xref: /openbmc/qemu/target/i386/tcg/emit.c.inc (revision f60033d3794427dfef30b2b1cd3485d3df459737)
1/*
2 * New-style TCG opcode generator for i386 instructions
3 *
4 *  Copyright (c) 2022 Red Hat, Inc.
5 *
6 * Author: Paolo Bonzini <pbonzini@redhat.com>
7 *
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 */
21
22/*
23 * Sometimes, knowing what the backend has can produce better code.
24 * The exact opcode to check depends on 32- vs. 64-bit.
25 */
26#ifdef TARGET_X86_64
27#define INDEX_op_extract2_tl            INDEX_op_extract2_i64
28#else
29#define INDEX_op_extract2_tl            INDEX_op_extract2_i32
30#endif
31
32#define MMX_OFFSET(reg)                        \
33  ({ assert((reg) >= 0 && (reg) <= 7);         \
34     offsetof(CPUX86State, fpregs[reg].mmx); })
35
36#define ZMM_OFFSET(reg)                        \
37  ({ assert((reg) >= 0 && (reg) <= 15);        \
38     offsetof(CPUX86State, xmm_regs[reg]); })
39
40typedef void (*SSEFunc_i_ep)(TCGv_i32 val, TCGv_ptr env, TCGv_ptr reg);
41typedef void (*SSEFunc_l_ep)(TCGv_i64 val, TCGv_ptr env, TCGv_ptr reg);
42typedef void (*SSEFunc_0_epp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b);
43typedef void (*SSEFunc_0_eppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
44                               TCGv_ptr reg_c);
45typedef void (*SSEFunc_0_epppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
46                                TCGv_ptr reg_c, TCGv_ptr reg_d);
47typedef void (*SSEFunc_0_eppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
48                               TCGv_i32 val);
49typedef void (*SSEFunc_0_epppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
50                                TCGv_ptr reg_c, TCGv_i32 val);
51typedef void (*SSEFunc_0_ppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_i32 val);
52typedef void (*SSEFunc_0_pppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_ptr reg_c,
53                               TCGv_i32 val);
54typedef void (*SSEFunc_0_eppt)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
55                               TCGv val);
56typedef void (*SSEFunc_0_epppti)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
57                                 TCGv_ptr reg_c, TCGv a0, TCGv_i32 scale);
58typedef void (*SSEFunc_0_eppppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
59                                  TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 flags);
60typedef void (*SSEFunc_0_eppppii)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
61                                  TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 even,
62                                  TCGv_i32 odd);
63
64static void gen_JMP_m(DisasContext *s, X86DecodedInsn *decode);
65static void gen_JMP(DisasContext *s, X86DecodedInsn *decode);
66
67static inline TCGv_i32 tcg_constant8u_i32(uint8_t val)
68{
69    return tcg_constant_i32(val);
70}
71
72static void gen_NM_exception(DisasContext *s)
73{
74    gen_exception(s, EXCP07_PREX);
75}
76
77static void gen_lea_modrm(DisasContext *s, X86DecodedInsn *decode)
78{
79    AddressParts *mem = &decode->mem;
80    TCGv ea;
81
82    ea = gen_lea_modrm_1(s, *mem, decode->e.vex_class == 12);
83    if (decode->e.special == X86_SPECIAL_BitTest) {
84        MemOp ot = decode->op[1].ot;
85        int poslen = 8 << ot;
86        int opn = decode->op[2].n;
87        TCGv ofs = tcg_temp_new();
88
89        /* Extract memory displacement from the second operand.  */
90        assert(decode->op[2].unit == X86_OP_INT && decode->op[2].ot != MO_8);
91        tcg_gen_sextract_tl(ofs, cpu_regs[opn], 3, poslen - 3);
92        tcg_gen_andi_tl(ofs, ofs, -1 << ot);
93        tcg_gen_add_tl(s->A0, ea, ofs);
94        ea = s->A0;
95    }
96
97    gen_lea_v_seg(s, ea, mem->def_seg, s->override);
98}
99
100static inline int mmx_offset(MemOp ot)
101{
102    switch (ot) {
103    case MO_8:
104        return offsetof(MMXReg, MMX_B(0));
105    case MO_16:
106        return offsetof(MMXReg, MMX_W(0));
107    case MO_32:
108        return offsetof(MMXReg, MMX_L(0));
109    case MO_64:
110        return offsetof(MMXReg, MMX_Q(0));
111    default:
112        g_assert_not_reached();
113    }
114}
115
116static inline int xmm_offset(MemOp ot)
117{
118    switch (ot) {
119    case MO_8:
120        return offsetof(ZMMReg, ZMM_B(0));
121    case MO_16:
122        return offsetof(ZMMReg, ZMM_W(0));
123    case MO_32:
124        return offsetof(ZMMReg, ZMM_L(0));
125    case MO_64:
126        return offsetof(ZMMReg, ZMM_Q(0));
127    case MO_128:
128        return offsetof(ZMMReg, ZMM_X(0));
129    case MO_256:
130        return offsetof(ZMMReg, ZMM_Y(0));
131    default:
132        g_assert_not_reached();
133    }
134}
135
136static int vector_reg_offset(X86DecodedOp *op)
137{
138    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
139
140    if (op->unit == X86_OP_MMX) {
141        return op->offset - mmx_offset(op->ot);
142    } else {
143        return op->offset - xmm_offset(op->ot);
144    }
145}
146
147static int vector_elem_offset(X86DecodedOp *op, MemOp ot, int n)
148{
149    int base_ofs = vector_reg_offset(op);
150    switch(ot) {
151    case MO_8:
152        if (op->unit == X86_OP_MMX) {
153            return base_ofs + offsetof(MMXReg, MMX_B(n));
154        } else {
155            return base_ofs + offsetof(ZMMReg, ZMM_B(n));
156        }
157    case MO_16:
158        if (op->unit == X86_OP_MMX) {
159            return base_ofs + offsetof(MMXReg, MMX_W(n));
160        } else {
161            return base_ofs + offsetof(ZMMReg, ZMM_W(n));
162        }
163    case MO_32:
164        if (op->unit == X86_OP_MMX) {
165            return base_ofs + offsetof(MMXReg, MMX_L(n));
166        } else {
167            return base_ofs + offsetof(ZMMReg, ZMM_L(n));
168        }
169    case MO_64:
170        if (op->unit == X86_OP_MMX) {
171            return base_ofs;
172        } else {
173            return base_ofs + offsetof(ZMMReg, ZMM_Q(n));
174        }
175    case MO_128:
176        assert(op->unit == X86_OP_SSE);
177        return base_ofs + offsetof(ZMMReg, ZMM_X(n));
178    case MO_256:
179        assert(op->unit == X86_OP_SSE);
180        return base_ofs + offsetof(ZMMReg, ZMM_Y(n));
181    default:
182        g_assert_not_reached();
183    }
184}
185
186static void compute_mmx_offset(X86DecodedOp *op)
187{
188    if (!op->has_ea) {
189        op->offset = MMX_OFFSET(op->n) + mmx_offset(op->ot);
190    } else {
191        op->offset = offsetof(CPUX86State, mmx_t0) + mmx_offset(op->ot);
192    }
193}
194
195static void compute_xmm_offset(X86DecodedOp *op)
196{
197    if (!op->has_ea) {
198        op->offset = ZMM_OFFSET(op->n) + xmm_offset(op->ot);
199    } else {
200        op->offset = offsetof(CPUX86State, xmm_t0) + xmm_offset(op->ot);
201    }
202}
203
204static void gen_load_sse(DisasContext *s, TCGv temp, MemOp ot, int dest_ofs, bool aligned)
205{
206    switch(ot) {
207    case MO_8:
208        gen_op_ld_v(s, MO_8, temp, s->A0);
209        tcg_gen_st8_tl(temp, tcg_env, dest_ofs);
210        break;
211    case MO_16:
212        gen_op_ld_v(s, MO_16, temp, s->A0);
213        tcg_gen_st16_tl(temp, tcg_env, dest_ofs);
214        break;
215    case MO_32:
216        gen_op_ld_v(s, MO_32, temp, s->A0);
217        tcg_gen_st32_tl(temp, tcg_env, dest_ofs);
218        break;
219    case MO_64:
220        gen_ldq_env_A0(s, dest_ofs);
221        break;
222    case MO_128:
223        gen_ldo_env_A0(s, dest_ofs, aligned);
224        break;
225    case MO_256:
226        gen_ldy_env_A0(s, dest_ofs, aligned);
227        break;
228    default:
229        g_assert_not_reached();
230    }
231}
232
233static bool sse_needs_alignment(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
234{
235    switch (decode->e.vex_class) {
236    case 2:
237    case 4:
238        if ((s->prefix & PREFIX_VEX) ||
239            decode->e.vex_special == X86_VEX_SSEUnaligned) {
240            /* MOST legacy SSE instructions require aligned memory operands, but not all.  */
241            return false;
242        }
243        /* fall through */
244    case 1:
245        return ot >= MO_128;
246
247    default:
248        return false;
249    }
250}
251
252static void gen_load(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
253{
254    X86DecodedOp *op = &decode->op[opn];
255
256    switch (op->unit) {
257    case X86_OP_SKIP:
258        return;
259    case X86_OP_SEG:
260        tcg_gen_ld32u_tl(v, tcg_env,
261                         offsetof(CPUX86State,segs[op->n].selector));
262        break;
263#ifndef CONFIG_USER_ONLY
264    case X86_OP_CR:
265        if (op->n == 8) {
266            translator_io_start(&s->base);
267            gen_helper_read_cr8(v, tcg_env);
268        } else {
269            tcg_gen_ld_tl(v, tcg_env, offsetof(CPUX86State, cr[op->n]));
270        }
271        break;
272    case X86_OP_DR:
273        /* CR4.DE tested in the helper.  */
274        gen_helper_get_dr(v, tcg_env, tcg_constant_i32(op->n));
275        break;
276#endif
277    case X86_OP_INT:
278        if (op->has_ea) {
279            if (v == s->T0 && decode->e.special == X86_SPECIAL_SExtT0) {
280                gen_op_ld_v(s, op->ot | MO_SIGN, v, s->A0);
281            } else {
282                gen_op_ld_v(s, op->ot, v, s->A0);
283            }
284
285        } else if (op->ot < MO_TL && v == s->T0 &&
286                   (decode->e.special == X86_SPECIAL_SExtT0 ||
287                    decode->e.special == X86_SPECIAL_ZExtT0)) {
288            if (op->ot == MO_8 && byte_reg_is_xH(s, op->n)) {
289                if (decode->e.special == X86_SPECIAL_SExtT0) {
290                    tcg_gen_sextract_tl(v, cpu_regs[op->n - 4], 8, 8);
291                } else {
292                    tcg_gen_extract_tl(v, cpu_regs[op->n - 4], 8, 8);
293                }
294            } else {
295                if (decode->e.special == X86_SPECIAL_SExtT0) {
296                    tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot | MO_SIGN);
297                } else {
298                    tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot);
299                }
300            }
301
302        } else {
303            gen_op_mov_v_reg(s, op->ot, v, op->n);
304        }
305        break;
306    case X86_OP_IMM:
307        tcg_gen_movi_tl(v, op->imm);
308        break;
309
310    case X86_OP_MMX:
311        compute_mmx_offset(op);
312        goto load_vector;
313
314    case X86_OP_SSE:
315        compute_xmm_offset(op);
316    load_vector:
317        if (op->has_ea) {
318            bool aligned = sse_needs_alignment(s, decode, op->ot);
319            gen_load_sse(s, v, op->ot, op->offset, aligned);
320        }
321        break;
322
323    default:
324        g_assert_not_reached();
325    }
326}
327
328static TCGv_ptr op_ptr(X86DecodedInsn *decode, int opn)
329{
330    X86DecodedOp *op = &decode->op[opn];
331
332    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
333    if (op->v_ptr) {
334        return op->v_ptr;
335    }
336    op->v_ptr = tcg_temp_new_ptr();
337
338    /* The temporary points to the MMXReg or ZMMReg.  */
339    tcg_gen_addi_ptr(op->v_ptr, tcg_env, vector_reg_offset(op));
340    return op->v_ptr;
341}
342
343#define OP_PTR0 op_ptr(decode, 0)
344#define OP_PTR1 op_ptr(decode, 1)
345#define OP_PTR2 op_ptr(decode, 2)
346
347static void gen_writeback(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
348{
349    X86DecodedOp *op = &decode->op[opn];
350    switch (op->unit) {
351    case X86_OP_SKIP:
352        break;
353    case X86_OP_SEG:
354        /* Note that gen_movl_seg takes care of interrupt shadow and TF.  */
355        gen_movl_seg(s, op->n, v, op->n == R_SS);
356        break;
357    case X86_OP_INT:
358        if (op->has_ea) {
359            gen_op_st_v(s, op->ot, v, s->A0);
360        } else {
361            gen_op_mov_reg_v(s, op->ot, op->n, v);
362        }
363        break;
364    case X86_OP_MMX:
365        break;
366    case X86_OP_SSE:
367        if (!op->has_ea && (s->prefix & PREFIX_VEX) && op->ot <= MO_128) {
368            tcg_gen_gvec_dup_imm(MO_64,
369                                 offsetof(CPUX86State, xmm_regs[op->n].ZMM_X(1)),
370                                 16, 16, 0);
371        }
372        break;
373#ifndef CONFIG_USER_ONLY
374    case X86_OP_CR:
375        if (op->n == 8) {
376            translator_io_start(&s->base);
377        }
378        gen_helper_write_crN(tcg_env, tcg_constant_i32(op->n), v);
379        s->base.is_jmp = DISAS_EOB_NEXT;
380        break;
381    case X86_OP_DR:
382        /* CR4.DE tested in the helper.  */
383        gen_helper_set_dr(tcg_env, tcg_constant_i32(op->n), v);
384        s->base.is_jmp = DISAS_EOB_NEXT;
385        break;
386#endif
387    default:
388        g_assert_not_reached();
389    }
390    op->unit = X86_OP_SKIP;
391}
392
393static inline int vector_len(DisasContext *s, X86DecodedInsn *decode)
394{
395    if (decode->e.special == X86_SPECIAL_MMX &&
396        !(s->prefix & (PREFIX_DATA | PREFIX_REPZ | PREFIX_REPNZ))) {
397        return 8;
398    }
399    return s->vex_l ? 32 : 16;
400}
401
402static void prepare_update1_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
403{
404    decode->cc_dst = s->T0;
405    decode->cc_op = op;
406}
407
408static void prepare_update2_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
409{
410    decode->cc_src = s->T1;
411    decode->cc_dst = s->T0;
412    decode->cc_op = op;
413}
414
415static void prepare_update_cc_incdec(X86DecodedInsn *decode, DisasContext *s, CCOp op)
416{
417    gen_compute_eflags_c(s, s->T1);
418    prepare_update2_cc(decode, s, op);
419}
420
421static void prepare_update3_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op, TCGv reg)
422{
423    decode->cc_src2 = reg;
424    decode->cc_src = s->T1;
425    decode->cc_dst = s->T0;
426    decode->cc_op = op;
427}
428
429/* Set up decode->cc_* to modify CF while keeping other flags unchanged.  */
430static void prepare_update_cf(X86DecodedInsn *decode, DisasContext *s, TCGv cf)
431{
432    switch (s->cc_op) {
433    case CC_OP_ADOX:
434    case CC_OP_ADCOX:
435        decode->cc_src2 = cpu_cc_src2;
436        decode->cc_src = cpu_cc_src;
437        decode->cc_op = CC_OP_ADCOX;
438        break;
439
440    case CC_OP_EFLAGS:
441    case CC_OP_ADCX:
442        decode->cc_src = cpu_cc_src;
443        decode->cc_op = CC_OP_ADCX;
444        break;
445
446    default:
447        decode->cc_src = tcg_temp_new();
448        gen_mov_eflags(s, decode->cc_src);
449        decode->cc_op = CC_OP_ADCX;
450        break;
451    }
452    decode->cc_dst = cf;
453}
454
455static void gen_store_sse(DisasContext *s, X86DecodedInsn *decode, int src_ofs)
456{
457    MemOp ot = decode->op[0].ot;
458    int vec_len = vector_len(s, decode);
459    bool aligned = sse_needs_alignment(s, decode, ot);
460
461    if (!decode->op[0].has_ea) {
462        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, vec_len, vec_len);
463        return;
464    }
465
466    switch (ot) {
467    case MO_64:
468        gen_stq_env_A0(s, src_ofs);
469        break;
470    case MO_128:
471        gen_sto_env_A0(s, src_ofs, aligned);
472        break;
473    case MO_256:
474        gen_sty_env_A0(s, src_ofs, aligned);
475        break;
476    default:
477        g_assert_not_reached();
478    }
479}
480
481static void gen_helper_pavgusb(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b)
482{
483    gen_helper_pavgb_mmx(env, reg_a, reg_a, reg_b);
484}
485
486#define FN_3DNOW_MOVE ((SSEFunc_0_epp) (uintptr_t) 1)
487static const SSEFunc_0_epp fns_3dnow[] = {
488    [0x0c] = gen_helper_pi2fw,
489    [0x0d] = gen_helper_pi2fd,
490    [0x1c] = gen_helper_pf2iw,
491    [0x1d] = gen_helper_pf2id,
492    [0x8a] = gen_helper_pfnacc,
493    [0x8e] = gen_helper_pfpnacc,
494    [0x90] = gen_helper_pfcmpge,
495    [0x94] = gen_helper_pfmin,
496    [0x96] = gen_helper_pfrcp,
497    [0x97] = gen_helper_pfrsqrt,
498    [0x9a] = gen_helper_pfsub,
499    [0x9e] = gen_helper_pfadd,
500    [0xa0] = gen_helper_pfcmpgt,
501    [0xa4] = gen_helper_pfmax,
502    [0xa6] = FN_3DNOW_MOVE, /* PFRCPIT1; no need to actually increase precision */
503    [0xa7] = FN_3DNOW_MOVE, /* PFRSQIT1 */
504    [0xb6] = FN_3DNOW_MOVE, /* PFRCPIT2 */
505    [0xaa] = gen_helper_pfsubr,
506    [0xae] = gen_helper_pfacc,
507    [0xb0] = gen_helper_pfcmpeq,
508    [0xb4] = gen_helper_pfmul,
509    [0xb7] = gen_helper_pmulhrw_mmx,
510    [0xbb] = gen_helper_pswapd,
511    [0xbf] = gen_helper_pavgusb,
512};
513
514static void gen_3dnow(DisasContext *s, X86DecodedInsn *decode)
515{
516    uint8_t b = decode->immediate;
517    SSEFunc_0_epp fn = b < ARRAY_SIZE(fns_3dnow) ? fns_3dnow[b] : NULL;
518
519    if (!fn) {
520        gen_illegal_opcode(s);
521        return;
522    }
523    if (s->flags & HF_TS_MASK) {
524        gen_NM_exception(s);
525        return;
526    }
527    if (s->flags & HF_EM_MASK) {
528        gen_illegal_opcode(s);
529        return;
530    }
531
532    gen_helper_enter_mmx(tcg_env);
533    if (fn == FN_3DNOW_MOVE) {
534       tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset);
535       tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset);
536    } else {
537       fn(tcg_env, OP_PTR0, OP_PTR1);
538    }
539}
540
541/*
542 * 00 = v*ps Vps, Hps, Wpd
543 * 66 = v*pd Vpd, Hpd, Wps
544 * f3 = v*ss Vss, Hss, Wps
545 * f2 = v*sd Vsd, Hsd, Wps
546 */
547static inline void gen_unary_fp_sse(DisasContext *s, X86DecodedInsn *decode,
548                              SSEFunc_0_epp pd_xmm, SSEFunc_0_epp ps_xmm,
549                              SSEFunc_0_epp pd_ymm, SSEFunc_0_epp ps_ymm,
550                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
551{
552    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
553        SSEFunc_0_eppp fn = s->prefix & PREFIX_REPZ ? ss : sd;
554        if (!fn) {
555            gen_illegal_opcode(s);
556            return;
557        }
558        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
559    } else {
560        SSEFunc_0_epp ps, pd, fn;
561        ps = s->vex_l ? ps_ymm : ps_xmm;
562        pd = s->vex_l ? pd_ymm : pd_xmm;
563        fn = s->prefix & PREFIX_DATA ? pd : ps;
564        if (!fn) {
565            gen_illegal_opcode(s);
566            return;
567        }
568        fn(tcg_env, OP_PTR0, OP_PTR2);
569    }
570}
571#define UNARY_FP_SSE(uname, lname)                                                 \
572static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
573{                                                                                  \
574    gen_unary_fp_sse(s, decode,                                                    \
575                     gen_helper_##lname##pd_xmm,                                   \
576                     gen_helper_##lname##ps_xmm,                                   \
577                     gen_helper_##lname##pd_ymm,                                   \
578                     gen_helper_##lname##ps_ymm,                                   \
579                     gen_helper_##lname##sd,                                       \
580                     gen_helper_##lname##ss);                                      \
581}
582UNARY_FP_SSE(VSQRT, sqrt)
583
584/*
585 * 00 = v*ps Vps, Hps, Wpd
586 * 66 = v*pd Vpd, Hpd, Wps
587 * f3 = v*ss Vss, Hss, Wps
588 * f2 = v*sd Vsd, Hsd, Wps
589 */
590static inline void gen_fp_sse(DisasContext *s, X86DecodedInsn *decode,
591                              SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
592                              SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm,
593                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
594{
595    SSEFunc_0_eppp ps, pd, fn;
596    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
597        fn = s->prefix & PREFIX_REPZ ? ss : sd;
598    } else {
599        ps = s->vex_l ? ps_ymm : ps_xmm;
600        pd = s->vex_l ? pd_ymm : pd_xmm;
601        fn = s->prefix & PREFIX_DATA ? pd : ps;
602    }
603    if (fn) {
604        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
605    } else {
606        gen_illegal_opcode(s);
607    }
608}
609
610#define FP_SSE(uname, lname)                                                       \
611static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
612{                                                                                  \
613    gen_fp_sse(s, decode,                                                          \
614               gen_helper_##lname##pd_xmm,                                         \
615               gen_helper_##lname##ps_xmm,                                         \
616               gen_helper_##lname##pd_ymm,                                         \
617               gen_helper_##lname##ps_ymm,                                         \
618               gen_helper_##lname##sd,                                             \
619               gen_helper_##lname##ss);                                            \
620}
621FP_SSE(VADD, add)
622FP_SSE(VMUL, mul)
623FP_SSE(VSUB, sub)
624FP_SSE(VMIN, min)
625FP_SSE(VDIV, div)
626FP_SSE(VMAX, max)
627
628#define FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, even, odd)                         \
629static void gen_##uname##Px(DisasContext *s, X86DecodedInsn *decode)               \
630{                                                                                  \
631    SSEFunc_0_eppppii xmm = s->vex_w ? gen_helper_fma4pd_xmm : gen_helper_fma4ps_xmm; \
632    SSEFunc_0_eppppii ymm = s->vex_w ? gen_helper_fma4pd_ymm : gen_helper_fma4ps_ymm; \
633    SSEFunc_0_eppppii fn = s->vex_l ? ymm : xmm;                                   \
634                                                                                   \
635    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
636       tcg_constant_i32(even),                                                     \
637       tcg_constant_i32((even) ^ (odd)));                                          \
638}
639
640#define FMA_SSE(uname, ptr0, ptr1, ptr2, flags)                                    \
641FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, flags, flags)                              \
642static void gen_##uname##Sx(DisasContext *s, X86DecodedInsn *decode)               \
643{                                                                                  \
644    SSEFunc_0_eppppi fn = s->vex_w ? gen_helper_fma4sd : gen_helper_fma4ss;        \
645                                                                                   \
646    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
647       tcg_constant_i32(flags));                                                   \
648}                                                                                  \
649
650FMA_SSE(VFMADD231,  OP_PTR1, OP_PTR2, OP_PTR0, 0)
651FMA_SSE(VFMADD213,  OP_PTR1, OP_PTR0, OP_PTR2, 0)
652FMA_SSE(VFMADD132,  OP_PTR0, OP_PTR2, OP_PTR1, 0)
653
654FMA_SSE(VFNMADD231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_product)
655FMA_SSE(VFNMADD213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_product)
656FMA_SSE(VFNMADD132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_product)
657
658FMA_SSE(VFMSUB231,  OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c)
659FMA_SSE(VFMSUB213,  OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c)
660FMA_SSE(VFMSUB132,  OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c)
661
662FMA_SSE(VFNMSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c|float_muladd_negate_product)
663FMA_SSE(VFNMSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c|float_muladd_negate_product)
664FMA_SSE(VFNMSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c|float_muladd_negate_product)
665
666FMA_SSE_PACKED(VFMADDSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c, 0)
667FMA_SSE_PACKED(VFMADDSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c, 0)
668FMA_SSE_PACKED(VFMADDSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c, 0)
669
670FMA_SSE_PACKED(VFMSUBADD231, OP_PTR1, OP_PTR2, OP_PTR0, 0, float_muladd_negate_c)
671FMA_SSE_PACKED(VFMSUBADD213, OP_PTR1, OP_PTR0, OP_PTR2, 0, float_muladd_negate_c)
672FMA_SSE_PACKED(VFMSUBADD132, OP_PTR0, OP_PTR2, OP_PTR1, 0, float_muladd_negate_c)
673
674#define FP_UNPACK_SSE(uname, lname)                                                \
675static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
676{                                                                                  \
677    /* PS maps to the DQ integer instruction, PD maps to QDQ.  */                  \
678    gen_fp_sse(s, decode,                                                          \
679               gen_helper_##lname##qdq_xmm,                                        \
680               gen_helper_##lname##dq_xmm,                                         \
681               gen_helper_##lname##qdq_ymm,                                        \
682               gen_helper_##lname##dq_ymm,                                         \
683               NULL, NULL);                                                        \
684}
685FP_UNPACK_SSE(VUNPCKLPx, punpckl)
686FP_UNPACK_SSE(VUNPCKHPx, punpckh)
687
688/*
689 * 00 = v*ps Vps, Wpd
690 * f3 = v*ss Vss, Wps
691 */
692static inline void gen_unary_fp32_sse(DisasContext *s, X86DecodedInsn *decode,
693                                      SSEFunc_0_epp ps_xmm,
694                                      SSEFunc_0_epp ps_ymm,
695                                      SSEFunc_0_eppp ss)
696{
697    if ((s->prefix & (PREFIX_DATA | PREFIX_REPNZ)) != 0) {
698        goto illegal_op;
699    } else if (s->prefix & PREFIX_REPZ) {
700        if (!ss) {
701            goto illegal_op;
702        }
703        ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
704    } else {
705        SSEFunc_0_epp fn = s->vex_l ? ps_ymm : ps_xmm;
706        if (!fn) {
707            goto illegal_op;
708        }
709        fn(tcg_env, OP_PTR0, OP_PTR2);
710    }
711    return;
712
713illegal_op:
714    gen_illegal_opcode(s);
715}
716#define UNARY_FP32_SSE(uname, lname)                                               \
717static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
718{                                                                                  \
719    gen_unary_fp32_sse(s, decode,                                                  \
720                       gen_helper_##lname##ps_xmm,                                 \
721                       gen_helper_##lname##ps_ymm,                                 \
722                       gen_helper_##lname##ss);                                    \
723}
724UNARY_FP32_SSE(VRSQRT, rsqrt)
725UNARY_FP32_SSE(VRCP, rcp)
726
727/*
728 * 66 = v*pd Vpd, Hpd, Wpd
729 * f2 = v*ps Vps, Hps, Wps
730 */
731static inline void gen_horizontal_fp_sse(DisasContext *s, X86DecodedInsn *decode,
732                                         SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
733                                         SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm)
734{
735    SSEFunc_0_eppp ps, pd, fn;
736    ps = s->vex_l ? ps_ymm : ps_xmm;
737    pd = s->vex_l ? pd_ymm : pd_xmm;
738    fn = s->prefix & PREFIX_DATA ? pd : ps;
739    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
740}
741#define HORIZONTAL_FP_SSE(uname, lname)                                            \
742static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
743{                                                                                  \
744    gen_horizontal_fp_sse(s, decode,                                               \
745                          gen_helper_##lname##pd_xmm, gen_helper_##lname##ps_xmm,  \
746                          gen_helper_##lname##pd_ymm, gen_helper_##lname##ps_ymm); \
747}
748HORIZONTAL_FP_SSE(VHADD, hadd)
749HORIZONTAL_FP_SSE(VHSUB, hsub)
750HORIZONTAL_FP_SSE(VADDSUB, addsub)
751
752static inline void gen_ternary_sse(DisasContext *s, X86DecodedInsn *decode,
753                                   int op3, SSEFunc_0_epppp xmm, SSEFunc_0_epppp ymm)
754{
755    SSEFunc_0_epppp fn = s->vex_l ? ymm : xmm;
756    TCGv_ptr ptr3 = tcg_temp_new_ptr();
757
758    /* The format of the fourth input is Lx */
759    tcg_gen_addi_ptr(ptr3, tcg_env, ZMM_OFFSET(op3));
760    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, ptr3);
761}
762#define TERNARY_SSE(uname, uvname, lname)                                          \
763static void gen_##uvname(DisasContext *s, X86DecodedInsn *decode)                  \
764{                                                                                  \
765    gen_ternary_sse(s, decode, (uint8_t)decode->immediate >> 4,                    \
766                    gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);           \
767}                                                                                  \
768static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
769{                                                                                  \
770    gen_ternary_sse(s, decode, 0,                                                  \
771                  gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);             \
772}
773TERNARY_SSE(BLENDVPS, VBLENDVPS, blendvps)
774TERNARY_SSE(BLENDVPD, VBLENDVPD, blendvpd)
775TERNARY_SSE(PBLENDVB, VPBLENDVB, pblendvb)
776
777static inline void gen_binary_imm_sse(DisasContext *s, X86DecodedInsn *decode,
778                                      SSEFunc_0_epppi xmm, SSEFunc_0_epppi ymm)
779{
780    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
781    if (!s->vex_l) {
782        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
783    } else {
784        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
785    }
786}
787
788#define BINARY_IMM_SSE(uname, lname)                                               \
789static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
790{                                                                                  \
791    gen_binary_imm_sse(s, decode,                                                  \
792                       gen_helper_##lname##_xmm,                                   \
793                       gen_helper_##lname##_ymm);                                  \
794}
795
796BINARY_IMM_SSE(VBLENDPD,   blendpd)
797BINARY_IMM_SSE(VBLENDPS,   blendps)
798BINARY_IMM_SSE(VPBLENDW,   pblendw)
799BINARY_IMM_SSE(VDDPS,      dpps)
800#define gen_helper_dppd_ymm NULL
801BINARY_IMM_SSE(VDDPD,      dppd)
802BINARY_IMM_SSE(VMPSADBW,   mpsadbw)
803BINARY_IMM_SSE(PCLMULQDQ,  pclmulqdq)
804
805
806#define UNARY_INT_GVEC(uname, func, ...)                                           \
807static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
808{                                                                                  \
809    int vec_len = vector_len(s, decode);                                          \
810                                                                                   \
811    func(__VA_ARGS__, decode->op[0].offset,                                        \
812         decode->op[2].offset, vec_len, vec_len);                                  \
813}
814UNARY_INT_GVEC(PABSB,          tcg_gen_gvec_abs, MO_8)
815UNARY_INT_GVEC(PABSW,          tcg_gen_gvec_abs, MO_16)
816UNARY_INT_GVEC(PABSD,          tcg_gen_gvec_abs, MO_32)
817UNARY_INT_GVEC(VBROADCASTx128, tcg_gen_gvec_dup_mem, MO_128)
818UNARY_INT_GVEC(VPBROADCASTB,   tcg_gen_gvec_dup_mem, MO_8)
819UNARY_INT_GVEC(VPBROADCASTW,   tcg_gen_gvec_dup_mem, MO_16)
820UNARY_INT_GVEC(VPBROADCASTD,   tcg_gen_gvec_dup_mem, MO_32)
821UNARY_INT_GVEC(VPBROADCASTQ,   tcg_gen_gvec_dup_mem, MO_64)
822
823
824#define BINARY_INT_GVEC(uname, func, ...)                                          \
825static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
826{                                                                                  \
827    int vec_len = vector_len(s, decode);                                          \
828                                                                                   \
829    func(__VA_ARGS__,                                                              \
830         decode->op[0].offset, decode->op[1].offset,                               \
831         decode->op[2].offset, vec_len, vec_len);                                  \
832}
833
834BINARY_INT_GVEC(PADDB,   tcg_gen_gvec_add, MO_8)
835BINARY_INT_GVEC(PADDW,   tcg_gen_gvec_add, MO_16)
836BINARY_INT_GVEC(PADDD,   tcg_gen_gvec_add, MO_32)
837BINARY_INT_GVEC(PADDQ,   tcg_gen_gvec_add, MO_64)
838BINARY_INT_GVEC(PADDSB,  tcg_gen_gvec_ssadd, MO_8)
839BINARY_INT_GVEC(PADDSW,  tcg_gen_gvec_ssadd, MO_16)
840BINARY_INT_GVEC(PADDUSB, tcg_gen_gvec_usadd, MO_8)
841BINARY_INT_GVEC(PADDUSW, tcg_gen_gvec_usadd, MO_16)
842BINARY_INT_GVEC(PAND,    tcg_gen_gvec_and, MO_64)
843BINARY_INT_GVEC(PCMPEQB, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_8)
844BINARY_INT_GVEC(PCMPEQD, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_32)
845BINARY_INT_GVEC(PCMPEQW, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_16)
846BINARY_INT_GVEC(PCMPEQQ, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_64)
847BINARY_INT_GVEC(PCMPGTB, tcg_gen_gvec_cmp, TCG_COND_GT, MO_8)
848BINARY_INT_GVEC(PCMPGTW, tcg_gen_gvec_cmp, TCG_COND_GT, MO_16)
849BINARY_INT_GVEC(PCMPGTD, tcg_gen_gvec_cmp, TCG_COND_GT, MO_32)
850BINARY_INT_GVEC(PCMPGTQ, tcg_gen_gvec_cmp, TCG_COND_GT, MO_64)
851BINARY_INT_GVEC(PMAXSB,  tcg_gen_gvec_smax, MO_8)
852BINARY_INT_GVEC(PMAXSW,  tcg_gen_gvec_smax, MO_16)
853BINARY_INT_GVEC(PMAXSD,  tcg_gen_gvec_smax, MO_32)
854BINARY_INT_GVEC(PMAXUB,  tcg_gen_gvec_umax, MO_8)
855BINARY_INT_GVEC(PMAXUW,  tcg_gen_gvec_umax, MO_16)
856BINARY_INT_GVEC(PMAXUD,  tcg_gen_gvec_umax, MO_32)
857BINARY_INT_GVEC(PMINSB,  tcg_gen_gvec_smin, MO_8)
858BINARY_INT_GVEC(PMINSW,  tcg_gen_gvec_smin, MO_16)
859BINARY_INT_GVEC(PMINSD,  tcg_gen_gvec_smin, MO_32)
860BINARY_INT_GVEC(PMINUB,  tcg_gen_gvec_umin, MO_8)
861BINARY_INT_GVEC(PMINUW,  tcg_gen_gvec_umin, MO_16)
862BINARY_INT_GVEC(PMINUD,  tcg_gen_gvec_umin, MO_32)
863BINARY_INT_GVEC(PMULLW,  tcg_gen_gvec_mul, MO_16)
864BINARY_INT_GVEC(PMULLD,  tcg_gen_gvec_mul, MO_32)
865BINARY_INT_GVEC(POR,     tcg_gen_gvec_or, MO_64)
866BINARY_INT_GVEC(PSUBB,   tcg_gen_gvec_sub, MO_8)
867BINARY_INT_GVEC(PSUBW,   tcg_gen_gvec_sub, MO_16)
868BINARY_INT_GVEC(PSUBD,   tcg_gen_gvec_sub, MO_32)
869BINARY_INT_GVEC(PSUBQ,   tcg_gen_gvec_sub, MO_64)
870BINARY_INT_GVEC(PSUBSB,  tcg_gen_gvec_sssub, MO_8)
871BINARY_INT_GVEC(PSUBSW,  tcg_gen_gvec_sssub, MO_16)
872BINARY_INT_GVEC(PSUBUSB, tcg_gen_gvec_ussub, MO_8)
873BINARY_INT_GVEC(PSUBUSW, tcg_gen_gvec_ussub, MO_16)
874BINARY_INT_GVEC(PXOR,    tcg_gen_gvec_xor, MO_64)
875
876
877/*
878 * 00 = p*  Pq, Qq (if mmx not NULL; no VEX)
879 * 66 = vp* Vx, Hx, Wx
880 *
881 * These are really the same encoding, because 1) V is the same as P when VEX.V
882 * is not present 2) P and Q are the same as H and W apart from MM/XMM
883 */
884static inline void gen_binary_int_sse(DisasContext *s, X86DecodedInsn *decode,
885                                      SSEFunc_0_eppp mmx, SSEFunc_0_eppp xmm, SSEFunc_0_eppp ymm)
886{
887    assert(!!mmx == !!(decode->e.special == X86_SPECIAL_MMX));
888
889    if (mmx && (s->prefix & PREFIX_VEX) && !(s->prefix & PREFIX_DATA)) {
890        /* VEX encoding is not applicable to MMX instructions.  */
891        gen_illegal_opcode(s);
892        return;
893    }
894    if (!(s->prefix & PREFIX_DATA)) {
895        mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
896    } else if (!s->vex_l) {
897        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
898    } else {
899        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
900    }
901}
902
903
904#define BINARY_INT_MMX(uname, lname)                                               \
905static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
906{                                                                                  \
907    gen_binary_int_sse(s, decode,                                                  \
908                          gen_helper_##lname##_mmx,                                \
909                          gen_helper_##lname##_xmm,                                \
910                          gen_helper_##lname##_ymm);                               \
911}
912BINARY_INT_MMX(PUNPCKLBW,  punpcklbw)
913BINARY_INT_MMX(PUNPCKLWD,  punpcklwd)
914BINARY_INT_MMX(PUNPCKLDQ,  punpckldq)
915BINARY_INT_MMX(PACKSSWB,   packsswb)
916BINARY_INT_MMX(PACKUSWB,   packuswb)
917BINARY_INT_MMX(PUNPCKHBW,  punpckhbw)
918BINARY_INT_MMX(PUNPCKHWD,  punpckhwd)
919BINARY_INT_MMX(PUNPCKHDQ,  punpckhdq)
920BINARY_INT_MMX(PACKSSDW,   packssdw)
921
922BINARY_INT_MMX(PAVGB,   pavgb)
923BINARY_INT_MMX(PAVGW,   pavgw)
924BINARY_INT_MMX(PMADDWD, pmaddwd)
925BINARY_INT_MMX(PMULHUW, pmulhuw)
926BINARY_INT_MMX(PMULHW,  pmulhw)
927BINARY_INT_MMX(PMULUDQ, pmuludq)
928BINARY_INT_MMX(PSADBW,  psadbw)
929
930BINARY_INT_MMX(PSLLW_r, psllw)
931BINARY_INT_MMX(PSLLD_r, pslld)
932BINARY_INT_MMX(PSLLQ_r, psllq)
933BINARY_INT_MMX(PSRLW_r, psrlw)
934BINARY_INT_MMX(PSRLD_r, psrld)
935BINARY_INT_MMX(PSRLQ_r, psrlq)
936BINARY_INT_MMX(PSRAW_r, psraw)
937BINARY_INT_MMX(PSRAD_r, psrad)
938
939BINARY_INT_MMX(PHADDW,    phaddw)
940BINARY_INT_MMX(PHADDSW,   phaddsw)
941BINARY_INT_MMX(PHADDD,    phaddd)
942BINARY_INT_MMX(PHSUBW,    phsubw)
943BINARY_INT_MMX(PHSUBSW,   phsubsw)
944BINARY_INT_MMX(PHSUBD,    phsubd)
945BINARY_INT_MMX(PMADDUBSW, pmaddubsw)
946BINARY_INT_MMX(PSHUFB,    pshufb)
947BINARY_INT_MMX(PSIGNB,    psignb)
948BINARY_INT_MMX(PSIGNW,    psignw)
949BINARY_INT_MMX(PSIGND,    psignd)
950BINARY_INT_MMX(PMULHRSW,  pmulhrsw)
951
952/* Instructions with no MMX equivalent.  */
953#define BINARY_INT_SSE(uname, lname)                                               \
954static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
955{                                                                                  \
956    gen_binary_int_sse(s, decode,                                                  \
957                          NULL,                                                    \
958                          gen_helper_##lname##_xmm,                                \
959                          gen_helper_##lname##_ymm);                               \
960}
961
962/* Instructions with no MMX equivalent.  */
963BINARY_INT_SSE(PUNPCKLQDQ, punpcklqdq)
964BINARY_INT_SSE(PUNPCKHQDQ, punpckhqdq)
965BINARY_INT_SSE(VPACKUSDW,  packusdw)
966BINARY_INT_SSE(VPERMILPS,  vpermilps)
967BINARY_INT_SSE(VPERMILPD,  vpermilpd)
968BINARY_INT_SSE(VMASKMOVPS, vpmaskmovd)
969BINARY_INT_SSE(VMASKMOVPD, vpmaskmovq)
970
971BINARY_INT_SSE(PMULDQ,    pmuldq)
972
973BINARY_INT_SSE(VAESDEC, aesdec)
974BINARY_INT_SSE(VAESDECLAST, aesdeclast)
975BINARY_INT_SSE(VAESENC, aesenc)
976BINARY_INT_SSE(VAESENCLAST, aesenclast)
977
978#define UNARY_CMP_SSE(uname, lname)                                                \
979static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
980{                                                                                  \
981    if (!s->vex_l) {                                                               \
982        gen_helper_##lname##_xmm(tcg_env, OP_PTR1, OP_PTR2);                       \
983    } else {                                                                       \
984        gen_helper_##lname##_ymm(tcg_env, OP_PTR1, OP_PTR2);                       \
985    }                                                                              \
986    assume_cc_op(s, CC_OP_EFLAGS);                                                  \
987}
988UNARY_CMP_SSE(VPTEST,     ptest)
989UNARY_CMP_SSE(VTESTPS,    vtestps)
990UNARY_CMP_SSE(VTESTPD,    vtestpd)
991
992static inline void gen_unary_int_sse(DisasContext *s, X86DecodedInsn *decode,
993                                     SSEFunc_0_epp xmm, SSEFunc_0_epp ymm)
994{
995    if (!s->vex_l) {
996        xmm(tcg_env, OP_PTR0, OP_PTR2);
997    } else {
998        ymm(tcg_env, OP_PTR0, OP_PTR2);
999    }
1000}
1001
1002#define UNARY_INT_SSE(uname, lname)                                                \
1003static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1004{                                                                                  \
1005    gen_unary_int_sse(s, decode,                                                   \
1006                      gen_helper_##lname##_xmm,                                    \
1007                      gen_helper_##lname##_ymm);                                   \
1008}
1009
1010UNARY_INT_SSE(VPMOVSXBW,    pmovsxbw)
1011UNARY_INT_SSE(VPMOVSXBD,    pmovsxbd)
1012UNARY_INT_SSE(VPMOVSXBQ,    pmovsxbq)
1013UNARY_INT_SSE(VPMOVSXWD,    pmovsxwd)
1014UNARY_INT_SSE(VPMOVSXWQ,    pmovsxwq)
1015UNARY_INT_SSE(VPMOVSXDQ,    pmovsxdq)
1016
1017UNARY_INT_SSE(VPMOVZXBW,    pmovzxbw)
1018UNARY_INT_SSE(VPMOVZXBD,    pmovzxbd)
1019UNARY_INT_SSE(VPMOVZXBQ,    pmovzxbq)
1020UNARY_INT_SSE(VPMOVZXWD,    pmovzxwd)
1021UNARY_INT_SSE(VPMOVZXWQ,    pmovzxwq)
1022UNARY_INT_SSE(VPMOVZXDQ,    pmovzxdq)
1023
1024UNARY_INT_SSE(VMOVSLDUP,    pmovsldup)
1025UNARY_INT_SSE(VMOVSHDUP,    pmovshdup)
1026UNARY_INT_SSE(VMOVDDUP,     pmovdldup)
1027
1028UNARY_INT_SSE(VCVTDQ2PD, cvtdq2pd)
1029UNARY_INT_SSE(VCVTPD2DQ, cvtpd2dq)
1030UNARY_INT_SSE(VCVTTPD2DQ, cvttpd2dq)
1031UNARY_INT_SSE(VCVTDQ2PS, cvtdq2ps)
1032UNARY_INT_SSE(VCVTPS2DQ, cvtps2dq)
1033UNARY_INT_SSE(VCVTTPS2DQ, cvttps2dq)
1034UNARY_INT_SSE(VCVTPH2PS, cvtph2ps)
1035
1036
1037static inline void gen_unary_imm_sse(DisasContext *s, X86DecodedInsn *decode,
1038                                     SSEFunc_0_ppi xmm, SSEFunc_0_ppi ymm)
1039{
1040    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
1041    if (!s->vex_l) {
1042        xmm(OP_PTR0, OP_PTR1, imm);
1043    } else {
1044        ymm(OP_PTR0, OP_PTR1, imm);
1045    }
1046}
1047
1048#define UNARY_IMM_SSE(uname, lname)                                                \
1049static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1050{                                                                                  \
1051    gen_unary_imm_sse(s, decode,                                                   \
1052                      gen_helper_##lname##_xmm,                                    \
1053                      gen_helper_##lname##_ymm);                                   \
1054}
1055
1056UNARY_IMM_SSE(PSHUFD,     pshufd)
1057UNARY_IMM_SSE(PSHUFHW,    pshufhw)
1058UNARY_IMM_SSE(PSHUFLW,    pshuflw)
1059#define gen_helper_vpermq_xmm NULL
1060UNARY_IMM_SSE(VPERMQ,      vpermq)
1061UNARY_IMM_SSE(VPERMILPS_i, vpermilps_imm)
1062UNARY_IMM_SSE(VPERMILPD_i, vpermilpd_imm)
1063
1064static inline void gen_unary_imm_fp_sse(DisasContext *s, X86DecodedInsn *decode,
1065                                        SSEFunc_0_eppi xmm, SSEFunc_0_eppi ymm)
1066{
1067    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
1068    if (!s->vex_l) {
1069        xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
1070    } else {
1071        ymm(tcg_env, OP_PTR0, OP_PTR1, imm);
1072    }
1073}
1074
1075#define UNARY_IMM_FP_SSE(uname, lname)                                             \
1076static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1077{                                                                                  \
1078    gen_unary_imm_fp_sse(s, decode,                                                \
1079                      gen_helper_##lname##_xmm,                                    \
1080                      gen_helper_##lname##_ymm);                                   \
1081}
1082
1083UNARY_IMM_FP_SSE(VROUNDPS,    roundps)
1084UNARY_IMM_FP_SSE(VROUNDPD,    roundpd)
1085
1086static inline void gen_vexw_avx(DisasContext *s, X86DecodedInsn *decode,
1087                                SSEFunc_0_eppp d_xmm, SSEFunc_0_eppp q_xmm,
1088                                SSEFunc_0_eppp d_ymm, SSEFunc_0_eppp q_ymm)
1089{
1090    SSEFunc_0_eppp d = s->vex_l ? d_ymm : d_xmm;
1091    SSEFunc_0_eppp q = s->vex_l ? q_ymm : q_xmm;
1092    SSEFunc_0_eppp fn = s->vex_w ? q : d;
1093    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
1094}
1095
1096/* VEX.W affects whether to operate on 32- or 64-bit elements.  */
1097#define VEXW_AVX(uname, lname)                                                     \
1098static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1099{                                                                                  \
1100    gen_vexw_avx(s, decode,                                                        \
1101                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1102                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1103}
1104VEXW_AVX(VPSLLV,    vpsllv)
1105VEXW_AVX(VPSRLV,    vpsrlv)
1106VEXW_AVX(VPSRAV,    vpsrav)
1107VEXW_AVX(VPMASKMOV, vpmaskmov)
1108
1109/* Same as above, but with extra arguments to the helper.  */
1110static inline void gen_vsib_avx(DisasContext *s, X86DecodedInsn *decode,
1111                                SSEFunc_0_epppti d_xmm, SSEFunc_0_epppti q_xmm,
1112                                SSEFunc_0_epppti d_ymm, SSEFunc_0_epppti q_ymm)
1113{
1114    SSEFunc_0_epppti d = s->vex_l ? d_ymm : d_xmm;
1115    SSEFunc_0_epppti q = s->vex_l ? q_ymm : q_xmm;
1116    SSEFunc_0_epppti fn = s->vex_w ? q : d;
1117    TCGv_i32 scale = tcg_constant_i32(decode->mem.scale);
1118    TCGv_ptr index = tcg_temp_new_ptr();
1119
1120    /* Pass third input as (index, base, scale) */
1121    tcg_gen_addi_ptr(index, tcg_env, ZMM_OFFSET(decode->mem.index));
1122    fn(tcg_env, OP_PTR0, OP_PTR1, index, s->A0, scale);
1123
1124    /*
1125     * There are two output operands, so zero OP1's high 128 bits
1126     * in the VEX.128 case.
1127     */
1128    if (!s->vex_l) {
1129        int ymmh_ofs = vector_elem_offset(&decode->op[1], MO_128, 1);
1130        tcg_gen_gvec_dup_imm(MO_64, ymmh_ofs, 16, 16, 0);
1131    }
1132}
1133#define VSIB_AVX(uname, lname)                                                     \
1134static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1135{                                                                                  \
1136    gen_vsib_avx(s, decode,                                                        \
1137                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1138                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1139}
1140VSIB_AVX(VPGATHERD, vpgatherd)
1141VSIB_AVX(VPGATHERQ, vpgatherq)
1142
1143static void gen_AAA(DisasContext *s, X86DecodedInsn *decode)
1144{
1145    gen_update_cc_op(s);
1146    gen_helper_aaa(tcg_env);
1147    assume_cc_op(s, CC_OP_EFLAGS);
1148}
1149
1150static void gen_AAD(DisasContext *s, X86DecodedInsn *decode)
1151{
1152    gen_helper_aad(s->T0, s->T0, s->T1);
1153    prepare_update1_cc(decode, s, CC_OP_LOGICB);
1154}
1155
1156static void gen_AAM(DisasContext *s, X86DecodedInsn *decode)
1157{
1158    if (decode->immediate == 0) {
1159        gen_exception(s, EXCP00_DIVZ);
1160    } else {
1161        gen_helper_aam(s->T0, s->T0, s->T1);
1162        prepare_update1_cc(decode, s, CC_OP_LOGICB);
1163    }
1164}
1165
1166static void gen_AAS(DisasContext *s, X86DecodedInsn *decode)
1167{
1168    gen_update_cc_op(s);
1169    gen_helper_aas(tcg_env);
1170    assume_cc_op(s, CC_OP_EFLAGS);
1171}
1172
1173static void gen_ADC(DisasContext *s, X86DecodedInsn *decode)
1174{
1175    MemOp ot = decode->op[1].ot;
1176    TCGv c_in = tcg_temp_new();
1177
1178    gen_compute_eflags_c(s, c_in);
1179    if (s->prefix & PREFIX_LOCK) {
1180        tcg_gen_add_tl(s->T0, c_in, s->T1);
1181        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
1182                                    s->mem_index, ot | MO_LE);
1183    } else {
1184        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1185        tcg_gen_add_tl(s->T0, s->T0, c_in);
1186    }
1187    prepare_update3_cc(decode, s, CC_OP_ADCB + ot, c_in);
1188}
1189
1190static void gen_ADCOX(DisasContext *s, X86DecodedInsn *decode, int cc_op)
1191{
1192    MemOp ot = decode->op[0].ot;
1193    TCGv carry_in = NULL;
1194    TCGv *carry_out = (cc_op == CC_OP_ADCX ? &decode->cc_dst : &decode->cc_src2);
1195    TCGv zero;
1196
1197    decode->cc_op = cc_op;
1198    *carry_out = tcg_temp_new();
1199    if (CC_OP_HAS_EFLAGS(s->cc_op)) {
1200        decode->cc_src = cpu_cc_src;
1201
1202        /* Re-use the carry-out from a previous round?  */
1203        if (s->cc_op == cc_op || s->cc_op == CC_OP_ADCOX) {
1204            carry_in = (cc_op == CC_OP_ADCX ? cpu_cc_dst : cpu_cc_src2);
1205        }
1206
1207        /* Preserve the opposite carry from previous rounds?  */
1208        if (s->cc_op != cc_op && s->cc_op != CC_OP_EFLAGS) {
1209            decode->cc_op = CC_OP_ADCOX;
1210            if (carry_out == &decode->cc_dst) {
1211                decode->cc_src2 = cpu_cc_src2;
1212            } else {
1213                decode->cc_dst = cpu_cc_dst;
1214            }
1215        }
1216    } else {
1217        decode->cc_src = tcg_temp_new();
1218        gen_mov_eflags(s, decode->cc_src);
1219    }
1220
1221    if (!carry_in) {
1222        /* Get carry_in out of EFLAGS.  */
1223        carry_in = tcg_temp_new();
1224        tcg_gen_extract_tl(carry_in, decode->cc_src,
1225            ctz32(cc_op == CC_OP_ADCX ? CC_C : CC_O), 1);
1226    }
1227
1228    switch (ot) {
1229#ifdef TARGET_X86_64
1230    case MO_32:
1231        /* If TL is 64-bit just do everything in 64-bit arithmetic.  */
1232        tcg_gen_ext32u_tl(s->T0, s->T0);
1233        tcg_gen_ext32u_tl(s->T1, s->T1);
1234        tcg_gen_add_i64(s->T0, s->T0, s->T1);
1235        tcg_gen_add_i64(s->T0, s->T0, carry_in);
1236        tcg_gen_shri_i64(*carry_out, s->T0, 32);
1237        break;
1238#endif
1239    default:
1240        zero = tcg_constant_tl(0);
1241        tcg_gen_add2_tl(s->T0, *carry_out, s->T0, zero, carry_in, zero);
1242        tcg_gen_add2_tl(s->T0, *carry_out, s->T0, *carry_out, s->T1, zero);
1243        break;
1244    }
1245}
1246
1247static void gen_ADCX(DisasContext *s, X86DecodedInsn *decode)
1248{
1249    gen_ADCOX(s, decode, CC_OP_ADCX);
1250}
1251
1252static void gen_ADD(DisasContext *s, X86DecodedInsn *decode)
1253{
1254    MemOp ot = decode->op[1].ot;
1255
1256    if (s->prefix & PREFIX_LOCK) {
1257        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1258                                    s->mem_index, ot | MO_LE);
1259    } else {
1260        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1261    }
1262    prepare_update2_cc(decode, s, CC_OP_ADDB + ot);
1263}
1264
1265static void gen_ADOX(DisasContext *s, X86DecodedInsn *decode)
1266{
1267    gen_ADCOX(s, decode, CC_OP_ADOX);
1268}
1269
1270static void gen_AND(DisasContext *s, X86DecodedInsn *decode)
1271{
1272    MemOp ot = decode->op[1].ot;
1273
1274    if (s->prefix & PREFIX_LOCK) {
1275        tcg_gen_atomic_and_fetch_tl(s->T0, s->A0, s->T1,
1276                                    s->mem_index, ot | MO_LE);
1277    } else {
1278        tcg_gen_and_tl(s->T0, s->T0, s->T1);
1279    }
1280    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1281}
1282
1283static void gen_ANDN(DisasContext *s, X86DecodedInsn *decode)
1284{
1285    MemOp ot = decode->op[0].ot;
1286
1287    tcg_gen_andc_tl(s->T0, s->T1, s->T0);
1288    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1289}
1290
1291static void gen_ARPL(DisasContext *s, X86DecodedInsn *decode)
1292{
1293    TCGv zf = tcg_temp_new();
1294    TCGv flags = tcg_temp_new();
1295
1296    gen_mov_eflags(s, flags);
1297
1298    /* Compute adjusted DST in T1, merging in SRC[RPL].  */
1299    tcg_gen_deposit_tl(s->T1, s->T0, s->T1, 0, 2);
1300
1301    /* Z flag set if DST[RPL] < SRC[RPL] */
1302    tcg_gen_setcond_tl(TCG_COND_LTU, zf, s->T0, s->T1);
1303    tcg_gen_deposit_tl(flags, flags, zf, ctz32(CC_Z), 1);
1304
1305    /* Place maximum RPL in DST */
1306    tcg_gen_umax_tl(s->T0, s->T0, s->T1);
1307
1308    decode->cc_src = flags;
1309    decode->cc_op = CC_OP_EFLAGS;
1310}
1311
1312static void gen_BEXTR(DisasContext *s, X86DecodedInsn *decode)
1313{
1314    MemOp ot = decode->op[0].ot;
1315    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1316    TCGv zero = tcg_constant_tl(0);
1317    TCGv mone = tcg_constant_tl(-1);
1318
1319    /*
1320     * Extract START, and shift the operand.
1321     * Shifts larger than operand size get zeros.
1322     */
1323    tcg_gen_ext8u_tl(s->A0, s->T1);
1324    tcg_gen_shr_tl(s->T0, s->T0, s->A0);
1325
1326    tcg_gen_movcond_tl(TCG_COND_LEU, s->T0, s->A0, bound, s->T0, zero);
1327
1328    /*
1329     * Extract the LEN into an inverse mask.  Lengths larger than
1330     * operand size get all zeros, length 0 gets all ones.
1331     */
1332    tcg_gen_extract_tl(s->A0, s->T1, 8, 8);
1333    tcg_gen_shl_tl(s->T1, mone, s->A0);
1334    tcg_gen_movcond_tl(TCG_COND_LEU, s->T1, s->A0, bound, s->T1, zero);
1335    tcg_gen_andc_tl(s->T0, s->T0, s->T1);
1336
1337    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1338}
1339
1340static void gen_BLSI(DisasContext *s, X86DecodedInsn *decode)
1341{
1342    MemOp ot = decode->op[0].ot;
1343
1344    /* input in T1, which is ready for prepare_update2_cc  */
1345    tcg_gen_neg_tl(s->T0, s->T1);
1346    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1347    prepare_update2_cc(decode, s, CC_OP_BLSIB + ot);
1348}
1349
1350static void gen_BLSMSK(DisasContext *s, X86DecodedInsn *decode)
1351{
1352    MemOp ot = decode->op[0].ot;
1353
1354    /* input in T1, which is ready for prepare_update2_cc  */
1355    tcg_gen_subi_tl(s->T0, s->T1, 1);
1356    tcg_gen_xor_tl(s->T0, s->T0, s->T1);
1357    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1358}
1359
1360static void gen_BLSR(DisasContext *s, X86DecodedInsn *decode)
1361{
1362    MemOp ot = decode->op[0].ot;
1363
1364    /* input in T1, which is ready for prepare_update2_cc  */
1365    tcg_gen_subi_tl(s->T0, s->T1, 1);
1366    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1367    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1368}
1369
1370static void gen_BOUND(DisasContext *s, X86DecodedInsn *decode)
1371{
1372    TCGv_i32 op = tcg_temp_new_i32();
1373    tcg_gen_trunc_tl_i32(op, s->T0);
1374    if (decode->op[1].ot == MO_16) {
1375        gen_helper_boundw(tcg_env, s->A0, op);
1376    } else {
1377        gen_helper_boundl(tcg_env, s->A0, op);
1378    }
1379}
1380
1381/* Non-standard convention - on entry T0 is zero-extended input, T1 is the output.  */
1382static void gen_BSF(DisasContext *s, X86DecodedInsn *decode)
1383{
1384    MemOp ot = decode->op[0].ot;
1385
1386    /* Only the Z bit is defined and it is related to the input.  */
1387    decode->cc_dst = tcg_temp_new();
1388    decode->cc_op = CC_OP_LOGICB + ot;
1389    tcg_gen_mov_tl(decode->cc_dst, s->T0);
1390
1391    /*
1392     * The manual says that the output is undefined when the
1393     * input is zero, but real hardware leaves it unchanged, and
1394     * real programs appear to depend on that.  Accomplish this
1395     * by passing the output as the value to return upon zero.
1396     */
1397    tcg_gen_ctz_tl(s->T0, s->T0, s->T1);
1398}
1399
1400/* Non-standard convention - on entry T0 is zero-extended input, T1 is the output.  */
1401static void gen_BSR(DisasContext *s, X86DecodedInsn *decode)
1402{
1403    MemOp ot = decode->op[0].ot;
1404
1405    /* Only the Z bit is defined and it is related to the input.  */
1406    decode->cc_dst = tcg_temp_new();
1407    decode->cc_op = CC_OP_LOGICB + ot;
1408    tcg_gen_mov_tl(decode->cc_dst, s->T0);
1409
1410    /*
1411     * The manual says that the output is undefined when the
1412     * input is zero, but real hardware leaves it unchanged, and
1413     * real programs appear to depend on that.  Accomplish this
1414     * by passing the output as the value to return upon zero.
1415     * Plus, return the bit index of the first 1 bit.
1416     */
1417    tcg_gen_xori_tl(s->T1, s->T1, TARGET_LONG_BITS - 1);
1418    tcg_gen_clz_tl(s->T0, s->T0, s->T1);
1419    tcg_gen_xori_tl(s->T0, s->T0, TARGET_LONG_BITS - 1);
1420}
1421
1422static void gen_BSWAP(DisasContext *s, X86DecodedInsn *decode)
1423{
1424#ifdef TARGET_X86_64
1425    if (s->dflag == MO_64) {
1426        tcg_gen_bswap64_i64(s->T0, s->T0);
1427        return;
1428    }
1429#endif
1430    tcg_gen_bswap32_tl(s->T0, s->T0, TCG_BSWAP_OZ);
1431}
1432
1433static TCGv gen_bt_mask(DisasContext *s, X86DecodedInsn *decode)
1434{
1435    MemOp ot = decode->op[1].ot;
1436    TCGv mask = tcg_temp_new();
1437
1438    tcg_gen_andi_tl(s->T1, s->T1, (8 << ot) - 1);
1439    tcg_gen_shl_tl(mask, tcg_constant_tl(1), s->T1);
1440    return mask;
1441}
1442
1443/* Expects truncated bit index in COUNT, 1 << COUNT in MASK.  */
1444static void gen_bt_flags(DisasContext *s, X86DecodedInsn *decode, TCGv src,
1445                         TCGv count, TCGv mask)
1446{
1447    TCGv cf;
1448
1449    /*
1450     * C is the result of the test, Z is unchanged, and the others
1451     * are all undefined.
1452     */
1453    if (s->cc_op == CC_OP_DYNAMIC || CC_OP_HAS_EFLAGS(s->cc_op)) {
1454        /* Generate EFLAGS and replace the C bit.  */
1455        cf = tcg_temp_new();
1456        tcg_gen_setcond_tl(TCG_COND_TSTNE, cf, src, mask);
1457        prepare_update_cf(decode, s, cf);
1458    } else {
1459        /*
1460         * Z was going to be computed from the non-zero status of CC_DST.
1461         * We can get that same Z value (and the new C value) by leaving
1462         * CC_DST alone, setting CC_SRC, and using a CC_OP_SAR of the
1463         * same width.
1464         */
1465        decode->cc_src = tcg_temp_new();
1466        decode->cc_dst = cpu_cc_dst;
1467        decode->cc_op = CC_OP_SARB + cc_op_size(s->cc_op);
1468        tcg_gen_shr_tl(decode->cc_src, src, count);
1469    }
1470}
1471
1472static void gen_BT(DisasContext *s, X86DecodedInsn *decode)
1473{
1474    TCGv count = s->T1;
1475    TCGv mask;
1476
1477    /*
1478     * Try to ensure that the rhs of the TSTNE condition is a constant (and a
1479     * power of two), as that is more readily available on most TCG backends.
1480     *
1481     * For immediate bit number gen_bt_mask()'s output is already a constant;
1482     * for register bit number, shift the source right and check bit 0.
1483     */
1484    if (decode->e.op2 == X86_TYPE_I) {
1485        mask = gen_bt_mask(s, decode);
1486    } else {
1487        MemOp ot = decode->op[1].ot;
1488
1489        tcg_gen_andi_tl(s->T1, s->T1, (8 << ot) - 1);
1490        tcg_gen_shr_tl(s->T0, s->T0, s->T1);
1491
1492        count = tcg_constant_tl(0);
1493        mask = tcg_constant_tl(1);
1494    }
1495    gen_bt_flags(s, decode, s->T0, count, mask);
1496}
1497
1498static void gen_BTC(DisasContext *s, X86DecodedInsn *decode)
1499{
1500    MemOp ot = decode->op[0].ot;
1501    TCGv old = tcg_temp_new();
1502    TCGv mask = gen_bt_mask(s, decode);
1503
1504    if (s->prefix & PREFIX_LOCK) {
1505        tcg_gen_atomic_fetch_xor_tl(old, s->A0, mask, s->mem_index, ot | MO_LE);
1506    } else {
1507        tcg_gen_mov_tl(old, s->T0);
1508        tcg_gen_xor_tl(s->T0, s->T0, mask);
1509    }
1510
1511    gen_bt_flags(s, decode, old, s->T1, mask);
1512}
1513
1514static void gen_BTR(DisasContext *s, X86DecodedInsn *decode)
1515{
1516    MemOp ot = decode->op[0].ot;
1517    TCGv old = tcg_temp_new();
1518    TCGv mask = gen_bt_mask(s, decode);
1519
1520    if (s->prefix & PREFIX_LOCK) {
1521        TCGv maskc = tcg_temp_new();
1522        tcg_gen_not_tl(maskc, mask);
1523        tcg_gen_atomic_fetch_and_tl(old, s->A0, maskc, s->mem_index, ot | MO_LE);
1524    } else {
1525        tcg_gen_mov_tl(old, s->T0);
1526        tcg_gen_andc_tl(s->T0, s->T0, mask);
1527    }
1528
1529    gen_bt_flags(s, decode, old, s->T1, mask);
1530}
1531
1532static void gen_BTS(DisasContext *s, X86DecodedInsn *decode)
1533{
1534    MemOp ot = decode->op[0].ot;
1535    TCGv old = tcg_temp_new();
1536    TCGv mask = gen_bt_mask(s, decode);
1537
1538    if (s->prefix & PREFIX_LOCK) {
1539        tcg_gen_atomic_fetch_or_tl(old, s->A0, mask, s->mem_index, ot | MO_LE);
1540    } else {
1541        tcg_gen_mov_tl(old, s->T0);
1542        tcg_gen_or_tl(s->T0, s->T0, mask);
1543    }
1544
1545    gen_bt_flags(s, decode, old, s->T1, mask);
1546}
1547
1548static void gen_BZHI(DisasContext *s, X86DecodedInsn *decode)
1549{
1550    MemOp ot = decode->op[0].ot;
1551    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1552    TCGv zero = tcg_constant_tl(0);
1553    TCGv mone = tcg_constant_tl(-1);
1554
1555    tcg_gen_ext8u_tl(s->T1, s->T1);
1556
1557    tcg_gen_shl_tl(s->A0, mone, s->T1);
1558    tcg_gen_movcond_tl(TCG_COND_LEU, s->A0, s->T1, bound, s->A0, zero);
1559    tcg_gen_andc_tl(s->T0, s->T0, s->A0);
1560    /*
1561     * Note that since we're using BMILG (in order to get O
1562     * cleared) we need to store the inverse into C.
1563     */
1564    tcg_gen_setcond_tl(TCG_COND_LEU, s->T1, s->T1, bound);
1565    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1566}
1567
1568static void gen_CALL(DisasContext *s, X86DecodedInsn *decode)
1569{
1570    gen_push_v(s, eip_next_tl(s));
1571    gen_JMP(s, decode);
1572}
1573
1574static void gen_CALL_m(DisasContext *s, X86DecodedInsn *decode)
1575{
1576    gen_push_v(s, eip_next_tl(s));
1577    gen_JMP_m(s, decode);
1578}
1579
1580static void gen_CALLF(DisasContext *s, X86DecodedInsn *decode)
1581{
1582    gen_far_call(s);
1583}
1584
1585static void gen_CALLF_m(DisasContext *s, X86DecodedInsn *decode)
1586{
1587    MemOp ot = decode->op[1].ot;
1588
1589    gen_op_ld_v(s, ot, s->T0, s->A0);
1590    gen_add_A0_im(s, 1 << ot);
1591    gen_op_ld_v(s, MO_16, s->T1, s->A0);
1592    gen_far_call(s);
1593}
1594
1595static void gen_CBW(DisasContext *s, X86DecodedInsn *decode)
1596{
1597    MemOp src_ot = decode->op[0].ot - 1;
1598
1599    tcg_gen_ext_tl(s->T0, s->T0, src_ot | MO_SIGN);
1600}
1601
1602static void gen_CLC(DisasContext *s, X86DecodedInsn *decode)
1603{
1604    gen_compute_eflags(s);
1605    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, ~CC_C);
1606}
1607
1608static void gen_CLD(DisasContext *s, X86DecodedInsn *decode)
1609{
1610    tcg_gen_st_i32(tcg_constant_i32(1), tcg_env, offsetof(CPUX86State, df));
1611}
1612
1613static void gen_CLI(DisasContext *s, X86DecodedInsn *decode)
1614{
1615    gen_reset_eflags(s, IF_MASK);
1616}
1617
1618static void gen_CLTS(DisasContext *s, X86DecodedInsn *decode)
1619{
1620    gen_helper_clts(tcg_env);
1621    /* abort block because static cpu state changed */
1622    s->base.is_jmp = DISAS_EOB_NEXT;
1623}
1624
1625static void gen_CMC(DisasContext *s, X86DecodedInsn *decode)
1626{
1627    gen_compute_eflags(s);
1628    tcg_gen_xori_tl(cpu_cc_src, cpu_cc_src, CC_C);
1629}
1630
1631static void gen_CMOVcc(DisasContext *s, X86DecodedInsn *decode)
1632{
1633    gen_cmovcc(s, decode->b & 0xf, s->T0, s->T1);
1634}
1635
1636static void gen_CMPccXADD(DisasContext *s, X86DecodedInsn *decode)
1637{
1638    TCGLabel *label_top = gen_new_label();
1639    TCGLabel *label_bottom = gen_new_label();
1640    TCGv oldv = tcg_temp_new();
1641    TCGv newv = tcg_temp_new();
1642    TCGv cmpv = tcg_temp_new();
1643    TCGCond cond;
1644
1645    TCGv cmp_lhs, cmp_rhs;
1646    MemOp ot, ot_full;
1647
1648    int jcc_op = (decode->b >> 1) & 7;
1649    static const TCGCond cond_table[8] = {
1650        [JCC_O] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1651        [JCC_B] = TCG_COND_LTU,
1652        [JCC_Z] = TCG_COND_EQ,
1653        [JCC_BE] = TCG_COND_LEU,
1654        [JCC_S] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1655        [JCC_P] = TCG_COND_TSTEQ,  /* even parity - tests low bit of popcount */
1656        [JCC_L] = TCG_COND_LT,
1657        [JCC_LE] = TCG_COND_LE,
1658    };
1659
1660    cond = cond_table[jcc_op];
1661    if (decode->b & 1) {
1662        cond = tcg_invert_cond(cond);
1663    }
1664
1665    ot = decode->op[0].ot;
1666    ot_full = ot | MO_LE;
1667    if (jcc_op >= JCC_S) {
1668        /*
1669         * Sign-extend values before subtracting for S, P (zero/sign extension
1670         * does not matter there) L, LE and their inverses.
1671         */
1672        ot_full |= MO_SIGN;
1673    }
1674
1675    /*
1676     * cmpv will be moved to cc_src *after* cpu_regs[] is written back, so use
1677     * tcg_gen_ext_tl instead of gen_ext_tl.
1678     */
1679    tcg_gen_ext_tl(cmpv, cpu_regs[decode->op[1].n], ot_full);
1680
1681    /*
1682     * Cmpxchg loop starts here.
1683     * - s->T1: addition operand (from decoder)
1684     * - s->A0: dest address (from decoder)
1685     * - s->cc_srcT: memory operand (lhs for comparison)
1686     * - cmpv: rhs for comparison
1687     */
1688    gen_set_label(label_top);
1689    gen_op_ld_v(s, ot_full, s->cc_srcT, s->A0);
1690    tcg_gen_sub_tl(s->T0, s->cc_srcT, cmpv);
1691
1692    /* Compute the comparison result by hand, to avoid clobbering cc_*.  */
1693    switch (jcc_op) {
1694    case JCC_O:
1695        /* (src1 ^ src2) & (src1 ^ dst). newv is only used here for a moment */
1696        tcg_gen_xor_tl(newv, s->cc_srcT, s->T0);
1697        tcg_gen_xor_tl(s->tmp0, s->cc_srcT, cmpv);
1698        tcg_gen_and_tl(s->tmp0, s->tmp0, newv);
1699        tcg_gen_sextract_tl(s->tmp0, s->tmp0, 0, 8 << ot);
1700        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(0);
1701        break;
1702
1703    case JCC_P:
1704        tcg_gen_ext8u_tl(s->tmp0, s->T0);
1705        tcg_gen_ctpop_tl(s->tmp0, s->tmp0);
1706        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(1);
1707        break;
1708
1709    case JCC_S:
1710        tcg_gen_sextract_tl(s->tmp0, s->T0, 0, 8 << ot);
1711        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(0);
1712        break;
1713
1714    default:
1715        cmp_lhs = s->cc_srcT, cmp_rhs = cmpv;
1716        break;
1717    }
1718
1719    /* Compute new value: if condition does not hold, just store back s->cc_srcT */
1720    tcg_gen_add_tl(newv, s->cc_srcT, s->T1);
1721    tcg_gen_movcond_tl(cond, newv, cmp_lhs, cmp_rhs, newv, s->cc_srcT);
1722    tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, s->cc_srcT, newv, s->mem_index, ot_full);
1723
1724    /* Exit unconditionally if cmpxchg succeeded.  */
1725    tcg_gen_brcond_tl(TCG_COND_EQ, oldv, s->cc_srcT, label_bottom);
1726
1727    /* Try again if there was actually a store to make.  */
1728    tcg_gen_brcond_tl(cond, cmp_lhs, cmp_rhs, label_top);
1729    gen_set_label(label_bottom);
1730
1731    /* Store old value to registers only after a successful store.  */
1732    gen_writeback(s, decode, 1, s->cc_srcT);
1733
1734    decode->cc_dst = s->T0;
1735    decode->cc_src = cmpv;
1736    decode->cc_op = CC_OP_SUBB + ot;
1737}
1738
1739static void gen_CMPS(DisasContext *s, X86DecodedInsn *decode)
1740{
1741    MemOp ot = decode->op[2].ot;
1742    gen_repz_nz(s, ot, gen_cmps);
1743}
1744
1745static void gen_CMPXCHG(DisasContext *s, X86DecodedInsn *decode)
1746{
1747    MemOp ot = decode->op[2].ot;
1748    TCGv cmpv = tcg_temp_new();
1749    TCGv oldv = tcg_temp_new();
1750    TCGv newv = tcg_temp_new();
1751    TCGv dest;
1752
1753    tcg_gen_ext_tl(cmpv, cpu_regs[R_EAX], ot);
1754    tcg_gen_ext_tl(newv, s->T1, ot);
1755    if (s->prefix & PREFIX_LOCK) {
1756        tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, cmpv, newv,
1757                                  s->mem_index, ot | MO_LE);
1758    } else {
1759        tcg_gen_ext_tl(oldv, s->T0, ot);
1760        if (decode->op[0].has_ea) {
1761            /*
1762             * Perform an unconditional store cycle like physical cpu;
1763             * must be before changing accumulator to ensure
1764             * idempotency if the store faults and the instruction
1765             * is restarted
1766             */
1767            tcg_gen_movcond_tl(TCG_COND_EQ, newv, oldv, cmpv, newv, oldv);
1768            gen_op_st_v(s, ot, newv, s->A0);
1769        } else {
1770            /*
1771             * Unlike the memory case, where "the destination operand receives
1772             * a write cycle without regard to the result of the comparison",
1773             * rm must not be touched altogether if the write fails, including
1774             * not zero-extending it on 64-bit processors.  So, precompute
1775             * the result of a successful writeback and perform the movcond
1776             * directly on cpu_regs.  In case rm is part of RAX, note that this
1777             * movcond and the one below are mutually exclusive is executed.
1778             */
1779            dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, newv, newv);
1780            tcg_gen_movcond_tl(TCG_COND_EQ, dest, oldv, cmpv, newv, dest);
1781        }
1782        decode->op[0].unit = X86_OP_SKIP;
1783    }
1784
1785    /* Write RAX only if the cmpxchg fails.  */
1786    dest = gen_op_deposit_reg_v(s, ot, R_EAX, s->T0, oldv);
1787    tcg_gen_movcond_tl(TCG_COND_NE, dest, oldv, cmpv, s->T0, dest);
1788
1789    tcg_gen_mov_tl(s->cc_srcT, cmpv);
1790    tcg_gen_sub_tl(cmpv, cmpv, oldv);
1791    decode->cc_dst = cmpv;
1792    decode->cc_src = oldv;
1793    decode->cc_op = CC_OP_SUBB + ot;
1794}
1795
1796static void gen_CMPXCHG16B(DisasContext *s, X86DecodedInsn *decode)
1797{
1798#ifdef TARGET_X86_64
1799    MemOp mop = MO_TE | MO_128 | MO_ALIGN;
1800    TCGv_i64 t0, t1;
1801    TCGv_i128 cmp, val;
1802
1803    cmp = tcg_temp_new_i128();
1804    val = tcg_temp_new_i128();
1805    tcg_gen_concat_i64_i128(cmp, cpu_regs[R_EAX], cpu_regs[R_EDX]);
1806    tcg_gen_concat_i64_i128(val, cpu_regs[R_EBX], cpu_regs[R_ECX]);
1807
1808    /* Only require atomic with LOCK; non-parallel handled in generator. */
1809    if (s->prefix & PREFIX_LOCK) {
1810        tcg_gen_atomic_cmpxchg_i128(val, s->A0, cmp, val, s->mem_index, mop);
1811    } else {
1812        tcg_gen_nonatomic_cmpxchg_i128(val, s->A0, cmp, val, s->mem_index, mop);
1813    }
1814
1815    tcg_gen_extr_i128_i64(s->T0, s->T1, val);
1816
1817    /* Determine success after the fact. */
1818    t0 = tcg_temp_new_i64();
1819    t1 = tcg_temp_new_i64();
1820    tcg_gen_xor_i64(t0, s->T0, cpu_regs[R_EAX]);
1821    tcg_gen_xor_i64(t1, s->T1, cpu_regs[R_EDX]);
1822    tcg_gen_or_i64(t0, t0, t1);
1823
1824    /* Update Z. */
1825    gen_compute_eflags(s);
1826    tcg_gen_setcondi_i64(TCG_COND_EQ, t0, t0, 0);
1827    tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, t0, ctz32(CC_Z), 1);
1828
1829    /*
1830     * Extract the result values for the register pair.  We may do this
1831     * unconditionally, because on success (Z=1), the old value matches
1832     * the previous value in RDX:RAX.
1833     */
1834    tcg_gen_mov_i64(cpu_regs[R_EAX], s->T0);
1835    tcg_gen_mov_i64(cpu_regs[R_EDX], s->T1);
1836#else
1837    abort();
1838#endif
1839}
1840
1841static void gen_CMPXCHG8B(DisasContext *s, X86DecodedInsn *decode)
1842{
1843    TCGv_i64 cmp, val, old;
1844    TCGv Z;
1845
1846    cmp = tcg_temp_new_i64();
1847    val = tcg_temp_new_i64();
1848    old = tcg_temp_new_i64();
1849
1850    /* Construct the comparison values from the register pair. */
1851    tcg_gen_concat_tl_i64(cmp, cpu_regs[R_EAX], cpu_regs[R_EDX]);
1852    tcg_gen_concat_tl_i64(val, cpu_regs[R_EBX], cpu_regs[R_ECX]);
1853
1854    /* Only require atomic with LOCK; non-parallel handled in generator. */
1855    if (s->prefix & PREFIX_LOCK) {
1856        tcg_gen_atomic_cmpxchg_i64(old, s->A0, cmp, val, s->mem_index, MO_TEUQ);
1857    } else {
1858        tcg_gen_nonatomic_cmpxchg_i64(old, s->A0, cmp, val,
1859                                      s->mem_index, MO_TEUQ);
1860    }
1861
1862    /* Set tmp0 to match the required value of Z. */
1863    tcg_gen_setcond_i64(TCG_COND_EQ, cmp, old, cmp);
1864    Z = tcg_temp_new();
1865    tcg_gen_trunc_i64_tl(Z, cmp);
1866
1867    /*
1868     * Extract the result values for the register pair.
1869     * For 32-bit, we may do this unconditionally, because on success (Z=1),
1870     * the old value matches the previous value in EDX:EAX.  For x86_64,
1871     * the store must be conditional, because we must leave the source
1872     * registers unchanged on success, and zero-extend the writeback
1873     * on failure (Z=0).
1874     */
1875    if (TARGET_LONG_BITS == 32) {
1876        tcg_gen_extr_i64_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], old);
1877    } else {
1878        TCGv zero = tcg_constant_tl(0);
1879
1880        tcg_gen_extr_i64_tl(s->T0, s->T1, old);
1881        tcg_gen_movcond_tl(TCG_COND_EQ, cpu_regs[R_EAX], Z, zero,
1882                           s->T0, cpu_regs[R_EAX]);
1883        tcg_gen_movcond_tl(TCG_COND_EQ, cpu_regs[R_EDX], Z, zero,
1884                           s->T1, cpu_regs[R_EDX]);
1885    }
1886
1887    /* Update Z. */
1888    gen_compute_eflags(s);
1889    tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, Z, ctz32(CC_Z), 1);
1890}
1891
1892static void gen_CPUID(DisasContext *s, X86DecodedInsn *decode)
1893{
1894    gen_update_cc_op(s);
1895    gen_update_eip_cur(s);
1896    gen_helper_cpuid(tcg_env);
1897}
1898
1899static void gen_CRC32(DisasContext *s, X86DecodedInsn *decode)
1900{
1901    MemOp ot = decode->op[2].ot;
1902
1903    tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
1904    gen_helper_crc32(s->T0, s->tmp2_i32, s->T1, tcg_constant_i32(8 << ot));
1905}
1906
1907static void gen_CVTPI2Px(DisasContext *s, X86DecodedInsn *decode)
1908{
1909    gen_helper_enter_mmx(tcg_env);
1910    if (s->prefix & PREFIX_DATA) {
1911        gen_helper_cvtpi2pd(tcg_env, OP_PTR0, OP_PTR2);
1912    } else {
1913        gen_helper_cvtpi2ps(tcg_env, OP_PTR0, OP_PTR2);
1914    }
1915}
1916
1917static void gen_CVTPx2PI(DisasContext *s, X86DecodedInsn *decode)
1918{
1919    gen_helper_enter_mmx(tcg_env);
1920    if (s->prefix & PREFIX_DATA) {
1921        gen_helper_cvtpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1922    } else {
1923        gen_helper_cvtps2pi(tcg_env, OP_PTR0, OP_PTR2);
1924    }
1925}
1926
1927static void gen_CVTTPx2PI(DisasContext *s, X86DecodedInsn *decode)
1928{
1929    gen_helper_enter_mmx(tcg_env);
1930    if (s->prefix & PREFIX_DATA) {
1931        gen_helper_cvttpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1932    } else {
1933        gen_helper_cvttps2pi(tcg_env, OP_PTR0, OP_PTR2);
1934    }
1935}
1936
1937static void gen_CWD(DisasContext *s, X86DecodedInsn *decode)
1938{
1939    int shift = 8 << decode->op[0].ot;
1940
1941    tcg_gen_sextract_tl(s->T0, s->T0, shift - 1, 1);
1942}
1943
1944static void gen_DAA(DisasContext *s, X86DecodedInsn *decode)
1945{
1946    gen_update_cc_op(s);
1947    gen_helper_daa(tcg_env);
1948    assume_cc_op(s, CC_OP_EFLAGS);
1949}
1950
1951static void gen_DAS(DisasContext *s, X86DecodedInsn *decode)
1952{
1953    gen_update_cc_op(s);
1954    gen_helper_das(tcg_env);
1955    assume_cc_op(s, CC_OP_EFLAGS);
1956}
1957
1958static void gen_DEC(DisasContext *s, X86DecodedInsn *decode)
1959{
1960    MemOp ot = decode->op[1].ot;
1961
1962    tcg_gen_movi_tl(s->T1, -1);
1963    if (s->prefix & PREFIX_LOCK) {
1964        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1965                                    s->mem_index, ot | MO_LE);
1966    } else {
1967        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1968    }
1969    prepare_update_cc_incdec(decode, s, CC_OP_DECB + ot);
1970}
1971
1972static void gen_DIV(DisasContext *s, X86DecodedInsn *decode)
1973{
1974    MemOp ot = decode->op[1].ot;
1975
1976    switch(ot) {
1977    case MO_8:
1978        gen_helper_divb_AL(tcg_env, s->T0);
1979        break;
1980    case MO_16:
1981        gen_helper_divw_AX(tcg_env, s->T0);
1982        break;
1983    default:
1984    case MO_32:
1985        gen_helper_divl_EAX(tcg_env, s->T0);
1986        break;
1987#ifdef TARGET_X86_64
1988    case MO_64:
1989        gen_helper_divq_EAX(tcg_env, s->T0);
1990        break;
1991#endif
1992    }
1993}
1994
1995static void gen_EMMS(DisasContext *s, X86DecodedInsn *decode)
1996{
1997    gen_helper_emms(tcg_env);
1998}
1999
2000static void gen_ENTER(DisasContext *s, X86DecodedInsn *decode)
2001{
2002   gen_enter(s, decode->op[1].imm, decode->op[2].imm);
2003}
2004
2005static void gen_EXTRQ_i(DisasContext *s, X86DecodedInsn *decode)
2006{
2007    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
2008    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
2009
2010    gen_helper_extrq_i(tcg_env, OP_PTR0, index, length);
2011}
2012
2013static void gen_EXTRQ_r(DisasContext *s, X86DecodedInsn *decode)
2014{
2015    gen_helper_extrq_r(tcg_env, OP_PTR0, OP_PTR2);
2016}
2017
2018static void gen_FXRSTOR(DisasContext *s, X86DecodedInsn *decode)
2019{
2020    if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
2021        gen_NM_exception(s);
2022    } else {
2023        gen_helper_fxrstor(tcg_env, s->A0);
2024    }
2025}
2026
2027static void gen_FXSAVE(DisasContext *s, X86DecodedInsn *decode)
2028{
2029    if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
2030        gen_NM_exception(s);
2031    } else {
2032        gen_helper_fxsave(tcg_env, s->A0);
2033    }
2034}
2035
2036static void gen_HLT(DisasContext *s, X86DecodedInsn *decode)
2037{
2038#ifdef CONFIG_SYSTEM_ONLY
2039    gen_update_cc_op(s);
2040    gen_update_eip_next(s);
2041    gen_helper_hlt(tcg_env);
2042    s->base.is_jmp = DISAS_NORETURN;
2043#endif
2044}
2045
2046static void gen_IDIV(DisasContext *s, X86DecodedInsn *decode)
2047{
2048    MemOp ot = decode->op[1].ot;
2049
2050    switch(ot) {
2051    case MO_8:
2052        gen_helper_idivb_AL(tcg_env, s->T0);
2053        break;
2054    case MO_16:
2055        gen_helper_idivw_AX(tcg_env, s->T0);
2056        break;
2057    default:
2058    case MO_32:
2059        gen_helper_idivl_EAX(tcg_env, s->T0);
2060        break;
2061#ifdef TARGET_X86_64
2062    case MO_64:
2063        gen_helper_idivq_EAX(tcg_env, s->T0);
2064        break;
2065#endif
2066    }
2067}
2068
2069static void gen_IMUL3(DisasContext *s, X86DecodedInsn *decode)
2070{
2071    MemOp ot = decode->op[0].ot;
2072    TCGv cc_src_rhs;
2073
2074    switch (ot) {
2075    case MO_16:
2076        /* s->T0 already sign-extended */
2077        tcg_gen_ext16s_tl(s->T1, s->T1);
2078        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2079        /* Compare the full result to the extension of the truncated result.  */
2080        tcg_gen_ext16s_tl(s->T1, s->T0);
2081        cc_src_rhs = s->T0;
2082        break;
2083
2084    case MO_32:
2085#ifdef TARGET_X86_64
2086        if (TCG_TARGET_REG_BITS == 64) {
2087            /*
2088             * This produces fewer TCG ops, and better code if flags are needed,
2089             * but it requires a 64-bit multiply even if they are not.  Use it
2090             * only if the target has 64-bits registers.
2091             *
2092             * s->T0 is already sign-extended.
2093             */
2094            tcg_gen_ext32s_tl(s->T1, s->T1);
2095            tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2096            /* Compare the full result to the extension of the truncated result.  */
2097            tcg_gen_ext32s_tl(s->T1, s->T0);
2098            cc_src_rhs = s->T0;
2099        } else {
2100            /* Variant that only needs a 32-bit widening multiply.  */
2101            TCGv_i32 hi = tcg_temp_new_i32();
2102            TCGv_i32 lo = tcg_temp_new_i32();
2103            tcg_gen_trunc_tl_i32(lo, s->T0);
2104            tcg_gen_trunc_tl_i32(hi, s->T1);
2105            tcg_gen_muls2_i32(lo, hi, lo, hi);
2106            tcg_gen_extu_i32_tl(s->T0, lo);
2107
2108            cc_src_rhs = tcg_temp_new();
2109            tcg_gen_extu_i32_tl(cc_src_rhs, hi);
2110            /* Compare the high part to the sign bit of the truncated result */
2111            tcg_gen_sari_i32(lo, lo, 31);
2112            tcg_gen_extu_i32_tl(s->T1, lo);
2113        }
2114        break;
2115
2116    case MO_64:
2117#endif
2118        cc_src_rhs = tcg_temp_new();
2119        tcg_gen_muls2_tl(s->T0, cc_src_rhs, s->T0, s->T1);
2120        /* Compare the high part to the sign bit of the truncated result */
2121        tcg_gen_sari_tl(s->T1, s->T0, TARGET_LONG_BITS - 1);
2122        break;
2123
2124    default:
2125        g_assert_not_reached();
2126    }
2127
2128    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
2129    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
2130}
2131
2132static void gen_IMUL(DisasContext *s, X86DecodedInsn *decode)
2133{
2134    MemOp ot = decode->op[1].ot;
2135    TCGv cc_src_rhs;
2136
2137    switch (ot) {
2138    case MO_8:
2139        /* s->T0 already sign-extended */
2140        tcg_gen_ext8s_tl(s->T1, s->T1);
2141        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2142        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2143        /* Compare the full result to the extension of the truncated result.  */
2144        tcg_gen_ext8s_tl(s->T1, s->T0);
2145        cc_src_rhs = s->T0;
2146        break;
2147
2148    case MO_16:
2149        /* s->T0 already sign-extended */
2150        tcg_gen_ext16s_tl(s->T1, s->T1);
2151        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2152        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2153        tcg_gen_shri_tl(s->T1, s->T0, 16);
2154        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
2155        /* Compare the full result to the extension of the truncated result.  */
2156        tcg_gen_ext16s_tl(s->T1, s->T0);
2157        cc_src_rhs = s->T0;
2158        break;
2159
2160    case MO_32:
2161#ifdef TARGET_X86_64
2162        /* s->T0 already sign-extended */
2163        tcg_gen_ext32s_tl(s->T1, s->T1);
2164        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2165        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
2166        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
2167        /* Compare the full result to the extension of the truncated result.  */
2168        tcg_gen_ext32s_tl(s->T1, s->T0);
2169        cc_src_rhs = s->T0;
2170        break;
2171
2172    case MO_64:
2173#endif
2174        tcg_gen_muls2_tl(s->T0, cpu_regs[R_EDX], s->T0, s->T1);
2175        tcg_gen_mov_tl(cpu_regs[R_EAX], s->T0);
2176
2177        /* Compare the high part to the sign bit of the truncated result */
2178        tcg_gen_negsetcondi_tl(TCG_COND_LT, s->T1, s->T0, 0);
2179        cc_src_rhs = cpu_regs[R_EDX];
2180        break;
2181
2182    default:
2183        g_assert_not_reached();
2184    }
2185
2186    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
2187    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
2188}
2189
2190static void gen_IN(DisasContext *s, X86DecodedInsn *decode)
2191{
2192    MemOp ot = decode->op[0].ot;
2193    TCGv_i32 port = tcg_temp_new_i32();
2194
2195    tcg_gen_trunc_tl_i32(port, s->T0);
2196    tcg_gen_ext16u_i32(port, port);
2197    if (!gen_check_io(s, ot, port, SVM_IOIO_TYPE_MASK)) {
2198        return;
2199    }
2200    translator_io_start(&s->base);
2201    gen_helper_in_func(ot, s->T0, port);
2202    gen_writeback(s, decode, 0, s->T0);
2203    gen_bpt_io(s, port, ot);
2204}
2205
2206static void gen_INC(DisasContext *s, X86DecodedInsn *decode)
2207{
2208    MemOp ot = decode->op[1].ot;
2209
2210    tcg_gen_movi_tl(s->T1, 1);
2211    if (s->prefix & PREFIX_LOCK) {
2212        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
2213                                    s->mem_index, ot | MO_LE);
2214    } else {
2215        tcg_gen_add_tl(s->T0, s->T0, s->T1);
2216    }
2217    prepare_update_cc_incdec(decode, s, CC_OP_INCB + ot);
2218}
2219
2220static void gen_INS(DisasContext *s, X86DecodedInsn *decode)
2221{
2222    MemOp ot = decode->op[1].ot;
2223    TCGv_i32 port = tcg_temp_new_i32();
2224
2225    tcg_gen_trunc_tl_i32(port, s->T1);
2226    tcg_gen_ext16u_i32(port, port);
2227    if (!gen_check_io(s, ot, port,
2228                      SVM_IOIO_TYPE_MASK | SVM_IOIO_STR_MASK)) {
2229        return;
2230    }
2231
2232    translator_io_start(&s->base);
2233    gen_repz(s, ot, gen_ins);
2234}
2235
2236static void gen_INSERTQ_i(DisasContext *s, X86DecodedInsn *decode)
2237{
2238    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
2239    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
2240
2241    gen_helper_insertq_i(tcg_env, OP_PTR0, OP_PTR1, index, length);
2242}
2243
2244static void gen_INSERTQ_r(DisasContext *s, X86DecodedInsn *decode)
2245{
2246    gen_helper_insertq_r(tcg_env, OP_PTR0, OP_PTR2);
2247}
2248
2249static void gen_INT(DisasContext *s, X86DecodedInsn *decode)
2250{
2251    gen_interrupt(s, decode->immediate);
2252}
2253
2254static void gen_INT1(DisasContext *s, X86DecodedInsn *decode)
2255{
2256    gen_update_cc_op(s);
2257    gen_update_eip_next(s);
2258    gen_helper_icebp(tcg_env);
2259    s->base.is_jmp = DISAS_NORETURN;
2260}
2261
2262static void gen_INT3(DisasContext *s, X86DecodedInsn *decode)
2263{
2264    gen_interrupt(s, EXCP03_INT3);
2265}
2266
2267static void gen_INTO(DisasContext *s, X86DecodedInsn *decode)
2268{
2269    gen_update_cc_op(s);
2270    gen_update_eip_cur(s);
2271    gen_helper_into(tcg_env, cur_insn_len_i32(s));
2272}
2273
2274static void gen_IRET(DisasContext *s, X86DecodedInsn *decode)
2275{
2276    if (!PE(s) || VM86(s)) {
2277        gen_helper_iret_real(tcg_env, tcg_constant_i32(s->dflag - 1));
2278    } else {
2279        gen_helper_iret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
2280                                  eip_next_i32(s));
2281    }
2282    assume_cc_op(s, CC_OP_EFLAGS);
2283    s->base.is_jmp = DISAS_EOB_ONLY;
2284}
2285
2286static void gen_Jcc(DisasContext *s, X86DecodedInsn *decode)
2287{
2288    TCGLabel *taken = gen_new_label();
2289
2290    gen_bnd_jmp(s);
2291    gen_jcc(s, decode->b & 0xf, taken);
2292    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2293}
2294
2295static void gen_JCXZ(DisasContext *s, X86DecodedInsn *decode)
2296{
2297    TCGLabel *taken = gen_new_label();
2298
2299    gen_update_cc_op(s);
2300    gen_op_jz_ecx(s, taken);
2301    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2302}
2303
2304static void gen_JMP(DisasContext *s, X86DecodedInsn *decode)
2305{
2306    gen_update_cc_op(s);
2307    gen_jmp_rel(s, s->dflag, decode->immediate, 0);
2308}
2309
2310static void gen_JMP_m(DisasContext *s, X86DecodedInsn *decode)
2311{
2312    gen_op_jmp_v(s, s->T0);
2313    gen_bnd_jmp(s);
2314    s->base.is_jmp = DISAS_JUMP;
2315}
2316
2317static void gen_JMPF(DisasContext *s, X86DecodedInsn *decode)
2318{
2319    gen_far_jmp(s);
2320}
2321
2322static void gen_JMPF_m(DisasContext *s, X86DecodedInsn *decode)
2323{
2324    MemOp ot = decode->op[1].ot;
2325
2326    gen_op_ld_v(s, ot, s->T0, s->A0);
2327    gen_add_A0_im(s, 1 << ot);
2328    gen_op_ld_v(s, MO_16, s->T1, s->A0);
2329    gen_far_jmp(s);
2330}
2331
2332static void gen_LAHF(DisasContext *s, X86DecodedInsn *decode)
2333{
2334    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
2335        return gen_illegal_opcode(s);
2336    }
2337    gen_compute_eflags(s);
2338    /* Note: gen_compute_eflags() only gives the condition codes */
2339    tcg_gen_ori_tl(s->T0, cpu_cc_src, 0x02);
2340    tcg_gen_deposit_tl(cpu_regs[R_EAX], cpu_regs[R_EAX], s->T0, 8, 8);
2341}
2342
2343static void gen_LAR(DisasContext *s, X86DecodedInsn *decode)
2344{
2345    MemOp ot = decode->op[0].ot;
2346    TCGv result = tcg_temp_new();
2347    TCGv dest;
2348
2349    gen_compute_eflags(s);
2350    gen_update_cc_op(s);
2351    gen_helper_lar(result, tcg_env, s->T0);
2352
2353    /* Perform writeback here to skip it if ZF=0.  */
2354    decode->op[0].unit = X86_OP_SKIP;
2355    dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, result, result);
2356    tcg_gen_movcond_tl(TCG_COND_TSTNE, dest, cpu_cc_src, tcg_constant_tl(CC_Z),
2357                       result, dest);
2358}
2359
2360static void gen_LDMXCSR(DisasContext *s, X86DecodedInsn *decode)
2361{
2362    tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
2363    gen_helper_ldmxcsr(tcg_env, s->tmp2_i32);
2364}
2365
2366static void gen_lxx_seg(DisasContext *s, X86DecodedInsn *decode, int seg)
2367{
2368    MemOp ot = decode->op[0].ot;
2369
2370    /* Offset already in s->T0.  */
2371    gen_add_A0_im(s, 1 << ot);
2372    gen_op_ld_v(s, MO_16, s->T1, s->A0);
2373
2374    /* load the segment here to handle exceptions properly */
2375    gen_movl_seg(s, seg, s->T1, false);
2376}
2377
2378static void gen_LDS(DisasContext *s, X86DecodedInsn *decode)
2379{
2380    gen_lxx_seg(s, decode, R_DS);
2381}
2382
2383static void gen_LEA(DisasContext *s, X86DecodedInsn *decode)
2384{
2385    TCGv ea = gen_lea_modrm_1(s, decode->mem, false);
2386    gen_lea_v_seg_dest(s, s->aflag, s->T0, ea, -1, -1);
2387}
2388
2389static void gen_LEAVE(DisasContext *s, X86DecodedInsn *decode)
2390{
2391    gen_leave(s);
2392}
2393
2394static void gen_LES(DisasContext *s, X86DecodedInsn *decode)
2395{
2396    gen_lxx_seg(s, decode, R_ES);
2397}
2398
2399static void gen_LFENCE(DisasContext *s, X86DecodedInsn *decode)
2400{
2401    tcg_gen_mb(TCG_MO_LD_LD | TCG_BAR_SC);
2402}
2403
2404static void gen_LFS(DisasContext *s, X86DecodedInsn *decode)
2405{
2406    gen_lxx_seg(s, decode, R_FS);
2407}
2408
2409static void gen_LGS(DisasContext *s, X86DecodedInsn *decode)
2410{
2411    gen_lxx_seg(s, decode, R_GS);
2412}
2413
2414static void gen_LODS(DisasContext *s, X86DecodedInsn *decode)
2415{
2416    MemOp ot = decode->op[1].ot;
2417    gen_repz(s, ot, gen_lods);
2418}
2419
2420static void gen_LOOP(DisasContext *s, X86DecodedInsn *decode)
2421{
2422    TCGLabel *taken = gen_new_label();
2423
2424    gen_update_cc_op(s);
2425    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2426    gen_op_jnz_ecx(s, taken);
2427    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2428}
2429
2430static void gen_LOOPE(DisasContext *s, X86DecodedInsn *decode)
2431{
2432    TCGLabel *taken = gen_new_label();
2433    TCGLabel *not_taken = gen_new_label();
2434
2435    gen_update_cc_op(s);
2436    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2437    gen_op_jz_ecx(s, not_taken);
2438    gen_jcc(s, (JCC_Z << 1), taken); /* jz taken */
2439    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2440}
2441
2442static void gen_LOOPNE(DisasContext *s, X86DecodedInsn *decode)
2443{
2444    TCGLabel *taken = gen_new_label();
2445    TCGLabel *not_taken = gen_new_label();
2446
2447    gen_update_cc_op(s);
2448    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2449    gen_op_jz_ecx(s, not_taken);
2450    gen_jcc(s, (JCC_Z << 1) | 1, taken); /* jnz taken */
2451    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2452}
2453
2454static void gen_LSL(DisasContext *s, X86DecodedInsn *decode)
2455{
2456    MemOp ot = decode->op[0].ot;
2457    TCGv result = tcg_temp_new();
2458    TCGv dest;
2459
2460    gen_compute_eflags(s);
2461    gen_update_cc_op(s);
2462    gen_helper_lsl(result, tcg_env, s->T0);
2463
2464    /* Perform writeback here to skip it if ZF=0.  */
2465    decode->op[0].unit = X86_OP_SKIP;
2466    dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, result, result);
2467    tcg_gen_movcond_tl(TCG_COND_TSTNE, dest, cpu_cc_src, tcg_constant_tl(CC_Z),
2468                       result, dest);
2469}
2470
2471static void gen_LSS(DisasContext *s, X86DecodedInsn *decode)
2472{
2473    gen_lxx_seg(s, decode, R_SS);
2474}
2475
2476static void gen_LZCNT(DisasContext *s, X86DecodedInsn *decode)
2477{
2478    MemOp ot = decode->op[0].ot;
2479
2480    /* C bit (cc_src) is defined related to the input.  */
2481    decode->cc_src = tcg_temp_new();
2482    decode->cc_dst = s->T0;
2483    decode->cc_op = CC_OP_BMILGB + ot;
2484    tcg_gen_mov_tl(decode->cc_src, s->T0);
2485
2486    /*
2487     * Reduce the target_ulong result by the number of zeros that
2488     * we expect to find at the top.
2489     */
2490    tcg_gen_clzi_tl(s->T0, s->T0, TARGET_LONG_BITS);
2491    tcg_gen_subi_tl(s->T0, s->T0, TARGET_LONG_BITS - (8 << ot));
2492}
2493
2494static void gen_MFENCE(DisasContext *s, X86DecodedInsn *decode)
2495{
2496    tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
2497}
2498
2499static void gen_MOV(DisasContext *s, X86DecodedInsn *decode)
2500{
2501    /* nothing to do! */
2502}
2503#define gen_NOP gen_MOV
2504
2505static void gen_MASKMOV(DisasContext *s, X86DecodedInsn *decode)
2506{
2507    gen_lea_v_seg(s, cpu_regs[R_EDI], R_DS, s->override);
2508
2509    if (s->prefix & PREFIX_DATA) {
2510        gen_helper_maskmov_xmm(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2511    } else {
2512        gen_helper_maskmov_mmx(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2513    }
2514}
2515
2516static void gen_MOVBE(DisasContext *s, X86DecodedInsn *decode)
2517{
2518    MemOp ot = decode->op[0].ot;
2519
2520    /* M operand type does not load/store */
2521    if (decode->e.op0 == X86_TYPE_M) {
2522        tcg_gen_qemu_st_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2523    } else {
2524        tcg_gen_qemu_ld_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2525    }
2526}
2527
2528static void gen_MOVD_from(DisasContext *s, X86DecodedInsn *decode)
2529{
2530    MemOp ot = decode->op[2].ot;
2531
2532    switch (ot) {
2533    case MO_32:
2534#ifdef TARGET_X86_64
2535        tcg_gen_ld32u_tl(s->T0, tcg_env, decode->op[2].offset);
2536        break;
2537    case MO_64:
2538#endif
2539        tcg_gen_ld_tl(s->T0, tcg_env, decode->op[2].offset);
2540        break;
2541    default:
2542        abort();
2543    }
2544}
2545
2546static void gen_MOVD_to(DisasContext *s, X86DecodedInsn *decode)
2547{
2548    MemOp ot = decode->op[2].ot;
2549    int vec_len = vector_len(s, decode);
2550    int lo_ofs = vector_elem_offset(&decode->op[0], ot, 0);
2551
2552    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2553
2554    switch (ot) {
2555    case MO_32:
2556#ifdef TARGET_X86_64
2557        tcg_gen_st32_tl(s->T1, tcg_env, lo_ofs);
2558        break;
2559    case MO_64:
2560#endif
2561        tcg_gen_st_tl(s->T1, tcg_env, lo_ofs);
2562        break;
2563    default:
2564        g_assert_not_reached();
2565    }
2566}
2567
2568static void gen_MOVDQ(DisasContext *s, X86DecodedInsn *decode)
2569{
2570    gen_store_sse(s, decode, decode->op[2].offset);
2571}
2572
2573static void gen_MOVMSK(DisasContext *s, X86DecodedInsn *decode)
2574{
2575    typeof(gen_helper_movmskps_ymm) *ps, *pd, *fn;
2576    ps = s->vex_l ? gen_helper_movmskps_ymm : gen_helper_movmskps_xmm;
2577    pd = s->vex_l ? gen_helper_movmskpd_ymm : gen_helper_movmskpd_xmm;
2578    fn = s->prefix & PREFIX_DATA ? pd : ps;
2579    fn(s->tmp2_i32, tcg_env, OP_PTR2);
2580    tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
2581}
2582
2583static void gen_MOVQ(DisasContext *s, X86DecodedInsn *decode)
2584{
2585    int vec_len = vector_len(s, decode);
2586    int lo_ofs = vector_elem_offset(&decode->op[0], MO_64, 0);
2587
2588    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
2589    if (decode->op[0].has_ea) {
2590        tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
2591    } else {
2592        /*
2593         * tcg_gen_gvec_dup_i64(MO_64, op0.offset, 8, vec_len, s->tmp1_64) would
2594         * seem to work, but it does not on big-endian platforms; the cleared parts
2595         * are always at higher addresses, but cross-endian emulation inverts the
2596         * byte order so that the cleared parts need to be at *lower* addresses.
2597         * Because oprsz is 8, we see this here even for SSE; but more in general,
2598         * it disqualifies using oprsz < maxsz to emulate VEX128.
2599         */
2600        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2601        tcg_gen_st_i64(s->tmp1_i64, tcg_env, lo_ofs);
2602    }
2603}
2604
2605static void gen_MOVq_dq(DisasContext *s, X86DecodedInsn *decode)
2606{
2607    gen_helper_enter_mmx(tcg_env);
2608    /* Otherwise the same as any other movq.  */
2609    return gen_MOVQ(s, decode);
2610}
2611
2612static void gen_MOVS(DisasContext *s, X86DecodedInsn *decode)
2613{
2614    MemOp ot = decode->op[2].ot;
2615    gen_repz(s, ot, gen_movs);
2616}
2617
2618static void gen_MUL(DisasContext *s, X86DecodedInsn *decode)
2619{
2620    MemOp ot = decode->op[1].ot;
2621
2622    switch (ot) {
2623    case MO_8:
2624        /* s->T0 already zero-extended */
2625        tcg_gen_ext8u_tl(s->T1, s->T1);
2626        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2627        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2628        tcg_gen_andi_tl(s->T1, s->T0, 0xff00);
2629        decode->cc_dst = s->T0;
2630        decode->cc_src = s->T1;
2631        break;
2632
2633    case MO_16:
2634        /* s->T0 already zero-extended */
2635        tcg_gen_ext16u_tl(s->T1, s->T1);
2636        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2637        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2638        tcg_gen_shri_tl(s->T1, s->T0, 16);
2639        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
2640        decode->cc_dst = s->T0;
2641        decode->cc_src = s->T1;
2642        break;
2643
2644    case MO_32:
2645#ifdef TARGET_X86_64
2646        /* s->T0 already zero-extended */
2647        tcg_gen_ext32u_tl(s->T1, s->T1);
2648        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2649        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
2650        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
2651        decode->cc_dst = cpu_regs[R_EAX];
2652        decode->cc_src = cpu_regs[R_EDX];
2653        break;
2654
2655    case MO_64:
2656#endif
2657        tcg_gen_mulu2_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], s->T0, s->T1);
2658        decode->cc_dst = cpu_regs[R_EAX];
2659        decode->cc_src = cpu_regs[R_EDX];
2660        break;
2661
2662    default:
2663        g_assert_not_reached();
2664    }
2665
2666    decode->cc_op = CC_OP_MULB + ot;
2667}
2668
2669static void gen_MULX(DisasContext *s, X86DecodedInsn *decode)
2670{
2671    MemOp ot = decode->op[0].ot;
2672
2673    /* low part of result in VEX.vvvv, high in MODRM */
2674    switch (ot) {
2675    case MO_32:
2676#ifdef TARGET_X86_64
2677        tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
2678        tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
2679        tcg_gen_mulu2_i32(s->tmp2_i32, s->tmp3_i32,
2680                          s->tmp2_i32, s->tmp3_i32);
2681        tcg_gen_extu_i32_tl(cpu_regs[s->vex_v], s->tmp2_i32);
2682        tcg_gen_extu_i32_tl(s->T0, s->tmp3_i32);
2683        break;
2684
2685    case MO_64:
2686#endif
2687        tcg_gen_mulu2_tl(cpu_regs[s->vex_v], s->T0, s->T0, s->T1);
2688        break;
2689
2690    default:
2691        g_assert_not_reached();
2692    }
2693}
2694
2695static void gen_NEG(DisasContext *s, X86DecodedInsn *decode)
2696{
2697    MemOp ot = decode->op[0].ot;
2698    TCGv oldv = tcg_temp_new();
2699
2700    if (s->prefix & PREFIX_LOCK) {
2701        TCGv newv = tcg_temp_new();
2702        TCGv cmpv = tcg_temp_new();
2703        TCGLabel *label1 = gen_new_label();
2704
2705        gen_set_label(label1);
2706        gen_op_ld_v(s, ot, oldv, s->A0);
2707        tcg_gen_neg_tl(newv, oldv);
2708        tcg_gen_atomic_cmpxchg_tl(cmpv, s->A0, oldv, newv,
2709                                  s->mem_index, ot | MO_LE);
2710        tcg_gen_brcond_tl(TCG_COND_NE, oldv, cmpv, label1);
2711    } else {
2712        tcg_gen_mov_tl(oldv, s->T0);
2713    }
2714    tcg_gen_neg_tl(s->T0, oldv);
2715
2716    decode->cc_dst = s->T0;
2717    decode->cc_src = oldv;
2718    tcg_gen_movi_tl(s->cc_srcT, 0);
2719    decode->cc_op = CC_OP_SUBB + ot;
2720}
2721
2722static void gen_NOT(DisasContext *s, X86DecodedInsn *decode)
2723{
2724    MemOp ot = decode->op[0].ot;
2725
2726    if (s->prefix & PREFIX_LOCK) {
2727        tcg_gen_movi_tl(s->T0, ~0);
2728        tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T0,
2729                                    s->mem_index, ot | MO_LE);
2730    } else {
2731        tcg_gen_not_tl(s->T0, s->T0);
2732    }
2733}
2734
2735static void gen_OR(DisasContext *s, X86DecodedInsn *decode)
2736{
2737    MemOp ot = decode->op[1].ot;
2738
2739    if (s->prefix & PREFIX_LOCK) {
2740        tcg_gen_atomic_or_fetch_tl(s->T0, s->A0, s->T1,
2741                                   s->mem_index, ot | MO_LE);
2742    } else {
2743        tcg_gen_or_tl(s->T0, s->T0, s->T1);
2744    }
2745    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
2746}
2747
2748static void gen_OUT(DisasContext *s, X86DecodedInsn *decode)
2749{
2750    MemOp ot = decode->op[1].ot;
2751    TCGv_i32 port = tcg_temp_new_i32();
2752    TCGv_i32 value = tcg_temp_new_i32();
2753
2754    tcg_gen_trunc_tl_i32(port, s->T1);
2755    tcg_gen_ext16u_i32(port, port);
2756    if (!gen_check_io(s, ot, port, 0)) {
2757        return;
2758    }
2759    tcg_gen_trunc_tl_i32(value, s->T0);
2760    translator_io_start(&s->base);
2761    gen_helper_out_func(ot, port, value);
2762    gen_bpt_io(s, port, ot);
2763}
2764
2765static void gen_OUTS(DisasContext *s, X86DecodedInsn *decode)
2766{
2767    MemOp ot = decode->op[1].ot;
2768    TCGv_i32 port = 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, SVM_IOIO_STR_MASK)) {
2773        return;
2774    }
2775
2776    translator_io_start(&s->base);
2777    gen_repz(s, ot, gen_outs);
2778}
2779
2780static void gen_PALIGNR(DisasContext *s, X86DecodedInsn *decode)
2781{
2782    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2783    if (!(s->prefix & PREFIX_DATA)) {
2784        gen_helper_palignr_mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2785    } else if (!s->vex_l) {
2786        gen_helper_palignr_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2787    } else {
2788        gen_helper_palignr_ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2789    }
2790}
2791
2792static void gen_PANDN(DisasContext *s, X86DecodedInsn *decode)
2793{
2794    int vec_len = vector_len(s, decode);
2795
2796    /* Careful, operand order is reversed!  */
2797    tcg_gen_gvec_andc(MO_64,
2798                      decode->op[0].offset, decode->op[2].offset,
2799                      decode->op[1].offset, vec_len, vec_len);
2800}
2801
2802static void gen_PAUSE(DisasContext *s, X86DecodedInsn *decode)
2803{
2804    gen_update_cc_op(s);
2805    gen_update_eip_next(s);
2806    gen_helper_pause(tcg_env);
2807    s->base.is_jmp = DISAS_NORETURN;
2808}
2809
2810static void gen_PCMPESTRI(DisasContext *s, X86DecodedInsn *decode)
2811{
2812    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2813    gen_helper_pcmpestri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2814    assume_cc_op(s, CC_OP_EFLAGS);
2815}
2816
2817static void gen_PCMPESTRM(DisasContext *s, X86DecodedInsn *decode)
2818{
2819    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2820    gen_helper_pcmpestrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2821    assume_cc_op(s, CC_OP_EFLAGS);
2822    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2823        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2824                             16, 16, 0);
2825    }
2826}
2827
2828static void gen_PCMPISTRI(DisasContext *s, X86DecodedInsn *decode)
2829{
2830    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2831    gen_helper_pcmpistri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2832    assume_cc_op(s, CC_OP_EFLAGS);
2833}
2834
2835static void gen_PCMPISTRM(DisasContext *s, X86DecodedInsn *decode)
2836{
2837    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2838    gen_helper_pcmpistrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2839    assume_cc_op(s, CC_OP_EFLAGS);
2840    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2841        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2842                             16, 16, 0);
2843    }
2844}
2845
2846static void gen_PDEP(DisasContext *s, X86DecodedInsn *decode)
2847{
2848    gen_helper_pdep(s->T0, s->T0, s->T1);
2849}
2850
2851static void gen_PEXT(DisasContext *s, X86DecodedInsn *decode)
2852{
2853    gen_helper_pext(s->T0, s->T0, s->T1);
2854}
2855
2856static inline void gen_pextr(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
2857{
2858    int vec_len = vector_len(s, decode);
2859    int mask = (vec_len >> ot) - 1;
2860    int val = decode->immediate & mask;
2861
2862    switch (ot) {
2863    case MO_8:
2864        tcg_gen_ld8u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2865        break;
2866    case MO_16:
2867        tcg_gen_ld16u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2868        break;
2869    case MO_32:
2870#ifdef TARGET_X86_64
2871        tcg_gen_ld32u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2872        break;
2873    case MO_64:
2874#endif
2875        tcg_gen_ld_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2876        break;
2877    default:
2878        abort();
2879    }
2880}
2881
2882static void gen_PEXTRB(DisasContext *s, X86DecodedInsn *decode)
2883{
2884    gen_pextr(s, decode, MO_8);
2885}
2886
2887static void gen_PEXTRW(DisasContext *s, X86DecodedInsn *decode)
2888{
2889    gen_pextr(s, decode, MO_16);
2890}
2891
2892static void gen_PEXTR(DisasContext *s, X86DecodedInsn *decode)
2893{
2894    MemOp ot = decode->op[0].ot;
2895    gen_pextr(s, decode, ot);
2896}
2897
2898static inline void gen_pinsr(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
2899{
2900    int vec_len = vector_len(s, decode);
2901    int mask = (vec_len >> ot) - 1;
2902    int val = decode->immediate & mask;
2903
2904    if (decode->op[1].offset != decode->op[0].offset) {
2905        assert(vec_len == 16);
2906        gen_store_sse(s, decode, decode->op[1].offset);
2907    }
2908
2909    switch (ot) {
2910    case MO_8:
2911        tcg_gen_st8_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2912        break;
2913    case MO_16:
2914        tcg_gen_st16_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2915        break;
2916    case MO_32:
2917#ifdef TARGET_X86_64
2918        tcg_gen_st32_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2919        break;
2920    case MO_64:
2921#endif
2922        tcg_gen_st_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2923        break;
2924    default:
2925        abort();
2926    }
2927}
2928
2929static void gen_PINSRB(DisasContext *s, X86DecodedInsn *decode)
2930{
2931    gen_pinsr(s, decode, MO_8);
2932}
2933
2934static void gen_PINSRW(DisasContext *s, X86DecodedInsn *decode)
2935{
2936    gen_pinsr(s, decode, MO_16);
2937}
2938
2939static void gen_PINSR(DisasContext *s, X86DecodedInsn *decode)
2940{
2941    gen_pinsr(s, decode, decode->op[2].ot);
2942}
2943
2944static void gen_pmovmskb_i64(TCGv_i64 d, TCGv_i64 s)
2945{
2946    TCGv_i64 t = tcg_temp_new_i64();
2947
2948    tcg_gen_andi_i64(d, s, 0x8080808080808080ull);
2949
2950    /*
2951     * After each shift+or pair:
2952     * 0:  a.......b.......c.......d.......e.......f.......g.......h.......
2953     * 7:  ab......bc......cd......de......ef......fg......gh......h.......
2954     * 14: abcd....bcde....cdef....defg....efgh....fgh.....gh......h.......
2955     * 28: abcdefghbcdefgh.cdefgh..defgh...efgh....fgh.....gh......h.......
2956     * The result is left in the high bits of the word.
2957     */
2958    tcg_gen_shli_i64(t, d, 7);
2959    tcg_gen_or_i64(d, d, t);
2960    tcg_gen_shli_i64(t, d, 14);
2961    tcg_gen_or_i64(d, d, t);
2962    tcg_gen_shli_i64(t, d, 28);
2963    tcg_gen_or_i64(d, d, t);
2964}
2965
2966static void gen_pmovmskb_vec(unsigned vece, TCGv_vec d, TCGv_vec s)
2967{
2968    TCGv_vec t = tcg_temp_new_vec_matching(d);
2969    TCGv_vec m = tcg_constant_vec_matching(d, MO_8, 0x80);
2970
2971    /* See above */
2972    tcg_gen_and_vec(vece, d, s, m);
2973    tcg_gen_shli_vec(vece, t, d, 7);
2974    tcg_gen_or_vec(vece, d, d, t);
2975    tcg_gen_shli_vec(vece, t, d, 14);
2976    tcg_gen_or_vec(vece, d, d, t);
2977    tcg_gen_shli_vec(vece, t, d, 28);
2978    tcg_gen_or_vec(vece, d, d, t);
2979}
2980
2981static void gen_PMOVMSKB(DisasContext *s, X86DecodedInsn *decode)
2982{
2983    static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
2984    static const GVecGen2 g = {
2985        .fni8 = gen_pmovmskb_i64,
2986        .fniv = gen_pmovmskb_vec,
2987        .opt_opc = vecop_list,
2988        .vece = MO_64,
2989        .prefer_i64 = TCG_TARGET_REG_BITS == 64
2990    };
2991    MemOp ot = decode->op[2].ot;
2992    int vec_len = vector_len(s, decode);
2993    TCGv t = tcg_temp_new();
2994
2995    tcg_gen_gvec_2(offsetof(CPUX86State, xmm_t0) + xmm_offset(ot), decode->op[2].offset,
2996                   vec_len, vec_len, &g);
2997    tcg_gen_ld8u_tl(s->T0, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
2998    while (vec_len > 8) {
2999        vec_len -= 8;
3000        if (tcg_op_supported(INDEX_op_extract2_tl, TCG_TYPE_TL, 0)) {
3001            /*
3002             * Load the next byte of the result into the high byte of T.
3003             * TCG does a similar expansion of deposit to shl+extract2; by
3004             * loading the whole word, the shift left is avoided.
3005             */
3006#ifdef TARGET_X86_64
3007            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_Q((vec_len - 1) / 8)));
3008#else
3009            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L((vec_len - 1) / 4)));
3010#endif
3011
3012            tcg_gen_extract2_tl(s->T0, t, s->T0, TARGET_LONG_BITS - 8);
3013        } else {
3014            /*
3015             * The _previous_ value is deposited into bits 8 and higher of t.  Because
3016             * those bits are known to be zero after ld8u, this becomes a shift+or
3017             * if deposit is not available.
3018             */
3019            tcg_gen_ld8u_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
3020            tcg_gen_deposit_tl(s->T0, t, s->T0, 8, TARGET_LONG_BITS - 8);
3021        }
3022    }
3023}
3024
3025static void gen_POP(DisasContext *s, X86DecodedInsn *decode)
3026{
3027    X86DecodedOp *op = &decode->op[0];
3028    MemOp ot = gen_pop_T0(s);
3029
3030    assert(ot >= op->ot);
3031    if (op->has_ea || op->unit == X86_OP_SEG) {
3032        /* NOTE: order is important for MMU exceptions */
3033        gen_writeback(s, decode, 0, s->T0);
3034    }
3035
3036    /* NOTE: writing back registers after update is important for pop %sp */
3037    gen_pop_update(s, ot);
3038}
3039
3040static void gen_POPA(DisasContext *s, X86DecodedInsn *decode)
3041{
3042    gen_popa(s);
3043}
3044
3045static void gen_POPCNT(DisasContext *s, X86DecodedInsn *decode)
3046{
3047    decode->cc_dst = tcg_temp_new();
3048    decode->cc_op = CC_OP_POPCNT;
3049
3050    tcg_gen_mov_tl(decode->cc_dst, s->T0);
3051    tcg_gen_ctpop_tl(s->T0, s->T0);
3052}
3053
3054static void gen_POPF(DisasContext *s, X86DecodedInsn *decode)
3055{
3056    MemOp ot;
3057    int mask = TF_MASK | AC_MASK | ID_MASK | NT_MASK;
3058
3059    if (CPL(s) == 0) {
3060        mask |= IF_MASK | IOPL_MASK;
3061    } else if (CPL(s) <= IOPL(s)) {
3062        mask |= IF_MASK;
3063    }
3064    if (s->dflag == MO_16) {
3065        mask &= 0xffff;
3066    }
3067
3068    ot = gen_pop_T0(s);
3069    gen_helper_write_eflags(tcg_env, s->T0, tcg_constant_i32(mask));
3070    gen_pop_update(s, ot);
3071    set_cc_op(s, CC_OP_EFLAGS);
3072    /* abort translation because TF/AC flag may change */
3073    s->base.is_jmp = DISAS_EOB_NEXT;
3074}
3075
3076static void gen_PSHUFW(DisasContext *s, X86DecodedInsn *decode)
3077{
3078    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
3079    gen_helper_pshufw_mmx(OP_PTR0, OP_PTR1, imm);
3080}
3081
3082static void gen_PSRLW_i(DisasContext *s, X86DecodedInsn *decode)
3083{
3084    int vec_len = vector_len(s, decode);
3085
3086    if (decode->immediate >= 16) {
3087        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3088    } else {
3089        tcg_gen_gvec_shri(MO_16,
3090                          decode->op[0].offset, decode->op[1].offset,
3091                          decode->immediate, vec_len, vec_len);
3092    }
3093}
3094
3095static void gen_PSLLW_i(DisasContext *s, X86DecodedInsn *decode)
3096{
3097    int vec_len = vector_len(s, decode);
3098
3099    if (decode->immediate >= 16) {
3100        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3101    } else {
3102        tcg_gen_gvec_shli(MO_16,
3103                          decode->op[0].offset, decode->op[1].offset,
3104                          decode->immediate, vec_len, vec_len);
3105    }
3106}
3107
3108static void gen_PSRAW_i(DisasContext *s, X86DecodedInsn *decode)
3109{
3110    int vec_len = vector_len(s, decode);
3111
3112    if (decode->immediate >= 16) {
3113        decode->immediate = 15;
3114    }
3115    tcg_gen_gvec_sari(MO_16,
3116                      decode->op[0].offset, decode->op[1].offset,
3117                      decode->immediate, vec_len, vec_len);
3118}
3119
3120static void gen_PSRLD_i(DisasContext *s, X86DecodedInsn *decode)
3121{
3122    int vec_len = vector_len(s, decode);
3123
3124    if (decode->immediate >= 32) {
3125        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3126    } else {
3127        tcg_gen_gvec_shri(MO_32,
3128                          decode->op[0].offset, decode->op[1].offset,
3129                          decode->immediate, vec_len, vec_len);
3130    }
3131}
3132
3133static void gen_PSLLD_i(DisasContext *s, X86DecodedInsn *decode)
3134{
3135    int vec_len = vector_len(s, decode);
3136
3137    if (decode->immediate >= 32) {
3138        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3139    } else {
3140        tcg_gen_gvec_shli(MO_32,
3141                          decode->op[0].offset, decode->op[1].offset,
3142                          decode->immediate, vec_len, vec_len);
3143    }
3144}
3145
3146static void gen_PSRAD_i(DisasContext *s, X86DecodedInsn *decode)
3147{
3148    int vec_len = vector_len(s, decode);
3149
3150    if (decode->immediate >= 32) {
3151        decode->immediate = 31;
3152    }
3153    tcg_gen_gvec_sari(MO_32,
3154                      decode->op[0].offset, decode->op[1].offset,
3155                      decode->immediate, vec_len, vec_len);
3156}
3157
3158static void gen_PSRLQ_i(DisasContext *s, X86DecodedInsn *decode)
3159{
3160    int vec_len = vector_len(s, decode);
3161
3162    if (decode->immediate >= 64) {
3163        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3164    } else {
3165        tcg_gen_gvec_shri(MO_64,
3166                          decode->op[0].offset, decode->op[1].offset,
3167                          decode->immediate, vec_len, vec_len);
3168    }
3169}
3170
3171static void gen_PSLLQ_i(DisasContext *s, X86DecodedInsn *decode)
3172{
3173    int vec_len = vector_len(s, decode);
3174
3175    if (decode->immediate >= 64) {
3176        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3177    } else {
3178        tcg_gen_gvec_shli(MO_64,
3179                          decode->op[0].offset, decode->op[1].offset,
3180                          decode->immediate, vec_len, vec_len);
3181    }
3182}
3183
3184static TCGv_ptr make_imm8u_xmm_vec(uint8_t imm, int vec_len)
3185{
3186    MemOp ot = vec_len == 16 ? MO_128 : MO_256;
3187    TCGv_i32 imm_v = tcg_constant8u_i32(imm);
3188    TCGv_ptr ptr = tcg_temp_new_ptr();
3189
3190    tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_t0) + xmm_offset(ot),
3191                         vec_len, vec_len, 0);
3192
3193    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_t0));
3194    tcg_gen_st_i32(imm_v, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L(0)));
3195    return ptr;
3196}
3197
3198static void gen_PSRLDQ_i(DisasContext *s, X86DecodedInsn *decode)
3199{
3200    int vec_len = vector_len(s, decode);
3201    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
3202
3203    if (s->vex_l) {
3204        gen_helper_psrldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3205    } else {
3206        gen_helper_psrldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3207    }
3208}
3209
3210static void gen_PSLLDQ_i(DisasContext *s, X86DecodedInsn *decode)
3211{
3212    int vec_len = vector_len(s, decode);
3213    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
3214
3215    if (s->vex_l) {
3216        gen_helper_pslldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3217    } else {
3218        gen_helper_pslldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
3219    }
3220}
3221
3222static void gen_PUSH(DisasContext *s, X86DecodedInsn *decode)
3223{
3224    gen_push_v(s, s->T0);
3225}
3226
3227static void gen_PUSHA(DisasContext *s, X86DecodedInsn *decode)
3228{
3229    gen_pusha(s);
3230}
3231
3232static void gen_PUSHF(DisasContext *s, X86DecodedInsn *decode)
3233{
3234    gen_update_cc_op(s);
3235    gen_helper_read_eflags(s->T0, tcg_env);
3236    gen_push_v(s, s->T0);
3237}
3238
3239static MemOp gen_shift_count(DisasContext *s, X86DecodedInsn *decode,
3240                             bool *can_be_zero, TCGv *count, int unit)
3241{
3242    MemOp ot = decode->op[0].ot;
3243    int mask = (ot <= MO_32 ? 0x1f : 0x3f);
3244
3245    *can_be_zero = false;
3246    switch (unit) {
3247    case X86_OP_INT:
3248        *count = tcg_temp_new();
3249        tcg_gen_andi_tl(*count, cpu_regs[R_ECX], mask);
3250        *can_be_zero = true;
3251        break;
3252
3253    case X86_OP_IMM:
3254        if ((decode->immediate & mask) == 0) {
3255            *count = NULL;
3256            break;
3257        }
3258        *count = tcg_temp_new();
3259        tcg_gen_movi_tl(*count, decode->immediate & mask);
3260        break;
3261
3262    case X86_OP_SKIP:
3263        *count = tcg_temp_new();
3264        tcg_gen_movi_tl(*count, 1);
3265        break;
3266
3267    default:
3268        g_assert_not_reached();
3269    }
3270
3271    return ot;
3272}
3273
3274/*
3275 * Compute existing flags in decode->cc_src, for gen_* functions that wants
3276 * to set the cc_op set to CC_OP_ADCOX.  In particular, this allows rotate
3277 * operations to compute the carry in decode->cc_dst and the overflow in
3278 * decode->cc_src2.
3279 *
3280 * If need_flags is true, decode->cc_dst and decode->cc_src2 are preloaded
3281 * with the value of CF and OF before the instruction, so that it is possible
3282 * to keep the flags unmodified.
3283 *
3284 * Return true if carry could be made available cheaply as a 1-bit value in
3285 * decode->cc_dst (trying a bit harder if want_carry is true).  If false is
3286 * returned, decode->cc_dst is uninitialized and the carry is only available
3287 * as bit 0 of decode->cc_src.
3288 */
3289static bool gen_eflags_adcox(DisasContext *s, X86DecodedInsn *decode, bool want_carry, bool need_flags)
3290{
3291    bool got_cf = false;
3292    bool got_of = false;
3293
3294    decode->cc_dst = tcg_temp_new();
3295    decode->cc_src = tcg_temp_new();
3296    decode->cc_src2 = tcg_temp_new();
3297    decode->cc_op = CC_OP_ADCOX;
3298
3299    /* A lot more cc_ops could be "optimized" to avoid the extracts at
3300     * the end (INC/DEC, BMILG, MUL), but they are all really unlikely
3301     * to be followed by rotations within the same basic block.
3302     */
3303    switch (s->cc_op) {
3304    case CC_OP_ADCOX:
3305        /* No need to compute the full EFLAGS, CF/OF are already isolated.  */
3306        tcg_gen_mov_tl(decode->cc_src, cpu_cc_src);
3307        if (need_flags) {
3308            tcg_gen_mov_tl(decode->cc_src2, cpu_cc_src2);
3309            got_of = true;
3310        }
3311        if (want_carry || need_flags) {
3312            tcg_gen_mov_tl(decode->cc_dst, cpu_cc_dst);
3313            got_cf = true;
3314        }
3315        break;
3316
3317    case CC_OP_LOGICB ... CC_OP_LOGICQ:
3318        /* CF and OF are zero, do it just because it's easy.  */
3319        gen_mov_eflags(s, decode->cc_src);
3320        if (need_flags) {
3321            tcg_gen_movi_tl(decode->cc_src2, 0);
3322            got_of = true;
3323        }
3324        if (want_carry || need_flags) {
3325            tcg_gen_movi_tl(decode->cc_dst, 0);
3326            got_cf = true;
3327        }
3328        break;
3329
3330    case CC_OP_SARB ... CC_OP_SARQ:
3331        /*
3332         * SHR/RCR/SHR/RCR/... is a relatively common occurrence of RCR.
3333         * By computing CF without using eflags, the calls to cc_compute_all
3334         * can be eliminated as dead code (except for the last RCR).
3335         */
3336        if (want_carry || need_flags) {
3337            tcg_gen_andi_tl(decode->cc_dst, cpu_cc_src, 1);
3338            got_cf = true;
3339        }
3340        gen_mov_eflags(s, decode->cc_src);
3341        break;
3342
3343    case CC_OP_SHLB ... CC_OP_SHLQ:
3344        /*
3345         * Likewise for SHL/RCL/SHL/RCL/... but, if CF is not in the sign
3346         * bit, we might as well fish CF out of EFLAGS and save a shift.
3347         */
3348        if (want_carry && (!need_flags || s->cc_op == CC_OP_SHLB + MO_TL)) {
3349            MemOp size = cc_op_size(s->cc_op);
3350            tcg_gen_shri_tl(decode->cc_dst, cpu_cc_src, (8 << size) - 1);
3351            got_cf = true;
3352        }
3353        gen_mov_eflags(s, decode->cc_src);
3354        break;
3355
3356    default:
3357        gen_mov_eflags(s, decode->cc_src);
3358        break;
3359    }
3360
3361    if (need_flags) {
3362        /* If the flags could be left unmodified, always load them.  */
3363        if (!got_of) {
3364            tcg_gen_extract_tl(decode->cc_src2, decode->cc_src, ctz32(CC_O), 1);
3365            got_of = true;
3366        }
3367        if (!got_cf) {
3368            tcg_gen_extract_tl(decode->cc_dst, decode->cc_src, ctz32(CC_C), 1);
3369            got_cf = true;
3370        }
3371    }
3372    return got_cf;
3373}
3374
3375static void gen_rot_overflow(X86DecodedInsn *decode, TCGv result, TCGv old,
3376                             bool can_be_zero, TCGv count)
3377{
3378    MemOp ot = decode->op[0].ot;
3379    TCGv temp = can_be_zero ? tcg_temp_new() : decode->cc_src2;
3380
3381    tcg_gen_xor_tl(temp, old, result);
3382    tcg_gen_extract_tl(temp, temp, (8 << ot) - 1, 1);
3383    if (can_be_zero) {
3384        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src2, count, tcg_constant_tl(0),
3385                           decode->cc_src2, temp);
3386    }
3387}
3388
3389/*
3390 * RCx operations are invariant modulo 8*operand_size+1.  For 8 and 16-bit operands,
3391 * this is less than 0x1f (the mask applied by gen_shift_count) so reduce further.
3392 */
3393static void gen_rotc_mod(MemOp ot, TCGv count)
3394{
3395    TCGv temp;
3396
3397    switch (ot) {
3398    case MO_8:
3399        temp = tcg_temp_new();
3400        tcg_gen_subi_tl(temp, count, 18);
3401        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3402        tcg_gen_subi_tl(temp, count, 9);
3403        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3404        break;
3405
3406    case MO_16:
3407        temp = tcg_temp_new();
3408        tcg_gen_subi_tl(temp, count, 17);
3409        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3410        break;
3411
3412    default:
3413        break;
3414    }
3415}
3416
3417/*
3418 * The idea here is that the bit to the right of the new bit 0 is the
3419 * new carry, and the bit to the right of the old bit 0 is the old carry.
3420 * Just like a regular rotation, the result of the rotation is composed
3421 * from a right shifted part and a left shifted part of s->T0.  The new carry
3422 * is extracted from the right-shifted portion, and the old carry is
3423 * inserted at the end of the left-shifted portion.
3424 *
3425 * Because of the separate shifts involving the carry, gen_RCL and gen_RCR
3426 * mostly operate on count-1.  This also comes in handy when computing
3427 * length - count, because (length-1) - (count-1) can be computed with
3428 * a XOR, and that is commutative unlike subtraction.
3429 */
3430static void gen_RCL(DisasContext *s, X86DecodedInsn *decode)
3431{
3432    bool have_1bit_cin, can_be_zero;
3433    TCGv count;
3434    TCGLabel *zero_label = NULL;
3435    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3436    TCGv low, high, low_count;
3437
3438    if (!count) {
3439        return;
3440    }
3441
3442    low = tcg_temp_new();
3443    high = tcg_temp_new();
3444    low_count = tcg_temp_new();
3445
3446    gen_rotc_mod(ot, count);
3447    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3448    if (can_be_zero) {
3449        zero_label = gen_new_label();
3450        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3451    }
3452
3453    /* Compute high part, including incoming carry.  */
3454    if (!have_1bit_cin || tcg_op_deposit_valid(TCG_TYPE_TL, 1, TARGET_LONG_BITS - 1)) {
3455        /* high = (T0 << 1) | cin */
3456        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3457        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3458    } else {
3459        /* Same as above but without deposit; cin in cc_dst.  */
3460        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3461        tcg_gen_add_tl(high, high, s->T0);
3462    }
3463    tcg_gen_subi_tl(count, count, 1);
3464    tcg_gen_shl_tl(high, high, count);
3465
3466    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3467    tcg_gen_xori_tl(low_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3468    tcg_gen_shr_tl(low, s->T0, low_count);
3469    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3470    tcg_gen_shri_tl(low, low, 1);
3471
3472    /* Compute result and outgoing overflow */
3473    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3474    tcg_gen_or_tl(s->T0, low, high);
3475    gen_rot_overflow(decode, s->T0, decode->cc_src2, false, NULL);
3476
3477    if (zero_label) {
3478        gen_set_label(zero_label);
3479    }
3480}
3481
3482static void gen_RCR(DisasContext *s, X86DecodedInsn *decode)
3483{
3484    bool have_1bit_cin, can_be_zero;
3485    TCGv count;
3486    TCGLabel *zero_label = NULL;
3487    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3488    TCGv low, high, high_count;
3489
3490    if (!count) {
3491        return;
3492    }
3493
3494    low = tcg_temp_new();
3495    high = tcg_temp_new();
3496    high_count = tcg_temp_new();
3497
3498    gen_rotc_mod(ot, count);
3499    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3500    if (can_be_zero) {
3501        zero_label = gen_new_label();
3502        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3503    }
3504
3505    /* Save incoming carry into high, it will be shifted later.  */
3506    if (!have_1bit_cin || tcg_op_deposit_valid(TCG_TYPE_TL, 1, TARGET_LONG_BITS - 1)) {
3507        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3508        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3509    } else {
3510        /* Same as above but without deposit; cin in cc_dst.  */
3511        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3512        tcg_gen_add_tl(high, high, s->T0);
3513    }
3514
3515    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3516    tcg_gen_subi_tl(count, count, 1);
3517    tcg_gen_shr_tl(low, s->T0, count);
3518    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3519    tcg_gen_shri_tl(low, low, 1);
3520
3521    /* Move high part to the right position */
3522    tcg_gen_xori_tl(high_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3523    tcg_gen_shl_tl(high, high, high_count);
3524
3525    /* Compute result and outgoing overflow */
3526    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3527    tcg_gen_or_tl(s->T0, low, high);
3528    gen_rot_overflow(decode, s->T0, decode->cc_src2, false, NULL);
3529
3530    if (zero_label) {
3531        gen_set_label(zero_label);
3532    }
3533}
3534
3535#ifdef CONFIG_USER_ONLY
3536static void gen_unreachable(DisasContext *s, X86DecodedInsn *decode)
3537{
3538    g_assert_not_reached();
3539}
3540#endif
3541
3542#ifndef CONFIG_USER_ONLY
3543static void gen_RDMSR(DisasContext *s, X86DecodedInsn *decode)
3544{
3545    gen_update_cc_op(s);
3546    gen_update_eip_cur(s);
3547    gen_helper_rdmsr(tcg_env);
3548}
3549#else
3550#define gen_RDMSR gen_unreachable
3551#endif
3552
3553static void gen_RDPMC(DisasContext *s, X86DecodedInsn *decode)
3554{
3555    gen_update_cc_op(s);
3556    gen_update_eip_cur(s);
3557    translator_io_start(&s->base);
3558    gen_helper_rdpmc(tcg_env);
3559    s->base.is_jmp = DISAS_NORETURN;
3560}
3561
3562static void gen_RDTSC(DisasContext *s, X86DecodedInsn *decode)
3563{
3564    gen_update_cc_op(s);
3565    gen_update_eip_cur(s);
3566    translator_io_start(&s->base);
3567    gen_helper_rdtsc(tcg_env);
3568}
3569
3570static void gen_RDxxBASE(DisasContext *s, X86DecodedInsn *decode)
3571{
3572    TCGv base = cpu_seg_base[s->modrm & 8 ? R_GS : R_FS];
3573
3574    /* Preserve hflags bits by testing CR4 at runtime.  */
3575    gen_helper_cr4_testbit(tcg_env, tcg_constant_i32(CR4_FSGSBASE_MASK));
3576    tcg_gen_mov_tl(s->T0, base);
3577}
3578
3579static void gen_RET(DisasContext *s, X86DecodedInsn *decode)
3580{
3581    int16_t adjust = decode->e.op1 == X86_TYPE_I ? decode->immediate : 0;
3582
3583    MemOp ot = gen_pop_T0(s);
3584    gen_stack_update(s, adjust + (1 << ot));
3585    gen_op_jmp_v(s, s->T0);
3586    gen_bnd_jmp(s);
3587    s->base.is_jmp = DISAS_JUMP;
3588}
3589
3590static void gen_RETF(DisasContext *s, X86DecodedInsn *decode)
3591{
3592    int16_t adjust = decode->e.op1 == X86_TYPE_I ? decode->immediate : 0;
3593
3594    if (!PE(s) || VM86(s)) {
3595        gen_lea_ss_ofs(s, s->A0, cpu_regs[R_ESP], 0);
3596        /* pop offset */
3597        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3598        /* NOTE: keeping EIP updated is not a problem in case of
3599           exception */
3600        gen_op_jmp_v(s, s->T0);
3601        /* pop selector */
3602        gen_add_A0_im(s, 1 << s->dflag);
3603        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3604        gen_op_movl_seg_real(s, R_CS, s->T0);
3605        /* add stack offset */
3606        gen_stack_update(s, adjust + (2 << s->dflag));
3607    } else {
3608        gen_update_cc_op(s);
3609        gen_update_eip_cur(s);
3610        gen_helper_lret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
3611                                  tcg_constant_i32(adjust));
3612    }
3613    s->base.is_jmp = DISAS_EOB_ONLY;
3614}
3615
3616/*
3617 * Return non-NULL if a 32-bit rotate works, after possibly replicating the input.
3618 * The input has already been zero-extended upon operand decode.
3619 */
3620static TCGv_i32 gen_rot_replicate(MemOp ot, TCGv in)
3621{
3622    TCGv_i32 temp;
3623    switch (ot) {
3624    case MO_8:
3625        temp = tcg_temp_new_i32();
3626        tcg_gen_trunc_tl_i32(temp, in);
3627        tcg_gen_muli_i32(temp, temp, 0x01010101);
3628        return temp;
3629
3630    case MO_16:
3631        temp = tcg_temp_new_i32();
3632        tcg_gen_trunc_tl_i32(temp, in);
3633        tcg_gen_deposit_i32(temp, temp, temp, 16, 16);
3634        return temp;
3635
3636#ifdef TARGET_X86_64
3637    case MO_32:
3638        temp = tcg_temp_new_i32();
3639        tcg_gen_trunc_tl_i32(temp, in);
3640        return temp;
3641#endif
3642
3643    default:
3644        return NULL;
3645    }
3646}
3647
3648static void gen_rot_carry(X86DecodedInsn *decode, TCGv result,
3649                          bool can_be_zero, TCGv count, int bit)
3650{
3651    if (!can_be_zero) {
3652        tcg_gen_extract_tl(decode->cc_dst, result, bit, 1);
3653    } else {
3654        TCGv temp = tcg_temp_new();
3655        tcg_gen_extract_tl(temp, result, bit, 1);
3656        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3657                           decode->cc_dst, temp);
3658    }
3659}
3660
3661static void gen_ROL(DisasContext *s, X86DecodedInsn *decode)
3662{
3663    bool can_be_zero;
3664    TCGv count;
3665    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3666    TCGv_i32 temp32, count32;
3667    TCGv old = tcg_temp_new();
3668
3669    if (!count) {
3670        return;
3671    }
3672
3673    gen_eflags_adcox(s, decode, false, can_be_zero);
3674    tcg_gen_mov_tl(old, s->T0);
3675    temp32 = gen_rot_replicate(ot, s->T0);
3676    if (temp32) {
3677        count32 = tcg_temp_new_i32();
3678        tcg_gen_trunc_tl_i32(count32, count);
3679        tcg_gen_rotl_i32(temp32, temp32, count32);
3680        /* Zero extend to facilitate later optimization.  */
3681        tcg_gen_extu_i32_tl(s->T0, temp32);
3682    } else {
3683        tcg_gen_rotl_tl(s->T0, s->T0, count);
3684    }
3685    gen_rot_carry(decode, s->T0, can_be_zero, count, 0);
3686    gen_rot_overflow(decode, s->T0, old, can_be_zero, count);
3687}
3688
3689static void gen_ROR(DisasContext *s, X86DecodedInsn *decode)
3690{
3691    bool can_be_zero;
3692    TCGv count;
3693    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3694    TCGv_i32 temp32, count32;
3695    TCGv old = tcg_temp_new();
3696
3697    if (!count) {
3698        return;
3699    }
3700
3701    gen_eflags_adcox(s, decode, false, can_be_zero);
3702    tcg_gen_mov_tl(old, s->T0);
3703    temp32 = gen_rot_replicate(ot, s->T0);
3704    if (temp32) {
3705        count32 = tcg_temp_new_i32();
3706        tcg_gen_trunc_tl_i32(count32, count);
3707        tcg_gen_rotr_i32(temp32, temp32, count32);
3708        /* Zero extend to facilitate later optimization.  */
3709        tcg_gen_extu_i32_tl(s->T0, temp32);
3710        gen_rot_carry(decode, s->T0, can_be_zero, count, 31);
3711    } else {
3712        tcg_gen_rotr_tl(s->T0, s->T0, count);
3713        gen_rot_carry(decode, s->T0, can_be_zero, count, TARGET_LONG_BITS - 1);
3714    }
3715    gen_rot_overflow(decode, s->T0, old, can_be_zero, count);
3716}
3717
3718static void gen_RORX(DisasContext *s, X86DecodedInsn *decode)
3719{
3720    MemOp ot = decode->op[0].ot;
3721    int mask = ot == MO_64 ? 63 : 31;
3722    int b = decode->immediate & mask;
3723
3724    switch (ot) {
3725    case MO_32:
3726#ifdef TARGET_X86_64
3727        tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3728        tcg_gen_rotri_i32(s->tmp2_i32, s->tmp2_i32, b);
3729        tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
3730        break;
3731
3732    case MO_64:
3733#endif
3734        tcg_gen_rotri_tl(s->T0, s->T0, b);
3735        break;
3736
3737    default:
3738        g_assert_not_reached();
3739    }
3740}
3741
3742#ifndef CONFIG_USER_ONLY
3743static void gen_RSM(DisasContext *s, X86DecodedInsn *decode)
3744{
3745    gen_helper_rsm(tcg_env);
3746    assume_cc_op(s, CC_OP_EFLAGS);
3747    s->base.is_jmp = DISAS_EOB_ONLY;
3748}
3749#else
3750#define gen_RSM gen_UD
3751#endif
3752
3753static void gen_SAHF(DisasContext *s, X86DecodedInsn *decode)
3754{
3755    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
3756        return gen_illegal_opcode(s);
3757    }
3758    tcg_gen_shri_tl(s->T0, cpu_regs[R_EAX], 8);
3759    gen_compute_eflags(s);
3760    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, CC_O);
3761    tcg_gen_andi_tl(s->T0, s->T0, CC_S | CC_Z | CC_A | CC_P | CC_C);
3762    tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, s->T0);
3763}
3764
3765static void gen_SALC(DisasContext *s, X86DecodedInsn *decode)
3766{
3767    gen_compute_eflags_c(s, s->T0);
3768    tcg_gen_neg_tl(s->T0, s->T0);
3769}
3770
3771static void gen_shift_dynamic_flags(DisasContext *s, X86DecodedInsn *decode, TCGv count, CCOp cc_op)
3772{
3773    TCGv_i32 count32 = tcg_temp_new_i32();
3774    TCGv_i32 old_cc_op;
3775
3776    decode->cc_op = CC_OP_DYNAMIC;
3777    decode->cc_op_dynamic = tcg_temp_new_i32();
3778
3779    assert(decode->cc_dst == s->T0);
3780    if (cc_op_live(s->cc_op) & USES_CC_DST) {
3781        decode->cc_dst = tcg_temp_new();
3782        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3783                           cpu_cc_dst, s->T0);
3784    }
3785
3786    if (cc_op_live(s->cc_op) & USES_CC_SRC) {
3787        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src, count, tcg_constant_tl(0),
3788                           cpu_cc_src, decode->cc_src);
3789    }
3790
3791    tcg_gen_trunc_tl_i32(count32, count);
3792    if (s->cc_op == CC_OP_DYNAMIC) {
3793        old_cc_op = cpu_cc_op;
3794    } else {
3795        old_cc_op = tcg_constant_i32(s->cc_op);
3796    }
3797    tcg_gen_movcond_i32(TCG_COND_EQ, decode->cc_op_dynamic, count32, tcg_constant_i32(0),
3798                        old_cc_op, tcg_constant_i32(cc_op));
3799}
3800
3801static void gen_SAR(DisasContext *s, X86DecodedInsn *decode)
3802{
3803    bool can_be_zero;
3804    TCGv count;
3805    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3806
3807    if (!count) {
3808        return;
3809    }
3810
3811    decode->cc_dst = s->T0;
3812    decode->cc_src = tcg_temp_new();
3813    tcg_gen_subi_tl(decode->cc_src, count, 1);
3814    tcg_gen_sar_tl(decode->cc_src, s->T0, decode->cc_src);
3815    tcg_gen_sar_tl(s->T0, s->T0, count);
3816    if (can_be_zero) {
3817        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3818    } else {
3819        decode->cc_op = CC_OP_SARB + ot;
3820    }
3821}
3822
3823static void gen_SARX(DisasContext *s, X86DecodedInsn *decode)
3824{
3825    MemOp ot = decode->op[0].ot;
3826    int mask;
3827
3828    mask = ot == MO_64 ? 63 : 31;
3829    tcg_gen_andi_tl(s->T1, s->T1, mask);
3830    tcg_gen_sar_tl(s->T0, s->T0, s->T1);
3831}
3832
3833static void gen_SBB(DisasContext *s, X86DecodedInsn *decode)
3834{
3835    MemOp ot = decode->op[0].ot;
3836    TCGv c_in = tcg_temp_new();
3837
3838    gen_compute_eflags_c(s, c_in);
3839    if (s->prefix & PREFIX_LOCK) {
3840        tcg_gen_add_tl(s->T0, s->T1, c_in);
3841        tcg_gen_neg_tl(s->T0, s->T0);
3842        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
3843                                    s->mem_index, ot | MO_LE);
3844    } else {
3845        /*
3846         * TODO: SBB reg, reg could use gen_prepare_eflags_c followed by
3847         * negsetcond, and CC_OP_SUBB as the cc_op.
3848         */
3849        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
3850        tcg_gen_sub_tl(s->T0, s->T0, c_in);
3851    }
3852    prepare_update3_cc(decode, s, CC_OP_SBBB + ot, c_in);
3853}
3854
3855static void gen_SCAS(DisasContext *s, X86DecodedInsn *decode)
3856{
3857    MemOp ot = decode->op[2].ot;
3858    gen_repz_nz(s, ot, gen_scas);
3859}
3860
3861static void gen_SETcc(DisasContext *s, X86DecodedInsn *decode)
3862{
3863    gen_setcc(s, decode->b & 0xf, s->T0);
3864}
3865
3866static void gen_SFENCE(DisasContext *s, X86DecodedInsn *decode)
3867{
3868    tcg_gen_mb(TCG_MO_ST_ST | TCG_BAR_SC);
3869}
3870
3871static void gen_SHA1NEXTE(DisasContext *s, X86DecodedInsn *decode)
3872{
3873    gen_helper_sha1nexte(OP_PTR0, OP_PTR1, OP_PTR2);
3874}
3875
3876static void gen_SHA1MSG1(DisasContext *s, X86DecodedInsn *decode)
3877{
3878    gen_helper_sha1msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3879}
3880
3881static void gen_SHA1MSG2(DisasContext *s, X86DecodedInsn *decode)
3882{
3883    gen_helper_sha1msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3884}
3885
3886static void gen_SHA1RNDS4(DisasContext *s, X86DecodedInsn *decode)
3887{
3888    switch(decode->immediate & 3) {
3889    case 0:
3890        gen_helper_sha1rnds4_f0(OP_PTR0, OP_PTR0, OP_PTR1);
3891        break;
3892    case 1:
3893        gen_helper_sha1rnds4_f1(OP_PTR0, OP_PTR0, OP_PTR1);
3894        break;
3895    case 2:
3896        gen_helper_sha1rnds4_f2(OP_PTR0, OP_PTR0, OP_PTR1);
3897        break;
3898    case 3:
3899        gen_helper_sha1rnds4_f3(OP_PTR0, OP_PTR0, OP_PTR1);
3900        break;
3901    }
3902}
3903
3904static void gen_SHA256MSG1(DisasContext *s, X86DecodedInsn *decode)
3905{
3906    gen_helper_sha256msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3907}
3908
3909static void gen_SHA256MSG2(DisasContext *s, X86DecodedInsn *decode)
3910{
3911    gen_helper_sha256msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3912}
3913
3914static void gen_SHA256RNDS2(DisasContext *s, X86DecodedInsn *decode)
3915{
3916    TCGv_i32 wk0 = tcg_temp_new_i32();
3917    TCGv_i32 wk1 = tcg_temp_new_i32();
3918
3919    tcg_gen_ld_i32(wk0, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(0)));
3920    tcg_gen_ld_i32(wk1, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(1)));
3921
3922    gen_helper_sha256rnds2(OP_PTR0, OP_PTR1, OP_PTR2, wk0, wk1);
3923}
3924
3925static void gen_SHL(DisasContext *s, X86DecodedInsn *decode)
3926{
3927    bool can_be_zero;
3928    TCGv count;
3929    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3930
3931    if (!count) {
3932        return;
3933    }
3934
3935    decode->cc_dst = s->T0;
3936    decode->cc_src = tcg_temp_new();
3937    tcg_gen_subi_tl(decode->cc_src, count, 1);
3938    tcg_gen_shl_tl(decode->cc_src, s->T0, decode->cc_src);
3939    tcg_gen_shl_tl(s->T0, s->T0, count);
3940    if (can_be_zero) {
3941        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
3942    } else {
3943        decode->cc_op = CC_OP_SHLB + ot;
3944    }
3945}
3946
3947static void gen_SHLD(DisasContext *s, X86DecodedInsn *decode)
3948{
3949    bool can_be_zero;
3950    TCGv count;
3951    int unit = decode->e.op3 == X86_TYPE_I ? X86_OP_IMM : X86_OP_INT;
3952    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, unit);
3953
3954    if (!count) {
3955        return;
3956    }
3957
3958    decode->cc_dst = s->T0;
3959    decode->cc_src = s->tmp0;
3960    gen_shiftd_rm_T1(s, ot, false, count);
3961    if (can_be_zero) {
3962        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
3963    } else {
3964        decode->cc_op = CC_OP_SHLB + ot;
3965    }
3966}
3967
3968static void gen_SHLX(DisasContext *s, X86DecodedInsn *decode)
3969{
3970    MemOp ot = decode->op[0].ot;
3971    int mask;
3972
3973    mask = ot == MO_64 ? 63 : 31;
3974    tcg_gen_andi_tl(s->T1, s->T1, mask);
3975    tcg_gen_shl_tl(s->T0, s->T0, s->T1);
3976}
3977
3978static void gen_SHR(DisasContext *s, X86DecodedInsn *decode)
3979{
3980    bool can_be_zero;
3981    TCGv count;
3982    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3983
3984    if (!count) {
3985        return;
3986    }
3987
3988    decode->cc_dst = s->T0;
3989    decode->cc_src = tcg_temp_new();
3990    tcg_gen_subi_tl(decode->cc_src, count, 1);
3991    tcg_gen_shr_tl(decode->cc_src, s->T0, decode->cc_src);
3992    tcg_gen_shr_tl(s->T0, s->T0, count);
3993    if (can_be_zero) {
3994        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3995    } else {
3996        decode->cc_op = CC_OP_SARB + ot;
3997    }
3998}
3999
4000static void gen_SHRD(DisasContext *s, X86DecodedInsn *decode)
4001{
4002    bool can_be_zero;
4003    TCGv count;
4004    int unit = decode->e.op3 == X86_TYPE_I ? X86_OP_IMM : X86_OP_INT;
4005    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, unit);
4006
4007    if (!count) {
4008        return;
4009    }
4010
4011    decode->cc_dst = s->T0;
4012    decode->cc_src = s->tmp0;
4013    gen_shiftd_rm_T1(s, ot, true, count);
4014    if (can_be_zero) {
4015        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
4016    } else {
4017        decode->cc_op = CC_OP_SARB + ot;
4018    }
4019}
4020
4021static void gen_SHRX(DisasContext *s, X86DecodedInsn *decode)
4022{
4023    MemOp ot = decode->op[0].ot;
4024    int mask;
4025
4026    mask = ot == MO_64 ? 63 : 31;
4027    tcg_gen_andi_tl(s->T1, s->T1, mask);
4028    tcg_gen_shr_tl(s->T0, s->T0, s->T1);
4029}
4030
4031static void gen_STC(DisasContext *s, X86DecodedInsn *decode)
4032{
4033    gen_compute_eflags(s);
4034    tcg_gen_ori_tl(cpu_cc_src, cpu_cc_src, CC_C);
4035}
4036
4037static void gen_STD(DisasContext *s, X86DecodedInsn *decode)
4038{
4039    tcg_gen_st_i32(tcg_constant_i32(-1), tcg_env, offsetof(CPUX86State, df));
4040}
4041
4042static void gen_STI(DisasContext *s, X86DecodedInsn *decode)
4043{
4044    gen_set_eflags(s, IF_MASK);
4045    s->base.is_jmp = DISAS_EOB_INHIBIT_IRQ;
4046}
4047
4048static void gen_VAESKEYGEN(DisasContext *s, X86DecodedInsn *decode)
4049{
4050    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4051    assert(!s->vex_l);
4052    gen_helper_aeskeygenassist_xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
4053}
4054
4055static void gen_STMXCSR(DisasContext *s, X86DecodedInsn *decode)
4056{
4057    gen_helper_update_mxcsr(tcg_env);
4058    tcg_gen_ld32u_tl(s->T0, tcg_env, offsetof(CPUX86State, mxcsr));
4059}
4060
4061static void gen_STOS(DisasContext *s, X86DecodedInsn *decode)
4062{
4063    MemOp ot = decode->op[1].ot;
4064    gen_repz(s, ot, gen_stos);
4065}
4066
4067static void gen_SUB(DisasContext *s, X86DecodedInsn *decode)
4068{
4069    MemOp ot = decode->op[1].ot;
4070
4071    if (s->prefix & PREFIX_LOCK) {
4072        tcg_gen_neg_tl(s->T0, s->T1);
4073        tcg_gen_atomic_fetch_add_tl(s->cc_srcT, s->A0, s->T0,
4074                                    s->mem_index, ot | MO_LE);
4075        tcg_gen_sub_tl(s->T0, s->cc_srcT, s->T1);
4076    } else {
4077        tcg_gen_mov_tl(s->cc_srcT, s->T0);
4078        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
4079    }
4080    prepare_update2_cc(decode, s, CC_OP_SUBB + ot);
4081}
4082
4083static void gen_SYSCALL(DisasContext *s, X86DecodedInsn *decode)
4084{
4085    gen_update_cc_op(s);
4086    gen_update_eip_cur(s);
4087    gen_helper_syscall(tcg_env, cur_insn_len_i32(s));
4088    if (LMA(s)) {
4089        assume_cc_op(s, CC_OP_EFLAGS);
4090    }
4091
4092    /*
4093     * TF handling for the syscall insn is different. The TF bit is checked
4094     * after the syscall insn completes. This allows #DB to not be
4095     * generated after one has entered CPL0 if TF is set in FMASK.
4096     */
4097    s->base.is_jmp = DISAS_EOB_RECHECK_TF;
4098}
4099
4100static void gen_SYSENTER(DisasContext *s, X86DecodedInsn *decode)
4101{
4102    gen_helper_sysenter(tcg_env);
4103    s->base.is_jmp = DISAS_EOB_ONLY;
4104}
4105
4106static void gen_SYSEXIT(DisasContext *s, X86DecodedInsn *decode)
4107{
4108    gen_helper_sysexit(tcg_env, tcg_constant_i32(s->dflag - 1));
4109    s->base.is_jmp = DISAS_EOB_ONLY;
4110}
4111
4112static void gen_SYSRET(DisasContext *s, X86DecodedInsn *decode)
4113{
4114    gen_helper_sysret(tcg_env, tcg_constant_i32(s->dflag - 1));
4115    if (LMA(s)) {
4116        assume_cc_op(s, CC_OP_EFLAGS);
4117    }
4118
4119    /*
4120     * TF handling for the sysret insn is different. The TF bit is checked
4121     * after the sysret insn completes. This allows #DB to be
4122     * generated "as if" the syscall insn in userspace has just
4123     * completed.
4124     */
4125    s->base.is_jmp = DISAS_EOB_RECHECK_TF;
4126}
4127
4128static void gen_TZCNT(DisasContext *s, X86DecodedInsn *decode)
4129{
4130    MemOp ot = decode->op[0].ot;
4131
4132    /* C bit (cc_src) is defined related to the input.  */
4133    decode->cc_src = tcg_temp_new();
4134    decode->cc_dst = s->T0;
4135    decode->cc_op = CC_OP_BMILGB + ot;
4136    tcg_gen_mov_tl(decode->cc_src, s->T0);
4137
4138    /* A zero input returns the operand size.  */
4139    tcg_gen_ctzi_tl(s->T0, s->T0, 8 << ot);
4140}
4141
4142static void gen_UD(DisasContext *s, X86DecodedInsn *decode)
4143{
4144    gen_illegal_opcode(s);
4145}
4146
4147static void gen_VAESIMC(DisasContext *s, X86DecodedInsn *decode)
4148{
4149    assert(!s->vex_l);
4150    gen_helper_aesimc_xmm(tcg_env, OP_PTR0, OP_PTR2);
4151}
4152
4153/*
4154 * 00 = v*ps Vps, Hps, Wpd
4155 * 66 = v*pd Vpd, Hpd, Wps
4156 * f3 = v*ss Vss, Hss, Wps
4157 * f2 = v*sd Vsd, Hsd, Wps
4158 */
4159#define SSE_CMP(x) { \
4160    gen_helper_ ## x ## ps ## _xmm, gen_helper_ ## x ## pd ## _xmm, \
4161    gen_helper_ ## x ## ss, gen_helper_ ## x ## sd, \
4162    gen_helper_ ## x ## ps ## _ymm, gen_helper_ ## x ## pd ## _ymm}
4163static const SSEFunc_0_eppp gen_helper_cmp_funcs[32][6] = {
4164    SSE_CMP(cmpeq),
4165    SSE_CMP(cmplt),
4166    SSE_CMP(cmple),
4167    SSE_CMP(cmpunord),
4168    SSE_CMP(cmpneq),
4169    SSE_CMP(cmpnlt),
4170    SSE_CMP(cmpnle),
4171    SSE_CMP(cmpord),
4172
4173    SSE_CMP(cmpequ),
4174    SSE_CMP(cmpnge),
4175    SSE_CMP(cmpngt),
4176    SSE_CMP(cmpfalse),
4177    SSE_CMP(cmpnequ),
4178    SSE_CMP(cmpge),
4179    SSE_CMP(cmpgt),
4180    SSE_CMP(cmptrue),
4181
4182    SSE_CMP(cmpeqs),
4183    SSE_CMP(cmpltq),
4184    SSE_CMP(cmpleq),
4185    SSE_CMP(cmpunords),
4186    SSE_CMP(cmpneqq),
4187    SSE_CMP(cmpnltq),
4188    SSE_CMP(cmpnleq),
4189    SSE_CMP(cmpords),
4190
4191    SSE_CMP(cmpequs),
4192    SSE_CMP(cmpngeq),
4193    SSE_CMP(cmpngtq),
4194    SSE_CMP(cmpfalses),
4195    SSE_CMP(cmpnequs),
4196    SSE_CMP(cmpgeq),
4197    SSE_CMP(cmpgtq),
4198    SSE_CMP(cmptrues),
4199};
4200#undef SSE_CMP
4201
4202static void gen_VCMP(DisasContext *s, X86DecodedInsn *decode)
4203{
4204    int index = decode->immediate & (s->prefix & PREFIX_VEX ? 31 : 7);
4205    int b =
4206        s->prefix & PREFIX_REPZ  ? 2 /* ss */ :
4207        s->prefix & PREFIX_REPNZ ? 3 /* sd */ :
4208        !!(s->prefix & PREFIX_DATA) /* pd */ + (s->vex_l << 2);
4209
4210    gen_helper_cmp_funcs[index][b](tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4211}
4212
4213static void gen_VCOMI(DisasContext *s, X86DecodedInsn *decode)
4214{
4215    SSEFunc_0_epp fn;
4216    fn = s->prefix & PREFIX_DATA ? gen_helper_comisd : gen_helper_comiss;
4217    fn(tcg_env, OP_PTR1, OP_PTR2);
4218    assume_cc_op(s, CC_OP_EFLAGS);
4219}
4220
4221static void gen_VCVTPD2PS(DisasContext *s, X86DecodedInsn *decode)
4222{
4223    if (s->vex_l) {
4224        gen_helper_cvtpd2ps_ymm(tcg_env, OP_PTR0, OP_PTR2);
4225    } else {
4226        gen_helper_cvtpd2ps_xmm(tcg_env, OP_PTR0, OP_PTR2);
4227    }
4228}
4229
4230static void gen_VCVTPS2PD(DisasContext *s, X86DecodedInsn *decode)
4231{
4232    if (s->vex_l) {
4233        gen_helper_cvtps2pd_ymm(tcg_env, OP_PTR0, OP_PTR2);
4234    } else {
4235        gen_helper_cvtps2pd_xmm(tcg_env, OP_PTR0, OP_PTR2);
4236    }
4237}
4238
4239static void gen_VCVTPS2PH(DisasContext *s, X86DecodedInsn *decode)
4240{
4241    gen_unary_imm_fp_sse(s, decode,
4242                      gen_helper_cvtps2ph_xmm,
4243                      gen_helper_cvtps2ph_ymm);
4244    /*
4245     * VCVTPS2PH is the only instruction that performs an operation on a
4246     * register source and then *stores* into memory.
4247     */
4248    if (decode->op[0].has_ea) {
4249        gen_store_sse(s, decode, decode->op[0].offset);
4250    }
4251}
4252
4253static void gen_VCVTSD2SS(DisasContext *s, X86DecodedInsn *decode)
4254{
4255    gen_helper_cvtsd2ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4256}
4257
4258static void gen_VCVTSS2SD(DisasContext *s, X86DecodedInsn *decode)
4259{
4260    gen_helper_cvtss2sd(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4261}
4262
4263static void gen_VCVTSI2Sx(DisasContext *s, X86DecodedInsn *decode)
4264{
4265    int vec_len = vector_len(s, decode);
4266    TCGv_i32 in;
4267
4268    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4269
4270#ifdef TARGET_X86_64
4271    MemOp ot = decode->op[2].ot;
4272    if (ot == MO_64) {
4273        if (s->prefix & PREFIX_REPNZ) {
4274            gen_helper_cvtsq2sd(tcg_env, OP_PTR0, s->T1);
4275        } else {
4276            gen_helper_cvtsq2ss(tcg_env, OP_PTR0, s->T1);
4277        }
4278        return;
4279    }
4280    in = s->tmp2_i32;
4281    tcg_gen_trunc_tl_i32(in, s->T1);
4282#else
4283    in = s->T1;
4284#endif
4285
4286    if (s->prefix & PREFIX_REPNZ) {
4287        gen_helper_cvtsi2sd(tcg_env, OP_PTR0, in);
4288    } else {
4289        gen_helper_cvtsi2ss(tcg_env, OP_PTR0, in);
4290    }
4291}
4292
4293static inline void gen_VCVTtSx2SI(DisasContext *s, X86DecodedInsn *decode,
4294                                  SSEFunc_i_ep ss2si, SSEFunc_l_ep ss2sq,
4295                                  SSEFunc_i_ep sd2si, SSEFunc_l_ep sd2sq)
4296{
4297    TCGv_i32 out;
4298
4299#ifdef TARGET_X86_64
4300    MemOp ot = decode->op[0].ot;
4301    if (ot == MO_64) {
4302        if (s->prefix & PREFIX_REPNZ) {
4303            sd2sq(s->T0, tcg_env, OP_PTR2);
4304        } else {
4305            ss2sq(s->T0, tcg_env, OP_PTR2);
4306        }
4307        return;
4308    }
4309
4310    out = s->tmp2_i32;
4311#else
4312    out = s->T0;
4313#endif
4314    if (s->prefix & PREFIX_REPNZ) {
4315        sd2si(out, tcg_env, OP_PTR2);
4316    } else {
4317        ss2si(out, tcg_env, OP_PTR2);
4318    }
4319#ifdef TARGET_X86_64
4320    tcg_gen_extu_i32_tl(s->T0, out);
4321#endif
4322}
4323
4324#ifndef TARGET_X86_64
4325#define gen_helper_cvtss2sq NULL
4326#define gen_helper_cvtsd2sq NULL
4327#define gen_helper_cvttss2sq NULL
4328#define gen_helper_cvttsd2sq NULL
4329#endif
4330
4331static void gen_VCVTSx2SI(DisasContext *s, X86DecodedInsn *decode)
4332{
4333    gen_VCVTtSx2SI(s, decode,
4334                   gen_helper_cvtss2si, gen_helper_cvtss2sq,
4335                   gen_helper_cvtsd2si, gen_helper_cvtsd2sq);
4336}
4337
4338static void gen_VCVTTSx2SI(DisasContext *s, X86DecodedInsn *decode)
4339{
4340    gen_VCVTtSx2SI(s, decode,
4341                   gen_helper_cvttss2si, gen_helper_cvttss2sq,
4342                   gen_helper_cvttsd2si, gen_helper_cvttsd2sq);
4343}
4344
4345static void gen_VEXTRACTx128(DisasContext *s, X86DecodedInsn *decode)
4346{
4347    int mask = decode->immediate & 1;
4348    int src_ofs = vector_elem_offset(&decode->op[1], MO_128, mask);
4349    if (decode->op[0].has_ea) {
4350        /* VEX-only instruction, no alignment requirements.  */
4351        gen_sto_env_A0(s, src_ofs, false);
4352    } else {
4353        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, 16, 16);
4354    }
4355}
4356
4357static void gen_VEXTRACTPS(DisasContext *s, X86DecodedInsn *decode)
4358{
4359    gen_pextr(s, decode, MO_32);
4360}
4361
4362static void gen_vinsertps(DisasContext *s, X86DecodedInsn *decode)
4363{
4364    int val = decode->immediate;
4365    int dest_word = (val >> 4) & 3;
4366    int new_mask = (val & 15) | (1 << dest_word);
4367    int vec_len = 16;
4368
4369    assert(!s->vex_l);
4370
4371    if (new_mask == 15) {
4372        /* All zeroes except possibly for the inserted element */
4373        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
4374    } else if (decode->op[1].offset != decode->op[0].offset) {
4375        gen_store_sse(s, decode, decode->op[1].offset);
4376    }
4377
4378    if (new_mask != (val & 15)) {
4379        tcg_gen_st_i32(s->tmp2_i32, tcg_env,
4380                       vector_elem_offset(&decode->op[0], MO_32, dest_word));
4381    }
4382
4383    if (new_mask != 15) {
4384        TCGv_i32 zero = tcg_constant_i32(0); /* float32_zero */
4385        int i;
4386        for (i = 0; i < 4; i++) {
4387            if ((val >> i) & 1) {
4388                tcg_gen_st_i32(zero, tcg_env,
4389                               vector_elem_offset(&decode->op[0], MO_32, i));
4390            }
4391        }
4392    }
4393}
4394
4395static void gen_VINSERTPS_r(DisasContext *s, X86DecodedInsn *decode)
4396{
4397    int val = decode->immediate;
4398    tcg_gen_ld_i32(s->tmp2_i32, tcg_env,
4399                   vector_elem_offset(&decode->op[2], MO_32, (val >> 6) & 3));
4400    gen_vinsertps(s, decode);
4401}
4402
4403static void gen_VINSERTPS_m(DisasContext *s, X86DecodedInsn *decode)
4404{
4405    tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4406    gen_vinsertps(s, decode);
4407}
4408
4409static void gen_VINSERTx128(DisasContext *s, X86DecodedInsn *decode)
4410{
4411    int mask = decode->immediate & 1;
4412    tcg_gen_gvec_mov(MO_64,
4413                     decode->op[0].offset + offsetof(YMMReg, YMM_X(mask)),
4414                     decode->op[2].offset + offsetof(YMMReg, YMM_X(0)), 16, 16);
4415    tcg_gen_gvec_mov(MO_64,
4416                     decode->op[0].offset + offsetof(YMMReg, YMM_X(!mask)),
4417                     decode->op[1].offset + offsetof(YMMReg, YMM_X(!mask)), 16, 16);
4418}
4419
4420static inline void gen_maskmov(DisasContext *s, X86DecodedInsn *decode,
4421                               SSEFunc_0_eppt xmm, SSEFunc_0_eppt ymm)
4422{
4423    if (!s->vex_l) {
4424        xmm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
4425    } else {
4426        ymm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
4427    }
4428}
4429
4430static void gen_VMASKMOVPD_st(DisasContext *s, X86DecodedInsn *decode)
4431{
4432    gen_maskmov(s, decode, gen_helper_vpmaskmovq_st_xmm, gen_helper_vpmaskmovq_st_ymm);
4433}
4434
4435static void gen_VMASKMOVPS_st(DisasContext *s, X86DecodedInsn *decode)
4436{
4437    gen_maskmov(s, decode, gen_helper_vpmaskmovd_st_xmm, gen_helper_vpmaskmovd_st_ymm);
4438}
4439
4440static void gen_VMOVHPx_ld(DisasContext *s, X86DecodedInsn *decode)
4441{
4442    gen_ldq_env_A0(s, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4443    if (decode->op[0].offset != decode->op[1].offset) {
4444        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4445        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4446    }
4447}
4448
4449static void gen_VMOVHPx_st(DisasContext *s, X86DecodedInsn *decode)
4450{
4451    gen_stq_env_A0(s, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4452}
4453
4454static void gen_VMOVHPx(DisasContext *s, X86DecodedInsn *decode)
4455{
4456    if (decode->op[0].offset != decode->op[2].offset) {
4457        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4458        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4459    }
4460    if (decode->op[0].offset != decode->op[1].offset) {
4461        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4462        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4463    }
4464}
4465
4466static void gen_VMOVHLPS(DisasContext *s, X86DecodedInsn *decode)
4467{
4468    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4469    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4470    if (decode->op[0].offset != decode->op[1].offset) {
4471        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(1)));
4472        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4473    }
4474}
4475
4476static void gen_VMOVLHPS(DisasContext *s, X86DecodedInsn *decode)
4477{
4478    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
4479    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4480    if (decode->op[0].offset != decode->op[1].offset) {
4481        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4482        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4483    }
4484}
4485
4486/*
4487 * Note that MOVLPx supports 256-bit operation unlike MOVHLPx, MOVLHPx, MOXHPx.
4488 * Use a gvec move to move everything above the bottom 64 bits.
4489 */
4490
4491static void gen_VMOVLPx(DisasContext *s, X86DecodedInsn *decode)
4492{
4493    int vec_len = vector_len(s, decode);
4494
4495    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(0)));
4496    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4497    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4498}
4499
4500static void gen_VMOVLPx_ld(DisasContext *s, X86DecodedInsn *decode)
4501{
4502    int vec_len = vector_len(s, decode);
4503
4504    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4505    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4506    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
4507}
4508
4509static void gen_VMOVLPx_st(DisasContext *s, X86DecodedInsn *decode)
4510{
4511    tcg_gen_ld_i64(s->tmp1_i64, OP_PTR2, offsetof(ZMMReg, ZMM_Q(0)));
4512    tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4513}
4514
4515static void gen_VMOVSD_ld(DisasContext *s, X86DecodedInsn *decode)
4516{
4517    TCGv_i64 zero = tcg_constant_i64(0);
4518
4519    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4520    tcg_gen_st_i64(zero, OP_PTR0, offsetof(ZMMReg, ZMM_Q(1)));
4521    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
4522}
4523
4524static void gen_VMOVSS(DisasContext *s, X86DecodedInsn *decode)
4525{
4526    int vec_len = vector_len(s, decode);
4527
4528    tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
4529    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4530    tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
4531}
4532
4533static void gen_VMOVSS_ld(DisasContext *s, X86DecodedInsn *decode)
4534{
4535    int vec_len = vector_len(s, decode);
4536
4537    tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4538    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
4539    tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
4540}
4541
4542static void gen_VMOVSS_st(DisasContext *s, X86DecodedInsn *decode)
4543{
4544    tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
4545    tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4546}
4547
4548static void gen_VPMASKMOV_st(DisasContext *s, X86DecodedInsn *decode)
4549{
4550    if (s->vex_w) {
4551        gen_VMASKMOVPD_st(s, decode);
4552    } else {
4553        gen_VMASKMOVPS_st(s, decode);
4554    }
4555}
4556
4557static void gen_VPERMD(DisasContext *s, X86DecodedInsn *decode)
4558{
4559    assert(s->vex_l);
4560    gen_helper_vpermd_ymm(OP_PTR0, OP_PTR1, OP_PTR2);
4561}
4562
4563static void gen_VPERM2x128(DisasContext *s, X86DecodedInsn *decode)
4564{
4565    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4566    assert(s->vex_l);
4567    gen_helper_vpermdq_ymm(OP_PTR0, OP_PTR1, OP_PTR2, imm);
4568}
4569
4570static void gen_VPHMINPOSUW(DisasContext *s, X86DecodedInsn *decode)
4571{
4572    assert(!s->vex_l);
4573    gen_helper_phminposuw_xmm(tcg_env, OP_PTR0, OP_PTR2);
4574}
4575
4576static void gen_VROUNDSD(DisasContext *s, X86DecodedInsn *decode)
4577{
4578    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4579    assert(!s->vex_l);
4580    gen_helper_roundsd_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
4581}
4582
4583static void gen_VROUNDSS(DisasContext *s, X86DecodedInsn *decode)
4584{
4585    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4586    assert(!s->vex_l);
4587    gen_helper_roundss_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
4588}
4589
4590static void gen_VSHUF(DisasContext *s, X86DecodedInsn *decode)
4591{
4592    TCGv_i32 imm = tcg_constant_i32(decode->immediate);
4593    SSEFunc_0_pppi ps, pd, fn;
4594    ps = s->vex_l ? gen_helper_shufps_ymm : gen_helper_shufps_xmm;
4595    pd = s->vex_l ? gen_helper_shufpd_ymm : gen_helper_shufpd_xmm;
4596    fn = s->prefix & PREFIX_DATA ? pd : ps;
4597    fn(OP_PTR0, OP_PTR1, OP_PTR2, imm);
4598}
4599
4600static void gen_VUCOMI(DisasContext *s, X86DecodedInsn *decode)
4601{
4602    SSEFunc_0_epp fn;
4603    fn = s->prefix & PREFIX_DATA ? gen_helper_ucomisd : gen_helper_ucomiss;
4604    fn(tcg_env, OP_PTR1, OP_PTR2);
4605    assume_cc_op(s, CC_OP_EFLAGS);
4606}
4607
4608static void gen_VZEROALL(DisasContext *s, X86DecodedInsn *decode)
4609{
4610    TCGv_ptr ptr = tcg_temp_new_ptr();
4611
4612    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_regs));
4613    gen_helper_memset(ptr, ptr, tcg_constant_i32(0),
4614                      tcg_constant_ptr(CPU_NB_REGS * sizeof(ZMMReg)));
4615}
4616
4617static void gen_VZEROUPPER(DisasContext *s, X86DecodedInsn *decode)
4618{
4619    int i;
4620
4621    for (i = 0; i < CPU_NB_REGS; i++) {
4622        int offset = offsetof(CPUX86State, xmm_regs[i].ZMM_X(1));
4623        tcg_gen_gvec_dup_imm(MO_64, offset, 16, 16, 0);
4624    }
4625}
4626
4627static void gen_WAIT(DisasContext *s, X86DecodedInsn *decode)
4628{
4629    if ((s->flags & (HF_MP_MASK | HF_TS_MASK)) == (HF_MP_MASK | HF_TS_MASK)) {
4630        gen_NM_exception(s);
4631    } else {
4632        /* needs to be treated as I/O because of ferr_irq */
4633        translator_io_start(&s->base);
4634        gen_helper_fwait(tcg_env);
4635    }
4636}
4637
4638#ifndef CONFIG_USER_ONLY
4639static void gen_WRMSR(DisasContext *s, X86DecodedInsn *decode)
4640{
4641    gen_update_cc_op(s);
4642    gen_update_eip_cur(s);
4643    gen_helper_wrmsr(tcg_env);
4644    s->base.is_jmp = DISAS_EOB_NEXT;
4645}
4646#else
4647#define gen_WRMSR gen_unreachable
4648#endif
4649
4650static void gen_WRxxBASE(DisasContext *s, X86DecodedInsn *decode)
4651{
4652    TCGv base = cpu_seg_base[s->modrm & 8 ? R_GS : R_FS];
4653
4654    /* Preserve hflags bits by testing CR4 at runtime.  */
4655    gen_helper_cr4_testbit(tcg_env, tcg_constant_i32(CR4_FSGSBASE_MASK));
4656    tcg_gen_mov_tl(base, s->T0);
4657}
4658
4659static void gen_XADD(DisasContext *s, X86DecodedInsn *decode)
4660{
4661    MemOp ot = decode->op[1].ot;
4662
4663    decode->cc_dst = tcg_temp_new();
4664    decode->cc_src = s->T1;
4665    decode->cc_op = CC_OP_ADDB + ot;
4666
4667    if (s->prefix & PREFIX_LOCK) {
4668        tcg_gen_atomic_fetch_add_tl(s->T0, s->A0, s->T1, s->mem_index, ot | MO_LE);
4669        tcg_gen_add_tl(decode->cc_dst, s->T0, s->T1);
4670    } else {
4671        tcg_gen_add_tl(decode->cc_dst, s->T0, s->T1);
4672        /*
4673         * NOTE: writing memory first is important for MMU exceptions,
4674         * but "new result" wins for XADD AX, AX.
4675         */
4676        gen_writeback(s, decode, 0, decode->cc_dst);
4677    }
4678    if (decode->op[0].has_ea || decode->op[2].n != decode->op[0].n) {
4679        gen_writeback(s, decode, 2, s->T0);
4680    }
4681}
4682
4683static void gen_XCHG(DisasContext *s, X86DecodedInsn *decode)
4684{
4685    if (s->prefix & PREFIX_LOCK) {
4686        tcg_gen_atomic_xchg_tl(s->T0, s->A0, s->T1,
4687                               s->mem_index, decode->op[0].ot | MO_LE);
4688        /* now store old value into register operand */
4689        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4690    } else {
4691        /* move destination value into source operand, source preserved in T1 */
4692        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4693        tcg_gen_mov_tl(s->T0, s->T1);
4694    }
4695}
4696
4697static void gen_XLAT(DisasContext *s, X86DecodedInsn *decode)
4698{
4699    /* AL is already zero-extended into s->T0.  */
4700    tcg_gen_add_tl(s->A0, cpu_regs[R_EBX], s->T0);
4701    gen_lea_v_seg(s, s->A0, R_DS, s->override);
4702    gen_op_ld_v(s, MO_8, s->T0, s->A0);
4703}
4704
4705static void gen_XOR(DisasContext *s, X86DecodedInsn *decode)
4706{
4707    /* special case XOR reg, reg */
4708    if (decode->op[1].unit == X86_OP_INT &&
4709        decode->op[2].unit == X86_OP_INT &&
4710        decode->op[1].n == decode->op[2].n) {
4711        tcg_gen_movi_tl(s->T0, 0);
4712        decode->cc_op = CC_OP_EFLAGS;
4713        decode->cc_src = tcg_constant_tl(CC_Z | CC_P);
4714    } else {
4715        MemOp ot = decode->op[1].ot;
4716
4717        if (s->prefix & PREFIX_LOCK) {
4718            tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T1,
4719                                        s->mem_index, ot | MO_LE);
4720        } else {
4721            tcg_gen_xor_tl(s->T0, s->T0, s->T1);
4722        }
4723        prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
4724    }
4725}
4726
4727static void gen_XRSTOR(DisasContext *s, X86DecodedInsn *decode)
4728{
4729    TCGv_i64 features = tcg_temp_new_i64();
4730
4731    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4732    gen_helper_xrstor(tcg_env, s->A0, features);
4733    if (s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_MPX) {
4734        /*
4735         * XRSTOR is how MPX is enabled, which changes how
4736         * we translate.  Thus we need to end the TB.
4737         */
4738        s->base.is_jmp = DISAS_EOB_NEXT;
4739    }
4740}
4741
4742static void gen_XSAVE(DisasContext *s, X86DecodedInsn *decode)
4743{
4744    TCGv_i64 features = tcg_temp_new_i64();
4745
4746    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4747    gen_helper_xsave(tcg_env, s->A0, features);
4748}
4749
4750static void gen_XSAVEOPT(DisasContext *s, X86DecodedInsn *decode)
4751{
4752    TCGv_i64 features = tcg_temp_new_i64();
4753
4754    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4755    gen_helper_xsave(tcg_env, s->A0, features);
4756}
4757