xref: /openbmc/qemu/target/i386/tcg/emit.c.inc (revision 4f7b1ecb)
1/*
2 * New-style TCG opcode generator for i386 instructions
3 *
4 *  Copyright (c) 2022 Red Hat, Inc.
5 *
6 * Author: Paolo Bonzini <pbonzini@redhat.com>
7 *
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 */
21
22/*
23 * Sometimes, knowing what the backend has can produce better code.
24 * The exact opcode to check depends on 32- vs. 64-bit.
25 */
26#ifdef TARGET_X86_64
27#define TCG_TARGET_HAS_extract2_tl      TCG_TARGET_HAS_extract2_i64
28#define TCG_TARGET_deposit_tl_valid     TCG_TARGET_deposit_i64_valid
29#define TCG_TARGET_extract_tl_valid     TCG_TARGET_extract_i64_valid
30#else
31#define TCG_TARGET_HAS_extract2_tl      TCG_TARGET_HAS_extract2_i32
32#define TCG_TARGET_deposit_tl_valid     TCG_TARGET_deposit_i32_valid
33#define TCG_TARGET_extract_tl_valid     TCG_TARGET_extract_i32_valid
34#endif
35
36
37#define ZMM_OFFSET(reg) offsetof(CPUX86State, xmm_regs[reg])
38
39typedef void (*SSEFunc_i_ep)(TCGv_i32 val, TCGv_ptr env, TCGv_ptr reg);
40typedef void (*SSEFunc_l_ep)(TCGv_i64 val, TCGv_ptr env, TCGv_ptr reg);
41typedef void (*SSEFunc_0_epp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b);
42typedef void (*SSEFunc_0_eppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
43                               TCGv_ptr reg_c);
44typedef void (*SSEFunc_0_epppp)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
45                                TCGv_ptr reg_c, TCGv_ptr reg_d);
46typedef void (*SSEFunc_0_eppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
47                               TCGv_i32 val);
48typedef void (*SSEFunc_0_epppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
49                                TCGv_ptr reg_c, TCGv_i32 val);
50typedef void (*SSEFunc_0_ppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_i32 val);
51typedef void (*SSEFunc_0_pppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_ptr reg_c,
52                               TCGv_i32 val);
53typedef void (*SSEFunc_0_eppt)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
54                               TCGv val);
55typedef void (*SSEFunc_0_epppti)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
56                                 TCGv_ptr reg_c, TCGv a0, TCGv_i32 scale);
57typedef void (*SSEFunc_0_eppppi)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
58                                  TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 flags);
59typedef void (*SSEFunc_0_eppppii)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
60                                  TCGv_ptr reg_c, TCGv_ptr reg_d, TCGv_i32 even,
61                                  TCGv_i32 odd);
62
63static void gen_JMP_m(DisasContext *s, X86DecodedInsn *decode);
64static void gen_JMP(DisasContext *s, X86DecodedInsn *decode);
65
66static inline TCGv_i32 tcg_constant8u_i32(uint8_t val)
67{
68    return tcg_constant_i32(val);
69}
70
71static void gen_NM_exception(DisasContext *s)
72{
73    gen_exception(s, EXCP07_PREX);
74}
75
76static void gen_load_ea(DisasContext *s, AddressParts *mem, bool is_vsib)
77{
78    TCGv ea = gen_lea_modrm_1(s, *mem, is_vsib);
79    gen_lea_v_seg(s, ea, mem->def_seg, s->override);
80}
81
82static inline int mmx_offset(MemOp ot)
83{
84    switch (ot) {
85    case MO_8:
86        return offsetof(MMXReg, MMX_B(0));
87    case MO_16:
88        return offsetof(MMXReg, MMX_W(0));
89    case MO_32:
90        return offsetof(MMXReg, MMX_L(0));
91    case MO_64:
92        return offsetof(MMXReg, MMX_Q(0));
93    default:
94        g_assert_not_reached();
95    }
96}
97
98static inline int xmm_offset(MemOp ot)
99{
100    switch (ot) {
101    case MO_8:
102        return offsetof(ZMMReg, ZMM_B(0));
103    case MO_16:
104        return offsetof(ZMMReg, ZMM_W(0));
105    case MO_32:
106        return offsetof(ZMMReg, ZMM_L(0));
107    case MO_64:
108        return offsetof(ZMMReg, ZMM_Q(0));
109    case MO_128:
110        return offsetof(ZMMReg, ZMM_X(0));
111    case MO_256:
112        return offsetof(ZMMReg, ZMM_Y(0));
113    default:
114        g_assert_not_reached();
115    }
116}
117
118static int vector_reg_offset(X86DecodedOp *op)
119{
120    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
121
122    if (op->unit == X86_OP_MMX) {
123        return op->offset - mmx_offset(op->ot);
124    } else {
125        return op->offset - xmm_offset(op->ot);
126    }
127}
128
129static int vector_elem_offset(X86DecodedOp *op, MemOp ot, int n)
130{
131    int base_ofs = vector_reg_offset(op);
132    switch(ot) {
133    case MO_8:
134        if (op->unit == X86_OP_MMX) {
135            return base_ofs + offsetof(MMXReg, MMX_B(n));
136        } else {
137            return base_ofs + offsetof(ZMMReg, ZMM_B(n));
138        }
139    case MO_16:
140        if (op->unit == X86_OP_MMX) {
141            return base_ofs + offsetof(MMXReg, MMX_W(n));
142        } else {
143            return base_ofs + offsetof(ZMMReg, ZMM_W(n));
144        }
145    case MO_32:
146        if (op->unit == X86_OP_MMX) {
147            return base_ofs + offsetof(MMXReg, MMX_L(n));
148        } else {
149            return base_ofs + offsetof(ZMMReg, ZMM_L(n));
150        }
151    case MO_64:
152        if (op->unit == X86_OP_MMX) {
153            return base_ofs;
154        } else {
155            return base_ofs + offsetof(ZMMReg, ZMM_Q(n));
156        }
157    case MO_128:
158        assert(op->unit == X86_OP_SSE);
159        return base_ofs + offsetof(ZMMReg, ZMM_X(n));
160    case MO_256:
161        assert(op->unit == X86_OP_SSE);
162        return base_ofs + offsetof(ZMMReg, ZMM_Y(n));
163    default:
164        g_assert_not_reached();
165    }
166}
167
168static void compute_mmx_offset(X86DecodedOp *op)
169{
170    if (!op->has_ea) {
171        op->offset = offsetof(CPUX86State, fpregs[op->n].mmx) + mmx_offset(op->ot);
172    } else {
173        op->offset = offsetof(CPUX86State, mmx_t0) + mmx_offset(op->ot);
174    }
175}
176
177static void compute_xmm_offset(X86DecodedOp *op)
178{
179    if (!op->has_ea) {
180        op->offset = ZMM_OFFSET(op->n) + xmm_offset(op->ot);
181    } else {
182        op->offset = offsetof(CPUX86State, xmm_t0) + xmm_offset(op->ot);
183    }
184}
185
186static void gen_load_sse(DisasContext *s, TCGv temp, MemOp ot, int dest_ofs, bool aligned)
187{
188    switch(ot) {
189    case MO_8:
190        gen_op_ld_v(s, MO_8, temp, s->A0);
191        tcg_gen_st8_tl(temp, tcg_env, dest_ofs);
192        break;
193    case MO_16:
194        gen_op_ld_v(s, MO_16, temp, s->A0);
195        tcg_gen_st16_tl(temp, tcg_env, dest_ofs);
196        break;
197    case MO_32:
198        gen_op_ld_v(s, MO_32, temp, s->A0);
199        tcg_gen_st32_tl(temp, tcg_env, dest_ofs);
200        break;
201    case MO_64:
202        gen_ldq_env_A0(s, dest_ofs);
203        break;
204    case MO_128:
205        gen_ldo_env_A0(s, dest_ofs, aligned);
206        break;
207    case MO_256:
208        gen_ldy_env_A0(s, dest_ofs, aligned);
209        break;
210    default:
211        g_assert_not_reached();
212    }
213}
214
215static bool sse_needs_alignment(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
216{
217    switch (decode->e.vex_class) {
218    case 2:
219    case 4:
220        if ((s->prefix & PREFIX_VEX) ||
221            decode->e.vex_special == X86_VEX_SSEUnaligned) {
222            /* MOST legacy SSE instructions require aligned memory operands, but not all.  */
223            return false;
224        }
225        /* fall through */
226    case 1:
227        return ot >= MO_128;
228
229    default:
230        return false;
231    }
232}
233
234static void gen_load(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
235{
236    X86DecodedOp *op = &decode->op[opn];
237
238    switch (op->unit) {
239    case X86_OP_SKIP:
240        return;
241    case X86_OP_SEG:
242        tcg_gen_ld32u_tl(v, tcg_env,
243                         offsetof(CPUX86State,segs[op->n].selector));
244        break;
245#ifndef CONFIG_USER_ONLY
246    case X86_OP_CR:
247        if (op->n == 8) {
248            translator_io_start(&s->base);
249            gen_helper_read_cr8(v, tcg_env);
250        } else {
251            tcg_gen_ld_tl(v, tcg_env, offsetof(CPUX86State, cr[op->n]));
252        }
253        break;
254    case X86_OP_DR:
255        /* CR4.DE tested in the helper.  */
256        gen_helper_get_dr(v, tcg_env, tcg_constant_i32(op->n));
257        break;
258#endif
259    case X86_OP_INT:
260        if (op->has_ea) {
261            if (v == s->T0 && decode->e.special == X86_SPECIAL_SExtT0) {
262                gen_op_ld_v(s, op->ot | MO_SIGN, v, s->A0);
263            } else {
264                gen_op_ld_v(s, op->ot, v, s->A0);
265            }
266
267        } else if (op->ot == MO_8 && byte_reg_is_xH(s, op->n)) {
268            if (v == s->T0 && decode->e.special == X86_SPECIAL_SExtT0) {
269                tcg_gen_sextract_tl(v, cpu_regs[op->n - 4], 8, 8);
270            } else {
271                tcg_gen_extract_tl(v, cpu_regs[op->n - 4], 8, 8);
272            }
273
274        } else if (op->ot < MO_TL && v == s->T0 &&
275                   (decode->e.special == X86_SPECIAL_SExtT0 ||
276                    decode->e.special == X86_SPECIAL_ZExtT0)) {
277            if (decode->e.special == X86_SPECIAL_SExtT0) {
278                tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot | MO_SIGN);
279            } else {
280                tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot);
281            }
282
283        } else {
284            tcg_gen_mov_tl(v, cpu_regs[op->n]);
285        }
286        break;
287    case X86_OP_IMM:
288        tcg_gen_movi_tl(v, op->imm);
289        break;
290
291    case X86_OP_MMX:
292        compute_mmx_offset(op);
293        goto load_vector;
294
295    case X86_OP_SSE:
296        compute_xmm_offset(op);
297    load_vector:
298        if (op->has_ea) {
299            bool aligned = sse_needs_alignment(s, decode, op->ot);
300            gen_load_sse(s, v, op->ot, op->offset, aligned);
301        }
302        break;
303
304    default:
305        g_assert_not_reached();
306    }
307}
308
309static TCGv_ptr op_ptr(X86DecodedInsn *decode, int opn)
310{
311    X86DecodedOp *op = &decode->op[opn];
312
313    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
314    if (op->v_ptr) {
315        return op->v_ptr;
316    }
317    op->v_ptr = tcg_temp_new_ptr();
318
319    /* The temporary points to the MMXReg or ZMMReg.  */
320    tcg_gen_addi_ptr(op->v_ptr, tcg_env, vector_reg_offset(op));
321    return op->v_ptr;
322}
323
324#define OP_PTR0 op_ptr(decode, 0)
325#define OP_PTR1 op_ptr(decode, 1)
326#define OP_PTR2 op_ptr(decode, 2)
327
328static void gen_writeback(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
329{
330    X86DecodedOp *op = &decode->op[opn];
331    switch (op->unit) {
332    case X86_OP_SKIP:
333        break;
334    case X86_OP_SEG:
335        /* Note that gen_movl_seg takes care of interrupt shadow and TF.  */
336        gen_movl_seg(s, op->n, s->T0);
337        break;
338    case X86_OP_INT:
339        if (op->has_ea) {
340            gen_op_st_v(s, op->ot, v, s->A0);
341        } else {
342            gen_op_mov_reg_v(s, op->ot, op->n, v);
343        }
344        break;
345    case X86_OP_MMX:
346        break;
347    case X86_OP_SSE:
348        if (!op->has_ea && (s->prefix & PREFIX_VEX) && op->ot <= MO_128) {
349            tcg_gen_gvec_dup_imm(MO_64,
350                                 offsetof(CPUX86State, xmm_regs[op->n].ZMM_X(1)),
351                                 16, 16, 0);
352        }
353        break;
354#ifndef CONFIG_USER_ONLY
355    case X86_OP_CR:
356        if (op->n == 8) {
357            translator_io_start(&s->base);
358        }
359        gen_helper_write_crN(tcg_env, tcg_constant_i32(op->n), v);
360        s->base.is_jmp = DISAS_EOB_NEXT;
361        break;
362    case X86_OP_DR:
363        /* CR4.DE tested in the helper.  */
364        gen_helper_set_dr(tcg_env, tcg_constant_i32(op->n), v);
365        s->base.is_jmp = DISAS_EOB_NEXT;
366        break;
367#endif
368    default:
369        g_assert_not_reached();
370    }
371    op->unit = X86_OP_SKIP;
372}
373
374static inline int vector_len(DisasContext *s, X86DecodedInsn *decode)
375{
376    if (decode->e.special == X86_SPECIAL_MMX &&
377        !(s->prefix & (PREFIX_DATA | PREFIX_REPZ | PREFIX_REPNZ))) {
378        return 8;
379    }
380    return s->vex_l ? 32 : 16;
381}
382
383static void prepare_update1_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
384{
385    decode->cc_dst = s->T0;
386    decode->cc_op = op;
387}
388
389static void prepare_update2_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
390{
391    decode->cc_src = s->T1;
392    decode->cc_dst = s->T0;
393    decode->cc_op = op;
394}
395
396static void prepare_update_cc_incdec(X86DecodedInsn *decode, DisasContext *s, CCOp op)
397{
398    gen_compute_eflags_c(s, s->T1);
399    prepare_update2_cc(decode, s, op);
400}
401
402static void prepare_update3_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op, TCGv reg)
403{
404    decode->cc_src2 = reg;
405    decode->cc_src = s->T1;
406    decode->cc_dst = s->T0;
407    decode->cc_op = op;
408}
409
410static void gen_store_sse(DisasContext *s, X86DecodedInsn *decode, int src_ofs)
411{
412    MemOp ot = decode->op[0].ot;
413    int vec_len = vector_len(s, decode);
414    bool aligned = sse_needs_alignment(s, decode, ot);
415
416    if (!decode->op[0].has_ea) {
417        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, vec_len, vec_len);
418        return;
419    }
420
421    switch (ot) {
422    case MO_64:
423        gen_stq_env_A0(s, src_ofs);
424        break;
425    case MO_128:
426        gen_sto_env_A0(s, src_ofs, aligned);
427        break;
428    case MO_256:
429        gen_sty_env_A0(s, src_ofs, aligned);
430        break;
431    default:
432        g_assert_not_reached();
433    }
434}
435
436static void gen_helper_pavgusb(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b)
437{
438    gen_helper_pavgb_mmx(env, reg_a, reg_a, reg_b);
439}
440
441#define FN_3DNOW_MOVE ((SSEFunc_0_epp) (uintptr_t) 1)
442static const SSEFunc_0_epp fns_3dnow[] = {
443    [0x0c] = gen_helper_pi2fw,
444    [0x0d] = gen_helper_pi2fd,
445    [0x1c] = gen_helper_pf2iw,
446    [0x1d] = gen_helper_pf2id,
447    [0x8a] = gen_helper_pfnacc,
448    [0x8e] = gen_helper_pfpnacc,
449    [0x90] = gen_helper_pfcmpge,
450    [0x94] = gen_helper_pfmin,
451    [0x96] = gen_helper_pfrcp,
452    [0x97] = gen_helper_pfrsqrt,
453    [0x9a] = gen_helper_pfsub,
454    [0x9e] = gen_helper_pfadd,
455    [0xa0] = gen_helper_pfcmpgt,
456    [0xa4] = gen_helper_pfmax,
457    [0xa6] = FN_3DNOW_MOVE, /* PFRCPIT1; no need to actually increase precision */
458    [0xa7] = FN_3DNOW_MOVE, /* PFRSQIT1 */
459    [0xb6] = FN_3DNOW_MOVE, /* PFRCPIT2 */
460    [0xaa] = gen_helper_pfsubr,
461    [0xae] = gen_helper_pfacc,
462    [0xb0] = gen_helper_pfcmpeq,
463    [0xb4] = gen_helper_pfmul,
464    [0xb7] = gen_helper_pmulhrw_mmx,
465    [0xbb] = gen_helper_pswapd,
466    [0xbf] = gen_helper_pavgusb,
467};
468
469static void gen_3dnow(DisasContext *s, X86DecodedInsn *decode)
470{
471    uint8_t b = decode->immediate;
472    SSEFunc_0_epp fn = b < ARRAY_SIZE(fns_3dnow) ? fns_3dnow[b] : NULL;
473
474    if (!fn) {
475        gen_illegal_opcode(s);
476        return;
477    }
478    if (s->flags & HF_TS_MASK) {
479        gen_NM_exception(s);
480        return;
481    }
482    if (s->flags & HF_EM_MASK) {
483        gen_illegal_opcode(s);
484        return;
485    }
486
487    gen_helper_enter_mmx(tcg_env);
488    if (fn == FN_3DNOW_MOVE) {
489       tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset);
490       tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset);
491    } else {
492       fn(tcg_env, OP_PTR0, OP_PTR1);
493    }
494}
495
496/*
497 * 00 = v*ps Vps, Hps, Wpd
498 * 66 = v*pd Vpd, Hpd, Wps
499 * f3 = v*ss Vss, Hss, Wps
500 * f2 = v*sd Vsd, Hsd, Wps
501 */
502static inline void gen_unary_fp_sse(DisasContext *s, X86DecodedInsn *decode,
503                              SSEFunc_0_epp pd_xmm, SSEFunc_0_epp ps_xmm,
504                              SSEFunc_0_epp pd_ymm, SSEFunc_0_epp ps_ymm,
505                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
506{
507    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
508        SSEFunc_0_eppp fn = s->prefix & PREFIX_REPZ ? ss : sd;
509        if (!fn) {
510            gen_illegal_opcode(s);
511            return;
512        }
513        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
514    } else {
515        SSEFunc_0_epp ps, pd, fn;
516        ps = s->vex_l ? ps_ymm : ps_xmm;
517        pd = s->vex_l ? pd_ymm : pd_xmm;
518        fn = s->prefix & PREFIX_DATA ? pd : ps;
519        if (!fn) {
520            gen_illegal_opcode(s);
521            return;
522        }
523        fn(tcg_env, OP_PTR0, OP_PTR2);
524    }
525}
526#define UNARY_FP_SSE(uname, lname)                                                 \
527static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
528{                                                                                  \
529    gen_unary_fp_sse(s, decode,                                                    \
530                     gen_helper_##lname##pd_xmm,                                   \
531                     gen_helper_##lname##ps_xmm,                                   \
532                     gen_helper_##lname##pd_ymm,                                   \
533                     gen_helper_##lname##ps_ymm,                                   \
534                     gen_helper_##lname##sd,                                       \
535                     gen_helper_##lname##ss);                                      \
536}
537UNARY_FP_SSE(VSQRT, sqrt)
538
539/*
540 * 00 = v*ps Vps, Hps, Wpd
541 * 66 = v*pd Vpd, Hpd, Wps
542 * f3 = v*ss Vss, Hss, Wps
543 * f2 = v*sd Vsd, Hsd, Wps
544 */
545static inline void gen_fp_sse(DisasContext *s, X86DecodedInsn *decode,
546                              SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
547                              SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm,
548                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
549{
550    SSEFunc_0_eppp ps, pd, fn;
551    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
552        fn = s->prefix & PREFIX_REPZ ? ss : sd;
553    } else {
554        ps = s->vex_l ? ps_ymm : ps_xmm;
555        pd = s->vex_l ? pd_ymm : pd_xmm;
556        fn = s->prefix & PREFIX_DATA ? pd : ps;
557    }
558    if (fn) {
559        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
560    } else {
561        gen_illegal_opcode(s);
562    }
563}
564
565#define FP_SSE(uname, lname)                                                       \
566static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
567{                                                                                  \
568    gen_fp_sse(s, decode,                                                          \
569               gen_helper_##lname##pd_xmm,                                         \
570               gen_helper_##lname##ps_xmm,                                         \
571               gen_helper_##lname##pd_ymm,                                         \
572               gen_helper_##lname##ps_ymm,                                         \
573               gen_helper_##lname##sd,                                             \
574               gen_helper_##lname##ss);                                            \
575}
576FP_SSE(VADD, add)
577FP_SSE(VMUL, mul)
578FP_SSE(VSUB, sub)
579FP_SSE(VMIN, min)
580FP_SSE(VDIV, div)
581FP_SSE(VMAX, max)
582
583#define FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, even, odd)                         \
584static void gen_##uname##Px(DisasContext *s, X86DecodedInsn *decode)               \
585{                                                                                  \
586    SSEFunc_0_eppppii xmm = s->vex_w ? gen_helper_fma4pd_xmm : gen_helper_fma4ps_xmm; \
587    SSEFunc_0_eppppii ymm = s->vex_w ? gen_helper_fma4pd_ymm : gen_helper_fma4ps_ymm; \
588    SSEFunc_0_eppppii fn = s->vex_l ? ymm : xmm;                                   \
589                                                                                   \
590    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
591       tcg_constant_i32(even),                                                     \
592       tcg_constant_i32((even) ^ (odd)));                                          \
593}
594
595#define FMA_SSE(uname, ptr0, ptr1, ptr2, flags)                                    \
596FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, flags, flags)                              \
597static void gen_##uname##Sx(DisasContext *s, X86DecodedInsn *decode)               \
598{                                                                                  \
599    SSEFunc_0_eppppi fn = s->vex_w ? gen_helper_fma4sd : gen_helper_fma4ss;        \
600                                                                                   \
601    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
602       tcg_constant_i32(flags));                                                   \
603}                                                                                  \
604
605FMA_SSE(VFMADD231,  OP_PTR1, OP_PTR2, OP_PTR0, 0)
606FMA_SSE(VFMADD213,  OP_PTR1, OP_PTR0, OP_PTR2, 0)
607FMA_SSE(VFMADD132,  OP_PTR0, OP_PTR2, OP_PTR1, 0)
608
609FMA_SSE(VFNMADD231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_product)
610FMA_SSE(VFNMADD213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_product)
611FMA_SSE(VFNMADD132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_product)
612
613FMA_SSE(VFMSUB231,  OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c)
614FMA_SSE(VFMSUB213,  OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c)
615FMA_SSE(VFMSUB132,  OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c)
616
617FMA_SSE(VFNMSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c|float_muladd_negate_product)
618FMA_SSE(VFNMSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c|float_muladd_negate_product)
619FMA_SSE(VFNMSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c|float_muladd_negate_product)
620
621FMA_SSE_PACKED(VFMADDSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c, 0)
622FMA_SSE_PACKED(VFMADDSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c, 0)
623FMA_SSE_PACKED(VFMADDSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c, 0)
624
625FMA_SSE_PACKED(VFMSUBADD231, OP_PTR1, OP_PTR2, OP_PTR0, 0, float_muladd_negate_c)
626FMA_SSE_PACKED(VFMSUBADD213, OP_PTR1, OP_PTR0, OP_PTR2, 0, float_muladd_negate_c)
627FMA_SSE_PACKED(VFMSUBADD132, OP_PTR0, OP_PTR2, OP_PTR1, 0, float_muladd_negate_c)
628
629#define FP_UNPACK_SSE(uname, lname)                                                \
630static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
631{                                                                                  \
632    /* PS maps to the DQ integer instruction, PD maps to QDQ.  */                  \
633    gen_fp_sse(s, decode,                                                          \
634               gen_helper_##lname##qdq_xmm,                                        \
635               gen_helper_##lname##dq_xmm,                                         \
636               gen_helper_##lname##qdq_ymm,                                        \
637               gen_helper_##lname##dq_ymm,                                         \
638               NULL, NULL);                                                        \
639}
640FP_UNPACK_SSE(VUNPCKLPx, punpckl)
641FP_UNPACK_SSE(VUNPCKHPx, punpckh)
642
643/*
644 * 00 = v*ps Vps, Wpd
645 * f3 = v*ss Vss, Wps
646 */
647static inline void gen_unary_fp32_sse(DisasContext *s, X86DecodedInsn *decode,
648                                      SSEFunc_0_epp ps_xmm,
649                                      SSEFunc_0_epp ps_ymm,
650                                      SSEFunc_0_eppp ss)
651{
652    if ((s->prefix & (PREFIX_DATA | PREFIX_REPNZ)) != 0) {
653        goto illegal_op;
654    } else if (s->prefix & PREFIX_REPZ) {
655        if (!ss) {
656            goto illegal_op;
657        }
658        ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
659    } else {
660        SSEFunc_0_epp fn = s->vex_l ? ps_ymm : ps_xmm;
661        if (!fn) {
662            goto illegal_op;
663        }
664        fn(tcg_env, OP_PTR0, OP_PTR2);
665    }
666    return;
667
668illegal_op:
669    gen_illegal_opcode(s);
670}
671#define UNARY_FP32_SSE(uname, lname)                                               \
672static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
673{                                                                                  \
674    gen_unary_fp32_sse(s, decode,                                                  \
675                       gen_helper_##lname##ps_xmm,                                 \
676                       gen_helper_##lname##ps_ymm,                                 \
677                       gen_helper_##lname##ss);                                    \
678}
679UNARY_FP32_SSE(VRSQRT, rsqrt)
680UNARY_FP32_SSE(VRCP, rcp)
681
682/*
683 * 66 = v*pd Vpd, Hpd, Wpd
684 * f2 = v*ps Vps, Hps, Wps
685 */
686static inline void gen_horizontal_fp_sse(DisasContext *s, X86DecodedInsn *decode,
687                                         SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
688                                         SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm)
689{
690    SSEFunc_0_eppp ps, pd, fn;
691    ps = s->vex_l ? ps_ymm : ps_xmm;
692    pd = s->vex_l ? pd_ymm : pd_xmm;
693    fn = s->prefix & PREFIX_DATA ? pd : ps;
694    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
695}
696#define HORIZONTAL_FP_SSE(uname, lname)                                            \
697static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
698{                                                                                  \
699    gen_horizontal_fp_sse(s, decode,                                               \
700                          gen_helper_##lname##pd_xmm, gen_helper_##lname##ps_xmm,  \
701                          gen_helper_##lname##pd_ymm, gen_helper_##lname##ps_ymm); \
702}
703HORIZONTAL_FP_SSE(VHADD, hadd)
704HORIZONTAL_FP_SSE(VHSUB, hsub)
705HORIZONTAL_FP_SSE(VADDSUB, addsub)
706
707static inline void gen_ternary_sse(DisasContext *s, X86DecodedInsn *decode,
708                                   int op3, SSEFunc_0_epppp xmm, SSEFunc_0_epppp ymm)
709{
710    SSEFunc_0_epppp fn = s->vex_l ? ymm : xmm;
711    TCGv_ptr ptr3 = tcg_temp_new_ptr();
712
713    /* The format of the fourth input is Lx */
714    tcg_gen_addi_ptr(ptr3, tcg_env, ZMM_OFFSET(op3));
715    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, ptr3);
716}
717#define TERNARY_SSE(uname, uvname, lname)                                          \
718static void gen_##uvname(DisasContext *s, X86DecodedInsn *decode)                  \
719{                                                                                  \
720    gen_ternary_sse(s, decode, (uint8_t)decode->immediate >> 4,                    \
721                    gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);           \
722}                                                                                  \
723static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
724{                                                                                  \
725    gen_ternary_sse(s, decode, 0,                                                  \
726                  gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);             \
727}
728TERNARY_SSE(BLENDVPS, VBLENDVPS, blendvps)
729TERNARY_SSE(BLENDVPD, VBLENDVPD, blendvpd)
730TERNARY_SSE(PBLENDVB, VPBLENDVB, pblendvb)
731
732static inline void gen_binary_imm_sse(DisasContext *s, X86DecodedInsn *decode,
733                                      SSEFunc_0_epppi xmm, SSEFunc_0_epppi ymm)
734{
735    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
736    if (!s->vex_l) {
737        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
738    } else {
739        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
740    }
741}
742
743#define BINARY_IMM_SSE(uname, lname)                                               \
744static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
745{                                                                                  \
746    gen_binary_imm_sse(s, decode,                                                  \
747                       gen_helper_##lname##_xmm,                                   \
748                       gen_helper_##lname##_ymm);                                  \
749}
750
751BINARY_IMM_SSE(VBLENDPD,   blendpd)
752BINARY_IMM_SSE(VBLENDPS,   blendps)
753BINARY_IMM_SSE(VPBLENDW,   pblendw)
754BINARY_IMM_SSE(VDDPS,      dpps)
755#define gen_helper_dppd_ymm NULL
756BINARY_IMM_SSE(VDDPD,      dppd)
757BINARY_IMM_SSE(VMPSADBW,   mpsadbw)
758BINARY_IMM_SSE(PCLMULQDQ,  pclmulqdq)
759
760
761#define UNARY_INT_GVEC(uname, func, ...)                                           \
762static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
763{                                                                                  \
764    int vec_len = vector_len(s, decode);                                          \
765                                                                                   \
766    func(__VA_ARGS__, decode->op[0].offset,                                        \
767         decode->op[2].offset, vec_len, vec_len);                                  \
768}
769UNARY_INT_GVEC(PABSB,          tcg_gen_gvec_abs, MO_8)
770UNARY_INT_GVEC(PABSW,          tcg_gen_gvec_abs, MO_16)
771UNARY_INT_GVEC(PABSD,          tcg_gen_gvec_abs, MO_32)
772UNARY_INT_GVEC(VBROADCASTx128, tcg_gen_gvec_dup_mem, MO_128)
773UNARY_INT_GVEC(VPBROADCASTB,   tcg_gen_gvec_dup_mem, MO_8)
774UNARY_INT_GVEC(VPBROADCASTW,   tcg_gen_gvec_dup_mem, MO_16)
775UNARY_INT_GVEC(VPBROADCASTD,   tcg_gen_gvec_dup_mem, MO_32)
776UNARY_INT_GVEC(VPBROADCASTQ,   tcg_gen_gvec_dup_mem, MO_64)
777
778
779#define BINARY_INT_GVEC(uname, func, ...)                                          \
780static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
781{                                                                                  \
782    int vec_len = vector_len(s, decode);                                          \
783                                                                                   \
784    func(__VA_ARGS__,                                                              \
785         decode->op[0].offset, decode->op[1].offset,                               \
786         decode->op[2].offset, vec_len, vec_len);                                  \
787}
788
789BINARY_INT_GVEC(PADDB,   tcg_gen_gvec_add, MO_8)
790BINARY_INT_GVEC(PADDW,   tcg_gen_gvec_add, MO_16)
791BINARY_INT_GVEC(PADDD,   tcg_gen_gvec_add, MO_32)
792BINARY_INT_GVEC(PADDQ,   tcg_gen_gvec_add, MO_64)
793BINARY_INT_GVEC(PADDSB,  tcg_gen_gvec_ssadd, MO_8)
794BINARY_INT_GVEC(PADDSW,  tcg_gen_gvec_ssadd, MO_16)
795BINARY_INT_GVEC(PADDUSB, tcg_gen_gvec_usadd, MO_8)
796BINARY_INT_GVEC(PADDUSW, tcg_gen_gvec_usadd, MO_16)
797BINARY_INT_GVEC(PAND,    tcg_gen_gvec_and, MO_64)
798BINARY_INT_GVEC(PCMPEQB, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_8)
799BINARY_INT_GVEC(PCMPEQD, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_32)
800BINARY_INT_GVEC(PCMPEQW, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_16)
801BINARY_INT_GVEC(PCMPEQQ, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_64)
802BINARY_INT_GVEC(PCMPGTB, tcg_gen_gvec_cmp, TCG_COND_GT, MO_8)
803BINARY_INT_GVEC(PCMPGTW, tcg_gen_gvec_cmp, TCG_COND_GT, MO_16)
804BINARY_INT_GVEC(PCMPGTD, tcg_gen_gvec_cmp, TCG_COND_GT, MO_32)
805BINARY_INT_GVEC(PCMPGTQ, tcg_gen_gvec_cmp, TCG_COND_GT, MO_64)
806BINARY_INT_GVEC(PMAXSB,  tcg_gen_gvec_smax, MO_8)
807BINARY_INT_GVEC(PMAXSW,  tcg_gen_gvec_smax, MO_16)
808BINARY_INT_GVEC(PMAXSD,  tcg_gen_gvec_smax, MO_32)
809BINARY_INT_GVEC(PMAXUB,  tcg_gen_gvec_umax, MO_8)
810BINARY_INT_GVEC(PMAXUW,  tcg_gen_gvec_umax, MO_16)
811BINARY_INT_GVEC(PMAXUD,  tcg_gen_gvec_umax, MO_32)
812BINARY_INT_GVEC(PMINSB,  tcg_gen_gvec_smin, MO_8)
813BINARY_INT_GVEC(PMINSW,  tcg_gen_gvec_smin, MO_16)
814BINARY_INT_GVEC(PMINSD,  tcg_gen_gvec_smin, MO_32)
815BINARY_INT_GVEC(PMINUB,  tcg_gen_gvec_umin, MO_8)
816BINARY_INT_GVEC(PMINUW,  tcg_gen_gvec_umin, MO_16)
817BINARY_INT_GVEC(PMINUD,  tcg_gen_gvec_umin, MO_32)
818BINARY_INT_GVEC(PMULLW,  tcg_gen_gvec_mul, MO_16)
819BINARY_INT_GVEC(PMULLD,  tcg_gen_gvec_mul, MO_32)
820BINARY_INT_GVEC(POR,     tcg_gen_gvec_or, MO_64)
821BINARY_INT_GVEC(PSUBB,   tcg_gen_gvec_sub, MO_8)
822BINARY_INT_GVEC(PSUBW,   tcg_gen_gvec_sub, MO_16)
823BINARY_INT_GVEC(PSUBD,   tcg_gen_gvec_sub, MO_32)
824BINARY_INT_GVEC(PSUBQ,   tcg_gen_gvec_sub, MO_64)
825BINARY_INT_GVEC(PSUBSB,  tcg_gen_gvec_sssub, MO_8)
826BINARY_INT_GVEC(PSUBSW,  tcg_gen_gvec_sssub, MO_16)
827BINARY_INT_GVEC(PSUBUSB, tcg_gen_gvec_ussub, MO_8)
828BINARY_INT_GVEC(PSUBUSW, tcg_gen_gvec_ussub, MO_16)
829BINARY_INT_GVEC(PXOR,    tcg_gen_gvec_xor, MO_64)
830
831
832/*
833 * 00 = p*  Pq, Qq (if mmx not NULL; no VEX)
834 * 66 = vp* Vx, Hx, Wx
835 *
836 * These are really the same encoding, because 1) V is the same as P when VEX.V
837 * is not present 2) P and Q are the same as H and W apart from MM/XMM
838 */
839static inline void gen_binary_int_sse(DisasContext *s, X86DecodedInsn *decode,
840                                      SSEFunc_0_eppp mmx, SSEFunc_0_eppp xmm, SSEFunc_0_eppp ymm)
841{
842    assert(!!mmx == !!(decode->e.special == X86_SPECIAL_MMX));
843
844    if (mmx && (s->prefix & PREFIX_VEX) && !(s->prefix & PREFIX_DATA)) {
845        /* VEX encoding is not applicable to MMX instructions.  */
846        gen_illegal_opcode(s);
847        return;
848    }
849    if (!(s->prefix & PREFIX_DATA)) {
850        mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
851    } else if (!s->vex_l) {
852        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
853    } else {
854        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
855    }
856}
857
858
859#define BINARY_INT_MMX(uname, lname)                                               \
860static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
861{                                                                                  \
862    gen_binary_int_sse(s, decode,                                                  \
863                          gen_helper_##lname##_mmx,                                \
864                          gen_helper_##lname##_xmm,                                \
865                          gen_helper_##lname##_ymm);                               \
866}
867BINARY_INT_MMX(PUNPCKLBW,  punpcklbw)
868BINARY_INT_MMX(PUNPCKLWD,  punpcklwd)
869BINARY_INT_MMX(PUNPCKLDQ,  punpckldq)
870BINARY_INT_MMX(PACKSSWB,   packsswb)
871BINARY_INT_MMX(PACKUSWB,   packuswb)
872BINARY_INT_MMX(PUNPCKHBW,  punpckhbw)
873BINARY_INT_MMX(PUNPCKHWD,  punpckhwd)
874BINARY_INT_MMX(PUNPCKHDQ,  punpckhdq)
875BINARY_INT_MMX(PACKSSDW,   packssdw)
876
877BINARY_INT_MMX(PAVGB,   pavgb)
878BINARY_INT_MMX(PAVGW,   pavgw)
879BINARY_INT_MMX(PMADDWD, pmaddwd)
880BINARY_INT_MMX(PMULHUW, pmulhuw)
881BINARY_INT_MMX(PMULHW,  pmulhw)
882BINARY_INT_MMX(PMULUDQ, pmuludq)
883BINARY_INT_MMX(PSADBW,  psadbw)
884
885BINARY_INT_MMX(PSLLW_r, psllw)
886BINARY_INT_MMX(PSLLD_r, pslld)
887BINARY_INT_MMX(PSLLQ_r, psllq)
888BINARY_INT_MMX(PSRLW_r, psrlw)
889BINARY_INT_MMX(PSRLD_r, psrld)
890BINARY_INT_MMX(PSRLQ_r, psrlq)
891BINARY_INT_MMX(PSRAW_r, psraw)
892BINARY_INT_MMX(PSRAD_r, psrad)
893
894BINARY_INT_MMX(PHADDW,    phaddw)
895BINARY_INT_MMX(PHADDSW,   phaddsw)
896BINARY_INT_MMX(PHADDD,    phaddd)
897BINARY_INT_MMX(PHSUBW,    phsubw)
898BINARY_INT_MMX(PHSUBSW,   phsubsw)
899BINARY_INT_MMX(PHSUBD,    phsubd)
900BINARY_INT_MMX(PMADDUBSW, pmaddubsw)
901BINARY_INT_MMX(PSHUFB,    pshufb)
902BINARY_INT_MMX(PSIGNB,    psignb)
903BINARY_INT_MMX(PSIGNW,    psignw)
904BINARY_INT_MMX(PSIGND,    psignd)
905BINARY_INT_MMX(PMULHRSW,  pmulhrsw)
906
907/* Instructions with no MMX equivalent.  */
908#define BINARY_INT_SSE(uname, lname)                                               \
909static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
910{                                                                                  \
911    gen_binary_int_sse(s, decode,                                                  \
912                          NULL,                                                    \
913                          gen_helper_##lname##_xmm,                                \
914                          gen_helper_##lname##_ymm);                               \
915}
916
917/* Instructions with no MMX equivalent.  */
918BINARY_INT_SSE(PUNPCKLQDQ, punpcklqdq)
919BINARY_INT_SSE(PUNPCKHQDQ, punpckhqdq)
920BINARY_INT_SSE(VPACKUSDW,  packusdw)
921BINARY_INT_SSE(VPERMILPS,  vpermilps)
922BINARY_INT_SSE(VPERMILPD,  vpermilpd)
923BINARY_INT_SSE(VMASKMOVPS, vpmaskmovd)
924BINARY_INT_SSE(VMASKMOVPD, vpmaskmovq)
925
926BINARY_INT_SSE(PMULDQ,    pmuldq)
927
928BINARY_INT_SSE(VAESDEC, aesdec)
929BINARY_INT_SSE(VAESDECLAST, aesdeclast)
930BINARY_INT_SSE(VAESENC, aesenc)
931BINARY_INT_SSE(VAESENCLAST, aesenclast)
932
933#define UNARY_CMP_SSE(uname, lname)                                                \
934static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
935{                                                                                  \
936    if (!s->vex_l) {                                                               \
937        gen_helper_##lname##_xmm(tcg_env, OP_PTR1, OP_PTR2);                       \
938    } else {                                                                       \
939        gen_helper_##lname##_ymm(tcg_env, OP_PTR1, OP_PTR2);                       \
940    }                                                                              \
941    assume_cc_op(s, CC_OP_EFLAGS);                                                  \
942}
943UNARY_CMP_SSE(VPTEST,     ptest)
944UNARY_CMP_SSE(VTESTPS,    vtestps)
945UNARY_CMP_SSE(VTESTPD,    vtestpd)
946
947static inline void gen_unary_int_sse(DisasContext *s, X86DecodedInsn *decode,
948                                     SSEFunc_0_epp xmm, SSEFunc_0_epp ymm)
949{
950    if (!s->vex_l) {
951        xmm(tcg_env, OP_PTR0, OP_PTR2);
952    } else {
953        ymm(tcg_env, OP_PTR0, OP_PTR2);
954    }
955}
956
957#define UNARY_INT_SSE(uname, lname)                                                \
958static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
959{                                                                                  \
960    gen_unary_int_sse(s, decode,                                                   \
961                      gen_helper_##lname##_xmm,                                    \
962                      gen_helper_##lname##_ymm);                                   \
963}
964
965UNARY_INT_SSE(VPMOVSXBW,    pmovsxbw)
966UNARY_INT_SSE(VPMOVSXBD,    pmovsxbd)
967UNARY_INT_SSE(VPMOVSXBQ,    pmovsxbq)
968UNARY_INT_SSE(VPMOVSXWD,    pmovsxwd)
969UNARY_INT_SSE(VPMOVSXWQ,    pmovsxwq)
970UNARY_INT_SSE(VPMOVSXDQ,    pmovsxdq)
971
972UNARY_INT_SSE(VPMOVZXBW,    pmovzxbw)
973UNARY_INT_SSE(VPMOVZXBD,    pmovzxbd)
974UNARY_INT_SSE(VPMOVZXBQ,    pmovzxbq)
975UNARY_INT_SSE(VPMOVZXWD,    pmovzxwd)
976UNARY_INT_SSE(VPMOVZXWQ,    pmovzxwq)
977UNARY_INT_SSE(VPMOVZXDQ,    pmovzxdq)
978
979UNARY_INT_SSE(VMOVSLDUP,    pmovsldup)
980UNARY_INT_SSE(VMOVSHDUP,    pmovshdup)
981UNARY_INT_SSE(VMOVDDUP,     pmovdldup)
982
983UNARY_INT_SSE(VCVTDQ2PD, cvtdq2pd)
984UNARY_INT_SSE(VCVTPD2DQ, cvtpd2dq)
985UNARY_INT_SSE(VCVTTPD2DQ, cvttpd2dq)
986UNARY_INT_SSE(VCVTDQ2PS, cvtdq2ps)
987UNARY_INT_SSE(VCVTPS2DQ, cvtps2dq)
988UNARY_INT_SSE(VCVTTPS2DQ, cvttps2dq)
989UNARY_INT_SSE(VCVTPH2PS, cvtph2ps)
990
991
992static inline void gen_unary_imm_sse(DisasContext *s, X86DecodedInsn *decode,
993                                     SSEFunc_0_ppi xmm, SSEFunc_0_ppi ymm)
994{
995    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
996    if (!s->vex_l) {
997        xmm(OP_PTR0, OP_PTR1, imm);
998    } else {
999        ymm(OP_PTR0, OP_PTR1, imm);
1000    }
1001}
1002
1003#define UNARY_IMM_SSE(uname, lname)                                                \
1004static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1005{                                                                                  \
1006    gen_unary_imm_sse(s, decode,                                                   \
1007                      gen_helper_##lname##_xmm,                                    \
1008                      gen_helper_##lname##_ymm);                                   \
1009}
1010
1011UNARY_IMM_SSE(PSHUFD,     pshufd)
1012UNARY_IMM_SSE(PSHUFHW,    pshufhw)
1013UNARY_IMM_SSE(PSHUFLW,    pshuflw)
1014#define gen_helper_vpermq_xmm NULL
1015UNARY_IMM_SSE(VPERMQ,      vpermq)
1016UNARY_IMM_SSE(VPERMILPS_i, vpermilps_imm)
1017UNARY_IMM_SSE(VPERMILPD_i, vpermilpd_imm)
1018
1019static inline void gen_unary_imm_fp_sse(DisasContext *s, X86DecodedInsn *decode,
1020                                        SSEFunc_0_eppi xmm, SSEFunc_0_eppi ymm)
1021{
1022    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
1023    if (!s->vex_l) {
1024        xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
1025    } else {
1026        ymm(tcg_env, OP_PTR0, OP_PTR1, imm);
1027    }
1028}
1029
1030#define UNARY_IMM_FP_SSE(uname, lname)                                             \
1031static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1032{                                                                                  \
1033    gen_unary_imm_fp_sse(s, decode,                                                \
1034                      gen_helper_##lname##_xmm,                                    \
1035                      gen_helper_##lname##_ymm);                                   \
1036}
1037
1038UNARY_IMM_FP_SSE(VROUNDPS,    roundps)
1039UNARY_IMM_FP_SSE(VROUNDPD,    roundpd)
1040
1041static inline void gen_vexw_avx(DisasContext *s, X86DecodedInsn *decode,
1042                                SSEFunc_0_eppp d_xmm, SSEFunc_0_eppp q_xmm,
1043                                SSEFunc_0_eppp d_ymm, SSEFunc_0_eppp q_ymm)
1044{
1045    SSEFunc_0_eppp d = s->vex_l ? d_ymm : d_xmm;
1046    SSEFunc_0_eppp q = s->vex_l ? q_ymm : q_xmm;
1047    SSEFunc_0_eppp fn = s->vex_w ? q : d;
1048    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
1049}
1050
1051/* VEX.W affects whether to operate on 32- or 64-bit elements.  */
1052#define VEXW_AVX(uname, lname)                                                     \
1053static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1054{                                                                                  \
1055    gen_vexw_avx(s, decode,                                                        \
1056                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1057                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1058}
1059VEXW_AVX(VPSLLV,    vpsllv)
1060VEXW_AVX(VPSRLV,    vpsrlv)
1061VEXW_AVX(VPSRAV,    vpsrav)
1062VEXW_AVX(VPMASKMOV, vpmaskmov)
1063
1064/* Same as above, but with extra arguments to the helper.  */
1065static inline void gen_vsib_avx(DisasContext *s, X86DecodedInsn *decode,
1066                                SSEFunc_0_epppti d_xmm, SSEFunc_0_epppti q_xmm,
1067                                SSEFunc_0_epppti d_ymm, SSEFunc_0_epppti q_ymm)
1068{
1069    SSEFunc_0_epppti d = s->vex_l ? d_ymm : d_xmm;
1070    SSEFunc_0_epppti q = s->vex_l ? q_ymm : q_xmm;
1071    SSEFunc_0_epppti fn = s->vex_w ? q : d;
1072    TCGv_i32 scale = tcg_constant_i32(decode->mem.scale);
1073    TCGv_ptr index = tcg_temp_new_ptr();
1074
1075    /* Pass third input as (index, base, scale) */
1076    tcg_gen_addi_ptr(index, tcg_env, ZMM_OFFSET(decode->mem.index));
1077    fn(tcg_env, OP_PTR0, OP_PTR1, index, s->A0, scale);
1078
1079    /*
1080     * There are two output operands, so zero OP1's high 128 bits
1081     * in the VEX.128 case.
1082     */
1083    if (!s->vex_l) {
1084        int ymmh_ofs = vector_elem_offset(&decode->op[1], MO_128, 1);
1085        tcg_gen_gvec_dup_imm(MO_64, ymmh_ofs, 16, 16, 0);
1086    }
1087}
1088#define VSIB_AVX(uname, lname)                                                     \
1089static void gen_##uname(DisasContext *s, X86DecodedInsn *decode)                   \
1090{                                                                                  \
1091    gen_vsib_avx(s, decode,                                                        \
1092                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1093                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1094}
1095VSIB_AVX(VPGATHERD, vpgatherd)
1096VSIB_AVX(VPGATHERQ, vpgatherq)
1097
1098static void gen_AAA(DisasContext *s, X86DecodedInsn *decode)
1099{
1100    gen_update_cc_op(s);
1101    gen_helper_aaa(tcg_env);
1102    assume_cc_op(s, CC_OP_EFLAGS);
1103}
1104
1105static void gen_AAD(DisasContext *s, X86DecodedInsn *decode)
1106{
1107    gen_helper_aad(s->T0, s->T0, s->T1);
1108    prepare_update1_cc(decode, s, CC_OP_LOGICB);
1109}
1110
1111static void gen_AAM(DisasContext *s, X86DecodedInsn *decode)
1112{
1113    if (decode->immediate == 0) {
1114        gen_exception(s, EXCP00_DIVZ);
1115    } else {
1116        gen_helper_aam(s->T0, s->T0, s->T1);
1117        prepare_update1_cc(decode, s, CC_OP_LOGICB);
1118    }
1119}
1120
1121static void gen_AAS(DisasContext *s, X86DecodedInsn *decode)
1122{
1123    gen_update_cc_op(s);
1124    gen_helper_aas(tcg_env);
1125    assume_cc_op(s, CC_OP_EFLAGS);
1126}
1127
1128static void gen_ADC(DisasContext *s, X86DecodedInsn *decode)
1129{
1130    MemOp ot = decode->op[1].ot;
1131    TCGv c_in = tcg_temp_new();
1132
1133    gen_compute_eflags_c(s, c_in);
1134    if (s->prefix & PREFIX_LOCK) {
1135        tcg_gen_add_tl(s->T0, c_in, s->T1);
1136        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
1137                                    s->mem_index, ot | MO_LE);
1138    } else {
1139        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1140        tcg_gen_add_tl(s->T0, s->T0, c_in);
1141    }
1142    prepare_update3_cc(decode, s, CC_OP_ADCB + ot, c_in);
1143}
1144
1145static void gen_ADCOX(DisasContext *s, X86DecodedInsn *decode, int cc_op)
1146{
1147    MemOp ot = decode->op[0].ot;
1148    TCGv carry_in = NULL;
1149    TCGv *carry_out = (cc_op == CC_OP_ADCX ? &decode->cc_dst : &decode->cc_src2);
1150    TCGv zero;
1151
1152    decode->cc_op = cc_op;
1153    *carry_out = tcg_temp_new();
1154    if (CC_OP_HAS_EFLAGS(s->cc_op)) {
1155        decode->cc_src = cpu_cc_src;
1156
1157        /* Re-use the carry-out from a previous round?  */
1158        if (s->cc_op == cc_op || s->cc_op == CC_OP_ADCOX) {
1159            carry_in = (cc_op == CC_OP_ADCX ? cpu_cc_dst : cpu_cc_src2);
1160        }
1161
1162        /* Preserve the opposite carry from previous rounds?  */
1163        if (s->cc_op != cc_op && s->cc_op != CC_OP_EFLAGS) {
1164            decode->cc_op = CC_OP_ADCOX;
1165            if (carry_out == &decode->cc_dst) {
1166                decode->cc_src2 = cpu_cc_src2;
1167            } else {
1168                decode->cc_dst = cpu_cc_dst;
1169            }
1170        }
1171    } else {
1172        decode->cc_src = tcg_temp_new();
1173        gen_mov_eflags(s, decode->cc_src);
1174    }
1175
1176    if (!carry_in) {
1177        /* Get carry_in out of EFLAGS.  */
1178        carry_in = tcg_temp_new();
1179        tcg_gen_extract_tl(carry_in, decode->cc_src,
1180            ctz32(cc_op == CC_OP_ADCX ? CC_C : CC_O), 1);
1181    }
1182
1183    switch (ot) {
1184#ifdef TARGET_X86_64
1185    case MO_32:
1186        /* If TL is 64-bit just do everything in 64-bit arithmetic.  */
1187        tcg_gen_ext32u_tl(s->T0, s->T0);
1188        tcg_gen_ext32u_tl(s->T1, s->T1);
1189        tcg_gen_add_i64(s->T0, s->T0, s->T1);
1190        tcg_gen_add_i64(s->T0, s->T0, carry_in);
1191        tcg_gen_shri_i64(*carry_out, s->T0, 32);
1192        break;
1193#endif
1194    default:
1195        zero = tcg_constant_tl(0);
1196        tcg_gen_add2_tl(s->T0, *carry_out, s->T0, zero, carry_in, zero);
1197        tcg_gen_add2_tl(s->T0, *carry_out, s->T0, *carry_out, s->T1, zero);
1198        break;
1199    }
1200}
1201
1202static void gen_ADCX(DisasContext *s, X86DecodedInsn *decode)
1203{
1204    gen_ADCOX(s, decode, CC_OP_ADCX);
1205}
1206
1207static void gen_ADD(DisasContext *s, X86DecodedInsn *decode)
1208{
1209    MemOp ot = decode->op[1].ot;
1210
1211    if (s->prefix & PREFIX_LOCK) {
1212        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1213                                    s->mem_index, ot | MO_LE);
1214    } else {
1215        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1216    }
1217    prepare_update2_cc(decode, s, CC_OP_ADDB + ot);
1218}
1219
1220static void gen_ADOX(DisasContext *s, X86DecodedInsn *decode)
1221{
1222    gen_ADCOX(s, decode, CC_OP_ADOX);
1223}
1224
1225static void gen_AND(DisasContext *s, X86DecodedInsn *decode)
1226{
1227    MemOp ot = decode->op[1].ot;
1228
1229    if (s->prefix & PREFIX_LOCK) {
1230        tcg_gen_atomic_and_fetch_tl(s->T0, s->A0, s->T1,
1231                                    s->mem_index, ot | MO_LE);
1232    } else {
1233        tcg_gen_and_tl(s->T0, s->T0, s->T1);
1234    }
1235    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1236}
1237
1238static void gen_ANDN(DisasContext *s, X86DecodedInsn *decode)
1239{
1240    MemOp ot = decode->op[0].ot;
1241
1242    tcg_gen_andc_tl(s->T0, s->T1, s->T0);
1243    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1244}
1245
1246static void gen_ARPL(DisasContext *s, X86DecodedInsn *decode)
1247{
1248    TCGv zf = tcg_temp_new();
1249    TCGv flags = tcg_temp_new();
1250
1251    gen_mov_eflags(s, flags);
1252
1253    /* Compute adjusted DST in T1, merging in SRC[RPL].  */
1254    tcg_gen_deposit_tl(s->T1, s->T0, s->T1, 0, 2);
1255
1256    /* Z flag set if DST[RPL] < SRC[RPL] */
1257    tcg_gen_setcond_tl(TCG_COND_LTU, zf, s->T0, s->T1);
1258    tcg_gen_deposit_tl(flags, flags, zf, ctz32(CC_Z), 1);
1259
1260    /* Place maximum RPL in DST */
1261    tcg_gen_umax_tl(s->T0, s->T0, s->T1);
1262
1263    decode->cc_src = flags;
1264    decode->cc_op = CC_OP_EFLAGS;
1265}
1266
1267static void gen_BEXTR(DisasContext *s, X86DecodedInsn *decode)
1268{
1269    MemOp ot = decode->op[0].ot;
1270    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1271    TCGv zero = tcg_constant_tl(0);
1272    TCGv mone = tcg_constant_tl(-1);
1273
1274    /*
1275     * Extract START, and shift the operand.
1276     * Shifts larger than operand size get zeros.
1277     */
1278    tcg_gen_ext8u_tl(s->A0, s->T1);
1279    tcg_gen_shr_tl(s->T0, s->T0, s->A0);
1280
1281    tcg_gen_movcond_tl(TCG_COND_LEU, s->T0, s->A0, bound, s->T0, zero);
1282
1283    /*
1284     * Extract the LEN into an inverse mask.  Lengths larger than
1285     * operand size get all zeros, length 0 gets all ones.
1286     */
1287    tcg_gen_extract_tl(s->A0, s->T1, 8, 8);
1288    tcg_gen_shl_tl(s->T1, mone, s->A0);
1289    tcg_gen_movcond_tl(TCG_COND_LEU, s->T1, s->A0, bound, s->T1, zero);
1290    tcg_gen_andc_tl(s->T0, s->T0, s->T1);
1291
1292    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1293}
1294
1295static void gen_BLSI(DisasContext *s, X86DecodedInsn *decode)
1296{
1297    MemOp ot = decode->op[0].ot;
1298
1299    /* input in T1, which is ready for prepare_update2_cc  */
1300    tcg_gen_neg_tl(s->T0, s->T1);
1301    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1302    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1303}
1304
1305static void gen_BLSMSK(DisasContext *s, X86DecodedInsn *decode)
1306{
1307    MemOp ot = decode->op[0].ot;
1308
1309    /* input in T1, which is ready for prepare_update2_cc  */
1310    tcg_gen_subi_tl(s->T0, s->T1, 1);
1311    tcg_gen_xor_tl(s->T0, s->T0, s->T1);
1312    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1313}
1314
1315static void gen_BLSR(DisasContext *s, X86DecodedInsn *decode)
1316{
1317    MemOp ot = decode->op[0].ot;
1318
1319    /* input in T1, which is ready for prepare_update2_cc  */
1320    tcg_gen_subi_tl(s->T0, s->T1, 1);
1321    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1322    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1323}
1324
1325static void gen_BOUND(DisasContext *s, X86DecodedInsn *decode)
1326{
1327    TCGv_i32 op = tcg_temp_new_i32();
1328    tcg_gen_trunc_tl_i32(op, s->T0);
1329    if (decode->op[1].ot == MO_16) {
1330        gen_helper_boundw(tcg_env, s->A0, op);
1331    } else {
1332        gen_helper_boundl(tcg_env, s->A0, op);
1333    }
1334}
1335
1336/* Non-standard convention - on entry T0 is zero-extended input, T1 is the output.  */
1337static void gen_BSF(DisasContext *s, X86DecodedInsn *decode)
1338{
1339    MemOp ot = decode->op[0].ot;
1340
1341    /* Only the Z bit is defined and it is related to the input.  */
1342    decode->cc_dst = tcg_temp_new();
1343    decode->cc_op = CC_OP_LOGICB + ot;
1344    tcg_gen_mov_tl(decode->cc_dst, s->T0);
1345
1346    /*
1347     * The manual says that the output is undefined when the
1348     * input is zero, but real hardware leaves it unchanged, and
1349     * real programs appear to depend on that.  Accomplish this
1350     * by passing the output as the value to return upon zero.
1351     */
1352    tcg_gen_ctz_tl(s->T0, s->T0, s->T1);
1353}
1354
1355/* Non-standard convention - on entry T0 is zero-extended input, T1 is the output.  */
1356static void gen_BSR(DisasContext *s, X86DecodedInsn *decode)
1357{
1358    MemOp ot = decode->op[0].ot;
1359
1360    /* Only the Z bit is defined and it is related to the input.  */
1361    decode->cc_dst = tcg_temp_new();
1362    decode->cc_op = CC_OP_LOGICB + ot;
1363    tcg_gen_mov_tl(decode->cc_dst, s->T0);
1364
1365    /*
1366     * The manual says that the output is undefined when the
1367     * input is zero, but real hardware leaves it unchanged, and
1368     * real programs appear to depend on that.  Accomplish this
1369     * by passing the output as the value to return upon zero.
1370     * Plus, return the bit index of the first 1 bit.
1371     */
1372    tcg_gen_xori_tl(s->T1, s->T1, TARGET_LONG_BITS - 1);
1373    tcg_gen_clz_tl(s->T0, s->T0, s->T1);
1374    tcg_gen_xori_tl(s->T0, s->T0, TARGET_LONG_BITS - 1);
1375}
1376
1377static void gen_BSWAP(DisasContext *s, X86DecodedInsn *decode)
1378{
1379#ifdef TARGET_X86_64
1380    if (s->dflag == MO_64) {
1381        tcg_gen_bswap64_i64(s->T0, s->T0);
1382        return;
1383    }
1384#endif
1385    tcg_gen_bswap32_tl(s->T0, s->T0, TCG_BSWAP_OZ);
1386}
1387
1388static void gen_BZHI(DisasContext *s, X86DecodedInsn *decode)
1389{
1390    MemOp ot = decode->op[0].ot;
1391    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1392    TCGv zero = tcg_constant_tl(0);
1393    TCGv mone = tcg_constant_tl(-1);
1394
1395    tcg_gen_ext8u_tl(s->T1, s->T1);
1396
1397    tcg_gen_shl_tl(s->A0, mone, s->T1);
1398    tcg_gen_movcond_tl(TCG_COND_LEU, s->A0, s->T1, bound, s->A0, zero);
1399    tcg_gen_andc_tl(s->T0, s->T0, s->A0);
1400    /*
1401     * Note that since we're using BMILG (in order to get O
1402     * cleared) we need to store the inverse into C.
1403     */
1404    tcg_gen_setcond_tl(TCG_COND_LEU, s->T1, s->T1, bound);
1405    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1406}
1407
1408static void gen_CALL(DisasContext *s, X86DecodedInsn *decode)
1409{
1410    gen_push_v(s, eip_next_tl(s));
1411    gen_JMP(s, decode);
1412}
1413
1414static void gen_CALL_m(DisasContext *s, X86DecodedInsn *decode)
1415{
1416    gen_push_v(s, eip_next_tl(s));
1417    gen_JMP_m(s, decode);
1418}
1419
1420static void gen_CALLF(DisasContext *s, X86DecodedInsn *decode)
1421{
1422    gen_far_call(s);
1423}
1424
1425static void gen_CALLF_m(DisasContext *s, X86DecodedInsn *decode)
1426{
1427    MemOp ot = decode->op[1].ot;
1428
1429    gen_op_ld_v(s, ot, s->T0, s->A0);
1430    gen_add_A0_im(s, 1 << ot);
1431    gen_op_ld_v(s, MO_16, s->T1, s->A0);
1432    gen_far_call(s);
1433}
1434
1435static void gen_CBW(DisasContext *s, X86DecodedInsn *decode)
1436{
1437    MemOp src_ot = decode->op[0].ot - 1;
1438
1439    tcg_gen_ext_tl(s->T0, s->T0, src_ot | MO_SIGN);
1440}
1441
1442static void gen_CLC(DisasContext *s, X86DecodedInsn *decode)
1443{
1444    gen_compute_eflags(s);
1445    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, ~CC_C);
1446}
1447
1448static void gen_CLD(DisasContext *s, X86DecodedInsn *decode)
1449{
1450    tcg_gen_st_i32(tcg_constant_i32(1), tcg_env, offsetof(CPUX86State, df));
1451}
1452
1453static void gen_CLI(DisasContext *s, X86DecodedInsn *decode)
1454{
1455    gen_reset_eflags(s, IF_MASK);
1456}
1457
1458static void gen_CLTS(DisasContext *s, X86DecodedInsn *decode)
1459{
1460    gen_helper_clts(tcg_env);
1461    /* abort block because static cpu state changed */
1462    s->base.is_jmp = DISAS_EOB_NEXT;
1463}
1464
1465static void gen_CMC(DisasContext *s, X86DecodedInsn *decode)
1466{
1467    gen_compute_eflags(s);
1468    tcg_gen_xori_tl(cpu_cc_src, cpu_cc_src, CC_C);
1469}
1470
1471static void gen_CMOVcc(DisasContext *s, X86DecodedInsn *decode)
1472{
1473    gen_cmovcc1(s, decode->b & 0xf, s->T0, s->T1);
1474}
1475
1476static void gen_CMPccXADD(DisasContext *s, X86DecodedInsn *decode)
1477{
1478    TCGLabel *label_top = gen_new_label();
1479    TCGLabel *label_bottom = gen_new_label();
1480    TCGv oldv = tcg_temp_new();
1481    TCGv newv = tcg_temp_new();
1482    TCGv cmpv = tcg_temp_new();
1483    TCGCond cond;
1484
1485    TCGv cmp_lhs, cmp_rhs;
1486    MemOp ot, ot_full;
1487
1488    int jcc_op = (decode->b >> 1) & 7;
1489    static const TCGCond cond_table[8] = {
1490        [JCC_O] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1491        [JCC_B] = TCG_COND_LTU,
1492        [JCC_Z] = TCG_COND_EQ,
1493        [JCC_BE] = TCG_COND_LEU,
1494        [JCC_S] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1495        [JCC_P] = TCG_COND_TSTEQ,  /* even parity - tests low bit of popcount */
1496        [JCC_L] = TCG_COND_LT,
1497        [JCC_LE] = TCG_COND_LE,
1498    };
1499
1500    cond = cond_table[jcc_op];
1501    if (decode->b & 1) {
1502        cond = tcg_invert_cond(cond);
1503    }
1504
1505    ot = decode->op[0].ot;
1506    ot_full = ot | MO_LE;
1507    if (jcc_op >= JCC_S) {
1508        /*
1509         * Sign-extend values before subtracting for S, P (zero/sign extension
1510         * does not matter there) L, LE and their inverses.
1511         */
1512        ot_full |= MO_SIGN;
1513    }
1514
1515    /*
1516     * cmpv will be moved to cc_src *after* cpu_regs[] is written back, so use
1517     * tcg_gen_ext_tl instead of gen_ext_tl.
1518     */
1519    tcg_gen_ext_tl(cmpv, cpu_regs[decode->op[1].n], ot_full);
1520
1521    /*
1522     * Cmpxchg loop starts here.
1523     * - s->T1: addition operand (from decoder)
1524     * - s->A0: dest address (from decoder)
1525     * - s->cc_srcT: memory operand (lhs for comparison)
1526     * - cmpv: rhs for comparison
1527     */
1528    gen_set_label(label_top);
1529    gen_op_ld_v(s, ot_full, s->cc_srcT, s->A0);
1530    tcg_gen_sub_tl(s->T0, s->cc_srcT, cmpv);
1531
1532    /* Compute the comparison result by hand, to avoid clobbering cc_*.  */
1533    switch (jcc_op) {
1534    case JCC_O:
1535        /* (src1 ^ src2) & (src1 ^ dst). newv is only used here for a moment */
1536        tcg_gen_xor_tl(newv, s->cc_srcT, s->T0);
1537        tcg_gen_xor_tl(s->tmp0, s->cc_srcT, cmpv);
1538        tcg_gen_and_tl(s->tmp0, s->tmp0, newv);
1539        tcg_gen_sextract_tl(s->tmp0, s->tmp0, 0, 8 << ot);
1540        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(0);
1541        break;
1542
1543    case JCC_P:
1544        tcg_gen_ext8u_tl(s->tmp0, s->T0);
1545        tcg_gen_ctpop_tl(s->tmp0, s->tmp0);
1546        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(1);
1547        break;
1548
1549    case JCC_S:
1550        tcg_gen_sextract_tl(s->tmp0, s->T0, 0, 8 << ot);
1551        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(0);
1552        break;
1553
1554    default:
1555        cmp_lhs = s->cc_srcT, cmp_rhs = cmpv;
1556        break;
1557    }
1558
1559    /* Compute new value: if condition does not hold, just store back s->cc_srcT */
1560    tcg_gen_add_tl(newv, s->cc_srcT, s->T1);
1561    tcg_gen_movcond_tl(cond, newv, cmp_lhs, cmp_rhs, newv, s->cc_srcT);
1562    tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, s->cc_srcT, newv, s->mem_index, ot_full);
1563
1564    /* Exit unconditionally if cmpxchg succeeded.  */
1565    tcg_gen_brcond_tl(TCG_COND_EQ, oldv, s->cc_srcT, label_bottom);
1566
1567    /* Try again if there was actually a store to make.  */
1568    tcg_gen_brcond_tl(cond, cmp_lhs, cmp_rhs, label_top);
1569    gen_set_label(label_bottom);
1570
1571    /* Store old value to registers only after a successful store.  */
1572    gen_writeback(s, decode, 1, s->cc_srcT);
1573
1574    decode->cc_dst = s->T0;
1575    decode->cc_src = cmpv;
1576    decode->cc_op = CC_OP_SUBB + ot;
1577}
1578
1579static void gen_CMPS(DisasContext *s, X86DecodedInsn *decode)
1580{
1581    MemOp ot = decode->op[2].ot;
1582    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
1583        gen_repz_nz(s, ot, gen_cmps);
1584    } else {
1585        gen_cmps(s, ot);
1586    }
1587}
1588
1589static void gen_CMPXCHG(DisasContext *s, X86DecodedInsn *decode)
1590{
1591    MemOp ot = decode->op[2].ot;
1592    TCGv cmpv = tcg_temp_new();
1593    TCGv oldv = tcg_temp_new();
1594    TCGv newv = tcg_temp_new();
1595    TCGv dest;
1596
1597    tcg_gen_ext_tl(cmpv, cpu_regs[R_EAX], ot);
1598    tcg_gen_ext_tl(newv, s->T1, ot);
1599    if (s->prefix & PREFIX_LOCK) {
1600        tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, cmpv, newv,
1601                                  s->mem_index, ot | MO_LE);
1602    } else {
1603        tcg_gen_ext_tl(oldv, s->T0, ot);
1604        if (decode->op[0].has_ea) {
1605            /*
1606             * Perform an unconditional store cycle like physical cpu;
1607             * must be before changing accumulator to ensure
1608             * idempotency if the store faults and the instruction
1609             * is restarted
1610             */
1611            tcg_gen_movcond_tl(TCG_COND_EQ, newv, oldv, cmpv, newv, oldv);
1612            gen_op_st_v(s, ot, newv, s->A0);
1613        } else {
1614            /*
1615             * Unlike the memory case, where "the destination operand receives
1616             * a write cycle without regard to the result of the comparison",
1617             * rm must not be touched altogether if the write fails, including
1618             * not zero-extending it on 64-bit processors.  So, precompute
1619             * the result of a successful writeback and perform the movcond
1620             * directly on cpu_regs.  In case rm is part of RAX, note that this
1621             * movcond and the one below are mutually exclusive is executed.
1622             */
1623            dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, newv, newv);
1624            tcg_gen_movcond_tl(TCG_COND_EQ, dest, oldv, cmpv, newv, dest);
1625        }
1626        decode->op[0].unit = X86_OP_SKIP;
1627    }
1628
1629    /* Write RAX only if the cmpxchg fails.  */
1630    dest = gen_op_deposit_reg_v(s, ot, R_EAX, s->T0, oldv);
1631    tcg_gen_movcond_tl(TCG_COND_NE, dest, oldv, cmpv, s->T0, dest);
1632
1633    tcg_gen_mov_tl(s->cc_srcT, cmpv);
1634    tcg_gen_sub_tl(cmpv, cmpv, oldv);
1635    decode->cc_dst = cmpv;
1636    decode->cc_src = oldv;
1637    decode->cc_op = CC_OP_SUBB + ot;
1638}
1639
1640static void gen_CPUID(DisasContext *s, X86DecodedInsn *decode)
1641{
1642    gen_update_cc_op(s);
1643    gen_update_eip_cur(s);
1644    gen_helper_cpuid(tcg_env);
1645}
1646
1647static void gen_CRC32(DisasContext *s, X86DecodedInsn *decode)
1648{
1649    MemOp ot = decode->op[2].ot;
1650
1651    tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
1652    gen_helper_crc32(s->T0, s->tmp2_i32, s->T1, tcg_constant_i32(8 << ot));
1653}
1654
1655static void gen_CVTPI2Px(DisasContext *s, X86DecodedInsn *decode)
1656{
1657    gen_helper_enter_mmx(tcg_env);
1658    if (s->prefix & PREFIX_DATA) {
1659        gen_helper_cvtpi2pd(tcg_env, OP_PTR0, OP_PTR2);
1660    } else {
1661        gen_helper_cvtpi2ps(tcg_env, OP_PTR0, OP_PTR2);
1662    }
1663}
1664
1665static void gen_CVTPx2PI(DisasContext *s, X86DecodedInsn *decode)
1666{
1667    gen_helper_enter_mmx(tcg_env);
1668    if (s->prefix & PREFIX_DATA) {
1669        gen_helper_cvtpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1670    } else {
1671        gen_helper_cvtps2pi(tcg_env, OP_PTR0, OP_PTR2);
1672    }
1673}
1674
1675static void gen_CVTTPx2PI(DisasContext *s, X86DecodedInsn *decode)
1676{
1677    gen_helper_enter_mmx(tcg_env);
1678    if (s->prefix & PREFIX_DATA) {
1679        gen_helper_cvttpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1680    } else {
1681        gen_helper_cvttps2pi(tcg_env, OP_PTR0, OP_PTR2);
1682    }
1683}
1684
1685static void gen_CWD(DisasContext *s, X86DecodedInsn *decode)
1686{
1687    int shift = 8 << decode->op[0].ot;
1688
1689    tcg_gen_sextract_tl(s->T0, s->T0, shift - 1, 1);
1690}
1691
1692static void gen_DAA(DisasContext *s, X86DecodedInsn *decode)
1693{
1694    gen_update_cc_op(s);
1695    gen_helper_daa(tcg_env);
1696    assume_cc_op(s, CC_OP_EFLAGS);
1697}
1698
1699static void gen_DAS(DisasContext *s, X86DecodedInsn *decode)
1700{
1701    gen_update_cc_op(s);
1702    gen_helper_das(tcg_env);
1703    assume_cc_op(s, CC_OP_EFLAGS);
1704}
1705
1706static void gen_DEC(DisasContext *s, X86DecodedInsn *decode)
1707{
1708    MemOp ot = decode->op[1].ot;
1709
1710    tcg_gen_movi_tl(s->T1, -1);
1711    if (s->prefix & PREFIX_LOCK) {
1712        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1713                                    s->mem_index, ot | MO_LE);
1714    } else {
1715        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1716    }
1717    prepare_update_cc_incdec(decode, s, CC_OP_DECB + ot);
1718}
1719
1720static void gen_DIV(DisasContext *s, X86DecodedInsn *decode)
1721{
1722    MemOp ot = decode->op[1].ot;
1723
1724    switch(ot) {
1725    case MO_8:
1726        gen_helper_divb_AL(tcg_env, s->T0);
1727        break;
1728    case MO_16:
1729        gen_helper_divw_AX(tcg_env, s->T0);
1730        break;
1731    default:
1732    case MO_32:
1733        gen_helper_divl_EAX(tcg_env, s->T0);
1734        break;
1735#ifdef TARGET_X86_64
1736    case MO_64:
1737        gen_helper_divq_EAX(tcg_env, s->T0);
1738        break;
1739#endif
1740    }
1741}
1742
1743static void gen_EMMS(DisasContext *s, X86DecodedInsn *decode)
1744{
1745    gen_helper_emms(tcg_env);
1746}
1747
1748static void gen_ENTER(DisasContext *s, X86DecodedInsn *decode)
1749{
1750   gen_enter(s, decode->op[1].imm, decode->op[2].imm);
1751}
1752
1753static void gen_EXTRQ_i(DisasContext *s, X86DecodedInsn *decode)
1754{
1755    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
1756    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
1757
1758    gen_helper_extrq_i(tcg_env, OP_PTR0, index, length);
1759}
1760
1761static void gen_EXTRQ_r(DisasContext *s, X86DecodedInsn *decode)
1762{
1763    gen_helper_extrq_r(tcg_env, OP_PTR0, OP_PTR2);
1764}
1765
1766static void gen_FXRSTOR(DisasContext *s, X86DecodedInsn *decode)
1767{
1768    if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
1769        gen_NM_exception(s);
1770    } else {
1771        gen_helper_fxrstor(tcg_env, s->A0);
1772    }
1773}
1774
1775static void gen_FXSAVE(DisasContext *s, X86DecodedInsn *decode)
1776{
1777    if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
1778        gen_NM_exception(s);
1779    } else {
1780        gen_helper_fxsave(tcg_env, s->A0);
1781    }
1782}
1783
1784static void gen_HLT(DisasContext *s, X86DecodedInsn *decode)
1785{
1786#ifdef CONFIG_SYSTEM_ONLY
1787    gen_update_cc_op(s);
1788    gen_update_eip_next(s);
1789    gen_helper_hlt(tcg_env);
1790    s->base.is_jmp = DISAS_NORETURN;
1791#endif
1792}
1793
1794static void gen_IDIV(DisasContext *s, X86DecodedInsn *decode)
1795{
1796    MemOp ot = decode->op[1].ot;
1797
1798    switch(ot) {
1799    case MO_8:
1800        gen_helper_idivb_AL(tcg_env, s->T0);
1801        break;
1802    case MO_16:
1803        gen_helper_idivw_AX(tcg_env, s->T0);
1804        break;
1805    default:
1806    case MO_32:
1807        gen_helper_idivl_EAX(tcg_env, s->T0);
1808        break;
1809#ifdef TARGET_X86_64
1810    case MO_64:
1811        gen_helper_idivq_EAX(tcg_env, s->T0);
1812        break;
1813#endif
1814    }
1815}
1816
1817static void gen_IMUL3(DisasContext *s, X86DecodedInsn *decode)
1818{
1819    MemOp ot = decode->op[0].ot;
1820    TCGv cc_src_rhs;
1821
1822    switch (ot) {
1823    case MO_16:
1824        /* s->T0 already sign-extended */
1825        tcg_gen_ext16s_tl(s->T1, s->T1);
1826        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1827        /* Compare the full result to the extension of the truncated result.  */
1828        tcg_gen_ext16s_tl(s->T1, s->T0);
1829        cc_src_rhs = s->T0;
1830        break;
1831
1832    case MO_32:
1833#ifdef TARGET_X86_64
1834        if (TCG_TARGET_REG_BITS == 64) {
1835            /*
1836             * This produces fewer TCG ops, and better code if flags are needed,
1837             * but it requires a 64-bit multiply even if they are not.  Use it
1838             * only if the target has 64-bits registers.
1839             *
1840             * s->T0 is already sign-extended.
1841             */
1842            tcg_gen_ext32s_tl(s->T1, s->T1);
1843            tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1844            /* Compare the full result to the extension of the truncated result.  */
1845            tcg_gen_ext32s_tl(s->T1, s->T0);
1846            cc_src_rhs = s->T0;
1847        } else {
1848            /* Variant that only needs a 32-bit widening multiply.  */
1849            TCGv_i32 hi = tcg_temp_new_i32();
1850            TCGv_i32 lo = tcg_temp_new_i32();
1851            tcg_gen_trunc_tl_i32(lo, s->T0);
1852            tcg_gen_trunc_tl_i32(hi, s->T1);
1853            tcg_gen_muls2_i32(lo, hi, lo, hi);
1854            tcg_gen_extu_i32_tl(s->T0, lo);
1855
1856            cc_src_rhs = tcg_temp_new();
1857            tcg_gen_extu_i32_tl(cc_src_rhs, hi);
1858            /* Compare the high part to the sign bit of the truncated result */
1859            tcg_gen_sari_i32(lo, lo, 31);
1860            tcg_gen_extu_i32_tl(s->T1, lo);
1861        }
1862        break;
1863
1864    case MO_64:
1865#endif
1866        cc_src_rhs = tcg_temp_new();
1867        tcg_gen_muls2_tl(s->T0, cc_src_rhs, s->T0, s->T1);
1868        /* Compare the high part to the sign bit of the truncated result */
1869        tcg_gen_sari_tl(s->T1, s->T0, TARGET_LONG_BITS - 1);
1870        break;
1871
1872    default:
1873        g_assert_not_reached();
1874    }
1875
1876    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
1877    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
1878}
1879
1880static void gen_IMUL(DisasContext *s, X86DecodedInsn *decode)
1881{
1882    MemOp ot = decode->op[1].ot;
1883    TCGv cc_src_rhs;
1884
1885    switch (ot) {
1886    case MO_8:
1887        /* s->T0 already sign-extended */
1888        tcg_gen_ext8s_tl(s->T1, s->T1);
1889        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1890        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
1891        /* Compare the full result to the extension of the truncated result.  */
1892        tcg_gen_ext8s_tl(s->T1, s->T0);
1893        cc_src_rhs = s->T0;
1894        break;
1895
1896    case MO_16:
1897        /* s->T0 already sign-extended */
1898        tcg_gen_ext16s_tl(s->T1, s->T1);
1899        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1900        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
1901        tcg_gen_shri_tl(s->T1, s->T0, 16);
1902        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
1903        /* Compare the full result to the extension of the truncated result.  */
1904        tcg_gen_ext16s_tl(s->T1, s->T0);
1905        cc_src_rhs = s->T0;
1906        break;
1907
1908    case MO_32:
1909#ifdef TARGET_X86_64
1910        /* s->T0 already sign-extended */
1911        tcg_gen_ext32s_tl(s->T1, s->T1);
1912        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1913        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
1914        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
1915        /* Compare the full result to the extension of the truncated result.  */
1916        tcg_gen_ext32s_tl(s->T1, s->T0);
1917        cc_src_rhs = s->T0;
1918        break;
1919
1920    case MO_64:
1921#endif
1922        tcg_gen_muls2_tl(s->T0, cpu_regs[R_EDX], s->T0, s->T1);
1923        tcg_gen_mov_tl(cpu_regs[R_EAX], s->T0);
1924
1925        /* Compare the high part to the sign bit of the truncated result */
1926        tcg_gen_negsetcondi_tl(TCG_COND_LT, s->T1, s->T0, 0);
1927        cc_src_rhs = cpu_regs[R_EDX];
1928        break;
1929
1930    default:
1931        g_assert_not_reached();
1932    }
1933
1934    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
1935    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
1936}
1937
1938static void gen_IN(DisasContext *s, X86DecodedInsn *decode)
1939{
1940    MemOp ot = decode->op[0].ot;
1941    TCGv_i32 port = tcg_temp_new_i32();
1942
1943    tcg_gen_trunc_tl_i32(port, s->T0);
1944    tcg_gen_ext16u_i32(port, port);
1945    if (!gen_check_io(s, ot, port, SVM_IOIO_TYPE_MASK)) {
1946        return;
1947    }
1948    translator_io_start(&s->base);
1949    gen_helper_in_func(ot, s->T0, port);
1950    gen_writeback(s, decode, 0, s->T0);
1951    gen_bpt_io(s, port, ot);
1952}
1953
1954static void gen_INC(DisasContext *s, X86DecodedInsn *decode)
1955{
1956    MemOp ot = decode->op[1].ot;
1957
1958    tcg_gen_movi_tl(s->T1, 1);
1959    if (s->prefix & PREFIX_LOCK) {
1960        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1961                                    s->mem_index, ot | MO_LE);
1962    } else {
1963        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1964    }
1965    prepare_update_cc_incdec(decode, s, CC_OP_INCB + ot);
1966}
1967
1968static void gen_INS(DisasContext *s, X86DecodedInsn *decode)
1969{
1970    MemOp ot = decode->op[1].ot;
1971    TCGv_i32 port = tcg_temp_new_i32();
1972
1973    tcg_gen_trunc_tl_i32(port, s->T1);
1974    tcg_gen_ext16u_i32(port, port);
1975    if (!gen_check_io(s, ot, port,
1976                      SVM_IOIO_TYPE_MASK | SVM_IOIO_STR_MASK)) {
1977        return;
1978    }
1979
1980    translator_io_start(&s->base);
1981    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
1982        gen_repz(s, ot, gen_ins);
1983    } else {
1984        gen_ins(s, ot);
1985    }
1986}
1987
1988static void gen_INSERTQ_i(DisasContext *s, X86DecodedInsn *decode)
1989{
1990    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
1991    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
1992
1993    gen_helper_insertq_i(tcg_env, OP_PTR0, OP_PTR1, index, length);
1994}
1995
1996static void gen_INSERTQ_r(DisasContext *s, X86DecodedInsn *decode)
1997{
1998    gen_helper_insertq_r(tcg_env, OP_PTR0, OP_PTR2);
1999}
2000
2001static void gen_INT(DisasContext *s, X86DecodedInsn *decode)
2002{
2003    gen_interrupt(s, decode->immediate);
2004}
2005
2006static void gen_INT1(DisasContext *s, X86DecodedInsn *decode)
2007{
2008    gen_update_cc_op(s);
2009    gen_update_eip_next(s);
2010    gen_helper_icebp(tcg_env);
2011    s->base.is_jmp = DISAS_NORETURN;
2012}
2013
2014static void gen_INT3(DisasContext *s, X86DecodedInsn *decode)
2015{
2016    gen_interrupt(s, EXCP03_INT3);
2017}
2018
2019static void gen_INTO(DisasContext *s, X86DecodedInsn *decode)
2020{
2021    gen_update_cc_op(s);
2022    gen_update_eip_cur(s);
2023    gen_helper_into(tcg_env, cur_insn_len_i32(s));
2024}
2025
2026static void gen_IRET(DisasContext *s, X86DecodedInsn *decode)
2027{
2028    if (!PE(s) || VM86(s)) {
2029        gen_helper_iret_real(tcg_env, tcg_constant_i32(s->dflag - 1));
2030    } else {
2031        gen_helper_iret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
2032                                  eip_next_i32(s));
2033    }
2034    assume_cc_op(s, CC_OP_EFLAGS);
2035    s->base.is_jmp = DISAS_EOB_ONLY;
2036}
2037
2038static void gen_Jcc(DisasContext *s, X86DecodedInsn *decode)
2039{
2040    gen_bnd_jmp(s);
2041    gen_jcc(s, decode->b & 0xf, decode->immediate);
2042}
2043
2044static void gen_JCXZ(DisasContext *s, X86DecodedInsn *decode)
2045{
2046    TCGLabel *taken = gen_new_label();
2047
2048    gen_update_cc_op(s);
2049    gen_op_jz_ecx(s, taken);
2050    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2051}
2052
2053static void gen_JMP(DisasContext *s, X86DecodedInsn *decode)
2054{
2055    gen_update_cc_op(s);
2056    gen_jmp_rel(s, s->dflag, decode->immediate, 0);
2057}
2058
2059static void gen_JMP_m(DisasContext *s, X86DecodedInsn *decode)
2060{
2061    gen_op_jmp_v(s, s->T0);
2062    gen_bnd_jmp(s);
2063    s->base.is_jmp = DISAS_JUMP;
2064}
2065
2066static void gen_JMPF(DisasContext *s, X86DecodedInsn *decode)
2067{
2068    gen_far_jmp(s);
2069}
2070
2071static void gen_JMPF_m(DisasContext *s, X86DecodedInsn *decode)
2072{
2073    MemOp ot = decode->op[1].ot;
2074
2075    gen_op_ld_v(s, ot, s->T0, s->A0);
2076    gen_add_A0_im(s, 1 << ot);
2077    gen_op_ld_v(s, MO_16, s->T1, s->A0);
2078    gen_far_jmp(s);
2079}
2080
2081static void gen_LAHF(DisasContext *s, X86DecodedInsn *decode)
2082{
2083    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
2084        return gen_illegal_opcode(s);
2085    }
2086    gen_compute_eflags(s);
2087    /* Note: gen_compute_eflags() only gives the condition codes */
2088    tcg_gen_ori_tl(s->T0, cpu_cc_src, 0x02);
2089    tcg_gen_deposit_tl(cpu_regs[R_EAX], cpu_regs[R_EAX], s->T0, 8, 8);
2090}
2091
2092static void gen_LAR(DisasContext *s, X86DecodedInsn *decode)
2093{
2094    MemOp ot = decode->op[0].ot;
2095    TCGv result = tcg_temp_new();
2096    TCGv dest;
2097
2098    gen_compute_eflags(s);
2099    gen_update_cc_op(s);
2100    gen_helper_lar(result, tcg_env, s->T0);
2101
2102    /* Perform writeback here to skip it if ZF=0.  */
2103    decode->op[0].unit = X86_OP_SKIP;
2104    dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, result, result);
2105    tcg_gen_movcond_tl(TCG_COND_TSTNE, dest, cpu_cc_src, tcg_constant_tl(CC_Z),
2106                       result, dest);
2107}
2108
2109static void gen_LDMXCSR(DisasContext *s, X86DecodedInsn *decode)
2110{
2111    tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
2112    gen_helper_ldmxcsr(tcg_env, s->tmp2_i32);
2113}
2114
2115static void gen_lxx_seg(DisasContext *s, X86DecodedInsn *decode, int seg)
2116{
2117    MemOp ot = decode->op[0].ot;
2118
2119    /* Offset already in s->T0.  */
2120    gen_add_A0_im(s, 1 << ot);
2121    gen_op_ld_v(s, MO_16, s->T1, s->A0);
2122
2123    /* load the segment here to handle exceptions properly */
2124    gen_movl_seg(s, seg, s->T1);
2125}
2126
2127static void gen_LDS(DisasContext *s, X86DecodedInsn *decode)
2128{
2129    gen_lxx_seg(s, decode, R_DS);
2130}
2131
2132static void gen_LEA(DisasContext *s, X86DecodedInsn *decode)
2133{
2134    TCGv ea = gen_lea_modrm_1(s, decode->mem, false);
2135    gen_lea_v_seg_dest(s, s->aflag, s->T0, ea, -1, -1);
2136}
2137
2138static void gen_LEAVE(DisasContext *s, X86DecodedInsn *decode)
2139{
2140    gen_leave(s);
2141}
2142
2143static void gen_LES(DisasContext *s, X86DecodedInsn *decode)
2144{
2145    gen_lxx_seg(s, decode, R_ES);
2146}
2147
2148static void gen_LFENCE(DisasContext *s, X86DecodedInsn *decode)
2149{
2150    tcg_gen_mb(TCG_MO_LD_LD | TCG_BAR_SC);
2151}
2152
2153static void gen_LFS(DisasContext *s, X86DecodedInsn *decode)
2154{
2155    gen_lxx_seg(s, decode, R_FS);
2156}
2157
2158static void gen_LGS(DisasContext *s, X86DecodedInsn *decode)
2159{
2160    gen_lxx_seg(s, decode, R_GS);
2161}
2162
2163static void gen_LODS(DisasContext *s, X86DecodedInsn *decode)
2164{
2165    MemOp ot = decode->op[1].ot;
2166    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2167        gen_repz(s, ot, gen_lods);
2168    } else {
2169        gen_lods(s, ot);
2170    }
2171}
2172
2173static void gen_LOOP(DisasContext *s, X86DecodedInsn *decode)
2174{
2175    TCGLabel *taken = gen_new_label();
2176
2177    gen_update_cc_op(s);
2178    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2179    gen_op_jnz_ecx(s, taken);
2180    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2181}
2182
2183static void gen_LOOPE(DisasContext *s, X86DecodedInsn *decode)
2184{
2185    TCGLabel *taken = gen_new_label();
2186    TCGLabel *not_taken = gen_new_label();
2187
2188    gen_update_cc_op(s);
2189    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2190    gen_op_jz_ecx(s, not_taken);
2191    gen_jcc1(s, (JCC_Z << 1), taken); /* jz taken */
2192    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2193}
2194
2195static void gen_LOOPNE(DisasContext *s, X86DecodedInsn *decode)
2196{
2197    TCGLabel *taken = gen_new_label();
2198    TCGLabel *not_taken = gen_new_label();
2199
2200    gen_update_cc_op(s);
2201    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2202    gen_op_jz_ecx(s, not_taken);
2203    gen_jcc1(s, (JCC_Z << 1) | 1, taken); /* jnz taken */
2204    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2205}
2206
2207static void gen_LSL(DisasContext *s, X86DecodedInsn *decode)
2208{
2209    MemOp ot = decode->op[0].ot;
2210    TCGv result = tcg_temp_new();
2211    TCGv dest;
2212
2213    gen_compute_eflags(s);
2214    gen_update_cc_op(s);
2215    gen_helper_lsl(result, tcg_env, s->T0);
2216
2217    /* Perform writeback here to skip it if ZF=0.  */
2218    decode->op[0].unit = X86_OP_SKIP;
2219    dest = gen_op_deposit_reg_v(s, ot, decode->op[0].n, result, result);
2220    tcg_gen_movcond_tl(TCG_COND_TSTNE, dest, cpu_cc_src, tcg_constant_tl(CC_Z),
2221                       result, dest);
2222}
2223
2224static void gen_LSS(DisasContext *s, X86DecodedInsn *decode)
2225{
2226    gen_lxx_seg(s, decode, R_SS);
2227}
2228
2229static void gen_LZCNT(DisasContext *s, X86DecodedInsn *decode)
2230{
2231    MemOp ot = decode->op[0].ot;
2232
2233    /* C bit (cc_src) is defined related to the input.  */
2234    decode->cc_src = tcg_temp_new();
2235    decode->cc_dst = s->T0;
2236    decode->cc_op = CC_OP_BMILGB + ot;
2237    tcg_gen_mov_tl(decode->cc_src, s->T0);
2238
2239    /*
2240     * Reduce the target_ulong result by the number of zeros that
2241     * we expect to find at the top.
2242     */
2243    tcg_gen_clzi_tl(s->T0, s->T0, TARGET_LONG_BITS);
2244    tcg_gen_subi_tl(s->T0, s->T0, TARGET_LONG_BITS - (8 << ot));
2245}
2246
2247static void gen_MFENCE(DisasContext *s, X86DecodedInsn *decode)
2248{
2249    tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
2250}
2251
2252static void gen_MOV(DisasContext *s, X86DecodedInsn *decode)
2253{
2254    /* nothing to do! */
2255}
2256#define gen_NOP gen_MOV
2257
2258static void gen_MASKMOV(DisasContext *s, X86DecodedInsn *decode)
2259{
2260    gen_lea_v_seg(s, cpu_regs[R_EDI], R_DS, s->override);
2261
2262    if (s->prefix & PREFIX_DATA) {
2263        gen_helper_maskmov_xmm(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2264    } else {
2265        gen_helper_maskmov_mmx(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2266    }
2267}
2268
2269static void gen_MOVBE(DisasContext *s, X86DecodedInsn *decode)
2270{
2271    MemOp ot = decode->op[0].ot;
2272
2273    /* M operand type does not load/store */
2274    if (decode->e.op0 == X86_TYPE_M) {
2275        tcg_gen_qemu_st_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2276    } else {
2277        tcg_gen_qemu_ld_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2278    }
2279}
2280
2281static void gen_MOVD_from(DisasContext *s, X86DecodedInsn *decode)
2282{
2283    MemOp ot = decode->op[2].ot;
2284
2285    switch (ot) {
2286    case MO_32:
2287#ifdef TARGET_X86_64
2288        tcg_gen_ld32u_tl(s->T0, tcg_env, decode->op[2].offset);
2289        break;
2290    case MO_64:
2291#endif
2292        tcg_gen_ld_tl(s->T0, tcg_env, decode->op[2].offset);
2293        break;
2294    default:
2295        abort();
2296    }
2297}
2298
2299static void gen_MOVD_to(DisasContext *s, X86DecodedInsn *decode)
2300{
2301    MemOp ot = decode->op[2].ot;
2302    int vec_len = vector_len(s, decode);
2303    int lo_ofs = vector_elem_offset(&decode->op[0], ot, 0);
2304
2305    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2306
2307    switch (ot) {
2308    case MO_32:
2309#ifdef TARGET_X86_64
2310        tcg_gen_st32_tl(s->T1, tcg_env, lo_ofs);
2311        break;
2312    case MO_64:
2313#endif
2314        tcg_gen_st_tl(s->T1, tcg_env, lo_ofs);
2315        break;
2316    default:
2317        g_assert_not_reached();
2318    }
2319}
2320
2321static void gen_MOVDQ(DisasContext *s, X86DecodedInsn *decode)
2322{
2323    gen_store_sse(s, decode, decode->op[2].offset);
2324}
2325
2326static void gen_MOVMSK(DisasContext *s, X86DecodedInsn *decode)
2327{
2328    typeof(gen_helper_movmskps_ymm) *ps, *pd, *fn;
2329    ps = s->vex_l ? gen_helper_movmskps_ymm : gen_helper_movmskps_xmm;
2330    pd = s->vex_l ? gen_helper_movmskpd_ymm : gen_helper_movmskpd_xmm;
2331    fn = s->prefix & PREFIX_DATA ? pd : ps;
2332    fn(s->tmp2_i32, tcg_env, OP_PTR2);
2333    tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
2334}
2335
2336static void gen_MOVQ(DisasContext *s, X86DecodedInsn *decode)
2337{
2338    int vec_len = vector_len(s, decode);
2339    int lo_ofs = vector_elem_offset(&decode->op[0], MO_64, 0);
2340
2341    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
2342    if (decode->op[0].has_ea) {
2343        tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
2344    } else {
2345        /*
2346         * tcg_gen_gvec_dup_i64(MO_64, op0.offset, 8, vec_len, s->tmp1_64) would
2347         * seem to work, but it does not on big-endian platforms; the cleared parts
2348         * are always at higher addresses, but cross-endian emulation inverts the
2349         * byte order so that the cleared parts need to be at *lower* addresses.
2350         * Because oprsz is 8, we see this here even for SSE; but more in general,
2351         * it disqualifies using oprsz < maxsz to emulate VEX128.
2352         */
2353        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2354        tcg_gen_st_i64(s->tmp1_i64, tcg_env, lo_ofs);
2355    }
2356}
2357
2358static void gen_MOVq_dq(DisasContext *s, X86DecodedInsn *decode)
2359{
2360    gen_helper_enter_mmx(tcg_env);
2361    /* Otherwise the same as any other movq.  */
2362    return gen_MOVQ(s, decode);
2363}
2364
2365static void gen_MOVS(DisasContext *s, X86DecodedInsn *decode)
2366{
2367    MemOp ot = decode->op[2].ot;
2368    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2369        gen_repz(s, ot, gen_movs);
2370    } else {
2371        gen_movs(s, ot);
2372    }
2373}
2374
2375static void gen_MUL(DisasContext *s, X86DecodedInsn *decode)
2376{
2377    MemOp ot = decode->op[1].ot;
2378
2379    switch (ot) {
2380    case MO_8:
2381        /* s->T0 already zero-extended */
2382        tcg_gen_ext8u_tl(s->T1, s->T1);
2383        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2384        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2385        tcg_gen_andi_tl(s->T1, s->T0, 0xff00);
2386        decode->cc_dst = s->T0;
2387        decode->cc_src = s->T1;
2388        break;
2389
2390    case MO_16:
2391        /* s->T0 already zero-extended */
2392        tcg_gen_ext16u_tl(s->T1, s->T1);
2393        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2394        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2395        tcg_gen_shri_tl(s->T1, s->T0, 16);
2396        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
2397        decode->cc_dst = s->T0;
2398        decode->cc_src = s->T1;
2399        break;
2400
2401    case MO_32:
2402#ifdef TARGET_X86_64
2403        /* s->T0 already zero-extended */
2404        tcg_gen_ext32u_tl(s->T1, s->T1);
2405        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2406        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
2407        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
2408        decode->cc_dst = cpu_regs[R_EAX];
2409        decode->cc_src = cpu_regs[R_EDX];
2410        break;
2411
2412    case MO_64:
2413#endif
2414        tcg_gen_mulu2_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], s->T0, s->T1);
2415        decode->cc_dst = cpu_regs[R_EAX];
2416        decode->cc_src = cpu_regs[R_EDX];
2417        break;
2418
2419    default:
2420        g_assert_not_reached();
2421    }
2422
2423    decode->cc_op = CC_OP_MULB + ot;
2424}
2425
2426static void gen_MULX(DisasContext *s, X86DecodedInsn *decode)
2427{
2428    MemOp ot = decode->op[0].ot;
2429
2430    /* low part of result in VEX.vvvv, high in MODRM */
2431    switch (ot) {
2432    case MO_32:
2433#ifdef TARGET_X86_64
2434        tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
2435        tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
2436        tcg_gen_mulu2_i32(s->tmp2_i32, s->tmp3_i32,
2437                          s->tmp2_i32, s->tmp3_i32);
2438        tcg_gen_extu_i32_tl(cpu_regs[s->vex_v], s->tmp2_i32);
2439        tcg_gen_extu_i32_tl(s->T0, s->tmp3_i32);
2440        break;
2441
2442    case MO_64:
2443#endif
2444        tcg_gen_mulu2_tl(cpu_regs[s->vex_v], s->T0, s->T0, s->T1);
2445        break;
2446
2447    default:
2448        g_assert_not_reached();
2449    }
2450}
2451
2452static void gen_NEG(DisasContext *s, X86DecodedInsn *decode)
2453{
2454    MemOp ot = decode->op[0].ot;
2455    TCGv oldv = tcg_temp_new();
2456
2457    if (s->prefix & PREFIX_LOCK) {
2458        TCGv newv = tcg_temp_new();
2459        TCGv cmpv = tcg_temp_new();
2460        TCGLabel *label1 = gen_new_label();
2461
2462        gen_set_label(label1);
2463        gen_op_ld_v(s, ot, oldv, s->A0);
2464        tcg_gen_neg_tl(newv, oldv);
2465        tcg_gen_atomic_cmpxchg_tl(cmpv, s->A0, oldv, newv,
2466                                  s->mem_index, ot | MO_LE);
2467        tcg_gen_brcond_tl(TCG_COND_NE, oldv, cmpv, label1);
2468    } else {
2469        tcg_gen_mov_tl(oldv, s->T0);
2470    }
2471    tcg_gen_neg_tl(s->T0, oldv);
2472
2473    decode->cc_dst = s->T0;
2474    decode->cc_src = oldv;
2475    tcg_gen_movi_tl(s->cc_srcT, 0);
2476    decode->cc_op = CC_OP_SUBB + ot;
2477}
2478
2479static void gen_NOT(DisasContext *s, X86DecodedInsn *decode)
2480{
2481    MemOp ot = decode->op[0].ot;
2482
2483    if (s->prefix & PREFIX_LOCK) {
2484        tcg_gen_movi_tl(s->T0, ~0);
2485        tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T0,
2486                                    s->mem_index, ot | MO_LE);
2487    } else {
2488        tcg_gen_not_tl(s->T0, s->T0);
2489    }
2490}
2491
2492static void gen_OR(DisasContext *s, X86DecodedInsn *decode)
2493{
2494    MemOp ot = decode->op[1].ot;
2495
2496    if (s->prefix & PREFIX_LOCK) {
2497        tcg_gen_atomic_or_fetch_tl(s->T0, s->A0, s->T1,
2498                                   s->mem_index, ot | MO_LE);
2499    } else {
2500        tcg_gen_or_tl(s->T0, s->T0, s->T1);
2501    }
2502    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
2503}
2504
2505static void gen_OUT(DisasContext *s, X86DecodedInsn *decode)
2506{
2507    MemOp ot = decode->op[1].ot;
2508    TCGv_i32 port = tcg_temp_new_i32();
2509    TCGv_i32 value = tcg_temp_new_i32();
2510
2511    tcg_gen_trunc_tl_i32(port, s->T1);
2512    tcg_gen_ext16u_i32(port, port);
2513    if (!gen_check_io(s, ot, port, 0)) {
2514        return;
2515    }
2516    tcg_gen_trunc_tl_i32(value, s->T0);
2517    translator_io_start(&s->base);
2518    gen_helper_out_func(ot, port, value);
2519    gen_bpt_io(s, port, ot);
2520}
2521
2522static void gen_OUTS(DisasContext *s, X86DecodedInsn *decode)
2523{
2524    MemOp ot = decode->op[1].ot;
2525    TCGv_i32 port = tcg_temp_new_i32();
2526
2527    tcg_gen_trunc_tl_i32(port, s->T1);
2528    tcg_gen_ext16u_i32(port, port);
2529    if (!gen_check_io(s, ot, port, SVM_IOIO_STR_MASK)) {
2530        return;
2531    }
2532
2533    translator_io_start(&s->base);
2534    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2535        gen_repz(s, ot, gen_outs);
2536    } else {
2537        gen_outs(s, ot);
2538    }
2539}
2540
2541static void gen_PALIGNR(DisasContext *s, X86DecodedInsn *decode)
2542{
2543    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2544    if (!(s->prefix & PREFIX_DATA)) {
2545        gen_helper_palignr_mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2546    } else if (!s->vex_l) {
2547        gen_helper_palignr_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2548    } else {
2549        gen_helper_palignr_ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2550    }
2551}
2552
2553static void gen_PANDN(DisasContext *s, X86DecodedInsn *decode)
2554{
2555    int vec_len = vector_len(s, decode);
2556
2557    /* Careful, operand order is reversed!  */
2558    tcg_gen_gvec_andc(MO_64,
2559                      decode->op[0].offset, decode->op[2].offset,
2560                      decode->op[1].offset, vec_len, vec_len);
2561}
2562
2563static void gen_PAUSE(DisasContext *s, X86DecodedInsn *decode)
2564{
2565    gen_update_cc_op(s);
2566    gen_update_eip_next(s);
2567    gen_helper_pause(tcg_env);
2568    s->base.is_jmp = DISAS_NORETURN;
2569}
2570
2571static void gen_PCMPESTRI(DisasContext *s, X86DecodedInsn *decode)
2572{
2573    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2574    gen_helper_pcmpestri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2575    assume_cc_op(s, CC_OP_EFLAGS);
2576}
2577
2578static void gen_PCMPESTRM(DisasContext *s, X86DecodedInsn *decode)
2579{
2580    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2581    gen_helper_pcmpestrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2582    assume_cc_op(s, CC_OP_EFLAGS);
2583    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2584        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2585                             16, 16, 0);
2586    }
2587}
2588
2589static void gen_PCMPISTRI(DisasContext *s, X86DecodedInsn *decode)
2590{
2591    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2592    gen_helper_pcmpistri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2593    assume_cc_op(s, CC_OP_EFLAGS);
2594}
2595
2596static void gen_PCMPISTRM(DisasContext *s, X86DecodedInsn *decode)
2597{
2598    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2599    gen_helper_pcmpistrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2600    assume_cc_op(s, CC_OP_EFLAGS);
2601    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2602        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2603                             16, 16, 0);
2604    }
2605}
2606
2607static void gen_PDEP(DisasContext *s, X86DecodedInsn *decode)
2608{
2609    gen_helper_pdep(s->T0, s->T0, s->T1);
2610}
2611
2612static void gen_PEXT(DisasContext *s, X86DecodedInsn *decode)
2613{
2614    gen_helper_pext(s->T0, s->T0, s->T1);
2615}
2616
2617static inline void gen_pextr(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
2618{
2619    int vec_len = vector_len(s, decode);
2620    int mask = (vec_len >> ot) - 1;
2621    int val = decode->immediate & mask;
2622
2623    switch (ot) {
2624    case MO_8:
2625        tcg_gen_ld8u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2626        break;
2627    case MO_16:
2628        tcg_gen_ld16u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2629        break;
2630    case MO_32:
2631#ifdef TARGET_X86_64
2632        tcg_gen_ld32u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2633        break;
2634    case MO_64:
2635#endif
2636        tcg_gen_ld_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2637        break;
2638    default:
2639        abort();
2640    }
2641}
2642
2643static void gen_PEXTRB(DisasContext *s, X86DecodedInsn *decode)
2644{
2645    gen_pextr(s, decode, MO_8);
2646}
2647
2648static void gen_PEXTRW(DisasContext *s, X86DecodedInsn *decode)
2649{
2650    gen_pextr(s, decode, MO_16);
2651}
2652
2653static void gen_PEXTR(DisasContext *s, X86DecodedInsn *decode)
2654{
2655    MemOp ot = decode->op[0].ot;
2656    gen_pextr(s, decode, ot);
2657}
2658
2659static inline void gen_pinsr(DisasContext *s, X86DecodedInsn *decode, MemOp ot)
2660{
2661    int vec_len = vector_len(s, decode);
2662    int mask = (vec_len >> ot) - 1;
2663    int val = decode->immediate & mask;
2664
2665    if (decode->op[1].offset != decode->op[0].offset) {
2666        assert(vec_len == 16);
2667        gen_store_sse(s, decode, decode->op[1].offset);
2668    }
2669
2670    switch (ot) {
2671    case MO_8:
2672        tcg_gen_st8_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2673        break;
2674    case MO_16:
2675        tcg_gen_st16_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2676        break;
2677    case MO_32:
2678#ifdef TARGET_X86_64
2679        tcg_gen_st32_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2680        break;
2681    case MO_64:
2682#endif
2683        tcg_gen_st_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2684        break;
2685    default:
2686        abort();
2687    }
2688}
2689
2690static void gen_PINSRB(DisasContext *s, X86DecodedInsn *decode)
2691{
2692    gen_pinsr(s, decode, MO_8);
2693}
2694
2695static void gen_PINSRW(DisasContext *s, X86DecodedInsn *decode)
2696{
2697    gen_pinsr(s, decode, MO_16);
2698}
2699
2700static void gen_PINSR(DisasContext *s, X86DecodedInsn *decode)
2701{
2702    gen_pinsr(s, decode, decode->op[2].ot);
2703}
2704
2705static void gen_pmovmskb_i64(TCGv_i64 d, TCGv_i64 s)
2706{
2707    TCGv_i64 t = tcg_temp_new_i64();
2708
2709    tcg_gen_andi_i64(d, s, 0x8080808080808080ull);
2710
2711    /*
2712     * After each shift+or pair:
2713     * 0:  a.......b.......c.......d.......e.......f.......g.......h.......
2714     * 7:  ab......bc......cd......de......ef......fg......gh......h.......
2715     * 14: abcd....bcde....cdef....defg....efgh....fgh.....gh......h.......
2716     * 28: abcdefghbcdefgh.cdefgh..defgh...efgh....fgh.....gh......h.......
2717     * The result is left in the high bits of the word.
2718     */
2719    tcg_gen_shli_i64(t, d, 7);
2720    tcg_gen_or_i64(d, d, t);
2721    tcg_gen_shli_i64(t, d, 14);
2722    tcg_gen_or_i64(d, d, t);
2723    tcg_gen_shli_i64(t, d, 28);
2724    tcg_gen_or_i64(d, d, t);
2725}
2726
2727static void gen_pmovmskb_vec(unsigned vece, TCGv_vec d, TCGv_vec s)
2728{
2729    TCGv_vec t = tcg_temp_new_vec_matching(d);
2730    TCGv_vec m = tcg_constant_vec_matching(d, MO_8, 0x80);
2731
2732    /* See above */
2733    tcg_gen_and_vec(vece, d, s, m);
2734    tcg_gen_shli_vec(vece, t, d, 7);
2735    tcg_gen_or_vec(vece, d, d, t);
2736    tcg_gen_shli_vec(vece, t, d, 14);
2737    tcg_gen_or_vec(vece, d, d, t);
2738    tcg_gen_shli_vec(vece, t, d, 28);
2739    tcg_gen_or_vec(vece, d, d, t);
2740}
2741
2742static void gen_PMOVMSKB(DisasContext *s, X86DecodedInsn *decode)
2743{
2744    static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
2745    static const GVecGen2 g = {
2746        .fni8 = gen_pmovmskb_i64,
2747        .fniv = gen_pmovmskb_vec,
2748        .opt_opc = vecop_list,
2749        .vece = MO_64,
2750        .prefer_i64 = TCG_TARGET_REG_BITS == 64
2751    };
2752    MemOp ot = decode->op[2].ot;
2753    int vec_len = vector_len(s, decode);
2754    TCGv t = tcg_temp_new();
2755
2756    tcg_gen_gvec_2(offsetof(CPUX86State, xmm_t0) + xmm_offset(ot), decode->op[2].offset,
2757                   vec_len, vec_len, &g);
2758    tcg_gen_ld8u_tl(s->T0, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
2759    while (vec_len > 8) {
2760        vec_len -= 8;
2761        if (TCG_TARGET_HAS_extract2_tl) {
2762            /*
2763             * Load the next byte of the result into the high byte of T.
2764             * TCG does a similar expansion of deposit to shl+extract2; by
2765             * loading the whole word, the shift left is avoided.
2766             */
2767#ifdef TARGET_X86_64
2768            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_Q((vec_len - 1) / 8)));
2769#else
2770            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L((vec_len - 1) / 4)));
2771#endif
2772
2773            tcg_gen_extract2_tl(s->T0, t, s->T0, TARGET_LONG_BITS - 8);
2774        } else {
2775            /*
2776             * The _previous_ value is deposited into bits 8 and higher of t.  Because
2777             * those bits are known to be zero after ld8u, this becomes a shift+or
2778             * if deposit is not available.
2779             */
2780            tcg_gen_ld8u_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
2781            tcg_gen_deposit_tl(s->T0, t, s->T0, 8, TARGET_LONG_BITS - 8);
2782        }
2783    }
2784}
2785
2786static void gen_POP(DisasContext *s, X86DecodedInsn *decode)
2787{
2788    X86DecodedOp *op = &decode->op[0];
2789    MemOp ot = gen_pop_T0(s);
2790
2791    if (op->has_ea || op->unit == X86_OP_SEG) {
2792        /* NOTE: order is important for MMU exceptions */
2793        gen_writeback(s, decode, 0, s->T0);
2794    }
2795
2796    /* NOTE: writing back registers after update is important for pop %sp */
2797    gen_pop_update(s, ot);
2798}
2799
2800static void gen_POPA(DisasContext *s, X86DecodedInsn *decode)
2801{
2802    gen_popa(s);
2803}
2804
2805static void gen_POPCNT(DisasContext *s, X86DecodedInsn *decode)
2806{
2807    decode->cc_dst = tcg_temp_new();
2808    decode->cc_op = CC_OP_POPCNT;
2809
2810    tcg_gen_mov_tl(decode->cc_dst, s->T0);
2811    tcg_gen_ctpop_tl(s->T0, s->T0);
2812}
2813
2814static void gen_POPF(DisasContext *s, X86DecodedInsn *decode)
2815{
2816    MemOp ot;
2817    int mask = TF_MASK | AC_MASK | ID_MASK | NT_MASK;
2818
2819    if (CPL(s) == 0) {
2820        mask |= IF_MASK | IOPL_MASK;
2821    } else if (CPL(s) <= IOPL(s)) {
2822        mask |= IF_MASK;
2823    }
2824    if (s->dflag == MO_16) {
2825        mask &= 0xffff;
2826    }
2827
2828    ot = gen_pop_T0(s);
2829    gen_helper_write_eflags(tcg_env, s->T0, tcg_constant_i32(mask));
2830    gen_pop_update(s, ot);
2831    set_cc_op(s, CC_OP_EFLAGS);
2832    /* abort translation because TF/AC flag may change */
2833    s->base.is_jmp = DISAS_EOB_NEXT;
2834}
2835
2836static void gen_PSHUFW(DisasContext *s, X86DecodedInsn *decode)
2837{
2838    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2839    gen_helper_pshufw_mmx(OP_PTR0, OP_PTR1, imm);
2840}
2841
2842static void gen_PSRLW_i(DisasContext *s, X86DecodedInsn *decode)
2843{
2844    int vec_len = vector_len(s, decode);
2845
2846    if (decode->immediate >= 16) {
2847        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2848    } else {
2849        tcg_gen_gvec_shri(MO_16,
2850                          decode->op[0].offset, decode->op[1].offset,
2851                          decode->immediate, vec_len, vec_len);
2852    }
2853}
2854
2855static void gen_PSLLW_i(DisasContext *s, X86DecodedInsn *decode)
2856{
2857    int vec_len = vector_len(s, decode);
2858
2859    if (decode->immediate >= 16) {
2860        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2861    } else {
2862        tcg_gen_gvec_shli(MO_16,
2863                          decode->op[0].offset, decode->op[1].offset,
2864                          decode->immediate, vec_len, vec_len);
2865    }
2866}
2867
2868static void gen_PSRAW_i(DisasContext *s, X86DecodedInsn *decode)
2869{
2870    int vec_len = vector_len(s, decode);
2871
2872    if (decode->immediate >= 16) {
2873        decode->immediate = 15;
2874    }
2875    tcg_gen_gvec_sari(MO_16,
2876                      decode->op[0].offset, decode->op[1].offset,
2877                      decode->immediate, vec_len, vec_len);
2878}
2879
2880static void gen_PSRLD_i(DisasContext *s, X86DecodedInsn *decode)
2881{
2882    int vec_len = vector_len(s, decode);
2883
2884    if (decode->immediate >= 32) {
2885        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2886    } else {
2887        tcg_gen_gvec_shri(MO_32,
2888                          decode->op[0].offset, decode->op[1].offset,
2889                          decode->immediate, vec_len, vec_len);
2890    }
2891}
2892
2893static void gen_PSLLD_i(DisasContext *s, X86DecodedInsn *decode)
2894{
2895    int vec_len = vector_len(s, decode);
2896
2897    if (decode->immediate >= 32) {
2898        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2899    } else {
2900        tcg_gen_gvec_shli(MO_32,
2901                          decode->op[0].offset, decode->op[1].offset,
2902                          decode->immediate, vec_len, vec_len);
2903    }
2904}
2905
2906static void gen_PSRAD_i(DisasContext *s, X86DecodedInsn *decode)
2907{
2908    int vec_len = vector_len(s, decode);
2909
2910    if (decode->immediate >= 32) {
2911        decode->immediate = 31;
2912    }
2913    tcg_gen_gvec_sari(MO_32,
2914                      decode->op[0].offset, decode->op[1].offset,
2915                      decode->immediate, vec_len, vec_len);
2916}
2917
2918static void gen_PSRLQ_i(DisasContext *s, X86DecodedInsn *decode)
2919{
2920    int vec_len = vector_len(s, decode);
2921
2922    if (decode->immediate >= 64) {
2923        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2924    } else {
2925        tcg_gen_gvec_shri(MO_64,
2926                          decode->op[0].offset, decode->op[1].offset,
2927                          decode->immediate, vec_len, vec_len);
2928    }
2929}
2930
2931static void gen_PSLLQ_i(DisasContext *s, X86DecodedInsn *decode)
2932{
2933    int vec_len = vector_len(s, decode);
2934
2935    if (decode->immediate >= 64) {
2936        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2937    } else {
2938        tcg_gen_gvec_shli(MO_64,
2939                          decode->op[0].offset, decode->op[1].offset,
2940                          decode->immediate, vec_len, vec_len);
2941    }
2942}
2943
2944static TCGv_ptr make_imm8u_xmm_vec(uint8_t imm, int vec_len)
2945{
2946    MemOp ot = vec_len == 16 ? MO_128 : MO_256;
2947    TCGv_i32 imm_v = tcg_constant8u_i32(imm);
2948    TCGv_ptr ptr = tcg_temp_new_ptr();
2949
2950    tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_t0) + xmm_offset(ot),
2951                         vec_len, vec_len, 0);
2952
2953    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_t0));
2954    tcg_gen_st_i32(imm_v, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L(0)));
2955    return ptr;
2956}
2957
2958static void gen_PSRLDQ_i(DisasContext *s, X86DecodedInsn *decode)
2959{
2960    int vec_len = vector_len(s, decode);
2961    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
2962
2963    if (s->vex_l) {
2964        gen_helper_psrldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
2965    } else {
2966        gen_helper_psrldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
2967    }
2968}
2969
2970static void gen_PSLLDQ_i(DisasContext *s, X86DecodedInsn *decode)
2971{
2972    int vec_len = vector_len(s, decode);
2973    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
2974
2975    if (s->vex_l) {
2976        gen_helper_pslldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
2977    } else {
2978        gen_helper_pslldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
2979    }
2980}
2981
2982static void gen_PUSH(DisasContext *s, X86DecodedInsn *decode)
2983{
2984    gen_push_v(s, s->T0);
2985}
2986
2987static void gen_PUSHA(DisasContext *s, X86DecodedInsn *decode)
2988{
2989    gen_pusha(s);
2990}
2991
2992static void gen_PUSHF(DisasContext *s, X86DecodedInsn *decode)
2993{
2994    gen_update_cc_op(s);
2995    gen_helper_read_eflags(s->T0, tcg_env);
2996    gen_push_v(s, s->T0);
2997}
2998
2999static MemOp gen_shift_count(DisasContext *s, X86DecodedInsn *decode,
3000                             bool *can_be_zero, TCGv *count, int unit)
3001{
3002    MemOp ot = decode->op[0].ot;
3003    int mask = (ot <= MO_32 ? 0x1f : 0x3f);
3004
3005    *can_be_zero = false;
3006    switch (unit) {
3007    case X86_OP_INT:
3008        *count = tcg_temp_new();
3009        tcg_gen_andi_tl(*count, cpu_regs[R_ECX], mask);
3010        *can_be_zero = true;
3011        break;
3012
3013    case X86_OP_IMM:
3014        if ((decode->immediate & mask) == 0) {
3015            *count = NULL;
3016            break;
3017        }
3018        *count = tcg_temp_new();
3019        tcg_gen_movi_tl(*count, decode->immediate & mask);
3020        break;
3021
3022    case X86_OP_SKIP:
3023        *count = tcg_temp_new();
3024        tcg_gen_movi_tl(*count, 1);
3025        break;
3026
3027    default:
3028        g_assert_not_reached();
3029    }
3030
3031    return ot;
3032}
3033
3034/*
3035 * Compute existing flags in decode->cc_src, for gen_* functions that wants
3036 * to set the cc_op set to CC_OP_ADCOX.  In particular, this allows rotate
3037 * operations to compute the carry in decode->cc_dst and the overflow in
3038 * decode->cc_src2.
3039 *
3040 * If need_flags is true, decode->cc_dst and decode->cc_src2 are preloaded
3041 * with the value of CF and OF before the instruction, so that it is possible
3042 * to keep the flags unmodified.
3043 *
3044 * Return true if carry could be made available cheaply as a 1-bit value in
3045 * decode->cc_dst (trying a bit harder if want_carry is true).  If false is
3046 * returned, decode->cc_dst is uninitialized and the carry is only available
3047 * as bit 0 of decode->cc_src.
3048 */
3049static bool gen_eflags_adcox(DisasContext *s, X86DecodedInsn *decode, bool want_carry, bool need_flags)
3050{
3051    bool got_cf = false;
3052    bool got_of = false;
3053
3054    decode->cc_dst = tcg_temp_new();
3055    decode->cc_src = tcg_temp_new();
3056    decode->cc_src2 = tcg_temp_new();
3057    decode->cc_op = CC_OP_ADCOX;
3058
3059    /* A lot more cc_ops could be "optimized" to avoid the extracts at
3060     * the end (INC/DEC, BMILG, MUL), but they are all really unlikely
3061     * to be followed by rotations within the same basic block.
3062     */
3063    switch (s->cc_op) {
3064    case CC_OP_ADCOX:
3065        /* No need to compute the full EFLAGS, CF/OF are already isolated.  */
3066        tcg_gen_mov_tl(decode->cc_src, cpu_cc_src);
3067        if (need_flags) {
3068            tcg_gen_mov_tl(decode->cc_src2, cpu_cc_src2);
3069            got_of = true;
3070        }
3071        if (want_carry || need_flags) {
3072            tcg_gen_mov_tl(decode->cc_dst, cpu_cc_dst);
3073            got_cf = true;
3074        }
3075        break;
3076
3077    case CC_OP_LOGICB ... CC_OP_LOGICQ:
3078        /* CF and OF are zero, do it just because it's easy.  */
3079        gen_mov_eflags(s, decode->cc_src);
3080        if (need_flags) {
3081            tcg_gen_movi_tl(decode->cc_src2, 0);
3082            got_of = true;
3083        }
3084        if (want_carry || need_flags) {
3085            tcg_gen_movi_tl(decode->cc_dst, 0);
3086            got_cf = true;
3087        }
3088        break;
3089
3090    case CC_OP_SARB ... CC_OP_SARQ:
3091        /*
3092         * SHR/RCR/SHR/RCR/... is a relatively common occurrence of RCR.
3093         * By computing CF without using eflags, the calls to cc_compute_all
3094         * can be eliminated as dead code (except for the last RCR).
3095         */
3096        if (want_carry || need_flags) {
3097            tcg_gen_andi_tl(decode->cc_dst, cpu_cc_src, 1);
3098            got_cf = true;
3099        }
3100        gen_mov_eflags(s, decode->cc_src);
3101        break;
3102
3103    case CC_OP_SHLB ... CC_OP_SHLQ:
3104        /*
3105         * Likewise for SHL/RCL/SHL/RCL/... but, if CF is not in the sign
3106         * bit, we might as well fish CF out of EFLAGS and save a shift.
3107         */
3108        if (want_carry && (!need_flags || s->cc_op == CC_OP_SHLB + MO_TL)) {
3109            tcg_gen_shri_tl(decode->cc_dst, cpu_cc_src, (8 << (s->cc_op - CC_OP_SHLB)) - 1);
3110            got_cf = true;
3111        }
3112        gen_mov_eflags(s, decode->cc_src);
3113        break;
3114
3115    default:
3116        gen_mov_eflags(s, decode->cc_src);
3117        break;
3118    }
3119
3120    if (need_flags) {
3121        /* If the flags could be left unmodified, always load them.  */
3122        if (!got_of) {
3123            tcg_gen_extract_tl(decode->cc_src2, decode->cc_src, ctz32(CC_O), 1);
3124            got_of = true;
3125        }
3126        if (!got_cf) {
3127            tcg_gen_extract_tl(decode->cc_dst, decode->cc_src, ctz32(CC_C), 1);
3128            got_cf = true;
3129        }
3130    }
3131    return got_cf;
3132}
3133
3134static void gen_rot_overflow(X86DecodedInsn *decode, TCGv result, TCGv old,
3135                             bool can_be_zero, TCGv count)
3136{
3137    MemOp ot = decode->op[0].ot;
3138    TCGv temp = can_be_zero ? tcg_temp_new() : decode->cc_src2;
3139
3140    tcg_gen_xor_tl(temp, old, result);
3141    tcg_gen_extract_tl(temp, temp, (8 << ot) - 1, 1);
3142    if (can_be_zero) {
3143        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src2, count, tcg_constant_tl(0),
3144                           decode->cc_src2, temp);
3145    }
3146}
3147
3148/*
3149 * RCx operations are invariant modulo 8*operand_size+1.  For 8 and 16-bit operands,
3150 * this is less than 0x1f (the mask applied by gen_shift_count) so reduce further.
3151 */
3152static void gen_rotc_mod(MemOp ot, TCGv count)
3153{
3154    TCGv temp;
3155
3156    switch (ot) {
3157    case MO_8:
3158        temp = tcg_temp_new();
3159        tcg_gen_subi_tl(temp, count, 18);
3160        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3161        tcg_gen_subi_tl(temp, count, 9);
3162        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3163        break;
3164
3165    case MO_16:
3166        temp = tcg_temp_new();
3167        tcg_gen_subi_tl(temp, count, 17);
3168        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
3169        break;
3170
3171    default:
3172        break;
3173    }
3174}
3175
3176/*
3177 * The idea here is that the bit to the right of the new bit 0 is the
3178 * new carry, and the bit to the right of the old bit 0 is the old carry.
3179 * Just like a regular rotation, the result of the rotation is composed
3180 * from a right shifted part and a left shifted part of s->T0.  The new carry
3181 * is extracted from the right-shifted portion, and the old carry is
3182 * inserted at the end of the left-shifted portion.
3183 *
3184 * Because of the separate shifts involving the carry, gen_RCL and gen_RCR
3185 * mostly operate on count-1.  This also comes in handy when computing
3186 * length - count, because (length-1) - (count-1) can be computed with
3187 * a XOR, and that is commutative unlike subtraction.
3188 */
3189static void gen_RCL(DisasContext *s, X86DecodedInsn *decode)
3190{
3191    bool have_1bit_cin, can_be_zero;
3192    TCGv count;
3193    TCGLabel *zero_label = NULL;
3194    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3195    TCGv low, high, low_count;
3196
3197    if (!count) {
3198        return;
3199    }
3200
3201    low = tcg_temp_new();
3202    high = tcg_temp_new();
3203    low_count = tcg_temp_new();
3204
3205    gen_rotc_mod(ot, count);
3206    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3207    if (can_be_zero) {
3208        zero_label = gen_new_label();
3209        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3210    }
3211
3212    /* Compute high part, including incoming carry.  */
3213    if (!have_1bit_cin || TCG_TARGET_deposit_tl_valid(1, TARGET_LONG_BITS - 1)) {
3214        /* high = (T0 << 1) | cin */
3215        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3216        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3217    } else {
3218        /* Same as above but without deposit; cin in cc_dst.  */
3219        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3220        tcg_gen_add_tl(high, high, s->T0);
3221    }
3222    tcg_gen_subi_tl(count, count, 1);
3223    tcg_gen_shl_tl(high, high, count);
3224
3225    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3226    tcg_gen_xori_tl(low_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3227    tcg_gen_shr_tl(low, s->T0, low_count);
3228    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3229    tcg_gen_shri_tl(low, low, 1);
3230
3231    /* Compute result and outgoing overflow */
3232    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3233    tcg_gen_or_tl(s->T0, low, high);
3234    gen_rot_overflow(decode, s->T0, decode->cc_src2, false, NULL);
3235
3236    if (zero_label) {
3237        gen_set_label(zero_label);
3238    }
3239}
3240
3241static void gen_RCR(DisasContext *s, X86DecodedInsn *decode)
3242{
3243    bool have_1bit_cin, can_be_zero;
3244    TCGv count;
3245    TCGLabel *zero_label = NULL;
3246    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3247    TCGv low, high, high_count;
3248
3249    if (!count) {
3250        return;
3251    }
3252
3253    low = tcg_temp_new();
3254    high = tcg_temp_new();
3255    high_count = tcg_temp_new();
3256
3257    gen_rotc_mod(ot, count);
3258    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3259    if (can_be_zero) {
3260        zero_label = gen_new_label();
3261        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3262    }
3263
3264    /* Save incoming carry into high, it will be shifted later.  */
3265    if (!have_1bit_cin || TCG_TARGET_deposit_tl_valid(1, TARGET_LONG_BITS - 1)) {
3266        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3267        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3268    } else {
3269        /* Same as above but without deposit; cin in cc_dst.  */
3270        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3271        tcg_gen_add_tl(high, high, s->T0);
3272    }
3273
3274    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3275    tcg_gen_subi_tl(count, count, 1);
3276    tcg_gen_shr_tl(low, s->T0, count);
3277    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3278    tcg_gen_shri_tl(low, low, 1);
3279
3280    /* Move high part to the right position */
3281    tcg_gen_xori_tl(high_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3282    tcg_gen_shl_tl(high, high, high_count);
3283
3284    /* Compute result and outgoing overflow */
3285    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3286    tcg_gen_or_tl(s->T0, low, high);
3287    gen_rot_overflow(decode, s->T0, decode->cc_src2, false, NULL);
3288
3289    if (zero_label) {
3290        gen_set_label(zero_label);
3291    }
3292}
3293
3294#ifdef CONFIG_USER_ONLY
3295static void gen_unreachable(DisasContext *s, X86DecodedInsn *decode)
3296{
3297    g_assert_not_reached();
3298}
3299#endif
3300
3301#ifndef CONFIG_USER_ONLY
3302static void gen_RDMSR(DisasContext *s, X86DecodedInsn *decode)
3303{
3304    gen_update_cc_op(s);
3305    gen_update_eip_cur(s);
3306    gen_helper_rdmsr(tcg_env);
3307}
3308#else
3309#define gen_RDMSR gen_unreachable
3310#endif
3311
3312static void gen_RDPMC(DisasContext *s, X86DecodedInsn *decode)
3313{
3314    gen_update_cc_op(s);
3315    gen_update_eip_cur(s);
3316    translator_io_start(&s->base);
3317    gen_helper_rdpmc(tcg_env);
3318    s->base.is_jmp = DISAS_NORETURN;
3319}
3320
3321static void gen_RDTSC(DisasContext *s, X86DecodedInsn *decode)
3322{
3323    gen_update_cc_op(s);
3324    gen_update_eip_cur(s);
3325    translator_io_start(&s->base);
3326    gen_helper_rdtsc(tcg_env);
3327}
3328
3329static void gen_RDxxBASE(DisasContext *s, X86DecodedInsn *decode)
3330{
3331    TCGv base = cpu_seg_base[s->modrm & 8 ? R_GS : R_FS];
3332
3333    /* Preserve hflags bits by testing CR4 at runtime.  */
3334    gen_helper_cr4_testbit(tcg_env, tcg_constant_i32(CR4_FSGSBASE_MASK));
3335    tcg_gen_mov_tl(s->T0, base);
3336}
3337
3338static void gen_RET(DisasContext *s, X86DecodedInsn *decode)
3339{
3340    int16_t adjust = decode->e.op1 == X86_TYPE_I ? decode->immediate : 0;
3341
3342    MemOp ot = gen_pop_T0(s);
3343    gen_stack_update(s, adjust + (1 << ot));
3344    gen_op_jmp_v(s, s->T0);
3345    gen_bnd_jmp(s);
3346    s->base.is_jmp = DISAS_JUMP;
3347}
3348
3349static void gen_RETF(DisasContext *s, X86DecodedInsn *decode)
3350{
3351    int16_t adjust = decode->e.op1 == X86_TYPE_I ? decode->immediate : 0;
3352
3353    if (!PE(s) || VM86(s)) {
3354        gen_lea_ss_ofs(s, s->A0, cpu_regs[R_ESP], 0);
3355        /* pop offset */
3356        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3357        /* NOTE: keeping EIP updated is not a problem in case of
3358           exception */
3359        gen_op_jmp_v(s, s->T0);
3360        /* pop selector */
3361        gen_add_A0_im(s, 1 << s->dflag);
3362        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3363        gen_op_movl_seg_real(s, R_CS, s->T0);
3364        /* add stack offset */
3365        gen_stack_update(s, adjust + (2 << s->dflag));
3366    } else {
3367        gen_update_cc_op(s);
3368        gen_update_eip_cur(s);
3369        gen_helper_lret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
3370                                  tcg_constant_i32(adjust));
3371    }
3372    s->base.is_jmp = DISAS_EOB_ONLY;
3373}
3374
3375/*
3376 * Return non-NULL if a 32-bit rotate works, after possibly replicating the input.
3377 * The input has already been zero-extended upon operand decode.
3378 */
3379static TCGv_i32 gen_rot_replicate(MemOp ot, TCGv in)
3380{
3381    TCGv_i32 temp;
3382    switch (ot) {
3383    case MO_8:
3384        temp = tcg_temp_new_i32();
3385        tcg_gen_trunc_tl_i32(temp, in);
3386        tcg_gen_muli_i32(temp, temp, 0x01010101);
3387        return temp;
3388
3389    case MO_16:
3390        temp = tcg_temp_new_i32();
3391        tcg_gen_trunc_tl_i32(temp, in);
3392        tcg_gen_deposit_i32(temp, temp, temp, 16, 16);
3393        return temp;
3394
3395#ifdef TARGET_X86_64
3396    case MO_32:
3397        temp = tcg_temp_new_i32();
3398        tcg_gen_trunc_tl_i32(temp, in);
3399        return temp;
3400#endif
3401
3402    default:
3403        return NULL;
3404    }
3405}
3406
3407static void gen_rot_carry(X86DecodedInsn *decode, TCGv result,
3408                          bool can_be_zero, TCGv count, int bit)
3409{
3410    if (!can_be_zero) {
3411        tcg_gen_extract_tl(decode->cc_dst, result, bit, 1);
3412    } else {
3413        TCGv temp = tcg_temp_new();
3414        tcg_gen_extract_tl(temp, result, bit, 1);
3415        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3416                           decode->cc_dst, temp);
3417    }
3418}
3419
3420static void gen_ROL(DisasContext *s, X86DecodedInsn *decode)
3421{
3422    bool can_be_zero;
3423    TCGv count;
3424    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3425    TCGv_i32 temp32, count32;
3426    TCGv old = tcg_temp_new();
3427
3428    if (!count) {
3429        return;
3430    }
3431
3432    gen_eflags_adcox(s, decode, false, can_be_zero);
3433    tcg_gen_mov_tl(old, s->T0);
3434    temp32 = gen_rot_replicate(ot, s->T0);
3435    if (temp32) {
3436        count32 = tcg_temp_new_i32();
3437        tcg_gen_trunc_tl_i32(count32, count);
3438        tcg_gen_rotl_i32(temp32, temp32, count32);
3439        /* Zero extend to facilitate later optimization.  */
3440        tcg_gen_extu_i32_tl(s->T0, temp32);
3441    } else {
3442        tcg_gen_rotl_tl(s->T0, s->T0, count);
3443    }
3444    gen_rot_carry(decode, s->T0, can_be_zero, count, 0);
3445    gen_rot_overflow(decode, s->T0, old, can_be_zero, count);
3446}
3447
3448static void gen_ROR(DisasContext *s, X86DecodedInsn *decode)
3449{
3450    bool can_be_zero;
3451    TCGv count;
3452    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3453    TCGv_i32 temp32, count32;
3454    TCGv old = tcg_temp_new();
3455
3456    if (!count) {
3457        return;
3458    }
3459
3460    gen_eflags_adcox(s, decode, false, can_be_zero);
3461    tcg_gen_mov_tl(old, s->T0);
3462    temp32 = gen_rot_replicate(ot, s->T0);
3463    if (temp32) {
3464        count32 = tcg_temp_new_i32();
3465        tcg_gen_trunc_tl_i32(count32, count);
3466        tcg_gen_rotr_i32(temp32, temp32, count32);
3467        /* Zero extend to facilitate later optimization.  */
3468        tcg_gen_extu_i32_tl(s->T0, temp32);
3469        gen_rot_carry(decode, s->T0, can_be_zero, count, 31);
3470    } else {
3471        tcg_gen_rotr_tl(s->T0, s->T0, count);
3472        gen_rot_carry(decode, s->T0, can_be_zero, count, TARGET_LONG_BITS - 1);
3473    }
3474    gen_rot_overflow(decode, s->T0, old, can_be_zero, count);
3475}
3476
3477static void gen_RORX(DisasContext *s, X86DecodedInsn *decode)
3478{
3479    MemOp ot = decode->op[0].ot;
3480    int mask = ot == MO_64 ? 63 : 31;
3481    int b = decode->immediate & mask;
3482
3483    switch (ot) {
3484    case MO_32:
3485#ifdef TARGET_X86_64
3486        tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3487        tcg_gen_rotri_i32(s->tmp2_i32, s->tmp2_i32, b);
3488        tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
3489        break;
3490
3491    case MO_64:
3492#endif
3493        tcg_gen_rotri_tl(s->T0, s->T0, b);
3494        break;
3495
3496    default:
3497        g_assert_not_reached();
3498    }
3499}
3500
3501#ifndef CONFIG_USER_ONLY
3502static void gen_RSM(DisasContext *s, X86DecodedInsn *decode)
3503{
3504    gen_helper_rsm(tcg_env);
3505    assume_cc_op(s, CC_OP_EFLAGS);
3506    s->base.is_jmp = DISAS_EOB_ONLY;
3507}
3508#else
3509#define gen_RSM gen_UD
3510#endif
3511
3512static void gen_SAHF(DisasContext *s, X86DecodedInsn *decode)
3513{
3514    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
3515        return gen_illegal_opcode(s);
3516    }
3517    tcg_gen_shri_tl(s->T0, cpu_regs[R_EAX], 8);
3518    gen_compute_eflags(s);
3519    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, CC_O);
3520    tcg_gen_andi_tl(s->T0, s->T0, CC_S | CC_Z | CC_A | CC_P | CC_C);
3521    tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, s->T0);
3522}
3523
3524static void gen_SALC(DisasContext *s, X86DecodedInsn *decode)
3525{
3526    gen_compute_eflags_c(s, s->T0);
3527    tcg_gen_neg_tl(s->T0, s->T0);
3528}
3529
3530static void gen_shift_dynamic_flags(DisasContext *s, X86DecodedInsn *decode, TCGv count, CCOp cc_op)
3531{
3532    TCGv_i32 count32 = tcg_temp_new_i32();
3533    TCGv_i32 old_cc_op;
3534
3535    decode->cc_op = CC_OP_DYNAMIC;
3536    decode->cc_op_dynamic = tcg_temp_new_i32();
3537
3538    assert(decode->cc_dst == s->T0);
3539    if (cc_op_live[s->cc_op] & USES_CC_DST) {
3540        decode->cc_dst = tcg_temp_new();
3541        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3542                           cpu_cc_dst, s->T0);
3543    }
3544
3545    if (cc_op_live[s->cc_op] & USES_CC_SRC) {
3546        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src, count, tcg_constant_tl(0),
3547                           cpu_cc_src, decode->cc_src);
3548    }
3549
3550    tcg_gen_trunc_tl_i32(count32, count);
3551    if (s->cc_op == CC_OP_DYNAMIC) {
3552        old_cc_op = cpu_cc_op;
3553    } else {
3554        old_cc_op = tcg_constant_i32(s->cc_op);
3555    }
3556    tcg_gen_movcond_i32(TCG_COND_EQ, decode->cc_op_dynamic, count32, tcg_constant_i32(0),
3557                        old_cc_op, tcg_constant_i32(cc_op));
3558}
3559
3560static void gen_SAR(DisasContext *s, X86DecodedInsn *decode)
3561{
3562    bool can_be_zero;
3563    TCGv count;
3564    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3565
3566    if (!count) {
3567        return;
3568    }
3569
3570    decode->cc_dst = s->T0;
3571    decode->cc_src = tcg_temp_new();
3572    tcg_gen_subi_tl(decode->cc_src, count, 1);
3573    tcg_gen_sar_tl(decode->cc_src, s->T0, decode->cc_src);
3574    tcg_gen_sar_tl(s->T0, s->T0, count);
3575    if (can_be_zero) {
3576        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3577    } else {
3578        decode->cc_op = CC_OP_SARB + ot;
3579    }
3580}
3581
3582static void gen_SARX(DisasContext *s, X86DecodedInsn *decode)
3583{
3584    MemOp ot = decode->op[0].ot;
3585    int mask;
3586
3587    mask = ot == MO_64 ? 63 : 31;
3588    tcg_gen_andi_tl(s->T1, s->T1, mask);
3589    tcg_gen_sar_tl(s->T0, s->T0, s->T1);
3590}
3591
3592static void gen_SBB(DisasContext *s, X86DecodedInsn *decode)
3593{
3594    MemOp ot = decode->op[0].ot;
3595    TCGv c_in = tcg_temp_new();
3596
3597    gen_compute_eflags_c(s, c_in);
3598    if (s->prefix & PREFIX_LOCK) {
3599        tcg_gen_add_tl(s->T0, s->T1, c_in);
3600        tcg_gen_neg_tl(s->T0, s->T0);
3601        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
3602                                    s->mem_index, ot | MO_LE);
3603    } else {
3604        /*
3605         * TODO: SBB reg, reg could use gen_prepare_eflags_c followed by
3606         * negsetcond, and CC_OP_SUBB as the cc_op.
3607         */
3608        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
3609        tcg_gen_sub_tl(s->T0, s->T0, c_in);
3610    }
3611    prepare_update3_cc(decode, s, CC_OP_SBBB + ot, c_in);
3612}
3613
3614static void gen_SCAS(DisasContext *s, X86DecodedInsn *decode)
3615{
3616    MemOp ot = decode->op[2].ot;
3617    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
3618        gen_repz_nz(s, ot, gen_scas);
3619    } else {
3620        gen_scas(s, ot);
3621    }
3622}
3623
3624static void gen_SETcc(DisasContext *s, X86DecodedInsn *decode)
3625{
3626    gen_setcc1(s, decode->b & 0xf, s->T0);
3627}
3628
3629static void gen_SFENCE(DisasContext *s, X86DecodedInsn *decode)
3630{
3631    tcg_gen_mb(TCG_MO_ST_ST | TCG_BAR_SC);
3632}
3633
3634static void gen_SHA1NEXTE(DisasContext *s, X86DecodedInsn *decode)
3635{
3636    gen_helper_sha1nexte(OP_PTR0, OP_PTR1, OP_PTR2);
3637}
3638
3639static void gen_SHA1MSG1(DisasContext *s, X86DecodedInsn *decode)
3640{
3641    gen_helper_sha1msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3642}
3643
3644static void gen_SHA1MSG2(DisasContext *s, X86DecodedInsn *decode)
3645{
3646    gen_helper_sha1msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3647}
3648
3649static void gen_SHA1RNDS4(DisasContext *s, X86DecodedInsn *decode)
3650{
3651    switch(decode->immediate & 3) {
3652    case 0:
3653        gen_helper_sha1rnds4_f0(OP_PTR0, OP_PTR0, OP_PTR1);
3654        break;
3655    case 1:
3656        gen_helper_sha1rnds4_f1(OP_PTR0, OP_PTR0, OP_PTR1);
3657        break;
3658    case 2:
3659        gen_helper_sha1rnds4_f2(OP_PTR0, OP_PTR0, OP_PTR1);
3660        break;
3661    case 3:
3662        gen_helper_sha1rnds4_f3(OP_PTR0, OP_PTR0, OP_PTR1);
3663        break;
3664    }
3665}
3666
3667static void gen_SHA256MSG1(DisasContext *s, X86DecodedInsn *decode)
3668{
3669    gen_helper_sha256msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3670}
3671
3672static void gen_SHA256MSG2(DisasContext *s, X86DecodedInsn *decode)
3673{
3674    gen_helper_sha256msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3675}
3676
3677static void gen_SHA256RNDS2(DisasContext *s, X86DecodedInsn *decode)
3678{
3679    TCGv_i32 wk0 = tcg_temp_new_i32();
3680    TCGv_i32 wk1 = tcg_temp_new_i32();
3681
3682    tcg_gen_ld_i32(wk0, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(0)));
3683    tcg_gen_ld_i32(wk1, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(1)));
3684
3685    gen_helper_sha256rnds2(OP_PTR0, OP_PTR1, OP_PTR2, wk0, wk1);
3686}
3687
3688static void gen_SHL(DisasContext *s, X86DecodedInsn *decode)
3689{
3690    bool can_be_zero;
3691    TCGv count;
3692    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3693
3694    if (!count) {
3695        return;
3696    }
3697
3698    decode->cc_dst = s->T0;
3699    decode->cc_src = tcg_temp_new();
3700    tcg_gen_subi_tl(decode->cc_src, count, 1);
3701    tcg_gen_shl_tl(decode->cc_src, s->T0, decode->cc_src);
3702    tcg_gen_shl_tl(s->T0, s->T0, count);
3703    if (can_be_zero) {
3704        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
3705    } else {
3706        decode->cc_op = CC_OP_SHLB + ot;
3707    }
3708}
3709
3710static void gen_SHLD(DisasContext *s, X86DecodedInsn *decode)
3711{
3712    bool can_be_zero;
3713    TCGv count;
3714    int unit = decode->e.op3 == X86_TYPE_I ? X86_OP_IMM : X86_OP_INT;
3715    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, unit);
3716
3717    if (!count) {
3718        return;
3719    }
3720
3721    decode->cc_dst = s->T0;
3722    decode->cc_src = s->tmp0;
3723    gen_shiftd_rm_T1(s, ot, false, count);
3724    if (can_be_zero) {
3725        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
3726    } else {
3727        decode->cc_op = CC_OP_SHLB + ot;
3728    }
3729}
3730
3731static void gen_SHLX(DisasContext *s, X86DecodedInsn *decode)
3732{
3733    MemOp ot = decode->op[0].ot;
3734    int mask;
3735
3736    mask = ot == MO_64 ? 63 : 31;
3737    tcg_gen_andi_tl(s->T1, s->T1, mask);
3738    tcg_gen_shl_tl(s->T0, s->T0, s->T1);
3739}
3740
3741static void gen_SHR(DisasContext *s, X86DecodedInsn *decode)
3742{
3743    bool can_be_zero;
3744    TCGv count;
3745    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, decode->op[2].unit);
3746
3747    if (!count) {
3748        return;
3749    }
3750
3751    decode->cc_dst = s->T0;
3752    decode->cc_src = tcg_temp_new();
3753    tcg_gen_subi_tl(decode->cc_src, count, 1);
3754    tcg_gen_shr_tl(decode->cc_src, s->T0, decode->cc_src);
3755    tcg_gen_shr_tl(s->T0, s->T0, count);
3756    if (can_be_zero) {
3757        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3758    } else {
3759        decode->cc_op = CC_OP_SARB + ot;
3760    }
3761}
3762
3763static void gen_SHRD(DisasContext *s, X86DecodedInsn *decode)
3764{
3765    bool can_be_zero;
3766    TCGv count;
3767    int unit = decode->e.op3 == X86_TYPE_I ? X86_OP_IMM : X86_OP_INT;
3768    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count, unit);
3769
3770    if (!count) {
3771        return;
3772    }
3773
3774    decode->cc_dst = s->T0;
3775    decode->cc_src = s->tmp0;
3776    gen_shiftd_rm_T1(s, ot, true, count);
3777    if (can_be_zero) {
3778        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3779    } else {
3780        decode->cc_op = CC_OP_SARB + ot;
3781    }
3782}
3783
3784static void gen_SHRX(DisasContext *s, X86DecodedInsn *decode)
3785{
3786    MemOp ot = decode->op[0].ot;
3787    int mask;
3788
3789    mask = ot == MO_64 ? 63 : 31;
3790    tcg_gen_andi_tl(s->T1, s->T1, mask);
3791    tcg_gen_shr_tl(s->T0, s->T0, s->T1);
3792}
3793
3794static void gen_STC(DisasContext *s, X86DecodedInsn *decode)
3795{
3796    gen_compute_eflags(s);
3797    tcg_gen_ori_tl(cpu_cc_src, cpu_cc_src, CC_C);
3798}
3799
3800static void gen_STD(DisasContext *s, X86DecodedInsn *decode)
3801{
3802    tcg_gen_st_i32(tcg_constant_i32(-1), tcg_env, offsetof(CPUX86State, df));
3803}
3804
3805static void gen_STI(DisasContext *s, X86DecodedInsn *decode)
3806{
3807    gen_set_eflags(s, IF_MASK);
3808    s->base.is_jmp = DISAS_EOB_INHIBIT_IRQ;
3809}
3810
3811static void gen_VAESKEYGEN(DisasContext *s, X86DecodedInsn *decode)
3812{
3813    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
3814    assert(!s->vex_l);
3815    gen_helper_aeskeygenassist_xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
3816}
3817
3818static void gen_STMXCSR(DisasContext *s, X86DecodedInsn *decode)
3819{
3820    gen_helper_update_mxcsr(tcg_env);
3821    tcg_gen_ld32u_tl(s->T0, tcg_env, offsetof(CPUX86State, mxcsr));
3822}
3823
3824static void gen_STOS(DisasContext *s, X86DecodedInsn *decode)
3825{
3826    MemOp ot = decode->op[1].ot;
3827    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
3828        gen_repz(s, ot, gen_stos);
3829    } else {
3830        gen_stos(s, ot);
3831    }
3832}
3833
3834static void gen_SUB(DisasContext *s, X86DecodedInsn *decode)
3835{
3836    MemOp ot = decode->op[1].ot;
3837
3838    if (s->prefix & PREFIX_LOCK) {
3839        tcg_gen_neg_tl(s->T0, s->T1);
3840        tcg_gen_atomic_fetch_add_tl(s->cc_srcT, s->A0, s->T0,
3841                                    s->mem_index, ot | MO_LE);
3842        tcg_gen_sub_tl(s->T0, s->cc_srcT, s->T1);
3843    } else {
3844        tcg_gen_mov_tl(s->cc_srcT, s->T0);
3845        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
3846    }
3847    prepare_update2_cc(decode, s, CC_OP_SUBB + ot);
3848}
3849
3850static void gen_SYSCALL(DisasContext *s, X86DecodedInsn *decode)
3851{
3852    gen_update_cc_op(s);
3853    gen_update_eip_cur(s);
3854    gen_helper_syscall(tcg_env, cur_insn_len_i32(s));
3855    if (LMA(s)) {
3856        assume_cc_op(s, CC_OP_EFLAGS);
3857    }
3858
3859    /*
3860     * TF handling for the syscall insn is different. The TF bit is checked
3861     * after the syscall insn completes. This allows #DB to not be
3862     * generated after one has entered CPL0 if TF is set in FMASK.
3863     */
3864    s->base.is_jmp = DISAS_EOB_RECHECK_TF;
3865}
3866
3867static void gen_SYSENTER(DisasContext *s, X86DecodedInsn *decode)
3868{
3869    gen_helper_sysenter(tcg_env);
3870    s->base.is_jmp = DISAS_EOB_ONLY;
3871}
3872
3873static void gen_SYSEXIT(DisasContext *s, X86DecodedInsn *decode)
3874{
3875    gen_helper_sysexit(tcg_env, tcg_constant_i32(s->dflag - 1));
3876    s->base.is_jmp = DISAS_EOB_ONLY;
3877}
3878
3879static void gen_SYSRET(DisasContext *s, X86DecodedInsn *decode)
3880{
3881    gen_helper_sysret(tcg_env, tcg_constant_i32(s->dflag - 1));
3882    if (LMA(s)) {
3883        assume_cc_op(s, CC_OP_EFLAGS);
3884    }
3885
3886    /*
3887     * TF handling for the sysret insn is different. The TF bit is checked
3888     * after the sysret insn completes. This allows #DB to be
3889     * generated "as if" the syscall insn in userspace has just
3890     * completed.
3891     */
3892    s->base.is_jmp = DISAS_EOB_RECHECK_TF;
3893}
3894
3895static void gen_TZCNT(DisasContext *s, X86DecodedInsn *decode)
3896{
3897    MemOp ot = decode->op[0].ot;
3898
3899    /* C bit (cc_src) is defined related to the input.  */
3900    decode->cc_src = tcg_temp_new();
3901    decode->cc_dst = s->T0;
3902    decode->cc_op = CC_OP_BMILGB + ot;
3903    tcg_gen_mov_tl(decode->cc_src, s->T0);
3904
3905    /* A zero input returns the operand size.  */
3906    tcg_gen_ctzi_tl(s->T0, s->T0, 8 << ot);
3907}
3908
3909static void gen_UD(DisasContext *s, X86DecodedInsn *decode)
3910{
3911    gen_illegal_opcode(s);
3912}
3913
3914static void gen_VAESIMC(DisasContext *s, X86DecodedInsn *decode)
3915{
3916    assert(!s->vex_l);
3917    gen_helper_aesimc_xmm(tcg_env, OP_PTR0, OP_PTR2);
3918}
3919
3920/*
3921 * 00 = v*ps Vps, Hps, Wpd
3922 * 66 = v*pd Vpd, Hpd, Wps
3923 * f3 = v*ss Vss, Hss, Wps
3924 * f2 = v*sd Vsd, Hsd, Wps
3925 */
3926#define SSE_CMP(x) { \
3927    gen_helper_ ## x ## ps ## _xmm, gen_helper_ ## x ## pd ## _xmm, \
3928    gen_helper_ ## x ## ss, gen_helper_ ## x ## sd, \
3929    gen_helper_ ## x ## ps ## _ymm, gen_helper_ ## x ## pd ## _ymm}
3930static const SSEFunc_0_eppp gen_helper_cmp_funcs[32][6] = {
3931    SSE_CMP(cmpeq),
3932    SSE_CMP(cmplt),
3933    SSE_CMP(cmple),
3934    SSE_CMP(cmpunord),
3935    SSE_CMP(cmpneq),
3936    SSE_CMP(cmpnlt),
3937    SSE_CMP(cmpnle),
3938    SSE_CMP(cmpord),
3939
3940    SSE_CMP(cmpequ),
3941    SSE_CMP(cmpnge),
3942    SSE_CMP(cmpngt),
3943    SSE_CMP(cmpfalse),
3944    SSE_CMP(cmpnequ),
3945    SSE_CMP(cmpge),
3946    SSE_CMP(cmpgt),
3947    SSE_CMP(cmptrue),
3948
3949    SSE_CMP(cmpeqs),
3950    SSE_CMP(cmpltq),
3951    SSE_CMP(cmpleq),
3952    SSE_CMP(cmpunords),
3953    SSE_CMP(cmpneqq),
3954    SSE_CMP(cmpnltq),
3955    SSE_CMP(cmpnleq),
3956    SSE_CMP(cmpords),
3957
3958    SSE_CMP(cmpequs),
3959    SSE_CMP(cmpngeq),
3960    SSE_CMP(cmpngtq),
3961    SSE_CMP(cmpfalses),
3962    SSE_CMP(cmpnequs),
3963    SSE_CMP(cmpgeq),
3964    SSE_CMP(cmpgtq),
3965    SSE_CMP(cmptrues),
3966};
3967#undef SSE_CMP
3968
3969static void gen_VCMP(DisasContext *s, X86DecodedInsn *decode)
3970{
3971    int index = decode->immediate & (s->prefix & PREFIX_VEX ? 31 : 7);
3972    int b =
3973        s->prefix & PREFIX_REPZ  ? 2 /* ss */ :
3974        s->prefix & PREFIX_REPNZ ? 3 /* sd */ :
3975        !!(s->prefix & PREFIX_DATA) /* pd */ + (s->vex_l << 2);
3976
3977    gen_helper_cmp_funcs[index][b](tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
3978}
3979
3980static void gen_VCOMI(DisasContext *s, X86DecodedInsn *decode)
3981{
3982    SSEFunc_0_epp fn;
3983    fn = s->prefix & PREFIX_DATA ? gen_helper_comisd : gen_helper_comiss;
3984    fn(tcg_env, OP_PTR1, OP_PTR2);
3985    assume_cc_op(s, CC_OP_EFLAGS);
3986}
3987
3988static void gen_VCVTPD2PS(DisasContext *s, X86DecodedInsn *decode)
3989{
3990    if (s->vex_l) {
3991        gen_helper_cvtpd2ps_ymm(tcg_env, OP_PTR0, OP_PTR2);
3992    } else {
3993        gen_helper_cvtpd2ps_xmm(tcg_env, OP_PTR0, OP_PTR2);
3994    }
3995}
3996
3997static void gen_VCVTPS2PD(DisasContext *s, X86DecodedInsn *decode)
3998{
3999    if (s->vex_l) {
4000        gen_helper_cvtps2pd_ymm(tcg_env, OP_PTR0, OP_PTR2);
4001    } else {
4002        gen_helper_cvtps2pd_xmm(tcg_env, OP_PTR0, OP_PTR2);
4003    }
4004}
4005
4006static void gen_VCVTPS2PH(DisasContext *s, X86DecodedInsn *decode)
4007{
4008    gen_unary_imm_fp_sse(s, decode,
4009                      gen_helper_cvtps2ph_xmm,
4010                      gen_helper_cvtps2ph_ymm);
4011    /*
4012     * VCVTPS2PH is the only instruction that performs an operation on a
4013     * register source and then *stores* into memory.
4014     */
4015    if (decode->op[0].has_ea) {
4016        gen_store_sse(s, decode, decode->op[0].offset);
4017    }
4018}
4019
4020static void gen_VCVTSD2SS(DisasContext *s, X86DecodedInsn *decode)
4021{
4022    gen_helper_cvtsd2ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4023}
4024
4025static void gen_VCVTSS2SD(DisasContext *s, X86DecodedInsn *decode)
4026{
4027    gen_helper_cvtss2sd(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
4028}
4029
4030static void gen_VCVTSI2Sx(DisasContext *s, X86DecodedInsn *decode)
4031{
4032    int vec_len = vector_len(s, decode);
4033    TCGv_i32 in;
4034
4035    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4036
4037#ifdef TARGET_X86_64
4038    MemOp ot = decode->op[2].ot;
4039    if (ot == MO_64) {
4040        if (s->prefix & PREFIX_REPNZ) {
4041            gen_helper_cvtsq2sd(tcg_env, OP_PTR0, s->T1);
4042        } else {
4043            gen_helper_cvtsq2ss(tcg_env, OP_PTR0, s->T1);
4044        }
4045        return;
4046    }
4047    in = s->tmp2_i32;
4048    tcg_gen_trunc_tl_i32(in, s->T1);
4049#else
4050    in = s->T1;
4051#endif
4052
4053    if (s->prefix & PREFIX_REPNZ) {
4054        gen_helper_cvtsi2sd(tcg_env, OP_PTR0, in);
4055    } else {
4056        gen_helper_cvtsi2ss(tcg_env, OP_PTR0, in);
4057    }
4058}
4059
4060static inline void gen_VCVTtSx2SI(DisasContext *s, X86DecodedInsn *decode,
4061                                  SSEFunc_i_ep ss2si, SSEFunc_l_ep ss2sq,
4062                                  SSEFunc_i_ep sd2si, SSEFunc_l_ep sd2sq)
4063{
4064    TCGv_i32 out;
4065
4066#ifdef TARGET_X86_64
4067    MemOp ot = decode->op[0].ot;
4068    if (ot == MO_64) {
4069        if (s->prefix & PREFIX_REPNZ) {
4070            sd2sq(s->T0, tcg_env, OP_PTR2);
4071        } else {
4072            ss2sq(s->T0, tcg_env, OP_PTR2);
4073        }
4074        return;
4075    }
4076
4077    out = s->tmp2_i32;
4078#else
4079    out = s->T0;
4080#endif
4081    if (s->prefix & PREFIX_REPNZ) {
4082        sd2si(out, tcg_env, OP_PTR2);
4083    } else {
4084        ss2si(out, tcg_env, OP_PTR2);
4085    }
4086#ifdef TARGET_X86_64
4087    tcg_gen_extu_i32_tl(s->T0, out);
4088#endif
4089}
4090
4091#ifndef TARGET_X86_64
4092#define gen_helper_cvtss2sq NULL
4093#define gen_helper_cvtsd2sq NULL
4094#define gen_helper_cvttss2sq NULL
4095#define gen_helper_cvttsd2sq NULL
4096#endif
4097
4098static void gen_VCVTSx2SI(DisasContext *s, X86DecodedInsn *decode)
4099{
4100    gen_VCVTtSx2SI(s, decode,
4101                   gen_helper_cvtss2si, gen_helper_cvtss2sq,
4102                   gen_helper_cvtsd2si, gen_helper_cvtsd2sq);
4103}
4104
4105static void gen_VCVTTSx2SI(DisasContext *s, X86DecodedInsn *decode)
4106{
4107    gen_VCVTtSx2SI(s, decode,
4108                   gen_helper_cvttss2si, gen_helper_cvttss2sq,
4109                   gen_helper_cvttsd2si, gen_helper_cvttsd2sq);
4110}
4111
4112static void gen_VEXTRACTx128(DisasContext *s, X86DecodedInsn *decode)
4113{
4114    int mask = decode->immediate & 1;
4115    int src_ofs = vector_elem_offset(&decode->op[1], MO_128, mask);
4116    if (decode->op[0].has_ea) {
4117        /* VEX-only instruction, no alignment requirements.  */
4118        gen_sto_env_A0(s, src_ofs, false);
4119    } else {
4120        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, 16, 16);
4121    }
4122}
4123
4124static void gen_VEXTRACTPS(DisasContext *s, X86DecodedInsn *decode)
4125{
4126    gen_pextr(s, decode, MO_32);
4127}
4128
4129static void gen_vinsertps(DisasContext *s, X86DecodedInsn *decode)
4130{
4131    int val = decode->immediate;
4132    int dest_word = (val >> 4) & 3;
4133    int new_mask = (val & 15) | (1 << dest_word);
4134    int vec_len = 16;
4135
4136    assert(!s->vex_l);
4137
4138    if (new_mask == 15) {
4139        /* All zeroes except possibly for the inserted element */
4140        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
4141    } else if (decode->op[1].offset != decode->op[0].offset) {
4142        gen_store_sse(s, decode, decode->op[1].offset);
4143    }
4144
4145    if (new_mask != (val & 15)) {
4146        tcg_gen_st_i32(s->tmp2_i32, tcg_env,
4147                       vector_elem_offset(&decode->op[0], MO_32, dest_word));
4148    }
4149
4150    if (new_mask != 15) {
4151        TCGv_i32 zero = tcg_constant_i32(0); /* float32_zero */
4152        int i;
4153        for (i = 0; i < 4; i++) {
4154            if ((val >> i) & 1) {
4155                tcg_gen_st_i32(zero, tcg_env,
4156                               vector_elem_offset(&decode->op[0], MO_32, i));
4157            }
4158        }
4159    }
4160}
4161
4162static void gen_VINSERTPS_r(DisasContext *s, X86DecodedInsn *decode)
4163{
4164    int val = decode->immediate;
4165    tcg_gen_ld_i32(s->tmp2_i32, tcg_env,
4166                   vector_elem_offset(&decode->op[2], MO_32, (val >> 6) & 3));
4167    gen_vinsertps(s, decode);
4168}
4169
4170static void gen_VINSERTPS_m(DisasContext *s, X86DecodedInsn *decode)
4171{
4172    tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4173    gen_vinsertps(s, decode);
4174}
4175
4176static void gen_VINSERTx128(DisasContext *s, X86DecodedInsn *decode)
4177{
4178    int mask = decode->immediate & 1;
4179    tcg_gen_gvec_mov(MO_64,
4180                     decode->op[0].offset + offsetof(YMMReg, YMM_X(mask)),
4181                     decode->op[2].offset + offsetof(YMMReg, YMM_X(0)), 16, 16);
4182    tcg_gen_gvec_mov(MO_64,
4183                     decode->op[0].offset + offsetof(YMMReg, YMM_X(!mask)),
4184                     decode->op[1].offset + offsetof(YMMReg, YMM_X(!mask)), 16, 16);
4185}
4186
4187static inline void gen_maskmov(DisasContext *s, X86DecodedInsn *decode,
4188                               SSEFunc_0_eppt xmm, SSEFunc_0_eppt ymm)
4189{
4190    if (!s->vex_l) {
4191        xmm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
4192    } else {
4193        ymm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
4194    }
4195}
4196
4197static void gen_VMASKMOVPD_st(DisasContext *s, X86DecodedInsn *decode)
4198{
4199    gen_maskmov(s, decode, gen_helper_vpmaskmovq_st_xmm, gen_helper_vpmaskmovq_st_ymm);
4200}
4201
4202static void gen_VMASKMOVPS_st(DisasContext *s, X86DecodedInsn *decode)
4203{
4204    gen_maskmov(s, decode, gen_helper_vpmaskmovd_st_xmm, gen_helper_vpmaskmovd_st_ymm);
4205}
4206
4207static void gen_VMOVHPx_ld(DisasContext *s, X86DecodedInsn *decode)
4208{
4209    gen_ldq_env_A0(s, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4210    if (decode->op[0].offset != decode->op[1].offset) {
4211        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4212        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4213    }
4214}
4215
4216static void gen_VMOVHPx_st(DisasContext *s, X86DecodedInsn *decode)
4217{
4218    gen_stq_env_A0(s, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4219}
4220
4221static void gen_VMOVHPx(DisasContext *s, X86DecodedInsn *decode)
4222{
4223    if (decode->op[0].offset != decode->op[2].offset) {
4224        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4225        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4226    }
4227    if (decode->op[0].offset != decode->op[1].offset) {
4228        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4229        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4230    }
4231}
4232
4233static void gen_VMOVHLPS(DisasContext *s, X86DecodedInsn *decode)
4234{
4235    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
4236    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4237    if (decode->op[0].offset != decode->op[1].offset) {
4238        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(1)));
4239        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4240    }
4241}
4242
4243static void gen_VMOVLHPS(DisasContext *s, X86DecodedInsn *decode)
4244{
4245    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
4246    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
4247    if (decode->op[0].offset != decode->op[1].offset) {
4248        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
4249        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4250    }
4251}
4252
4253/*
4254 * Note that MOVLPx supports 256-bit operation unlike MOVHLPx, MOVLHPx, MOXHPx.
4255 * Use a gvec move to move everything above the bottom 64 bits.
4256 */
4257
4258static void gen_VMOVLPx(DisasContext *s, X86DecodedInsn *decode)
4259{
4260    int vec_len = vector_len(s, decode);
4261
4262    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(0)));
4263    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4264    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
4265}
4266
4267static void gen_VMOVLPx_ld(DisasContext *s, X86DecodedInsn *decode)
4268{
4269    int vec_len = vector_len(s, decode);
4270
4271    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4272    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4273    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
4274}
4275
4276static void gen_VMOVLPx_st(DisasContext *s, X86DecodedInsn *decode)
4277{
4278    tcg_gen_ld_i64(s->tmp1_i64, OP_PTR2, offsetof(ZMMReg, ZMM_Q(0)));
4279    tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4280}
4281
4282static void gen_VMOVSD_ld(DisasContext *s, X86DecodedInsn *decode)
4283{
4284    TCGv_i64 zero = tcg_constant_i64(0);
4285
4286    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
4287    tcg_gen_st_i64(zero, OP_PTR0, offsetof(ZMMReg, ZMM_Q(1)));
4288    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
4289}
4290
4291static void gen_VMOVSS(DisasContext *s, X86DecodedInsn *decode)
4292{
4293    int vec_len = vector_len(s, decode);
4294
4295    tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
4296    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
4297    tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
4298}
4299
4300static void gen_VMOVSS_ld(DisasContext *s, X86DecodedInsn *decode)
4301{
4302    int vec_len = vector_len(s, decode);
4303
4304    tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4305    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
4306    tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
4307}
4308
4309static void gen_VMOVSS_st(DisasContext *s, X86DecodedInsn *decode)
4310{
4311    tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
4312    tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
4313}
4314
4315static void gen_VPMASKMOV_st(DisasContext *s, X86DecodedInsn *decode)
4316{
4317    if (s->vex_w) {
4318        gen_VMASKMOVPD_st(s, decode);
4319    } else {
4320        gen_VMASKMOVPS_st(s, decode);
4321    }
4322}
4323
4324static void gen_VPERMD(DisasContext *s, X86DecodedInsn *decode)
4325{
4326    assert(s->vex_l);
4327    gen_helper_vpermd_ymm(OP_PTR0, OP_PTR1, OP_PTR2);
4328}
4329
4330static void gen_VPERM2x128(DisasContext *s, X86DecodedInsn *decode)
4331{
4332    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4333    assert(s->vex_l);
4334    gen_helper_vpermdq_ymm(OP_PTR0, OP_PTR1, OP_PTR2, imm);
4335}
4336
4337static void gen_VPHMINPOSUW(DisasContext *s, X86DecodedInsn *decode)
4338{
4339    assert(!s->vex_l);
4340    gen_helper_phminposuw_xmm(tcg_env, OP_PTR0, OP_PTR2);
4341}
4342
4343static void gen_VROUNDSD(DisasContext *s, X86DecodedInsn *decode)
4344{
4345    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4346    assert(!s->vex_l);
4347    gen_helper_roundsd_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
4348}
4349
4350static void gen_VROUNDSS(DisasContext *s, X86DecodedInsn *decode)
4351{
4352    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
4353    assert(!s->vex_l);
4354    gen_helper_roundss_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
4355}
4356
4357static void gen_VSHUF(DisasContext *s, X86DecodedInsn *decode)
4358{
4359    TCGv_i32 imm = tcg_constant_i32(decode->immediate);
4360    SSEFunc_0_pppi ps, pd, fn;
4361    ps = s->vex_l ? gen_helper_shufps_ymm : gen_helper_shufps_xmm;
4362    pd = s->vex_l ? gen_helper_shufpd_ymm : gen_helper_shufpd_xmm;
4363    fn = s->prefix & PREFIX_DATA ? pd : ps;
4364    fn(OP_PTR0, OP_PTR1, OP_PTR2, imm);
4365}
4366
4367static void gen_VUCOMI(DisasContext *s, X86DecodedInsn *decode)
4368{
4369    SSEFunc_0_epp fn;
4370    fn = s->prefix & PREFIX_DATA ? gen_helper_ucomisd : gen_helper_ucomiss;
4371    fn(tcg_env, OP_PTR1, OP_PTR2);
4372    assume_cc_op(s, CC_OP_EFLAGS);
4373}
4374
4375static void gen_VZEROALL(DisasContext *s, X86DecodedInsn *decode)
4376{
4377    TCGv_ptr ptr = tcg_temp_new_ptr();
4378
4379    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_regs));
4380    gen_helper_memset(ptr, ptr, tcg_constant_i32(0),
4381                      tcg_constant_ptr(CPU_NB_REGS * sizeof(ZMMReg)));
4382}
4383
4384static void gen_VZEROUPPER(DisasContext *s, X86DecodedInsn *decode)
4385{
4386    int i;
4387
4388    for (i = 0; i < CPU_NB_REGS; i++) {
4389        int offset = offsetof(CPUX86State, xmm_regs[i].ZMM_X(1));
4390        tcg_gen_gvec_dup_imm(MO_64, offset, 16, 16, 0);
4391    }
4392}
4393
4394static void gen_WAIT(DisasContext *s, X86DecodedInsn *decode)
4395{
4396    if ((s->flags & (HF_MP_MASK | HF_TS_MASK)) == (HF_MP_MASK | HF_TS_MASK)) {
4397        gen_NM_exception(s);
4398    } else {
4399        /* needs to be treated as I/O because of ferr_irq */
4400        translator_io_start(&s->base);
4401        gen_helper_fwait(tcg_env);
4402    }
4403}
4404
4405#ifndef CONFIG_USER_ONLY
4406static void gen_WRMSR(DisasContext *s, X86DecodedInsn *decode)
4407{
4408    gen_update_cc_op(s);
4409    gen_update_eip_cur(s);
4410    gen_helper_wrmsr(tcg_env);
4411    s->base.is_jmp = DISAS_EOB_NEXT;
4412}
4413#else
4414#define gen_WRMSR gen_unreachable
4415#endif
4416
4417static void gen_WRxxBASE(DisasContext *s, X86DecodedInsn *decode)
4418{
4419    TCGv base = cpu_seg_base[s->modrm & 8 ? R_GS : R_FS];
4420
4421    /* Preserve hflags bits by testing CR4 at runtime.  */
4422    gen_helper_cr4_testbit(tcg_env, tcg_constant_i32(CR4_FSGSBASE_MASK));
4423    tcg_gen_mov_tl(base, s->T0);
4424}
4425
4426static void gen_XADD(DisasContext *s, X86DecodedInsn *decode)
4427{
4428    MemOp ot = decode->op[1].ot;
4429
4430    decode->cc_dst = tcg_temp_new();
4431    decode->cc_src = s->T1;
4432    decode->cc_op = CC_OP_ADDB + ot;
4433
4434    if (s->prefix & PREFIX_LOCK) {
4435        tcg_gen_atomic_fetch_add_tl(s->T0, s->A0, s->T1, s->mem_index, ot | MO_LE);
4436        tcg_gen_add_tl(decode->cc_dst, s->T0, s->T1);
4437    } else {
4438        tcg_gen_add_tl(decode->cc_dst, s->T0, s->T1);
4439        /*
4440         * NOTE: writing memory first is important for MMU exceptions,
4441         * but "new result" wins for XADD AX, AX.
4442         */
4443        gen_writeback(s, decode, 0, decode->cc_dst);
4444    }
4445    if (decode->op[0].has_ea || decode->op[2].n != decode->op[0].n) {
4446        gen_writeback(s, decode, 2, s->T0);
4447    }
4448}
4449
4450static void gen_XCHG(DisasContext *s, X86DecodedInsn *decode)
4451{
4452    if (s->prefix & PREFIX_LOCK) {
4453        tcg_gen_atomic_xchg_tl(s->T0, s->A0, s->T1,
4454                               s->mem_index, decode->op[0].ot | MO_LE);
4455        /* now store old value into register operand */
4456        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4457    } else {
4458        /* move destination value into source operand, source preserved in T1 */
4459        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4460        tcg_gen_mov_tl(s->T0, s->T1);
4461    }
4462}
4463
4464static void gen_XLAT(DisasContext *s, X86DecodedInsn *decode)
4465{
4466    /* AL is already zero-extended into s->T0.  */
4467    tcg_gen_add_tl(s->A0, cpu_regs[R_EBX], s->T0);
4468    gen_lea_v_seg(s, s->A0, R_DS, s->override);
4469    gen_op_ld_v(s, MO_8, s->T0, s->A0);
4470}
4471
4472static void gen_XOR(DisasContext *s, X86DecodedInsn *decode)
4473{
4474    /* special case XOR reg, reg */
4475    if (decode->op[1].unit == X86_OP_INT &&
4476        decode->op[2].unit == X86_OP_INT &&
4477        decode->op[1].n == decode->op[2].n) {
4478        tcg_gen_movi_tl(s->T0, 0);
4479        decode->cc_op = CC_OP_CLR;
4480    } else {
4481        MemOp ot = decode->op[1].ot;
4482
4483        if (s->prefix & PREFIX_LOCK) {
4484            tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T1,
4485                                        s->mem_index, ot | MO_LE);
4486        } else {
4487            tcg_gen_xor_tl(s->T0, s->T0, s->T1);
4488        }
4489        prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
4490    }
4491}
4492
4493static void gen_XRSTOR(DisasContext *s, X86DecodedInsn *decode)
4494{
4495    TCGv_i64 features = tcg_temp_new_i64();
4496
4497    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4498    gen_helper_xrstor(tcg_env, s->A0, features);
4499    if (s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_MPX) {
4500        /*
4501         * XRSTOR is how MPX is enabled, which changes how
4502         * we translate.  Thus we need to end the TB.
4503         */
4504        s->base.is_jmp = DISAS_EOB_NEXT;
4505    }
4506}
4507
4508static void gen_XSAVE(DisasContext *s, X86DecodedInsn *decode)
4509{
4510    TCGv_i64 features = tcg_temp_new_i64();
4511
4512    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4513    gen_helper_xsave(tcg_env, s->A0, features);
4514}
4515
4516static void gen_XSAVEOPT(DisasContext *s, X86DecodedInsn *decode)
4517{
4518    TCGv_i64 features = tcg_temp_new_i64();
4519
4520    tcg_gen_concat_tl_i64(features, cpu_regs[R_EAX], cpu_regs[R_EDX]);
4521    gen_helper_xsave(tcg_env, s->A0, features);
4522}
4523