s->T0, mask);
1516    }
1517
1518    gen_bt_flags(s, decode, old, s->T1, mask);
1519}
1520
1521static void gen_BTR(DisasContext *s, X86DecodedInsn *decode)
1522{
1523    MemOp ot = decode->op[0].ot;
1524    TCGv old = tcg_temp_new();
1525    TCGv mask = gen_bt_mask(s, decode);
1526
1527    if (s->prefix & PREFIX_LOCK) {
1528        TCGv maskc = tcg_temp_new();
1529        tcg_gen_not_tl(maskc, mask);
1530        tcg_gen_atomic_fetch_and_tl(old, s->A0, maskc, s->mem_index, ot | MO_LE);
1531    } else {
1532        tcg_gen_mov_tl(old, s->T0);
1533        tcg_gen_andc_tl(s->T0, s->T0, mask);
1534    }
1535
1536    gen_bt_flags(s, decode, old, s->T1, mask);
1537}
1538
1539static void gen_BTS(DisasContext *s, X86DecodedInsn *decode)
1540{
1541    MemOp ot = decode->op[0].ot;
1542    TCGv old = tcg_temp_new();
1543    TCGv mask = gen_bt_mask(s, decode);
1544
1545    if (s->prefix & PREFIX_LOCK) {
1546        tcg_gen_atomic_fetch_or_tl(old, s->A0, mask, s->mem_index, ot | MO_LE);
1547    } else {
1548        tcg_gen_mov_tl(old, s->T0);
1549        tcg_gen_or_tl(s->T0, s->T0, mask);
1550    }
1551
1552    gen_bt_flags(s, decode, old, s->T1, mask);
1553}
1554
1555static void gen_BZHI(DisasContext *s, X86DecodedInsn *decode)
1556{
1557    MemOp ot = decode->op[0].ot;
1558    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1559    TCGv zero = tcg_constant_tl(0);
1560    TCGv mone = tcg_constant_tl(-1);
1561
1562    tcg_gen_ext8u_tl(s->T1, s->T1);
1563
1564    tcg_gen_shl_tl(s->A0, mone, s->T1);
1565    tcg_gen_movcond_tl(TCG_COND_LEU, s->A0, s->T1, bound, s->A0, zero);
1566    tcg_gen_andc_tl(s->T0, s->T0, s->A0);
1567    /*
1568     * Note that since we're using BMILG (in order to get O
1569     * cleared) we need to store the inverse into C.
1570     */
1571    tcg_gen_setcond_tl(TCG_COND_LEU, s->T1, s->T1, bound);
1572    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1573}
1574
1575static void gen_CALL(DisasContext *s, X86DecodedInsn *decode)
1576{
1577    gen_push_v(s, eip_next_tl(s));
1578    gen_JMP(s, decode);
1579}
1580
1581static void gen_CALL_m(DisasContext *s, X86DecodedInsn *decode)
1582{
1583    gen_push_v(s, eip_next_tl(s));
1584    gen_JMP_m(s, decode);
1585}
1586
1587static void gen_CALLF(DisasContext *s, X86DecodedInsn *decode)
1588{
1589    gen_far_call(s);
1590}
1591
1592static void gen_CALLF_m(DisasContext *s, X86DecodedInsn *decode)
1593{
1594    MemOp ot = decode->op[1].ot;
1595
1596    gen_op_ld_v(s, ot, s->T0, s->A0);
1597    gen_add_A0_im(s, 1 << ot);
1598    gen_op_ld_v(s, MO_16, s->T1, s->A0);
1599    gen_far_call(s);
1600}
1601
1602static void gen_CBW(DisasContext *s, X86DecodedInsn *decode)
1603{
1604    MemOp src_ot = decode->op[0].ot - 1;
1605
1606    tcg_gen_ext_tl(s->T0, s->T0, src_ot | MO_SIGN);
1607}
1608
1609static void gen_CLC(DisasContext *s, X86DecodedInsn *decode)
1610{
1611    gen_compute_eflags(s);
1612    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, ~CC_C);
1613}
1614
1615static void gen_CLD(DisasContext *s, X86DecodedInsn *decode)
1616{
1617    tcg_gen_st_i32(tcg_constant_i32(1), tcg_env, offsetof(CPUX86State, df));
1618}
1619
1620static void gen_CLI(DisasContext *s, X86DecodedInsn *decode)
1621{
1622    gen_reset_eflags(s, IF_MASK);
1623}
1624
1625static void gen_CLTS(DisasContext *s, X86DecodedInsn *decode)
1626{
1627    gen_helper_clts(tcg_env);
1628    /* abort block because static cpu state changed */
1629    s->base.is_jmp = DISAS_EOB_NEXT;
1630}
1631
1632static void gen_CMC(DisasContext *s, X86DecodedInsn *decode)
1633{
1634    gen_compute_eflags(s);
1635    tcg_gen_xori_tl(cpu_cc_src, cpu_cc_src, CC_C);
1636}
1637
1638static void gen_CMOVcc(DisasContext *s, X86DecodedInsn *decode)
1639{
1640    gen_cmovcc(s, decode->b & 0xf, s->T0, s->T1);
1641}
1642
1643static void gen_CMPccXADD(DisasContext *s, X86DecodedInsn *decode)
1644{
1645    TCGLabel *label_top = gen_new_label();
1646    TCGLabel *label_bottom = gen_new_label();
1647    TCGv oldv = tcg_temp_new();
1648    TCGv newv = tcg_temp_new();
1649    TCGv cmpv = tcg_temp_new();
1650    TCGCond cond;
1651
1652    TCGv cmp_lhs, cmp_rhs;
1653    MemOp ot, ot_full;
1654
1655    int jcc_op = (decode->b >> 1) & 7;
1656    static const TCGCond cond_table[8] = {
1657        [JCC_O] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1658        [JCC_B] = TCG_COND_LTU,
1659        [JCC_Z] = TCG_COND_EQ,
1660        [JCC_BE] = TCG_COND_LEU,
1661        [JCC_S] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1662        [JCC_P] = TCG_COND_TSTEQ,  /* even parity - tests low bit of popcount */
1663        [JCC_L] = TCG_COND_LT,
1664        [JCC_LE] = TCG_COND_LE,
1665    };
1666
1667    cond = cond_table[jcc_op];
1668    if (decode->b & 1) {
1669        cond = tcg_invert_cond(cond);
1670    }
1671
1672    ot = decode->op[0].ot;
1673    ot_full = ot | MO_LE;
1674    if (jcc_op >= JCC_S) {
1675        /*
1676         * Sign-extend values before subtracting for S, P (zero/sign extension
1677         * does not matter there) L, LE and their inverses.
1678         */
1679        ot_full |= MO_SIGN;
1680    }
1681
1682    /*
1683     * cmpv will be moved to cc_src *after* cpu_regs[] is written back, so use
1684     * tcg_gen_ext_tl instead of gen_ext_tl.
1685     */
1686    tcg_gen_ext_tl(cmpv, cpu_regs[decode->op[1].n], ot_full);
1687
1688    /*
1689     * Cmpxchg loop starts here.
1690     * - s->T1: addition operand (from decoder)
1691     * - s->A0: dest address (from decoder)
1692     * - s->cc_srcT: memory operand (lhs for comparison)
1693     * - cmpv: rhs for comparison
1694     */
1695    gen_set_label(label_top);
1696    gen_op_ld_v(s, ot_full, s->cc_srcT, s->A0);
1697    tcg_gen_sub_tl(s->T0, s->cc_srcT, cmpv);
1698
1699    /* Compute the comparison result by hand, to avoid clobbering cc_*.  */
1700    switch (jcc_op) {
1701    case JCC_O:
1702        /* (src1 ^ src2) & (src1 ^ dst). newv is only used here for a moment */
1703        cmp_lhs = tcg_temp_new(), cmp_rhs = tcg_constant_tl(0);
1704        tcg_gen_xor_tl(newv, s->cc_srcT, s->T0);
1705        tcg_gen_xor_tl(cmp_lhs, s->cc_srcT, cmpv);
1706        tcg_gen_and_tl(cmp_lhs, cmp_lhs, newv);
1707        tcg_gen_sextract_tl(cmp_lhs, cmp_lhs, 0, 8 << ot);
1708        break;
1709
1710    case JCC_P:
1711        cmp_lhs = tcg_temp_new(), cmp_rhs = tcg_constant_tl(1);
1712        tcg_gen_ext8u_tl(cmp_lhs, s->T0);
1713        tcg_gen_ctpop_tl(cmp_lhs, cmp_lhs);
1714        break;
1715
1716    case JCC_S:
1717        cmp_lhs = tcg_temp_new(), cmp_rhs = tcg_constant_tl(0);
1718        tcg_gen_sextract_tl(cmp_lhs, s->T0, 0, 8 << ot);
1719        break;
1720
1721    default:
1722        cmp_lhs = s->cc_srcT, cmp_rhs = cmpv;
1723        break;
1724    }
1725
1726    /* Compute new value: if condition does not hold, just store back s->cc_srcT */
1727    tcg_gen_add_tl(newv, s->cc_srcT, s->T1);
1728    tcg_gen_movcond_tl(cond, newv, cmp_lhs, cmp_rhs, newv, s->cc_srcT);
1729    tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, s->cc_srcT, newv, s->mem_index, ot_full);
1730
1731    /* Exit unconditionally if cmpxchg succeeded.  */
1732    tcg_gen_brcond_tl(TCG_COND_EQ, oldv, s->cc_srcT, label_bottom);
1733
1734    /* Try again if there was actually a store to make.  */
1735    tcg_gen_brcond_tl(cond, cmp_lhs, cmp_rhs, label_top);
1736    gen_set_label(label_bottom);
1737
1738    /* Store old value to registers only after a successful store.  */
1739    gen_writeback(s, decode, 1, s->cc_srcT);
1740
1741    decode->cc_dst = s->T0;
1742    decode->cc_src = cmpv;
1743    decode->cc_op = CC_OP_SUBB + ot;
1744}
1745
1746static void gen_CMPS(DisasContext *s, X86DecodedInsn *decode)
1747{
1748    MemOp ot = decode->op[2].ot;
1749    gen_repz_nz(s, ot, gen_cmps);
1750}
1751
1752static void gen_CMPXCHG(DisasContext *s, X86DecodedInsn *decode)
1753{
1754    MemOp ot = decode->op[2].ot;
1755    TCGv cmpv = tcg_temp_new();
1756    TCGv oldv = tcg_temp_new();
1757    TCGv newv = tcg_temp_new();
1758    TCGv dest;
1759
1760    tcg_gen_ext_tl(cmpv, cpu_regs[R_EAX], ot);
1761    tcg_gen_ext_tl(newv, s->T1, ot);
1762    if (s->prefix & PREFIX_LOCK) {
1763        tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, cmpv, newv,
1764                                  s->mem_index, ot | MO_LE);
1765    } else {
1766        tcg_gen_ext_tl(oldv, s->T0, ot);
1767        if (decode->op[0].has_ea) {
1768            /*
1769             * Perform an unconditional store cycle like physical cpu;
1770             * must be before changing accumulator to ensure
1771             * idempotency if the store faults and the instruction
1772             * is restarted
1773             */
1774            tcg_gen_movcond_tl(TCG_COND_EQ, newv, oldv, cmpv, newv, oldv);
1775            gen_op_st_v(s, ot, newv, s->A0);
1776        } else {
1777            /*
1778             * Unlike the memory case, where "the destination operand receives
1779             * a write cycle without regard to the result of the comparison",
1780             * rm must not be touched altogether if the write fails, including
1781             * not zero-extending it on 64-bit processors.  So, precompute
1782             * the result of a successful writeback and perform the movcond
1783             * directly on cpu_regs.  In case rm is part of RAX, note that this
1784             * movcond and the one below are mutually exclusive is executed.
1785             */
1786            dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, newv, newv);
1787            tcg_gen_movcond_tl(TCG_COND_EQ, dest, oldv, cmpv, newv, dest);
1788        }
1789        decode->op[0].unit = X86_OP_SKIP;
1790    }
1791
1792    /* Write RAX only if the cmpxchg fails.  */
1793    dest = gen_op_deposit_reg_v(s, ot, R_EAX, s->T0, oldv);
1794    tcg_gen_movcond_tl(TCG_COND_NE, dest, oldv, cmpv, s->T0, dest);
1795
1796    tcg_gen_mov_tl(s->cc_srcT, cmpv);
1797    tcg_gen_sub_tl(cmpv, cmpv, oldv);
1798    decode->cc_dst = cmpv;
1799    decode->cc_src = oldv;
1800    decode->cc_op = CC_OP_SUBB + ot;
1801}
1802
1803static void gen_CMPXCHG16B(DisasContext *s, X86DecodedInsn *decode)
1804{
1805#ifdef TARGET_X86_64
1806    MemOp mop = MO_LE | MO_128 | MO_ALIGN;
1807    TCGv_i64 t0, t1;
1808    TCGv_i128 cmp, val;
1809
1810    cmp = tcg_temp_new_i128();
1811    val = tcg_temp_new_i128();
1812    tcg_gen_concat_i64_i128(cmp, cpu_regs[R_EAX], cpu_regs[R_EDX]);
1813    tcg_gen_concat_i64_i128(val, cpu_regs[R_EBX], cpu_regs[R_ECX]);
1814
1815    /* Only require atomic with LOCK; non-parallel handled in generator. */
1816    if (s->prefix & PREFIX_LOCK) {
1817        tcg_gen_atomic_cmpxchg_i128(val, s->A0, cmp, val, s->mem_index, mop);
1818    } else {
1819        tcg_gen_nonatomic_cmpxchg_i128(val, s->A0, cmp, val, s->mem_index, mop);
1820    }
1821
1822    tcg_gen_extr_i128_i64(s->T0, s->T1, val);
1823
1824    /* Determine success after the fact. */
1825    t0 = tcg_temp_new_i64();
1826    t1 = tcg_temp_new_i64();
1827    tcg_gen_xor_i64(t0, s->T0, cpu_regs[R_EAX]);
1828    tcg_gen_xor_i64(t1, s->T1, cpu_regs[R_EDX]);
1829    tcg_gen_or_i64(t0, t0, t1);
1830
1831    /* Update Z. */
1832    gen_compute_eflags(s);
1833    tcg_gen_setcondi_i64(TCG_COND_EQ, t0, t0, 0);
1834    tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, t0, ctz32(CC_Z), 1);
1835
1836    /*
1837     * Extract the result values for the register pair.  We may do this
1838     * unconditionally, because on success (Z=1), the old value matches
1839     * the previous value in RDX:RAX.
1840     */
1841    tcg_gen_mov_i64(cpu_regs[R_EAX], s->T0);
1842    tcg_gen_mov_i64(cpu_regs[R_EDX], s->T1);
1843#else
1844    abort();
1845#endif
1846}
1847
1848static void gen_CMPXCHG8B(DisasContext *s, X86DecodedInsn *decode)
1849{
1850    TCGv_i64 cmp, val, old;
1851    TCGv Z;
1852
1853    cmp = tcg_temp_new_i64();
1854    val = tcg_temp_new_i64();
1855    old = tcg_temp_new_i64();
1856
1857    /* Construct the comparison values from the register pair. */
1858    tcg_gen_concat_tl_i64(cmp, cpu_regs[R_EAX], cpu_regs[R_EDX]);
1859    tcg_gen_concat_tl_i64(val, cpu_regs[R_EBX], cpu_regs[R_ECX]);
1860
1861    /* Only require atomic with LOCK; non-parallel handled in generator. */
1862    if (s->prefix & PREFIX_LOCK) {
1863        tcg_gen_atomic_cmpxchg_i64(old, s->A0, cmp, val, s->mem_index, MO_LEUQ);
1864    } else {
1865        tcg_gen_nonatomic_cmpxchg_i64(old, s->A0, cmp, val,
1866                                      s->mem_index, MO_LEUQ);
1867    }
1868
1869    /* Compute the required value of Z. */
1870    tcg_gen_setcond_i64(TCG_COND_EQ, cmp, old, cmp);
1871    Z = tcg_temp_new();
1872    tcg_gen_trunc_i64_tl(Z, cmp);
1873
1874    /*
1875     * Extract the result values for the register pair.
1876     * For 32-bit, we may do this unconditionally, because on success (Z=1),
1877     * the old value matches the previous value in EDX:EAX.  For x86_64,
1878     * the store must be conditional, because we must leave the source
1879     * registers unchanged on success, and zero-extend the writeback
1880     * on failure (Z=0).
1881     */
1882    if (TARGET_LONG_BITS == 32) {
1883        tcg_gen_extr_i64_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], old);
1884    } else {
1885        TCGv zero = tcg_constant_tl(0);
1886
1887        tcg_gen_extr_i64_tl(s->T0, s->T1, old);
1888        tcg_gen_movcond_tl(TCG_COND_EQ, cpu_regs[R_EAX], Z, zero,
1889                           s->T0, cpu_regs[R_EAX]);
1890        tcg_gen_movcond_tl(TCG_COND_EQ, cpu_regs[R_EDX], Z, zero,
1891                           s->T1, cpu_regs[R_EDX]);
1892    }
1893
1894    /* Update Z. */
1895    gen_compute_eflags(s);
1896    tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, Z, ctz32(CC_Z), 1);
1897}
1898
1899static void gen_CPUID(DisasContext *s, X86DecodedInsn *decode)
1900{
1901    gen_update_cc_op(s);
1902    gen_update_eip_cur(s);
1903    gen_helper_cpuid(tcg_env);
1904}
1905
1906static void gen_CRC32(DisasContext *s, X86DecodedInsn *decode)
1907{
1908    MemOp ot = decode->op[2].ot;
1909    TCGv_i32 tmp = tcg_temp_new_i32();
1910
1911    tcg_gen_trunc_tl_i32(tmp, s->T0);
1912    gen_helper_crc32(s->T0, tmp, s->T1, tcg_constant_i32(8 << ot));
1913}
1914
1915static void gen_CVTPI2Px(DisasContext *s, X86DecodedInsn *decode)
1916{
1917    gen_helper_enter_mmx(tcg_env);
1918    if (s->prefix & PREFIX_DATA) {
1919        gen_helper_cvtpi2pd(tcg_env, OP_PTR0, OP_PTR2);
1920    } else {
1921        gen_helper_cvtpi2ps(tcg_env, OP_PTR0, OP_PTR2);
1922    }
1923}
1924
1925static void gen_CVTPx2PI(DisasContext *s, X86DecodedInsn *decode)
1926{
1927    gen_helper_enter_mmx(tcg_env);
1928    if (s->prefix & PREFIX_DATA) {
1929        gen_helper_cvtpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1930    } else {
1931        gen_helper_cvtps2pi(tcg_env, OP_PTR0, OP_PTR2);
1932    }
1933}
1934
1935static void gen_CVTTPx2PI(DisasContext *s, X86DecodedInsn *decode)
1936{
1937    gen_helper_enter_mmx(tcg_env);
1938    if (s->prefix & PREFIX_DATA) {
1939        gen_helper_cvttpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1940    } else {
1941        gen_helper_cvttps2pi(tcg_env, OP_PTR0, OP_PTR2);
1942    }
1943}
1944
1945static void gen_CWD(DisasContext *s, X86DecodedInsn *decode)
1946{
1947    int shift = 8 << decode->op[0].ot;
1948
1949    tcg_gen_sextract_tl(s->T0, s->T0, shift - 1, 1);
1950}
1951
1952static void gen_DAA(DisasContext *s, X86DecodedInsn *decode)
1953{
1954    gen_update_cc_op(s);
1955    gen_helper_daa(tcg_env);
1956    assume_cc_op(s, CC_OP_EFLAGS);
1957}
1958
1959static void gen_DAS(DisasContext *s, X86DecodedInsn *decode)
1960{
1961    gen_update_cc_op(s);
1962    gen_helper_das(tcg_env);
1963    assume_cc_op(s, CC_OP_EFLAGS);
1964}
1965
1966static void gen_DEC(DisasContext *s, X86DecodedInsn *decode)
1967{
1968    MemOp ot = decode->op[1].ot;
1969
1970    tcg_gen_movi_tl(s->T1, -1);
1971    if (s->prefix & PREFIX_LOCK) {
1972        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1973                                    s->mem_index, ot | MO_LE);
1974    } else {
1975        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1976    }
1977    prepare_update_cc_incdec(decode, s, CC_OP_DECB + ot);
1978}
1979
1980static void gen_DIV(DisasContext *s, X86DecodedInsn *decode)
1981{
1982    MemOp ot = decode->op[1].ot;
1983
1984    switch(ot) {
1985    case MO_8:
1986        gen_helper_divb_AL(tcg_env, s->T0);
1987        break;
1988    case MO_16:
1989        gen_helper_divw_AX(tcg_env, s->T0);
1990        break;
1991    default:
1992    case MO_32:
1993        gen_helper_divl_EAX(tcg_env, s->T0);
1994        break;
1995#ifdef TARGET_X86_64
1996    case MO_64:
1997        gen_helper_divq_EAX(tcg_env, s->T0);
1998        break;
1999#endif
2000    }
2001}
2002
2003static void gen_EMMS(DisasContext *s, X86DecodedInsn *decode)
2004{
2005    gen_helper_emms(tcg_env);
2006}
2007
2008static void gen_ENTER(DisasContext *s, X86DecodedInsn *decode)
2009{
2010   gen_enter(s, decode->op[1].imm, decode->op[2].imm);
2011}
2012
2013static void gen_EXTRQ_i(DisasContext *s, X86DecodedInsn *decode)
2014{
2015    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
2016    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
2017
2018    gen_helper_extrq_i(tcg_env, OP_PTR0, index, length);
2019}
2020
2021static void gen_EXTRQ_r(DisasContext *s, X86DecodedInsn *decode)
2022{
2023    gen_helper_extrq_r(tcg_env, OP_PTR0, OP_PTR2);
2024}
2025
2026static void gen_FXRSTOR(DisasContext *s, X86DecodedInsn *decode)
2027{
2028    if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
2029        gen_NM_exception(s);
2030    } else {
2031        gen_helper_fxrstor(tcg_env, s->A0);
2032    }
2033}
2034
2035static void gen_FXSAVE(DisasContext *s, X86DecodedInsn *decode)
2036{
2037    if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
2038        gen_NM_exception(s);
2039    } else {
2040        gen_helper_fxsave(tcg_env, s->A0);
2041    }
2042}
2043
2044static void gen_HLT(DisasContext *s, X86DecodedInsn *decode)
2045{
2046#ifdef CONFIG_SYSTEM_ONLY
2047    gen_update_cc_op(s);
2048    gen_update_eip_next(s);
2049    gen_helper_hlt(tcg_env);
2050    s->base.is_jmp = DISAS_NORETURN;
2051#endif
2052}
2053
2054static void gen_IDIV(DisasContext *s, X86DecodedInsn *decode)
2055{
2056    MemOp ot = decode->op[1].ot;
2057
2058    switch(ot) {
2059    case MO_8:
2060        gen_helper_idivb_AL(tcg_env, s->T0);
2061        break;
2062    case MO_16:
2063        gen_helper_idivw_AX(tcg_env, s->T0);
2064        break;
2065    default:
2066    case MO_32:
2067        gen_helper_idivl_EAX(tcg_env, s->T0);
2068        break;
2069#ifdef TARGET_X86_64
2070    case MO_64:
2071        gen_helper_idivq_EAX(tcg_env, s->T0);
2072        break;
2073#endif
2074    }
2075}
2076
2077static void gen_IMUL3(DisasContext *s, X86DecodedInsn *decode)
2078{
2079    MemOp ot = decode->op[0].ot;
2080    TCGv cc_src_rhs;
2081
2082    switch (ot) {
2083    case MO_16:
2084        /* s->T0 already sign-extended */
2085        tcg_gen_ext16s_tl(s->T1, s->T1);
2086        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2087        /* Compare the full result to the extension of the truncated result.  */
2088        tcg_gen_ext16s_tl(s->T1, s->T0);
2089        cc_src_rhs = s->T0;
2090        break;
2091
2092    case MO_32:
2093#ifdef TARGET_X86_64
2094        if (TCG_TARGET_REG_BITS == 64) {
2095            /*
2096             * This produces fewer TCG ops, and better code if flags are needed,
2097             * but it requires a 64-bit multiply even if they are not.  Use it
2098             * only if the target has 64-bits registers.
2099             *
2100             * s->T0 is already sign-extended.
2101             */
2102            tcg_gen_ext32s_tl(s->T1, s->T1);
2103            tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2104            /* Compare the full result to the extension of the truncated result.  */
2105            tcg_gen_ext32s_tl(s->T1, s->T0);
2106            cc_src_rhs = s->T0;
2107        } else {
2108            /* Variant that only needs a 32-bit widening multiply.  */
2109            TCGv_i32 hi = tcg_temp_new_i32();
2110            TCGv_i32 lo = tcg_temp_new_i32();
2111            tcg_gen_trunc_tl_i32(lo, s->T0);
2112            tcg_gen_trunc_tl_i32(hi, s->T1);
2113            tcg_gen_muls2_i32(lo, hi, lo, hi);
2114            tcg_gen_extu_i32_tl(s->T0, lo);
2115
2116            cc_src_rhs = tcg_temp_new();
2117            tcg_gen_extu_i32_tl(cc_src_rhs, hi);
2118            /* Compare the high part to the sign bit of the truncated result */
2119            tcg_gen_sari_i32(lo, lo, 31);
2120            tcg_gen_extu_i32_tl(s->T1, lo);
2121        }
2122        break;
2123
2124    case MO_64:
2125#endif
2126        cc_src_rhs = tcg_temp_new();
2127        tcg_gen_muls2_tl(s->T0, cc_src_rhs, s->T0, s->T1);
2128        /* Compare the high part to the sign bit of the truncated result */
2129        tcg_gen_sari_tl(s->T1, s->T0, TARGET_LONG_BITS - 1);
2130        break;
2131
2132    default:
2133        g_assert_not_reached();
2134    }
2135
2136    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
2137    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
2138}
2139
2140static void gen_IMUL(DisasContext *s, X86DecodedInsn *decode)
2141{
2142    MemOp ot = decode->op[1].ot;
2143    TCGv cc_src_rhs;
2144
2145    switch (ot) {
2146    case MO_8:
2147        /* s->T0 already sign-extended */
2148        tcg_gen_ext8s_tl(s->T1, s->T1);
2149        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2150        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2151        /* Compare the full result to the extension of the truncated result.  */
2152        tcg_gen_ext8s_tl(s->T1, s->T0);
2153        cc_src_rhs = s->T0;
2154        break;
2155
2156    case MO_16:
2157        /* s->T0 already sign-extended */
2158        tcg_gen_ext16s_tl(s->T1, s->T1);
2159        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2160        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2161        tcg_gen_shri_tl(s->T1, s->T0, 16);
2162        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
2163        /* Compare the full result to the extension of the truncated result.  */
2164        tcg_gen_ext16s_tl(s->T1, s->T0);
2165        cc_src_rhs = s->T0;
2166        break;
2167
2168    case MO_32:
2169#ifdef TARGET_X86_64
2170        /* s->T0 already sign-extended */
2171        tcg_gen_ext32s_tl(s->T1, s->T1);
2172        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2173        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
2174        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
2175        /* Compare the full result to the extension of the truncated result.  */
2176        tcg_gen_ext32s_tl(s->T1, s->T0);
2177        cc_src_rhs = s->T0;
2178        break;
2179
2180    case MO_64:
2181#endif
2182        tcg_gen_muls2_tl(s->T0, cpu_regs[R_EDX], s->T0, s->T1);
2183        tcg_gen_mov_tl(cpu_regs[R_EAX], s->T0);
2184
2185        /* Compare the high part to the sign bit of the truncated result */
2186        tcg_gen_negsetcondi_tl(TCG_COND_LT, s->T1, s->T0, 0);
2187        cc_src_rhs = cpu_regs[R_EDX];
2188        break;
2189
2190    default:
2191        g_assert_not_reached();
2192    }
2193
2194    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
2195    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
2196}
2197
2198static void gen_IN(DisasContext *s, X86DecodedInsn *decode)
2199{
2200    MemOp ot = decode->op[0].ot;
2201    TCGv_i32 port = tcg_temp_new_i32();
2202
2203    tcg_gen_trunc_tl_i32(port, s->T0);
2204    tcg_gen_ext16u_i32(port, port);
2205    if (!gen_check_io(s, ot, port, SVM_IOIO_TYPE_MASK)) {
2206        return;
2207    }
2208    translator_io_start(&s->base);
2209    gen_helper_in_func(ot, s->T0, port);
2210    gen_writeback(s, decode, 0, s->T0);
2211    gen_bpt_io(s, port, ot);
2212}
2213
2214static void gen_INC(DisasContext *s, X86DecodedInsn *decode)
2215{
2216    MemOp ot = decode->op[1].ot;
2217
2218    tcg_gen_movi_tl(s->T1, 1);
2219    if (s->prefix & PREFIX_LOCK) {
2220        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
2221                                    s->mem_index, ot | MO_LE);
2222    } else {
2223        tcg_gen_add_tl(s->T0, s->T0, s->T1);
2224    }
2225    prepare_update_cc_incdec(decode, s, CC_OP_INCB + ot);
2226}
2227
2228static void gen_INS(DisasContext *s, X86DecodedInsn *decode)
2229{
2230    MemOp ot = decode->op[1].ot;
2231    TCGv_i32 port = tcg_temp_new_i32();
2232
2233    tcg_gen_trunc_tl_i32(port, s->T1);
2234    tcg_gen_ext16u_i32(port, port);
2235    if (!gen_check_io(s, ot, port,
2236                      SVM_IOIO_TYPE_MASK | SVM_IOIO_STR_MASK)) {
2237        return;
2238    }
2239
2240    translator_io_start(&s->base);
2241    gen_repz(s, ot, gen_ins);
2242}
2243
2244static void gen_INSERTQ_i(DisasContext *s, X86DecodedInsn *decode)
2245{
2246    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
2247    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
2248
2249    gen_helper_insertq_i(tcg_env, OP_PTR0, OP_PTR1, index, length);
2250}
2251
2252static void gen_INSERTQ_r(DisasContext *s, X86DecodedInsn *decode)
2253{
2254    gen_helper_insertq_r(tcg_env, OP_PTR0, OP_PTR2);
2255}
2256
2257static void gen_INT(DisasContext *s, X86DecodedInsn *decode)
2258{
2259    gen_interrupt(s, decode->immediate);
2260}
2261
2262static void gen_INT1(DisasContext *s, X86DecodedInsn *decode)
2263{
2264    gen_update_cc_op(s);
2265    gen_update_eip_next(s);
2266    gen_helper_icebp(tcg_env);
2267    s->base.is_jmp = DISAS_NORETURN;
2268}
2269
2270static void gen_INT3(DisasContext *s, X86DecodedInsn *decode)
2271{
2272    gen_interrupt(s, EXCP03_INT3);
2273}
2274
2275static void gen_INTO(DisasContext *s, X86DecodedInsn *decode)
2276{
2277    gen_update_cc_op(s);
2278    gen_update_eip_cur(s);
2279    gen_helper_into(tcg_env, cur_insn_len_i32(s));
2280}
2281
2282static void gen_IRET(DisasContext *s, X86DecodedInsn *decode)
2283{
2284    if (!PE(s) || VM86(s)) {
2285        gen_helper_iret_real(tcg_env, tcg_constant_i32(s->dflag - 1));
2286    } else {
2287        gen_helper_iret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
2288                                  eip_next_i32(s));
2289    }
2290    assume_cc_op(s, CC_OP_EFLAGS);
2291    s->base.is_jmp = DISAS_EOB_ONLY;
2292}
2293
2294static void gen_Jcc(DisasContext *s, X86DecodedInsn *decode)
2295{
2296    TCGLabel *taken = gen_new_label();
2297
2298    gen_bnd_jmp(s);
2299    gen_jcc(s, decode->b & 0xf, taken);
2300    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2301}
2302
2303static void gen_JCXZ(DisasContext *s, X86DecodedInsn *decode)
2304{
2305    TCGLabel *taken = gen_new_label();
2306
2307    gen_update_cc_op(s);
2308    gen_op_jz_ecx(s, taken);
2309    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2310}
2311
2312static void gen_JMP(DisasContext *s, X86DecodedInsn *decode)
2313{
2314    gen_update_cc_op(s);
2315    gen_jmp_rel(s, s->dflag, decode->immediate, 0);
2316}
2317
2318static void gen_JMP_m(DisasContext *s, X86DecodedInsn *decode)
2319{
2320    gen_op_jmp_v(s, s->T0);
2321    gen_bnd_jmp(s);
2322    s->base.is_jmp = DISAS_JUMP;
2323}
2324
2325static void gen_JMPF(DisasContext *s, X86DecodedInsn *decode)
2326{
2327    gen_far_jmp(s);
2328}
2329
2330static void gen_JMPF_m(DisasContext *s, X86DecodedInsn *decode)
2331{
2332    MemOp ot = decode->op[1].ot;
2333
2334    gen_op_ld_v(s, ot, s->T0, s->A0);
2335    gen_add_A0_im(s, 1 << ot);
2336    gen_op_ld_v(s, MO_16, s->T1, s->A0);
2337    gen_far_jmp(s);
2338}
2339
2340static void gen_LAHF(DisasContext *s, X86DecodedInsn *decode)
2341{
2342    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
2343        return gen_illegal_opcode(s);
2344    }
2345    gen_compute_eflags(s);
2346    /* Note: gen_compute_eflags() only gives the condition codes */
2347    tcg_gen_ori_tl(s->T0, cpu_cc_src, 0x02);
2348    tcg_gen_deposit_tl(cpu_regs[R_EAX], cpu_regs[R_EAX], s->T0, 8, 8);
2349}
2350
2351static void gen_LAR(DisasContext *s, X86DecodedInsn *decode)
2352{
2353    MemOp ot = decode->op[0].ot;
2354    TCGv result = tcg_temp_new();
2355    TCGv dest;
2356
2357    gen_compute_eflags(s);
2358    gen_update_cc_op(s);
2359    gen_helper_lar(result, tcg_env, s->T0);
2360
2361    /* Perform writeback here to skip it if ZF=0.  */
2362    decode->op[0].unit = X86_OP_SKIP;
2363    dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, result, result);
2364    tcg_gen_movcond_tl(TCG_COND_TSTNE, dest, cpu_cc_src, tcg_constant_tl(CC_Z),
2365                       result, dest);
2366}
2367
2368static void gen_LDMXCSR(DisasContext *s, X86DecodedInsn *decode)
2369{
2370    TCGv_i32 tmp = tcg_temp_new_i32();
2371
2372    tcg_gen_trunc_tl_i32(tmp, s->T0);
2373    gen_helper_ldmxcsr(tcg_env, tmp);
2374}
2375
2376static void gen_lxx_seg(DisasContext *s, X86DecodedInsn *decode, int seg)
2377{
2378    MemOp ot = decode->op[0].ot;
2379
2380    /* Offset already in s->T0.  */
2381    gen_add_A0_im(s, 1 << ot);
2382    gen_op_ld_v(s, MO_16, s->T1, s->A0);
2383
2384    /* load the segment here to handle exceptions properly */
2385    gen_movl_seg(s, seg, s->T1, false);
2386}
2387
2388static void gen_LDS(DisasContext *s, X86DecodedInsn *decode)
2389{
2390    gen_lxx_seg(s, decode, R_DS);
2391}
2392
2393static void gen_LEA(DisasContext *s, X86DecodedInsn *decode)
2394{
2395    TCGv ea = gen_lea_modrm_1(s, decode->mem, false);
2396    gen_lea_v_seg_dest(s, s->aflag, s->T0, ea, -1, -1);
2397}
2398
2399static void gen_LEAVE(DisasContext *s, X86DecodedInsn *decode)
2400{
2401    gen_leave(s);
2402}
2403
2404static void gen_LES(DisasContext *s, X86DecodedInsn *decode)
2405{
2406    gen_lxx_seg(s, decode, R_ES);
2407}
2408
2409static void gen_LFENCE(DisasContext *s, X86DecodedInsn *decode)
2410{
2411    tcg_gen_mb(TCG_MO_LD_LD | TCG_BAR_SC);
2412}
2413
2414static void gen_LFS(DisasContext *s, X86DecodedInsn *decode)
2415{
2416    gen_lxx_seg(s, decode, R_FS);
2417}
2418
2419static void gen_LGS(DisasContext *s, X86DecodedInsn *decode)
2420{
2421    gen_lxx_seg(s, decode, R_GS);
2422}
2423
2424static void gen_LODS(DisasContext *s, X86DecodedInsn *decode)
2425{
2426    MemOp ot = decode->op[1].ot;
2427    gen_repz(s, ot, gen_lods);
2428}
2429
2430static void gen_LOOP(DisasContext *s, X86DecodedInsn *decode)
2431{
2432    TCGLabel *taken = gen_new_label();
2433
2434    gen_update_cc_op(s);
2435    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2436    gen_op_jnz_ecx(s, taken);
2437    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2438}
2439
2440static void gen_LOOPE(DisasContext *s, X86DecodedInsn *decode)
2441{
2442    TCGLabel *taken = gen_new_label();
2443    TCGLabel *not_taken = gen_new_label();
2444
2445    gen_update_cc_op(s);
2446    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2447    gen_op_jz_ecx(s, not_taken);
2448    gen_jcc(s, (JCC_Z << 1), taken); /* jz taken */
2449    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2450}
2451
2452static void gen_LOOPNE(DisasContext *s, X86DecodedInsn *decode)
2453{
2454    TCGLabel *taken = gen_new_label();
2455    TCGLabel *not_taken = gen_new_label();
2456
2457    gen_update_cc_op(s);
2458    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2459    gen_op_jz_ecx(s, not_taken);
2460    gen_jcc(s, (JCC_Z << 1) | 1, taken); /* jnz taken */
2461    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2462}
2463
2464static void gen_LSL(DisasContext *s, X86DecodedInsn *decode)
2465{
2466    MemOp ot = decode->op[0].ot;
2467    TCGv result = tcg_temp_new();
2468    TCGv dest;
2469
2470    gen_compute_eflags(s);
2471    gen_update_cc_op(s);
2472    gen_helper_lsl(result, tcg_env, s->T0);
2473
2474    /* Perform writeback here to skip it if ZF=0.  */
2475    decode->op[0].unit = X86_OP_SKIP;
2476    dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, result, result);
2477    tcg_gen_movcond_tl(TCG_COND_TSTNE, dest, cpu_cc_src, tcg_constant_tl(CC_Z),
2478                       result, dest);
2479}
2480
2481static void gen_LSS(DisasContext *s, X86DecodedInsn *decode)
2482{
2483    gen_lxx_seg(s, decode, R_SS);
2484}
2485
2486static void gen_LZCNT(DisasContext *s, X86DecodedInsn *decode)
2487{
2488    MemOp ot = decode->op[0].ot;
2489
2490    /* C bit (cc_src) is defined related to the input.  */
2491    decode->cc_src = tcg_temp_new();
2492    decode->cc_dst = s->T0;
2493    decode->cc_op = CC_OP_BMILGB + ot;
2494    tcg_gen_mov_tl(decode->cc_src, s->T0);
2495
2496    /*
2497     * Reduce the target_ulong result by the number of zeros that
2498     * we expect to find at the top.
2499     */
2500    tcg_gen_clzi_tl(s->T0, s->T0, TARGET_LONG_BITS);
2501    tcg_gen_subi_tl(s->T0, s->T0, TARGET_LONG_BITS - (8 << ot));
2502}
2503
2504static void gen_MFENCE(DisasContext *s, X86DecodedInsn *decode)
2505{
2506    tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
2507}
2508
2509static void gen_MOV(DisasContext *s, X86DecodedInsn *decode)
2510{
2511    /* nothing to do! */
2512}
2513#define gen_NOP gen_MOV
2514
2515static void gen_MASKMOV(DisasContext *s, X86DecodedInsn *decode)
2516{
2517    gen_lea_v_seg(s, cpu_regs[R_EDI], R_DS, s->override);
2518
2519    if (s->prefix & PREFIX_DATA) {
2520        gen_helper_maskmov_xmm(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2521    } else {
2522        gen_helper_maskmov_mmx(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2523    }
2524}
2525
2526static void gen_MOVBE(DisasContext *s, X86DecodedInsn *decode)
2527{
2528    MemOp ot = decode->op[0].ot;
2529
2530    /* M operand type does not load/store */
2531    if (decode->e.op0 == X86_TYPE_M) {
2532        tcg_gen_qemu_st_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2533    } else {
2534        tcg_gen_qemu_ld_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2535    }
2536}
2537
2538static void gen_MOVD_from(DisasContext *s, X86DecodedInsn *decode)
2539{
2540    MemOp ot = decode->op[2].ot;
2541
2542    switch (ot) {
2543    case MO_32:
2544#ifdef TARGET_X86_64
2545        tcg_gen_ld32u_tl(s->T0, tcg_env, decode->op[2].offset);
2546        break;
2547    case MO_64:
2548#endif
2549        tcg_gen_ld_tl(s->T0, tcg_env, decode->op[2].offset);
2550        break;
2551    default:
2552        abort();
2553    }
2554}
2555
2556static void gen_MOVD_to(DisasContext *s, X86DecodedInsn *decode)
2557{
2558    MemOp ot = decode->op[2].ot;
2559    int vec_len = vector_len(s, decode);
2560    int lo_ofs = vector_elem_offset(&decode->op[0], ot, 0);
2561
2562    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2563
2564    switch (ot) {
2565    case MO_32:
2566#ifdef TARGET_X86_64
2567        tcg_gen_st32_tl(s->T1, tcg_env, lo_ofs);
2568        break;
2569    case MO_64:
2570#endif
2571        tcg_gen_st_tl(s->T1, tcg_env, lo_ofs);
2572        break;
2573    default:
2574        g_assert_not_reached();
2575    }
2576}
2577
2578static void gen_MOVDQ(DisasContext *s, X86DecodedInsn *decode)
2579{
2580    gen_store_sse(s, decode, decode->op[2].offset);
2581}
2582
2583static void gen_MOVMSK(DisasContext *s, X86DecodedInsn *decode)
2584{
2585    typeof(gen_helper_movmskps_ymm) *ps, *pd, *fn;
2586    TCGv_i32 tmp = tcg_temp_new_i32();
2587
2588    ps = s->vex_l ? gen_helper_movmskps_ymm : gen_helper_movmskps_xmm;
2589    pd = s->vex_l ? gen_helper_movmskpd_ymm : gen_helper_movmskpd_xmm;
2590    fn = s->prefix & PREFIX_DATA ? pd : ps;
2591    fn(tmp, tcg_env, OP_PTR2);
2592    tcg_gen_extu_i32_tl(s->T0, tmp);
2593}
2594
2595static void gen_MOVQ(DisasContext *s, X86DecodedInsn *decode)
2596{
2597    int vec_len = vector_len(s, decode);
2598    int lo_ofs = vector_elem_offset(&decode->op[0], MO_64, 0);
2599
2600    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
2601    if (decode->op[0].has_ea) {
2602        tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
2603    } else {
2604        /*
2605         * tcg_gen_gvec_dup_i64(MO_64, op0.offset, 8, vec_len, s->tmp1_64) would
2606         * seem to work, but it does not on big-endian platforms; the cleared parts
2607         * are always at higher addresses, but cross-endian emulation inverts the
2608         * byte order so that the cleared parts need to be at *lower* addresses.
2609         * Because oprsz is 8, we see this here even for SSE; but more in general,
2610         * it disqualifies using oprsz < maxsz to emulate VEX128.
2611         */
2612        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2613        tcg_gen_st_i64(s->tmp1_i64, tcg_env, lo_ofs);
2614    }
2615}
2616
2617static void gen_MOVq_dq(DisasContext *s, X86DecodedInsn *decode)
2618{
2619    gen_helper_enter_mmx(tcg_env);
2620    /* Otherwise the same as any other movq.  */
2621    return gen_MOVQ(s, decode);
2622}
2623
2624static void gen_MOVS(DisasContext *s, X86DecodedInsn *decode)
2625{
2626    MemOp ot = decode->op[2].ot;
2627    gen_repz(s, ot, gen_movs);
2628}
2629
2630static void gen_MUL(DisasContext *s, X86DecodedInsn *decode)
2631{
2632    MemOp ot = decode->op[1].ot;
2633
2634    switch (ot) {
2635    case MO_8:
2636        /* s->T0 already zero-extended */
2637        tcg_gen_ext8u_tl(s->T1, s->T1);
2638        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2639        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2640        tcg_gen_andi_tl(s->T1, s->T0, 0xff00);
2641        decode->cc_dst = s->T0;
2642        decode->cc_src = s->T1;
2643        break;
2644
2645    case MO_16:
2646        /* s->T0 already zero-extended */
2647        tcg_gen_ext16u_tl(s->T1, s->T1);
2648        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2649        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2650        tcg_gen_shri_tl(s->T1, s->T0, 16);
2651        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
2652        decode->cc_dst = s->T0;
2653        decode->cc_src = s->T1;
2654        break;
2655
2656    case MO_32:
2657#ifdef TARGET_X86_64
2658        /* s->T0 already zero-extended */
2659        tcg_gen_ext32u_tl(s->T1, s->T1);
2660        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2661        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
2662        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
2663        decode->cc_dst = cpu_regs[R_EAX];
2664        decode->cc_src = cpu_regs[R_EDX];
2665        break;
2666
2667    case MO_64:
2668#endif
2669        tcg_gen_mulu2_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], s->T0, s->T1);
2670        decode->cc_dst = cpu_regs[R_EAX];
2671        decode->cc_src = cpu_regs[R_EDX];
2672        break;
2673
2674    default:
2675        g_assert_not_reached();
2676    }
2677
2678    decode->cc_op = CC_OP_MULB + ot;
2679}
2680
2681static void gen_MULX(DisasContext *s, X86DecodedInsn *decode)
2682{
2683    MemOp ot = decode->op[0].ot;
2684
2685    /* low part of result in VEX.vvvv, high in MODRM */
2686    switch (ot) {
2687    case MO_32:
2688#ifdef TARGET_X86_64
2689        {
2690            TCGv_i32 t0 = tcg_temp_new_i32();
2691            TCGv_i32 t1 = tcg_temp_new_i32();
2692
2693            tcg_gen_trunc_tl_i32(t0, s->T0);
2694            tcg_gen_trunc_tl_i32(t1, s->T1);
2695            tcg_gen_mulu2_i32(t0, t1, t0, t1);
2696            tcg_gen_extu_i32_tl(cpu_regs[s->vex_v], t0);
2697            tcg_gen_extu_i32_tl(s->T0, t1);
2698            break;
2699        }
2700
2701    case MO_64:
2702#endif
2703        tcg_gen_mulu2_tl(cpu_regs[s->vex_v], s->T0, s->T0, s->T1);
2704        break;
2705
2706    default:
2707        g_assert_not_reached();
2708    }
2709}
2710
2711static void gen_NEG(DisasContext *s, X86DecodedInsn *decode)
2712{
2713    MemOp ot = decode->op[0].ot;
2714    TCGv oldv = tcg_temp_new();
2715
2716    if (s->prefix & PREFIX_LOCK) {
2717        TCGv newv = tcg_temp_new();
2718        TCGv cmpv = tcg_temp_new();
2719        TCGLabel *label1 = gen_new_label();
2720
2721        gen_set_label(label1);
2722        gen_op_ld_v(s, ot, oldv, s->A0);
2723        tcg_gen_neg_tl(newv, oldv);
2724        tcg_gen_atomic_cmpxchg_tl(cmpv, s->A0, oldv, newv,
2725                                  s->mem_index, ot | MO_LE);
2726        tcg_gen_brcond_tl(TCG_COND_NE, oldv, cmpv, label1);
2727    } else {
2728        tcg_gen_mov_tl(oldv, s->T0);
2729    }
2730    tcg_gen_neg_tl(s->T0, oldv);
2731
2732    decode->cc_dst = s->T0;
2733    decode->cc_src = oldv;
2734    tcg_gen_movi_tl(s->cc_srcT, 0);
2735    decode->cc_op = CC_OP_SUBB + ot;
2736}
2737
2738static void gen_NOT(DisasContext *s, X86DecodedInsn *decode)
2739{
2740    MemOp ot = decode->op[0].ot;
2741
2742    if (s->prefix & PREFIX_LOCK) {
2743        tcg_gen_movi_tl(s->T0, ~0);
2744        tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T0,
2745                                    s->mem_index, ot | MO_LE);
2746    } else {
2747        tcg_gen_not_tl(s->T0, s->T0);
2748    }
2749}
2750
2751static void gen_OR(DisasContext *s, X86DecodedInsn *decode)
2752{
2753    MemOp ot = decode->op[1].ot;
2754
2755    if (s->prefix & PREFIX_LOCK) {
2756        tcg_gen_atomic_or_fetch_tl(s->T0, s->A0, s->T1,
2757                                   s->mem_index, ot | MO_LE);
2758    } else {
2759        tcg_gen_or_tl(s->T0, s->T0, s->T1);
2760    }
2761    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
2762}
2763
2764static void gen_OUT(DisasContext *s, X86DecodedInsn *decode)
2765{
2766    MemOp ot = decode->op[1].ot;
2767    TCGv_i32 port = tcg_temp_new_i32();
2768    TCGv_i32 value = tcg_temp_new_i32();
2769
2770    tcg_gen_trunc_tl_i32(port, s->T1);
2771    tcg_gen_ext16u_i32(port, port);
2772    if (!gen_check_io(s, ot, port, 0)) {
2773        return;
2774    }
2775    tcg_gen_trunc_tl_i32(value, s->T0);
2776    translator_io_start(&s->base);
2777    gen_helper_out_func(ot, port, value);
2778    gen_bpt_io(s, port, ot);
2779}
2780
2781static void gen_OUTS(DisasContext *s, X86DecodedInsn *decode)
2782{
2783    MemOp ot = decode->op[1].ot;
2784    TCGv_i32 port = tcg_temp_new_i32();
2785
2786    tcg_gen_trunc_tl_i32(port, s->T1);
2787    tcg_gen_ext16u_i32(port, port);
2788    if (!gen_check_io(s, ot, port, SVM_IOIO_STR_MASK)) {
2789        return;
2790    }
2791
2792    translator_io_start(&s->base);
2793    gen_repz(s, ot, gen_outs);
2794}
2795
2796static void gen_PALIGNR(DisasContext *s, X86DecodedInsn *decode)
2797{
2798    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2799    if (!(s->prefix & PREFIX_DATA)) {
2800        gen_helper_palignr_mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2801    } else if (!s->vex_l) {
2802        gen_helper_palignr_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2803    } else {
2804        gen_helper_palignr_ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2805    }
2806}
2807
2808static void gen_PANDN(DisasContext *s, X86DecodedInsn *decode)
2809{
2810    int vec_len = vector_len(s, decode);
2811
2812    /* Careful, operand order is reversed!  */
2813    tcg_gen_gvec_andc(MO_64,
2814                      decode->op[0].offset, decode->op[2].offset,
2815                      decode->op[1].offset, vec_len, vec_len);
2816}
2817
2818static void gen_PAUSE(DisasContext *s, X86DecodedInsn *decode)
2819{
2820    gen_update_cc_op(s);
2821    gen_update_eip_next(s);
2822    gen_helper_pause(tcg_env);
2823    s->base.is_jmp = DISAS_NORETURN;
2824}
2825
2826static void gen_PCMPESTRI(DisasContext *s, X86DecodedInsn *decode)
2827{
2828    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2829    gen_helper_pcmpestri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2830    assume_cc_op(s, CC_OP_EFLAGS);
2831}
2832
2833static void gen_PCMPESTRM(DisasContext *s, X86DecodedInsn *decode)
2834{
2835    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2836    gen_helper_pcmpestrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2837    assume_cc_op(s, CC_OP_EFLAGS);
2838    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2839        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2840                             16, 16, 0);
2841    }
2842}
2843
2844static void gen_PCMPISTRI(DisasContext *s, X86DecodedInsn *decode)
2845{
2846    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2847    gen_helper_pcmpistri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2848    assume_cc_op(s, CC_OP_EFLAGS);
2849}
2850
2851static void gen_PCMPISTRM(DisasContext *s, X86DecodedInsn *decode)
2852{
2853    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2854    gen_helper_pcmpistrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2855    assume_cc_op(s, CC_OP_EFLAGS);
2856    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2857        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2858                             16, 16, 0);
2859    }
2860}
2861
2862static void gen_PDEP(DisasContext *s, X86DecodedInsn *decode)
2863{
2864    gen_helper_pdep(s->T0, s->T0, s->T1);
2865}
2866
2867static void gen_PEXT(DisasContext *s, X86DecodedInsn *decode)
2868{
2869    gen_helper_pext(s->T0, s->T0, s->T1);
2870}
2871
2872static inline void gen_pextr(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
2873{
2874    int vec_len = vector_len(s, decode);
2875    int mask = (vec_len >> ot) - 1;
2876    int val = decode->immediate & mask;
2877
2878    switch (ot) {
2879    case MO_8:
2880        tcg_gen_ld8u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2881        break;
2882    case MO_16:
2883        tcg_gen_ld16u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2884        break;
2885    case MO_32:
2886#ifdef TARGET_X86_64
2887        tcg_gen_ld32u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2888        break;
2889    case MO_64:
2890#endif
2891        tcg_gen_ld_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2892        break;
2893    default:
2894        abort();
2895    }
2896}
2897
2898static void gen_PEXTRB(DisasContext *s, X86DecodedInsn *decode)
2899{
2900    gen_pextr(s, decode, MO_8);
2901}
2902
2903static void gen_PEXTRW(DisasContext *s, X86DecodedInsn *decode)
2904{
2905    gen_pextr(s, decode, MO_16);
2906}
2907
2908static void gen_PEXTR(DisasContext *s, X86DecodedInsn *decode)
2909{
2910    MemOp ot = decode->op[0].ot;
2911    gen_pextr(s, decode, ot);
2912}
2913
2914static inline void gen_pinsr(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
2915{
2916    int vec_len = vector_len(s, decode);
2917    int mask = (vec_len >> ot) - 1;
2918    int val = decode->immediate & mask;
2919
2920    if (decode->op[1].offset != decode->op[0].offset) {
2921        assert(vec_len == 16);
2922        gen_store_sse(s, decode, decode->op[1].offset);
2923    }
2924
2925    switch (ot) {
2926    case MO_8:
2927        tcg_gen_st8_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2928        break;
2929    case MO_16:
2930        tcg_gen_st16_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2931        break;
2932    case MO_32:
2933#ifdef TARGET_X86_64
2934        tcg_gen_st32_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2935        break;
2936    case MO_64:
2937#endif
2938        tcg_gen_st_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2939        break;
2940    default:
2941        abort();
2942    }
2943}
2944
2945static void gen_PINSRB(DisasContext *s, X86DecodedInsn *decode)
2946{
2947    gen_pinsr(s, decode, MO_8);
2948}
2949
2950static void gen_PINSRW(DisasContext *s, X86DecodedInsn *decode)
2951{
2952    gen_pinsr(s, decode, MO_16);
2953}
2954
2955static void gen_PINSR(DisasContext *s, X86DecodedInsn *decode)
2956{
2957    gen_pinsr(s, decode, decode->op[2].ot);
2958}
2959
2960static void gen_pmovmskb_i64(TCGv_i64 d, TCGv_i64 s)
2961{
2962    TCGv_i64 t = tcg_temp_new_i64();
2963
2964    tcg_gen_andi_i64(d, s, 0x8080808080808080ull);
2965
2966    /*
2967     * After each shift+or pair:
2968     * 0:  a.......b.......c.......d.......e.......f.......g.......h.......
2969     * 7:  ab......bc......cd......de......ef......fg......gh......h.......
2970     * 14: abcd....bcde....cdef....defg....efgh....fgh.....gh......h.......
2971     * 28: abcdefghbcdefgh.cdefgh..defgh...efgh....fgh.....gh......h.......
2972     * The result is left in the high bits of the word.
2973     */
2974    tcg_gen_shli_i64(t, d, 7);
2975    tcg_gen_or_i64(d, d, t);
2976    tcg_gen_shli_i64(t, d, 14);
2977    tcg_gen_or_i64(d, d, t);
2978    tcg_gen_shli_i64(t, d, 28);
2979    tcg_gen_or_i64(d, d, t);
2980}
2981
2982static void gen_pmovmskb_vec(unsigned vece, TCGv_vec d, TCGv_vec s)
2983{
2984    TCGv_vec t = tcg_temp_new_vec_matching(d);
2985    TCGv_vec m = tcg_constant_vec_matching(d, MO_8, 0x80);
2986
2987    /* See above */
2988    tcg_gen_and_vec(vece, d, s, m);
2989    tcg_gen_shli_vec(vece, t, d, 7);
2990    tcg_gen_or_vec(vece, d, d, t);
2991    tcg_gen_shli_vec(vece, t, d, 14);
2992    tcg_gen_or_vec(vece, d, d, t);
2993    tcg_gen_shli_vec(vece, t, d, 28);
2994    tcg_gen_or_vec(vece, d, d, t);
2995}
2996
2997static void gen_PMOVMSKB(DisasContext *s, X86DecodedInsn *decode)
2998{
2999    static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
3000    static const GVecGen2 g = {
3001        .fni8 = gen_pmovmskb_i64,
3002        .fniv = gen_pmovmskb_vec,
3003        .opt_opc = vecop_list,
3004        .vece = MO_64,
3005        .prefer_i64 = TCG_TARGET_REG_BITS == 64
3006    };
3007    MemOp ot = decode->op[2].ot;
3008    int vec_len = vector_len(s, decode);
3009    TCGv t = tcg_temp_new();
3010
3011    tcg_gen_gvec_2(offsetof(CPUX86State, xmm_t0) + xmm_offset(ot), decode->op[2].offset,
3012                   vec_len, vec_len, &g);
3013    tcg_gen_ld8u_tl(s->T0, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
3014    while (vec_len > 8) {
3015        vec_len -= 8;
3016        if (tcg_op_supported(INDEX_op_extract2, TCG_TYPE_TL, 0)) {
3017            /*
3018             * Load the next byte of the result into the high byte of T.
3019             * TCG does a similar expansion of deposit to shl+extract2; by
3020             * loading the whole word, the shift left is avoided.
3021             */
3022#ifdef TARGET_X86_64
3023            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_Q((vec_len - 1) / 8)));
3024#else
3025            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L((vec_len - 1) / 4)));
3026#endif
3027
3028            tcg_gen_extract2_tl(s->T0, t, s->T0, TARGET_LONG_BITS - 8);
3029        } else {
3030            /*
3031             * The _previous_ value is deposited into bits 8 and higher of t.  Because
3032             * those bits are known to be zero after ld8u, this becomes a shift+or
3033             * if deposit is not available.
3034             */
3035            tcg_gen_ld8u_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
3036            tcg_gen_deposit_tl(s->T0, t, s->T0, 8, TARGET_LONG_BITS - 8);
3037        }
3038    }
3039}
3040
3041static void gen_POP(DisasContext *s, X86DecodedInsn *decode)
3042{
3043    X86DecodedOp *op = &decode->op[0];
3044    MemOp ot = gen_pop_T0(s);
3045
3046    assert(ot >= op->ot);
3047    if (op->has_ea || op->unit == X86_OP_SEG) {
3048        /* NOTE: order is important for MMU exceptions */
3049        gen_writeback(s, decode, 0, s->T0);
3050    }
3051
3052    /* NOTE: writing back registers after update is important for pop %sp */
3053    gen_pop_update(s, ot);
3054}
3055
3056static void gen_POPA(DisasContext *s, X86DecodedInsn *decode)
3057{
3058    gen_popa(s);
3059}
3060
3061static void gen_POPCNT(DisasContext *s, X86DecodedInsn *decode)
3062{
3063    decode->cc_dst = tcg_temp_new();
3064    decode->cc_op = CC_OP_POPCNT;
3065
3066    tcg_gen_mov_tl(decode->cc_dst, s->T0);
3067    tcg_gen_ctpop_tl(s->T0, s->T0);
3068}
3069
3070static void gen_POPF(DisasContext *s, X86DecodedInsn *decode)
3071{
3072    MemOp ot;
3073    int mask = TF_MASK | AC_MASK | ID_MASK | NT_MASK;
3074
3075    if (CPL(s) == 0) {
3076        mask |= IF_MASK | IOPL_MASK;
3077    } else if (CPL(s) <= IOPL(s)) {
3078        mask |= IF_MASK;
3079    }
3080    if (s->dflag == MO_16) {
3081        mask &= 0xffff;
3082    }
3083
3084    ot = gen_pop_T0(s);
3085    gen_helper_write_eflags(tcg_env, s->T0, tcg_constant_i32(mask));
3086    gen_pop_update(s, ot);
3087    set_cc_op(s, CC_OP_EFLAGS);
3088    /* abort translation because TF/AC flag may change */
3089    s->base.is_jmp = DISAS_EOB_NEXT;
3090}
3091
3092static void gen_PSHUFW(DisasContext *s, X86DecodedInsn *decode)
3093{
3094    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
3095    gen_helper_pshufw_mmx(OP_PTR0, OP_PTR1, imm);
3096}
3097
3098static void gen_PSRLW_i(DisasContext *s, X86DecodedInsn *decode)
3099{
3100    int vec_len = vector_len(s, decode);
3101
3102    if (decode->immediate >= 16) {
3103        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3104    } else {
3105        tcg_gen_gvec_shri(MO_16,
3106                          decode->op[0].offset, decode->op[1].offset,
3107                          decode->immediate, vec_len, vec_len);
3108    }
3109}
3110
3111static void gen_PSLLW_i(DisasContext *s, X86DecodedInsn *decode)
3112{
3113    int vec_len = vector_len(s, decode);
3114
3115    if (decode->immediate >= 16) {
3116        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3117    } else {
3118        tcg_gen_gvec_shli(MO_16,
3119                          decode->op[0].offset, decode->op[1].offset,
3120                          decode->immediate, vec_len, vec_len);
3121    }
3122}
3123
3124static void gen_PSRAW_i(DisasContext *s, X86DecodedInsn *decode)
3125{
3126    int vec_len = vector_len(s, decode);
3127
3128    if (decode->immediate >= 16) {
3129        decode->immediate = 15;
3130    }
3131    tcg_gen_gvec_sari(MO_16,
3132                      decode->op[0].offset, decode->op[1].offset,
3133                      decode->immediate, vec_len, vec_len);
3134}
3135
3136static void gen_PSRLD_i(DisasContext *s, X86DecodedInsn *decode)
3137{
3138    int vec_len = vector_len(s, decode);
3139
3140    if (decode->immediate >= 32) {
3141        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3142    } else {
3143        tcg_gen_gvec_shri(MO_32,
3144                          decode->op[0].offset, decode->op[1].offset,
3145                          decode->immediate, vec_len, vec_len);
3146    }
3147}
3148
3149static void gen_PSLLD_i(DisasContext *s, X86DecodedInsn *decode)
3150{
3151    int vec_len = vector_len(s, decode);
3152
3153    if (decode->immediate >= 32) {
3154        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3155    } else {
3156        tcg_gen_gvec_shli(MO_32,
3157                          decode->op[0].offset, decode->op[1].offset,
3158                          decode->immediate, vec_len, vec_len);
3159    }
3160}
3161
3162static void gen_PSRAD_i(DisasContext *s, X86DecodedInsn *decode)
3163{
3164    int vec_len = vector_len(s, decode);
3165
3166    if (decode->immediate >= 32) {
3167        decode->immediate = 31;
3168    }
3169    tcg_gen_gvec_sari(MO_32,
3170                      decode->op[0].offset, decode->op[1].offset,
3171                      decode->immediate, vec_len, vec_len);
3172}
3173
3174static void gen_PSRLQ_i(DisasContext *s, X86DecodedInsn *decode)
3175{
3176    int vec_len = vector_len(s, decode);
3177
3178    if (decode->immediate >= 64) {
3179        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3180    } else {
3181        tcg_gen_gvec_shri(MO_64,
3182                          decode->op[0].offset, decode->op[1].offset,
3183                          decode->immediate, vec_len, vec_len);
3184    }
3185}
3186
3187static void gen_PSLLQ_i(DisasContext *s, X86DecodedInsn *decode)
3188{
3189    int vec_len = vector_len(s, decode);
3190
3191    if (decode->immediate >= 64) {
3192        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3193    } else {
3194        tcg_gen_gvec_shli(MO_64,
3195                          decode->op[0].offset, decode->op[1].offset,
3196                          decode->immediate, vec_len, vec_len);
3197    }
3198}
3199
3200static TCGv_ptr make_imm8u_xmm_vec(uint8_t imm, int vec_len)
3201{
3202    MemOp ot = vec_len == 16 ? MO_128 : MO_256;
3203    TCGv_i32 imm_v = tcg_constant8u_i32(imm);
3204    TCGv_ptr ptr = tcg_temp_new_ptr();
3205
3206    tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_t0) + xmm_offset(ot),
3207                         vec_len, vec_len, 0);
3208
3209    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_t0));
3210    tcg_gen_st_i32(imm_v, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L(0)));
3211    return ptr;
3212}
3213
3214static void gen_PSRLDQ_i(DisasContext *s, X86DecodedInsn *decode)
3215{
3216    int vec_len = vector_len(s, decode);
3217    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
3218
3219    if (s->vex_l) {
3220        gen_helper_psrldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3221    } else {
3222        gen_helper_psrldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3223    }
3224}
3225
3226static void gen_PSLLDQ_i(DisasContext *s, X86DecodedInsn *decode)
3227{
3228    int vec_len = vector_len(s, decode);
3229    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
3230
3231    if (s->vex_l) {
3232        gen_helper_pslldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3233    } else {
3234        gen_helper_pslldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3235    }
3236}
3237
3238static void gen_PUSH(DisasContext *s, X86DecodedInsn *decode)
3239{
3240    gen_push_v(s, s->T0);
3241}
3242
3243static void gen_PUSHA(DisasContext *s, X86DecodedInsn *decode)
3244{
3245    gen_pusha(s);
3246}
3247
3248static void gen_PUSHF(DisasContext *s, X86DecodedInsn *decode)
3249{
3250    gen_update_cc_op(s);
3251    gen_helper_read_eflags(s->T0, tcg_env);
3252    gen_push_v(s, s->T0);
3253}
3254
3255static MemOp gen_shift_count(DisasContext *s, X86DecodedInsn *decode,
3256                             bool *can_be_zero, TCGv *count, int unit)
3257{
3258    MemOp ot = decode->op[0].ot;
3259    int mask = (ot <= MO_32 ? 0x1f : 0x3f);
3260
3261    *can_be_zero = false;
3262    switch (unit) {
3263    case X86_OP_INT:
3264        *count = tcg_temp_new();
3265        tcg_gen_andi_tl(*count, cpu_regs[R_ECX], mask);
3266        *can_be_zero = true;
3267        break;
3268
3269    case X86_OP_IMM:
3270        if ((decode->immediate & mask) == 0) {
3271            *count = NULL;
3272            break;
3273        }
3274        *count = tcg_temp_new();
3275        tcg_gen_movi_tl(*count, decode->immediate & mask);
3276        break;
3277
3278    case X86_OP_SKIP:
3279        *count = tcg_temp_new();
3280        tcg_gen_movi_tl(*count, 1);
3281        break;
3282
3283    default:
3284        g_assert_not_reached();
3285    }
3286
3287    return ot;
3288}
3289
3290/*
3291 * Compute existing flags in decode->cc_src, for gen_* functions that wants
3292 * to set the cc_op set to CC_OP_ADCOX.  In particular, this allows rotate
3293 * operations to compute the carry in decode->cc_dst and the overflow in
3294 * decode->cc_src2.
3295 *
3296 * If need_flags is true, decode->cc_dst and decode->cc_src2 are preloaded
3297 * with the value of CF and OF before the instruction, so that it is possible
3298 * to keep the flags unmodified.
3299 *
3300 * Return true if carry could be made available cheaply as a 1-bit value in
3301 * decode->cc_dst (trying a bit harder if want_carry is true).  If false is
3302 * returned, decode->cc_dst is uninitialized and the carry is only available
3303 * as bit 0 of decode->cc_src.
3304 */
3305static bool gen_eflags_adcox(DisasContext *s, X86DecodedInsn *decode, bool want_carry, bool need_flags)
3306{
3307    bool got_cf = false;
3308    bool got_of = false;
3309
3310    decode->cc_dst = tcg_temp_new();
3311    decode->cc_src = tcg_temp_new();
3312    decode->cc_src2 = tcg_temp_new();
3313    decode->cc_op = CC_OP_ADCOX;
3314
3315    /* A lot more cc_ops could be "optimized" to avoid the extracts at
3316     * the end (INC/DEC, BMILG, MUL), but they are all really unlikely
3317     * to be followed by rotations within the same basic block.
3318     */
3319    switch (s->cc_op) {
3320    case CC_OP_ADCOX:
3321        /* No need to compute the full EFLAGS, CF/OF are already isolated.  */
3322        tcg_gen_mov_tl(decode->cc_src, cpu_cc_src);
3323        if (need_flags) {
3324            tcg_gen_mov_tl(decode->cc_src2, cpu_cc_src2);
3325            got_of = true;
3326        }
3327        if (want_carry || need_flags) {
3328            tcg_gen_mov_tl(decode->cc_dst, cpu_cc_dst);
3329            got_cf = true;
3330        }
3331        break;
3332
3333    case CC_OP_LOGICB ... CC_OP_LOGICQ:
3334        /* CF and OF are zero, do it just because it's easy.  */
3335        gen_mov_eflags(s, decode->cc_src);
3336        if (need_flags) {
3337            tcg_gen_movi_tl(decode->cc_src2, 0);
3338            got_of = true;
3339        }
3340        if (want_carry || need_flags) {
3341            tcg_gen_movi_tl(decode->cc_dst, 0);
3342            got_cf = true;
3343        }
3344        break;
3345
3346    case CC_OP_SARB ... CC_OP_SARQ:
3347        /*
3348         * SHR/RCR/SHR/RCR/... is a relatively common occurrence of RCR.
3349         * By computing CF without using eflags, the calls to cc_compute_all
3350         * can be eliminated as dead code (except for the last RCR).
3351         */
3352        if (want_carry || need_flags) {
3353            tcg_gen_andi_tl(decode->cc_dst, cpu_cc_src, 1);
3354            got_cf = true;
3355        }
3356        gen_mov_eflags(s, decode->cc_src);
3357        break;
3358
3359    case CC_OP_SHLB ... CC_OP_SHLQ:
3360        /*
3361         * Likewise for SHL/RCL/SHL/RCL/... but, if CF is not in the sign
3362         * bit, we might as well fish CF out of EFLAGS and save a shift.
3363         */
3364        if (want_carry && (!need_flags || s->cc_op == CC_OP_SHLB + MO_TL)) {
3365            MemOp size = cc_op_size(s->cc_op);
3366            tcg_gen_shri_tl(decode->cc_dst, cpu_cc_src, (8 << size) - 1);
3367            got_cf = true;
3368        }
3369        gen_mov_eflags(s, decode->cc_src);
3370        break;
3371
3372    default:
3373        gen_mov_eflags(s, decode->cc_src);
3374        break;
3375    }
3376
3377    if (need_flags) {
3378        /* If the flags could be left unmodified, always load them.  */
3379        if (!got_of) {
3380            tcg_gen_extract_tl(decode->cc_src2, decode->cc_src, ctz32(CC_O), 1);
3381            got_of = true;
3382        }
3383        if (!got_cf) {
3384            tcg_gen_extract_tl(decode->cc_dst, decode->cc_src, ctz32(CC_C), 1);
3385            got_cf = true;
3386        }
3387    }
3388    return got_cf;
3389}
3390
3391static void gen_rot_overflow(X86DecodedInsn *decode, TCGv result, TCGv old,
3392                             bool can_be_zero, TCGv count)
3393{
3394    MemOp ot = decode->op[0].ot;
3395    TCGv temp = can_be_zero ? tcg_temp_new() : decode->cc_src2;
3396
3397    tcg_gen_xor_tl(temp, old, result);
3398    tcg_gen_extract_tl(temp, temp, (8 << ot) - 1, 1);
3399    if (can_be_zero) {
3400        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src2, count, tcg_constant_tl(0),
3401                           decode->cc_src2, temp);
3402    }
3403}
3404
3405/*
3406 * RCx operations are invariant modulo 8*operand_size+1.  For 8 and 16-bit operands,
3407 * this is less than 0x1f (the mask applied by gen_shift_count) so reduce further.
3408 */
3409static void gen_rotc_mod(MemOp ot, TCGv count)
3410{
3411    TCGv temp;
3412
3413    switch (ot) {
3414    case MO_8:
3415        temp = tcg_temp_new();
3416        tcg_gen_subi_tl(temp, count, 18);
3417        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3418        tcg_gen_subi_tl(temp, count, 9);
3419        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3420        break;
3421
3422    case MO_16:
3423        temp = tcg_temp_new();
3424        tcg_gen_subi_tl(temp, count, 17);
3425        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3426        break;
3427
3428    default:
3429        break;
3430    }
3431}
3432
3433/*
3434 * The idea here is that the bit to the right of the new bit 0 is the
3435 * new carry, and the bit to the right of the old bit 0 is the old carry.
3436 * Just like a regular rotation, the result of the rotation is composed
3437 * from a right shifted part and a left shifted part of s->T0.  The new carry
3438 * is extracted from the right-shifted portion, and the old carry is
3439 * inserted at the end of the left-shifted portion.
3440 *
3441 * Because of the separate shifts involving the carry, gen_RCL and gen_RCR
3442 * mostly operate on count-1.  This also comes in handy when computing
3443 * length - count, because (length-1) - (count-1) can be computed with
3444 * a XOR, and that is commutative unlike subtraction.
3445 */
3446static void gen_RCL(DisasContext *s, X86DecodedInsn *decode)
3447{
3448    bool have_1bit_cin, can_be_zero;
3449    TCGv count;
3450    TCGLabel *zero_label = NULL;
3451    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3452    TCGv low, high, low_count;
3453
3454    if (!count) {
3455        return;
3456    }
3457
3458    low = tcg_temp_new();
3459    high = tcg_temp_new();
3460    low_count = tcg_temp_new();
3461
3462    gen_rotc_mod(ot, count);
3463    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3464    if (can_be_zero) {
3465        zero_label = gen_new_label();
3466        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3467    }
3468
3469    /* Compute high part, including incoming carry.  */
3470    if (!have_1bit_cin || tcg_op_deposit_valid(TCG_TYPE_TL, 1, TARGET_LONG_BITS - 1)) {
3471        /* high = (T0 << 1) | cin */
3472        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3473        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3474    } else {
3475        /* Same as above but without deposit; cin in cc_dst.  */
3476        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3477        tcg_gen_add_tl(high, high, s->T0);
3478    }
3479    tcg_gen_subi_tl(count, count, 1);
3480    tcg_gen_shl_tl(high, high, count);
3481
3482    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3483    tcg_gen_xori_tl(low_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3484    tcg_gen_shr_tl(low, s->T0, low_count);
3485    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3486    tcg_gen_shri_tl(low, low, 1);
3487
3488    /* Compute result and outgoing overflow */
3489    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3490    tcg_gen_or_tl(s->T0, low, high);
3491    gen_rot_overflow(decode, s->T0, decode->cc_src2, false, NULL);
3492
3493    if (zero_label) {
3494        gen_set_label(zero_label);
3495    }
3496}
3497
3498static void gen_RCR(DisasContext *s, X86DecodedInsn *decode)
3499{
3500    bool have_1bit_cin, can_be_zero;
3501    TCGv count;
3502    TCGLabel *zero_label = NULL;
3503    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3504    TCGv low, high, high_count;
3505
3506    if (!count) {
3507        return;
3508    }
3509
3510    low = tcg_temp_new();
3511    high = tcg_temp_new();
3512    high_count = tcg_temp_new();
3513
3514    gen_rotc_mod(ot, count);
3515    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3516    if (can_be_zero) {
3517        zero_label = gen_new_label();
3518        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3519    }
3520
3521    /* Save incoming carry into high, it will be shifted later.  */
3522    if (!have_1bit_cin || tcg_op_deposit_valid(TCG_TYPE_TL, 1, TARGET_LONG_BITS - 1)) {
3523        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3524        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3525    } else {
3526        /* Same as above but without deposit; cin in cc_dst.  */
3527        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3528        tcg_gen_add_tl(high, high, s->T0);
3529    }
3530
3531    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3532    tcg_gen_subi_tl(count, count, 1);
3533    tcg_gen_shr_tl(low, s->T0, count);
3534    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3535    tcg_gen_shri_tl(low, low, 1);
3536
3537    /* Move high part to the right position */
3538    tcg_gen_xori_tl(high_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3539    tcg_gen_shl_tl(high, high, high_count);
3540
3541    /* Compute result and outgoing overflow */
3542    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3543    tcg_gen_or_tl(s->T0, low, high);
3544    gen_rot_overflow(decode, s->T0, decode->cc_src2, false, NULL);
3545
3546    if (zero_label) {
3547        gen_set_label(zero_label);
3548    }
3549}
3550
3551#ifdef CONFIG_USER_ONLY
3552static void gen_unreachable(DisasContext *s, X86DecodedInsn *decode)
3553{
3554    g_assert_not_reached();
3555}
3556#endif
3557
3558#ifndef CONFIG_USER_ONLY
3559static void gen_RDMSR(DisasContext *s, X86DecodedInsn *decode)
3560{
3561    gen_update_cc_op(s);
3562    gen_update_eip_cur(s);
3563    gen_helper_rdmsr(tcg_env);
3564}
3565#else
3566#define gen_RDMSR gen_unreachable
3567#endif
3568
3569static void gen_RDPMC(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_rdpmc(tcg_env);
3575    s->base.is_jmp = DISAS_NORETURN;
3576}
3577
3578static void gen_RDTSC(DisasContext *s, X86DecodedInsn *decode)
3579{
3580    gen_update_cc_op(s);
3581    gen_update_eip_cur(s);
3582    translator_io_start(&s->base);
3583    gen_helper_rdtsc(tcg_env);
3584}
3585
3586static void gen_RDxxBASE(DisasContext *s, X86DecodedInsn *decode)
3587{
3588    TCGv base = cpu_seg_base[s->modrm & 8 ? R_GS : R_FS];
3589
3590    /* Preserve hflags bits by testing CR4 at runtime.  */
3591    gen_helper_cr4_testbit(tcg_env, tcg_constant_i32(CR4_FSGSBASE_MASK));
3592    tcg_gen_mov_tl(s->T0, base);
3593}
3594
3595static void gen_RET(DisasContext *s, X86DecodedInsn *decode)
3596{
3597    int16_t adjust = decode->e.op1 == X86_TYPE_I ? decode->immediate : 0;
3598
3599    MemOp ot = gen_pop_T0(s);
3600    gen_stack_update(s, adjust + (1 << ot));
3601    gen_op_jmp_v(s, s->T0);
3602    gen_bnd_jmp(s);
3603    s->base.is_jmp = DISAS_JUMP;
3604}
3605
3606static void gen_RETF(DisasContext *s, X86DecodedInsn *decode)
3607{
3608    int16_t adjust = decode->e.op1 == X86_TYPE_I ? decode->immediate : 0;
3609
3610    if (!PE(s) || VM86(s)) {
3611        gen_lea_ss_ofs(s, s->A0, cpu_regs[R_ESP], 0);
3612        /* pop offset */
3613        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3614        /* NOTE: keeping EIP updated is not a problem in case of
3615           exception */
3616        gen_op_jmp_v(s, s->T0);
3617        /* pop selector */
3618        gen_add_A0_im(s, 1 << s->dflag);
3619        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3620        gen_op_movl_seg_real(s, R_CS, s->T0);
3621        /* add stack offset */
3622        gen_stack_update(s, adjust + (2 << s->dflag));
3623    } else {
3624        gen_update_cc_op(s);
3625        gen_update_eip_cur(s);
3626        gen_helper_lret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
3627                                  tcg_constant_i32(adjust));
3628    }
3629    s->base.is_jmp = DISAS_EOB_ONLY;
3630}
3631
3632/*
3633 * Return non-NULL if a 32-bit rotate works, after possibly replicating the input.
3634 * The input has already been zero-extended upon operand decode.
3635 */
3636static TCGv_i32 gen_rot_replicate(MemOp ot, TCGv in)
3637{
3638    TCGv_i32 temp;
3639    switch (ot) {
3640    case MO_8:
3641        temp = tcg_temp_new_i32();
3642        tcg_gen_trunc_tl_i32(temp, in);
3643        tcg_gen_muli_i32(temp, temp, 0x01010101);
3644        return temp;
3645
3646    case MO_16:
3647        temp = tcg_temp_new_i32();
3648        tcg_gen_trunc_tl_i32(temp, in);
3649        tcg_gen_deposit_i32(temp, temp, temp, 16, 16);
3650        return temp;
3651
3652#ifdef TARGET_X86_64
3653    case MO_32:
3654        temp = tcg_temp_new_i32();
3655        tcg_gen_trunc_tl_i32(temp, in);
3656        return temp;
3657#endif
3658
3659    default:
3660        return NULL;
3661    }
3662}
3663
3664static void gen_rot_carry(X86DecodedInsn *decode, TCGv result,
3665                          bool can_be_zero, TCGv count, int bit)
3666{
3667    if (!can_be_zero) {
3668        tcg_gen_extract_tl(decode->cc_dst, result, bit, 1);
3669    } else {
3670        TCGv temp = tcg_temp_new();
3671        tcg_gen_extract_tl(temp, result, bit, 1);
3672        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3673                           decode->cc_dst, temp);
3674    }
3675}
3676
3677static void gen_ROL(DisasContext *s, X86DecodedInsn *decode)
3678{
3679    bool can_be_zero;
3680    TCGv count;
3681    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3682    TCGv_i32 temp32, count32;
3683    TCGv old = tcg_temp_new();
3684
3685    if (!count) {
3686        return;
3687    }
3688
3689    gen_eflags_adcox(s, decode, false, can_be_zero);
3690    tcg_gen_mov_tl(old, s->T0);
3691    temp32 = gen_rot_replicate(ot, s->T0);
3692    if (temp32) {
3693        count32 = tcg_temp_new_i32();
3694        tcg_gen_trunc_tl_i32(count32, count);
3695        tcg_gen_rotl_i32(temp32, temp32, count32);
3696        /* Zero extend to facilitate later optimization.  */
3697        tcg_gen_extu_i32_tl(s->T0, temp32);
3698    } else {
3699        tcg_gen_rotl_tl(s->T0, s->T0, count);
3700    }
3701    gen_rot_carry(decode, s->T0, can_be_zero, count, 0);
3702    gen_rot_overflow(decode, s->T0, old, can_be_zero, count);
3703}
3704
3705static void gen_ROR(DisasContext *s, X86DecodedInsn *decode)
3706{
3707    bool can_be_zero;
3708    TCGv count;
3709    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3710    TCGv_i32 temp32, count32;
3711    TCGv old = tcg_temp_new();
3712
3713    if (!count) {
3714        return;
3715    }
3716
3717    gen_eflags_adcox(s, decode, false, can_be_zero);
3718    tcg_gen_mov_tl(old, s->T0);
3719    temp32 = gen_rot_replicate(ot, s->T0);
3720    if (temp32) {
3721        count32 = tcg_temp_new_i32();
3722        tcg_gen_trunc_tl_i32(count32, count);
3723        tcg_gen_rotr_i32(temp32, temp32, count32);
3724        /* Zero extend to facilitate later optimization.  */
3725        tcg_gen_extu_i32_tl(s->T0, temp32);
3726        gen_rot_carry(decode, s->T0, can_be_zero, count, 31);
3727    } else {
3728        tcg_gen_rotr_tl(s->T0, s->T0, count);
3729        gen_rot_carry(decode, s->T0, can_be_zero, count, TARGET_LONG_BITS - 1);
3730    }
3731    gen_rot_overflow(decode, s->T0, old, can_be_zero, count);
3732}
3733
3734static void gen_RORX(DisasContext *s, X86DecodedInsn *decode)
3735{
3736    MemOp ot = decode->op[0].ot;
3737    int mask = ot == MO_64 ? 63 : 31;
3738    int b = decode->immediate & mask;
3739
3740    switch (ot) {
3741    case MO_32:
3742#ifdef TARGET_X86_64
3743        {
3744            TCGv_i32 tmp = tcg_temp_new_i32();
3745
3746            tcg_gen_trunc_tl_i32(tmp, s->T0);
3747            tcg_gen_rotri_i32(tmp, tmp, b);
3748            tcg_gen_extu_i32_tl(s->T0, tmp);
3749            break;
3750        }
3751
3752    case MO_64:
3753#endif
3754        tcg_gen_rotri_tl(s->T0, s->T0, b);
3755        break;
3756
3757    default:
3758        g_assert_not_reached();
3759    }
3760}
3761
3762#ifndef CONFIG_USER_ONLY
3763static void gen_RSM(DisasContext *s, X86DecodedInsn *decode)
3764{
3765    gen_helper_rsm(tcg_env);
3766    assume_cc_op(s, CC_OP_EFLAGS);
3767    s->base.is_jmp = DISAS_EOB_ONLY;
3768}
3769#else
3770#define gen_RSM gen_UD
3771#endif
3772
3773static void gen_SAHF(DisasContext *s, X86DecodedInsn *decode)
3774{
3775    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
3776        return gen_illegal_opcode(s);
3777    }
3778    tcg_gen_shri_tl(s->T0, cpu_regs[R_EAX], 8);
3779    gen_compute_eflags(s);
3780    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, CC_O);
3781    tcg_gen_andi_tl(s->T0, s->T0, CC_S | CC_Z | CC_A | CC_P | CC_C);
3782    tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, s->T0);
3783}
3784
3785static void gen_SALC(DisasContext *s, X86DecodedInsn *decode)
3786{
3787    gen_compute_eflags_c(s, s->T0);
3788    tcg_gen_neg_tl(s->T0, s->T0);
3789}
3790
3791static void gen_shift_dynamic_flags(DisasContext *s, X86DecodedInsn *decode, TCGv count, CCOp cc_op)
3792{
3793    TCGv_i32 count32 = tcg_temp_new_i32();
3794    TCGv_i32 old_cc_op;
3795
3796    decode->cc_op = CC_OP_DYNAMIC;
3797    decode->cc_op_dynamic = tcg_temp_new_i32();
3798
3799    assert(decode->cc_dst == s->T0);
3800    if (cc_op_live(s->cc_op) & USES_CC_DST) {
3801        decode->cc_dst = tcg_temp_new();
3802        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3803                           cpu_cc_dst, s->T0);
3804    }
3805
3806    if (cc_op_live(s->cc_op) & USES_CC_SRC) {
3807        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src, count, tcg_constant_tl(0),
3808                           cpu_cc_src, decode->cc_src);
3809    }
3810
3811    tcg_gen_trunc_tl_i32(count32, count);
3812    if (s->cc_op == CC_OP_DYNAMIC) {
3813        old_cc_op = cpu_cc_op;
3814    } else {
3815        old_cc_op = tcg_constant_i32(s->cc_op);
3816    }
3817    tcg_gen_movcond_i32(TCG_COND_EQ, decode->cc_op_dynamic, count32, tcg_constant_i32(0),
3818                        old_cc_op, tcg_constant_i32(cc_op));
3819}
3820
3821static void gen_SAR(DisasContext *s, X86DecodedInsn *decode)
3822{
3823    bool can_be_zero;
3824    TCGv count;
3825    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3826
3827    if (!count) {
3828        return;
3829    }
3830
3831    decode->cc_dst = s->T0;
3832    decode->cc_src = tcg_temp_new();
3833    tcg_gen_subi_tl(decode->cc_src, count, 1);
3834    tcg_gen_sar_tl(decode->cc_src, s->T0, decode->cc_src);
3835    tcg_gen_sar_tl(s->T0, s->T0, count);
3836    if (can_be_zero) {
3837        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3838    } else {
3839        decode->cc_op = CC_OP_SARB + ot;
3840    }
3841}
3842
3843static void gen_SARX(DisasContext *s, X86DecodedInsn *decode)
3844{
3845    MemOp ot = decode->op[0].ot;
3846    int mask;
3847
3848    mask = ot == MO_64 ? 63 : 31;
3849    tcg_gen_andi_tl(s->T1, s->T1, mask);
3850    tcg_gen_sar_tl(s->T0, s->T0, s->T1);
3851}
3852
3853static void gen_SUB(DisasContext *s, X86DecodedInsn *decode);
3854static void gen_SBB(DisasContext *s, X86DecodedInsn *decode)
3855{
3856    MemOp ot = decode->op[0].ot;
3857    TCGv c_in;
3858
3859    /*
3860     * Try to avoid CC_OP_SBB by transforming as follows:
3861     * CC_SBB: src1 = dst + c_in, src2 = 0, src3 = c_in
3862     * CC_SUB: src1 = dst + c_in, src2 = c_in (no src3)
3863     *
3864     * In general src2 vs. src3 matters when computing AF and OF, but not here:
3865     * - AF is bit 4 of dst^src1^src2, which is bit 4 of dst^src1 in both cases
3866     * - OF is a function of the two MSBs, and in both cases they are zero for src2
3867     */
3868    if (decode->e.op2 == X86_TYPE_I && decode->immediate == 0) {
3869        gen_compute_eflags_c(s, s->T1);
3870        gen_SUB(s, decode);
3871        return;
3872    }
3873
3874    c_in = tcg_temp_new();
3875    gen_compute_eflags_c(s, c_in);
3876
3877    /*
3878     * Here the change is as follows:
3879     * CC_SBB: src1 = T0, src2 = T0, src3 = c_in
3880     * CC_SUB: src1 = 0, src2 = c_in (no src3)
3881     *
3882     * The difference also does not matter:
3883     * - AF is bit 4 of dst^src1^src2, but bit 4 of src1^src2 is zero in both cases
3884     *   therefore AF comes straight from dst (in fact it is c_in)
3885     * - for OF, src1 and src2 have the same sign in both cases, meaning there
3886     *   can be no overflow
3887     */
3888    if (decode->e.op2 != X86_TYPE_I && !decode->op[0].has_ea && decode->op[0].n == decode->op[2].n) {
3889        if (s->cc_op == CC_OP_DYNAMIC) {
3890            tcg_gen_neg_tl(s->T0, c_in);
3891        } else {
3892            /*
3893             * Do not negate c_in because it will often be dead and only the
3894             * instruction generated by negsetcond will survive.
3895             */
3896            gen_neg_setcc(s, JCC_B << 1, s->T0);
3897        }
3898        tcg_gen_movi_tl(s->cc_srcT, 0);
3899        decode->cc_src = c_in;
3900        decode->cc_dst = s->T0;
3901        decode->cc_op = CC_OP_SUBB + ot;
3902        return;
3903    }
3904
3905    if (s->prefix & PREFIX_LOCK) {
3906        tcg_gen_add_tl(s->T0, s->T1, c_in);
3907        tcg_gen_neg_tl(s->T0, s->T0);
3908        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
3909                                    s->mem_index, ot | MO_LE);
3910    } else {
3911        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
3912        tcg_gen_sub_tl(s->T0, s->T0, c_in);
3913    }
3914    prepare_update3_cc(decode, s, CC_OP_SBBB + ot, c_in);
3915}
3916
3917static void gen_SCAS(DisasContext *s, X86DecodedInsn *decode)
3918{
3919    MemOp ot = decode->op[2].ot;
3920    gen_repz_nz(s, ot, gen_scas);
3921}
3922
3923static void gen_SETcc(DisasContext *s, X86DecodedInsn *decode)
3924{
3925    gen_setcc(s, decode->b & 0xf, s->T0);
3926}
3927
3928static void gen_SFENCE(DisasContext *s, X86DecodedInsn *decode)
3929{
3930    tcg_gen_mb(TCG_MO_ST_ST | TCG_BAR_SC);
3931}
3932
3933static void gen_SHA1NEXTE(DisasContext *s, X86DecodedInsn *decode)
3934{
3935    gen_helper_sha1nexte(OP_PTR0, OP_PTR1, OP_PTR2);
3936}
3937
3938static void gen_SHA1MSG1(DisasContext *s, X86DecodedInsn *decode)
3939{
3940    gen_helper_sha1msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3941}
3942
3943static void gen_SHA1MSG2(DisasContext *s, X86DecodedInsn *decode)
3944{
3945    gen_helper_sha1msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3946}
3947
3948static void gen_SHA1RNDS4(DisasContext *s, X86DecodedInsn *decode)
3949{
3950    switch(decode->immediate & 3) {
3951    case 0:
3952        gen_helper_sha1rnds4_f0(OP_PTR0, OP_PTR0, OP_PTR1);
3953        break;
3954    case 1:
3955        gen_helper_sha1rnds4_f1(OP_PTR0, OP_PTR0, OP_PTR1);
3956        break;
3957    case 2:
3958        gen_helper_sha1rnds4_f2(OP_PTR0, OP_PTR0, OP_PTR1);
3959        break;
3960    case 3:
3961        gen_helper_sha1rnds4_f3(OP_PTR0, OP_PTR0, OP_PTR1);
3962        break;
3963    }
3964}
3965
3966static void gen_SHA256MSG1(DisasContext *s, X86DecodedInsn *decode)
3967{
3968    gen_helper_sha256msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3969}
3970
3971static void gen_SHA256MSG2(DisasContext *s, X86DecodedInsn *decode)
3972{
3973    gen_helper_sha256msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3974}
3975
3976static void gen_SHA256RNDS2(DisasContext *s, X86DecodedInsn *decode)
3977{
3978    TCGv_i32 wk0 = tcg_temp_new_i32();
3979    TCGv_i32 wk1 = tcg_temp_new_i32();
3980
3981    tcg_gen_ld_i32(wk0, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(0)));
3982    tcg_gen_ld_i32(wk1, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(1)));
3983
3984    gen_helper_sha256rnds2(OP_PTR0, OP_PTR1, OP_PTR2, wk0, wk1);
3985}
3986
3987static void gen_SHL(DisasContext *s, X86DecodedInsn *decode)
3988{
3989    bool can_be_zero;
3990    TCGv count;
3991    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3992
3993    if (!count) {
3994        return;
3995    }
3996
3997    decode->cc_dst = s->T0;
3998    decode->cc_src = tcg_temp_new();
3999    tcg_gen_subi_tl(decode->cc_src, count, 1);
4000    tcg_gen_shl_tl(decode->cc_src, s->T0, decode->cc_src);
4001    tcg_gen_shl_tl(s->T0, s->T0, count);
4002    if (can_be_zero) {
4003        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
4004    } else {
4005        decode->cc_op = CC_OP_SHLB + ot;
4006    }
4007}
4008
4009static void gen_SHLD(DisasContext *s, X86DecodedInsn *decode)
4010{
4011    bool can_be_zero;
4012    TCGv count;
4013    int unit = decode->e.op3 == X86_TYPE_I ? X86_OP_IMM : X86_OP_INT;
4014    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, unit);
4015
4016    if (!count) {
4017        return;
4018    }
4019
4020    decode->cc_dst = s->T0;
4021    decode->cc_src = gen_shiftd_rm_T1(s, ot, false, count);
4022    if (can_be_zero) {
4023        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
4024    } else {
4025        decode->cc_op = CC_OP_SHLB + ot;
4026    }
4027}
4028
4029static void gen_SHLX(DisasContext *s, X86DecodedInsn *decode)
4030{
4031    MemOp ot = decode->op[0].ot;
4032    int mask;
4033
4034    mask = ot == MO_64 ? 63 : 31;
4035    tcg_gen_andi_tl(s->T1, s->T1, mask);
4036    tcg_gen_shl_tl(s->T0, s->T0, s->T1);
4037}
4038
4039static void gen_SHR(DisasContext *s, X86DecodedInsn *decode)
4040{
4041    bool can_be_zero;
4042    TCGv count;
4043    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
4044
4045    if (!count) {
4046        return;
4047    }
4048
4049    decode->cc_dst = s->T0;
4050    decode->cc_src = tcg_temp_new();
4051    tcg_gen_subi_tl(decode->cc_src, count, 1);
4052    tcg_gen_shr_tl(decode->cc_src, s->T0, decode->cc_src);
4053    tcg_gen_shr_tl(s->T0, s->T0, count);
4054    if (can_be_zero) {
4055        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
4056    } else {
4057        decode->cc_op = CC_OP_SARB + ot;
4058    }
4059}
4060
4061static void gen_SHRD(DisasContext *s, X86DecodedInsn *decode)
4062{
4063    bool can_be_zero;
4064    TCGv count;
4065    int unit = decode->e.op3 == X86_TYPE_I ? X86_OP_IMM : X86_OP_INT;
4066    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, unit);
4067
4068    if (!count) {
4069        return;
4070    }
4071
4072    decode->cc_dst = s->T0;
4073    decode->cc_src = gen_shiftd_rm_T1(s, ot, true, count);
4074    if (can_be_zero) {
4075        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
4076    } else {
4077        decode->cc_op = CC_OP_SARB + ot;
4078    }
4079}
4080
4081static void gen_SHRX(DisasContext *s, X86DecodedInsn *decode)
4082{
4083    MemOp ot = decode->op[0].ot;
4084    int mask;
4085
4086    mask = ot == MO_64 ? 63 : 31;
4087    tcg_gen_andi_tl(s->T1, s->T1, mask);
4088    tcg_gen_shr_tl(s->T0, s->T0, s->T1);
4089}
4090
4091static void gen_STC(DisasContext *s, X86DecodedInsn *decode)
4092{
4093    gen_compute_eflags(s);
4094    tcg_gen_ori_tl(cpu_cc_src, cpu_cc_src, CC_C);
4095}
4096
4097static void gen_STD(DisasContext *s, X86DecodedInsn *decode)
4098{
4099    tcg_gen_st_i32(tcg_constant_i32(-1), tcg_env, offsetof(CPUX86State, df));
4100}
4101
4102static void gen_STI(DisasContext *s, X86DecodedInsn *decode)
4103{
4104    gen_set_eflags(s, IF_MASK);
4105    s->base.is_jmp = DISAS_EOB_INHIBIT_IRQ;
4106}
4107
4108static void gen_VAESKEYGEN(DisasContext *s, X86DecodedInsn *decode)
4109{
4110    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4111    assert(!s->vex_l);
4112    gen_helper_aeskeygenassist_xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
4113}
4114
4115static void gen_STMXCSR(DisasContext *s, X86DecodedInsn *decode)
4116{
4117    gen_helper_update_mxcsr(tcg_env);
4118    tcg_gen_ld32u_tl(s->T0, tcg_env, offsetof(CPUX86State, mxcsr));
4119}
4120
4121static void gen_STOS(DisasContext *s, X86DecodedInsn *decode)
4122{
4123    MemOp ot = decode->op[1].ot;
4124    gen_repz(s, ot, gen_stos);
4125}
4126
4127static void gen_SUB(DisasContext *s, X86DecodedInsn *decode)
4128{
4129    MemOp ot = decode->op[1].ot;
4130
4131    if (s->prefix & PREFIX_LOCK) {
4132        tcg_gen_neg_tl(s->T0, s->T1);
4133        tcg_gen_atomic_fetch_add_tl(s->cc_srcT, s->A0, s->T0,
4134                                    s->mem_index, ot | MO_LE);
4135        tcg_gen_sub_tl(s->T0, s->cc_srcT, s->T1);
4136    } else {
4137        tcg_gen_mov_tl(s->cc_srcT, s->T0);
4138        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
4139    }
4140    prepare_update2_cc(decode, s, CC_OP_SUBB + ot);
4141}
4142
4143static void gen_SYSCALL(DisasContext *s, X86DecodedInsn *decode)
4144{
4145    gen_update_cc_op(s);
4146    gen_update_eip_cur(s);
4147    gen_helper_syscall(tcg_env, cur_insn_len_i32(s));
4148    if (LMA(s)) {
4149        assume_cc_op(s, CC_OP_EFLAGS);
4150    }
4151
4152    /*
4153     * TF handling for the syscall insn is different. The TF bit is checked
4154     * after the syscall insn completes. This allows #DB to not be
4155     * generated after one has entered CPL0 if TF is set in FMASK.
4156     */
4157    s->base.is_jmp = DISAS_EOB_RECHECK_TF;
4158}
4159
4160static void gen_SYSENTER(DisasContext *s, X86DecodedInsn *decode)
4161{
4162    gen_helper_sysenter(tcg_env);
4163    s->base.is_jmp = DISAS_EOB_ONLY;
4164}
4165
4166static void gen_SYSEXIT(DisasContext *s, X86DecodedInsn *decode)
4167{
4168    gen_helper_sysexit(tcg_env, tcg_constant_i32(s->dflag - 1));
4169    s->base.is_jmp = DISAS_EOB_ONLY;
4170}
4171
4172static void gen_SYSRET(DisasContext *s, X86DecodedInsn *decode)
4173{
4174    gen_helper_sysret(tcg_env, tcg_constant_i32(s->dflag - 1));
4175    if (LMA(s)) {
4176        assume_cc_op(s, CC_OP_EFLAGS);
4177    }
4178
4179    /*
4180     * TF handling for the sysret insn is different. The TF bit is checked
4181     * after the sysret insn completes. This allows #DB to be
4182     * generated "as if" the syscall insn in userspace has just
4183     * completed.
4184     */
4185    s->base.is_jmp = DISAS_EOB_RECHECK_TF;
4186}
4187
4188static void gen_TZCNT(DisasContext *s, X86DecodedInsn *decode)
4189{
4190    MemOp ot = decode->op[0].ot;
4191
4192    /* C bit (cc_src) is defined related to the input.  */
4193    decode->cc_src = tcg_temp_new();
4194    decode->cc_dst = s->T0;
4195    decode->cc_op = CC_OP_BMILGB + ot;
4196    tcg_gen_mov_tl(decode->cc_src, s->T0);
4197
4198    /* A zero input returns the operand size.  */
4199    tcg_gen_ctzi_tl(s->T0, s->T0, 8 << ot);
4200}
4201
4202static void gen_UD(DisasContext *s, X86DecodedInsn *decode)
4203{
4204    gen_illegal_opcode(s);
4205}
4206
4207static void gen_VAESIMC(DisasContext *s, X86DecodedInsn *decode)
4208{
4209    assert(!s->vex_l);
4210    gen_helper_aesimc_xmm(tcg_env, OP_PTR0, OP_PTR2);
4211}
4212
4213/*
4214 * 00 = v*ps Vps, Hps, Wpd
4215 * 66 = v*pd Vpd, Hpd, Wps
4216 * f3 = v*ss Vss, Hss, Wps
4217 * f2 = v*sd Vsd, Hsd, Wps
4218 */
4219#define SSE_CMP(x) { \
4220    gen_helper_ ## x ## ps ## _xmm, gen_helper_ ## x ## pd ## _xmm, \
4221    gen_helper_ ## x ## ss, gen_helper_ ## x ## sd, \
4222    gen_helper_ ## x ## ps ## _ymm, gen_helper_ ## x ## pd ## _ymm}
4223static const SSEFunc_0_eppp gen_helper_cmp_funcs[32][6] = {
4224    SSE_CMP(cmpeq),
4225    SSE_CMP(cmplt),
4226    SSE_CMP(cmple),
4227    SSE_CMP(cmpunord),
4228    SSE_CMP(cmpneq),
4229    SSE_CMP(cmpnlt),
4230    SSE_CMP(cmpnle),
4231    SSE_CMP(cmpord),
4232
4233    SSE_CMP(cmpequ),
4234    SSE_CMP(cmpnge),
4235    SSE_CMP(cmpngt),
4236    SSE_CMP(cmpfalse),
4237    SSE_CMP(cmpnequ),
4238    SSE_CMP(cmpge),
4239    SSE_CMP(cmpgt),
4240    SSE_CMP(cmptrue),
4241
4242    SSE_CMP(cmpeqs),
4243    SSE_CMP(cmpltq),
4244    SSE_CMP(cmpleq),
4245    SSE_CMP(cmpunords),
4246    SSE_CMP(cmpneqq),
4247    SSE_CMP(cmpnltq),
4248    SSE_CMP(cmpnleq),
4249    SSE_CMP(cmpords),
4250
4251    SSE_CMP(cmpequs),
4252    SSE_CMP(cmpngeq),
4253    SSE_CMP(cmpngtq),
4254    SSE_CMP(cmpfalses),
4255    SSE_CMP(cmpnequs),
4256    SSE_CMP(cmpgeq),
4257    SSE_CMP(cmpgtq),
4258    SSE_CMP(cmptrues),
4259};
4260#undef SSE_CMP
4261
4262static void gen_VCMP(DisasContext *s, X86DecodedInsn *decode)
4263{
4264    int index = decode->immediate & (s->prefix & PREFIX_VEX ? 31 : 7);
4265    int b =
4266        s->prefix & PREFIX_REPZ  ? 2 /* ss */ :
4267        s->prefix & PREFIX_REPNZ ? 3 /* sd */ :
4268        !!(s->prefix & PREFIX_DATA) /* pd */ + (s->vex_l << 2);
4269
4270    gen_helper_cmp_funcs[index][b](tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4271}
4272
4273static void gen_VCOMI(DisasContext *s, X86DecodedInsn *decode)
4274{
4275    SSEFunc_0_epp fn;
4276    fn = s->prefix & PREFIX_DATA ? gen_helper_comisd : gen_helper_comiss;
4277    fn(tcg_env, OP_PTR1, OP_PTR2);
4278    assume_cc_op(s, CC_OP_EFLAGS);
4279}
4280
4281static void gen_VCVTPD2PS(DisasContext *s, X86DecodedInsn *decode)
4282{
4283    if (s->vex_l) {
4284        gen_helper_cvtpd2ps_ymm(tcg_env, OP_PTR0, OP_PTR2);
4285    } else {
4286        gen_helper_cvtpd2ps_xmm(tcg_env, OP_PTR0, OP_PTR2);
4287    }
4288}
4289
4290static void gen_VCVTPS2PD(DisasContext *s, X86DecodedInsn *decode)
4291{
4292    if (s->vex_l) {
4293        gen_helper_cvtps2pd_ymm(tcg_env, OP_PTR0, OP_PTR2);
4294    } else {
4295        gen_helper_cvtps2pd_xmm(tcg_env, OP_PTR0, OP_PTR2);
4296    }
4297}
4298
4299static void gen_VCVTPS2PH(DisasContext *s, X86DecodedInsn *decode)
4300{
4301    gen_unary_imm_fp_sse(s, decode,
4302                      gen_helper_cvtps2ph_xmm,
4303                      gen_helper_cvtps2ph_ymm);
4304    /*
4305     * VCVTPS2PH is the only instruction that performs an operation on a
4306     * register source and then *stores* into memory.
4307     */
4308    if (decode->op[0].has_ea) {
4309        gen_store_sse(s, decode, decode->op[0].offset);
4310    }
4311}
4312
4313static void gen_VCVTSD2SS(DisasContext *s, X86DecodedInsn *decode)
4314{
4315    gen_helper_cvtsd2ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4316}
4317
4318static void gen_VCVTSS2SD(DisasContext *s, X86DecodedInsn *decode)
4319{
4320    gen_helper_cvtss2sd(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4321}
4322
4323static void gen_VCVTSI2Sx(DisasContext *s, X86DecodedInsn *decode)
4324{
4325    int vec_len = vector_len(s, decode);
4326    TCGv_i32 in;
4327
4328    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4329
4330#ifdef TARGET_X86_64
4331    MemOp ot = decode->op[2].ot;
4332    if (ot == MO_64) {
4333        if (s->prefix & PREFIX_REPNZ) {
4334            gen_helper_cvtsq2sd(tcg_env, OP_PTR0, s->T1);
4335        } else {
4336            gen_helper_cvtsq2ss(tcg_env, OP_PTR0, s->T1);
4337        }
4338        return;
4339    }
4340    in = tcg_temp_new_i32();
4341    tcg_gen_trunc_tl_i32(in, s->T1);
4342#else
4343    in = s->T1;
4344#endif
4345
4346    if (s->prefix & PREFIX_REPNZ) {
4347        gen_helper_cvtsi2sd(tcg_env, OP_PTR0, in);
4348    } else {
4349        gen_helper_cvtsi2ss(tcg_env, OP_PTR0, in);
4350    }
4351}
4352
4353static inline void gen_VCVTtSx2SI(DisasContext *s, X86DecodedInsn *decode,
4354                                  SSEFunc_i_ep ss2si, SSEFunc_l_ep ss2sq,
4355                                  SSEFunc_i_ep sd2si, SSEFunc_l_ep sd2sq)
4356{
4357    TCGv_i32 out;
4358
4359#ifdef TARGET_X86_64
4360    MemOp ot = decode->op[0].ot;
4361    if (ot == MO_64) {
4362        if (s->prefix & PREFIX_REPNZ) {
4363            sd2sq(s->T0, tcg_env, OP_PTR2);
4364        } else {
4365            ss2sq(s->T0, tcg_env, OP_PTR2);
4366        }
4367        return;
4368    }
4369
4370    out = tcg_temp_new_i32();
4371#else
4372    out = s->T0;
4373#endif
4374    if (s->prefix & PREFIX_REPNZ) {
4375        sd2si(out, tcg_env, OP_PTR2);
4376    } else {
4377        ss2si(out, tcg_env, OP_PTR2);
4378    }
4379#ifdef TARGET_X86_64
4380    tcg_gen_extu_i32_tl(s->T0, out);
4381#endif
4382}
4383
4384#ifndef TARGET_X86_64
4385#define gen_helper_cvtss2sq NULL
4386#define gen_helper_cvtsd2sq NULL
4387#define gen_helper_cvttss2sq NULL
4388#define gen_helper_cvttsd2sq NULL
4389#endif
4390
4391static void gen_VCVTSx2SI(DisasContext *s, X86DecodedInsn *decode)
4392{
4393    gen_VCVTtSx2SI(s, decode,
4394                   gen_helper_cvtss2si, gen_helper_cvtss2sq,
4395                   gen_helper_cvtsd2si, gen_helper_cvtsd2sq);
4396}
4397
4398static void gen_VCVTTSx2SI(DisasContext *s, X86DecodedInsn *decode)
4399{
4400    gen_VCVTtSx2SI(s, decode,
4401                   gen_helper_cvttss2si, gen_helper_cvttss2sq,
4402                   gen_helper_cvttsd2si, gen_helper_cvttsd2sq);
4403}
4404
4405static void gen_VEXTRACTx128(DisasContext *s, X86DecodedInsn *decode)
4406{
4407    int mask = decode->immediate & 1;
4408    int src_ofs = vector_elem_offset(&decode->op[1], MO_128, mask);
4409    if (decode->op[0].has_ea) {
4410        /* VEX-only instruction, no alignment requirements.  */
4411        gen_sto_env_A0(s, src_ofs, false);
4412    } else {
4413        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, 16, 16);
4414    }
4415}
4416
4417static void gen_VEXTRACTPS(DisasContext *s, X86DecodedInsn *decode)
4418{
4419    gen_pextr(s, decode, MO_32);
4420}
4421
4422static void gen_vinsertps(DisasContext *s, X86DecodedInsn *decode, TCGv_i32 tmp)
4423{
4424    int val = decode->immediate;
4425    int dest_word = (val >> 4) & 3;
4426    int new_mask = (val & 15) | (1 << dest_word);
4427    int vec_len = 16;
4428
4429    assert(!s->vex_l);
4430
4431    if (new_mask == 15) {
4432        /* All zeroes except possibly for the inserted element */
4433        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
4434    } else if (decode->op[1].offset != decode->op[0].offset) {
4435        gen_store_sse(s, decode, decode->op[1].offset);
4436    }
4437
4438    if (new_mask != (val & 15)) {
4439        tcg_gen_st_i32(tmp, tcg_env,
4440                       vector_elem_offset(&decode->op[0], MO_32, dest_word));
4441    }
4442
4443    if (new_mask != 15) {
4444        TCGv_i32 zero = tcg_constant_i32(0); /* float32_zero */
4445        int i;
4446        for (i = 0; i < 4; i++) {
4447            if ((val >> i) & 1) {
4448                tcg_gen_st_i32(zero, tcg_env,
4449                               vector_elem_offset(&decode->op[0], MO_32, i));
4450            }
4451        }
4452    }
4453}
4454
4455static void gen_VINSERTPS_r(DisasContext *s, X86DecodedInsn *decode)
4456{
4457    int val = decode->immediate;
4458    TCGv_i32 tmp = tcg_temp_new_i32();
4459
4460    tcg_gen_ld_i32(tmp, tcg_env,
4461                   vector_elem_offset(&decode->op[2], MO_32, (val >> 6) & 3));
4462    gen_vinsertps(s, decode, tmp);
4463}
4464
4465static void gen_VINSERTPS_m(DisasContext *s, X86DecodedInsn *decode)
4466{
4467    TCGv_i32 tmp = tcg_temp_new_i32();
4468
4469    tcg_gen_qemu_ld_i32(tmp, s->A0, s->mem_index, MO_LEUL);
4470    gen_vinsertps(s, decode, tmp);
4471}
4472
4473static void gen_VINSERTx128(DisasContext *s, X86DecodedInsn *decode)
4474{
4475    int mask = decode->immediate & 1;
4476    tcg_gen_gvec_mov(MO_64,
4477                     decode->op[0].offset + offsetof(YMMReg, YMM_X(mask)),
4478                     decode->op[2].offset + offsetof(YMMReg, YMM_X(0)), 16, 16);
4479    tcg_gen_gvec_mov(MO_64,
4480                     decode->op[0].offset + offsetof(YMMReg, YMM_X(!mask)),
4481                     decode->op[1].offset + offsetof(YMMReg, YMM_X(!mask)), 16, 16);
4482}
4483
4484static inline void gen_maskmov(DisasContext *s, X86DecodedInsn *decode,
4485                               SSEFunc_0_eppt xmm, SSEFunc_0_eppt ymm)
4486{
4487    if (!s->vex_l) {
4488        xmm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
4489    } else {
4490        ymm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
4491    }
4492}
4493
4494static void gen_VMASKMOVPD_st(DisasContext *s, X86DecodedInsn *decode)
4495{
4496    gen_maskmov(s, decode, gen_helper_vpmaskmovq_st_xmm, gen_helper_vpmaskmovq_st_ymm);
4497}
4498
4499static void gen_VMASKMOVPS_st(DisasContext *s, X86DecodedInsn *decode)
4500{
4501    gen_maskmov(s, decode, gen_helper_vpmaskmovd_st_xmm, gen_helper_vpmaskmovd_st_ymm);
4502}
4503
4504static void gen_VMOVHPx_ld(DisasContext *s, X86DecodedInsn *decode)
4505{
4506    gen_ldq_env_A0(s, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4507    if (decode->op[0].offset != decode->op[1].offset) {
4508        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4509        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4510    }
4511}
4512
4513static void gen_VMOVHPx_st(DisasContext *s, X86DecodedInsn *decode)
4514{
4515    gen_stq_env_A0(s, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4516}
4517
4518static void gen_VMOVHPx(DisasContext *s, X86DecodedInsn *decode)
4519{
4520    if (decode->op[0].offset != decode->op[2].offset) {
4521        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4522        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4523    }
4524    if (decode->op[0].offset != decode->op[1].offset) {
4525        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4526        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4527    }
4528}
4529
4530static void gen_VMOVHLPS(DisasContext *s, X86DecodedInsn *decode)
4531{
4532    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4533    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4534    if (decode->op[0].offset != decode->op[1].offset) {
4535        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(1)));
4536        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4537    }
4538}
4539
4540static void gen_VMOVLHPS(DisasContext *s, X86DecodedInsn *decode)
4541{
4542    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
4543    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4544    if (decode->op[0].offset != decode->op[1].offset) {
4545        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4546        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4547    }
4548}
4549
4550/*
4551 * Note that MOVLPx supports 256-bit operation unlike MOVHLPx, MOVLHPx, MOXHPx.
4552 * Use a gvec move to move everything above the bottom 64 bits.
4553 */
4554
4555static void gen_VMOVLPx(DisasContext *s, X86DecodedInsn *decode)
4556{
4557    int vec_len = vector_len(s, decode);
4558
4559    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(0)));
4560    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4561    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4562}
4563
4564static void gen_VMOVLPx_ld(DisasContext *s, X86DecodedInsn *decode)
4565{
4566    int vec_len = vector_len(s, decode);
4567
4568    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4569    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4570    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
4571}
4572
4573static void gen_VMOVLPx_st(DisasContext *s, X86DecodedInsn *decode)
4574{
4575    tcg_gen_ld_i64(s->tmp1_i64, OP_PTR2, offsetof(ZMMReg, ZMM_Q(0)));
4576    tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4577}
4578
4579static void gen_VMOVSD_ld(DisasContext *s, X86DecodedInsn *decode)
4580{
4581    TCGv_i64 zero = tcg_constant_i64(0);
4582
4583    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4584    tcg_gen_st_i64(zero, OP_PTR0, offsetof(ZMMReg, ZMM_Q(1)));
4585    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
4586}
4587
4588static void gen_VMOVSS(DisasContext *s, X86DecodedInsn *decode)
4589{
4590    int vec_len = vector_len(s, decode);
4591    TCGv_i32 tmp = tcg_temp_new_i32();
4592
4593    tcg_gen_ld_i32(tmp, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
4594    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4595    tcg_gen_st_i32(tmp, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
4596}
4597
4598static void gen_VMOVSS_ld(DisasContext *s, X86DecodedInsn *decode)
4599{
4600    int vec_len = vector_len(s, decode);
4601    TCGv_i32 tmp = tcg_temp_new_i32();
4602
4603    tcg_gen_qemu_ld_i32(tmp, s->A0, s->mem_index, MO_LEUL);
4604    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
4605    tcg_gen_st_i32(tmp, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
4606}
4607
4608static void gen_VMOVSS_st(DisasContext *s, X86DecodedInsn *decode)
4609{
4610    TCGv_i32 tmp = tcg_temp_new_i32();
4611
4612    tcg_gen_ld_i32(tmp, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
4613    tcg_gen_qemu_st_i32(tmp, s->A0, s->mem_index, MO_LEUL);
4614}
4615
4616static void gen_VPMASKMOV_st(DisasContext *s, X86DecodedInsn *decode)
4617{
4618    if (s->vex_w) {
4619        gen_VMASKMOVPD_st(s, decode);
4620    } else {
4621        gen_VMASKMOVPS_st(s, decode);
4622    }
4623}
4624
4625static void gen_VPERMD(DisasContext *s, X86DecodedInsn *decode)
4626{
4627    assert(s->vex_l);
4628    gen_helper_vpermd_ymm(OP_PTR0, OP_PTR1, OP_PTR2);
4629}
4630
4631static void gen_VPERM2x128(DisasContext *s, X86DecodedInsn *decode)
4632{
4633    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4634    assert(s->vex_l);
4635    gen_helper_vpermdq_ymm(OP_PTR0, OP_PTR1, OP_PTR2, imm);
4636}
4637
4638static void gen_VPHMINPOSUW(DisasContext *s, X86DecodedInsn *decode)
4639{
4640    assert(!s->vex_l);
4641    gen_helper_phminposuw_xmm(tcg_env, OP_PTR0, OP_PTR2);
4642}
4643
4644static void gen_VROUNDSD(DisasContext *s, X86DecodedInsn *decode)
4645{
4646    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4647    assert(!s->vex_l);
4648    gen_helper_roundsd_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
4649}
4650
4651static void gen_VROUNDSS(DisasContext *s, X86DecodedInsn *decode)
4652{
4653    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4654    assert(!s->vex_l);
4655    gen_helper_roundss_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
4656}
4657
4658static void gen_VSHUF(DisasContext *s, X86DecodedInsn *decode)
4659{
4660    TCGv_i32 imm = tcg_constant_i32(decode->immediate);
4661    SSEFunc_0_pppi ps, pd, fn;
4662    ps = s->vex_l ? gen_helper_shufps_ymm : gen_helper_shufps_xmm;
4663    pd = s->vex_l ? gen_helper_shufpd_ymm : gen_helper_shufpd_xmm;
4664    fn = s->prefix & PREFIX_DATA ? pd : ps;
4665    fn(OP_PTR0, OP_PTR1, OP_PTR2, imm);
4666}
4667
4668static void gen_VUCOMI(DisasContext *s, X86DecodedInsn *decode)
4669{
4670    SSEFunc_0_epp fn;
4671    fn = s->prefix & PREFIX_DATA ? gen_helper_ucomisd : gen_helper_ucomiss;
4672    fn(tcg_env, OP_PTR1, OP_PTR2);
4673    assume_cc_op(s, CC_OP_EFLAGS);
4674}
4675
4676static void gen_VZEROALL(DisasContext *s, X86DecodedInsn *decode)
4677{
4678    TCGv_ptr ptr = tcg_temp_new_ptr();
4679
4680    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_regs));
4681    gen_helper_memset(ptr, ptr, tcg_constant_i32(0),
4682                      tcg_constant_ptr(CPU_NB_REGS * sizeof(ZMMReg)));
4683}
4684
4685static void gen_VZEROUPPER(DisasContext *s, X86DecodedInsn *decode)
4686{
4687    int i;
4688
4689    for (i = 0; i < CPU_NB_REGS; i++) {
4690        int offset = offsetof(CPUX86State, xmm_regs[i].ZMM_X(1));
4691        tcg_gen_gvec_dup_imm(MO_64, offset, 16, 16, 0);
4692    }
4693}
4694
4695static void gen_WAIT(DisasContext *s, X86DecodedInsn *decode)
4696{
4697    if ((s->flags & (HF_MP_MASK | HF_TS_MASK)) == (HF_MP_MASK | HF_TS_MASK)) {
4698        gen_NM_exception(s);
4699    } else {
4700        /* needs to be treated as I/O because of ferr_irq */
4701        translator_io_start(&s->base);
4702        gen_helper_fwait(tcg_env);
4703    }
4704}
4705
4706#ifndef CONFIG_USER_ONLY
4707static void gen_WRMSR(DisasContext *s, X86DecodedInsn *decode)
4708{
4709    gen_update_cc_op(s);
4710    gen_update_eip_cur(s);
4711    gen_helper_wrmsr(tcg_env);
4712    s->base.is_jmp = DISAS_EOB_NEXT;
4713}
4714#else
4715#define gen_WRMSR gen_unreachable
4716#endif
4717
4718static void gen_WRxxBASE(DisasContext *s, X86DecodedInsn *decode)
4719{
4720    TCGv base = cpu_seg_base[s->modrm & 8 ? R_GS : R_FS];
4721
4722    /* Preserve hflags bits by testing CR4 at runtime.  */
4723    gen_helper_cr4_testbit(tcg_env, tcg_constant_i32(CR4_FSGSBASE_MASK));
4724    tcg_gen_mov_tl(base, s->T0);
4725}
4726
4727static void gen_XADD(DisasContext *s, X86DecodedInsn *decode)
4728{
4729    MemOp ot = decode->op[1].ot;
4730
4731    decode->cc_dst = tcg_temp_new();
4732    decode->cc_src = s->T1;
4733    decode->cc_op = CC_OP_ADDB + ot;
4734
4735    if (s->prefix & PREFIX_LOCK) {
4736        tcg_gen_atomic_fetch_add_tl(s->T0, s->A0, s->T1, s->mem_index, ot | MO_LE);
4737        tcg_gen_add_tl(decode->cc_dst, s->T0, s->T1);
4738    } else {
4739        tcg_gen_add_tl(decode->cc_dst, s->T0, s->T1);
4740        /*
4741         * NOTE: writing memory first is important for MMU exceptions,
4742         * but "new result" wins for XADD AX, AX.
4743         */
4744        gen_writeback(s, decode, 0, decode->cc_dst);
4745    }
4746    if (decode->op[0].has_ea || decode->op[2].n != decode->op[0].n) {
4747        gen_writeback(s, decode, 2, s->T0);
4748    }
4749}
4750
4751static void gen_XCHG(DisasContext *s, X86DecodedInsn *decode)
4752{
4753    if (s->prefix & PREFIX_LOCK) {
4754        tcg_gen_atomic_xchg_tl(s->T0, s->A0, s->T1,
4755                               s->mem_index, decode->op[0].ot | MO_LE);
4756        /* now store old value into register operand */
4757        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4758    } else {
4759        /* move destination value into source operand, source preserved in T1 */
4760        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4761        tcg_gen_mov_tl(s->T0, s->T1);
4762    }
4763}
4764
4765static void gen_XLAT(DisasContext *s, X86DecodedInsn *decode)
4766{
4767    /* AL is already zero-extended into s->T0.  */
4768    tcg_gen_add_tl(s->A0, cpu_regs[R_EBX], s->T0);
4769    gen_lea_v_seg(s, s->A0, R_DS, s->override);
4770    gen_op_ld_v(s, MO_8, s->T0, s->A0);
4771}
4772
4773static void gen_XOR(DisasContext *s, X86DecodedInsn *decode)
4774{
4775    /* special case XOR reg, reg */
4776    if (decode->op[1].unit == X86_OP_INT &&
4777        decode->op[2].unit == X86_OP_INT &&
4778        decode->op[1].n == decode->op[2].n) {
4779        tcg_gen_movi_tl(s->T0, 0);
4780        decode->cc_op = CC_OP_EFLAGS;
4781        decode->cc_src = tcg_constant_tl(CC_Z | CC_P);
4782    } else {
4783        MemOp ot = decode->op[1].ot;
4784
4785        if (s->prefix & PREFIX_LOCK) {
4786            tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T1,
4787                                        s->mem_index, ot | MO_LE);
4788        } else {
4789            tcg_gen_xor_tl(s->T0, s->T0, s->T1);
4790        }
4791        prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
4792    }
4793}
4794
4795static void gen_XRSTOR(DisasContext *s, X86DecodedInsn *decode)
4796{
4797    TCGv_i64 features = tcg_temp_new_i64();
4798
4799    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4800    gen_helper_xrstor(tcg_env, s->A0, features);
4801    if (s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_MPX) {
4802        /*
4803         * XRSTOR is how MPX is enabled, which changes how
4804         * we translate.  Thus we need to end the TB.
4805         */
4806        s->base.is_jmp = DISAS_EOB_NEXT;
4807    }
4808}
4809
4810static void gen_XSAVE(DisasContext *s, X86DecodedInsn *decode)
4811{
4812    TCGv_i64 features = tcg_temp_new_i64();
4813
4814    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4815    gen_helper_xsave(tcg_env, s->A0, features);
4816}
4817
4818static void gen_XSAVEOPT(DisasContext *s, X86DecodedInsn *decode)
4819{
4820    TCGv_i64 features = tcg_temp_new_i64();
4821
4822    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4823    gen_helper_xsave(tcg_env, s->A0, features);
4824}
4825