xref: /openbmc/qemu/target/i386/tcg/emit.c.inc (revision c0d691ab)
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, CPUX86State *env, X86DecodedInsn *decode);
64static void gen_JMP(DisasContext *s, CPUX86State *env, 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, s->aflag, 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    case X86_OP_CR:
246        tcg_gen_ld_tl(v, tcg_env, offsetof(CPUX86State, cr[op->n]));
247        break;
248    case X86_OP_DR:
249        tcg_gen_ld_tl(v, tcg_env, offsetof(CPUX86State, dr[op->n]));
250        break;
251    case X86_OP_INT:
252        if (op->has_ea) {
253            if (v == s->T0 && decode->e.special == X86_SPECIAL_SExtT0) {
254                gen_op_ld_v(s, op->ot | MO_SIGN, v, s->A0);
255            } else {
256                gen_op_ld_v(s, op->ot, v, s->A0);
257            }
258
259        } else if (op->ot == MO_8 && byte_reg_is_xH(s, op->n)) {
260            if (v == s->T0 && decode->e.special == X86_SPECIAL_SExtT0) {
261                tcg_gen_sextract_tl(v, cpu_regs[op->n - 4], 8, 8);
262            } else {
263                tcg_gen_extract_tl(v, cpu_regs[op->n - 4], 8, 8);
264            }
265
266        } else if (op->ot < MO_TL && v == s->T0 &&
267                   (decode->e.special == X86_SPECIAL_SExtT0 ||
268                    decode->e.special == X86_SPECIAL_ZExtT0)) {
269            if (decode->e.special == X86_SPECIAL_SExtT0) {
270                tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot | MO_SIGN);
271            } else {
272                tcg_gen_ext_tl(v, cpu_regs[op->n], op->ot);
273            }
274
275        } else {
276            tcg_gen_mov_tl(v, cpu_regs[op->n]);
277        }
278        break;
279    case X86_OP_IMM:
280        tcg_gen_movi_tl(v, op->imm);
281        break;
282
283    case X86_OP_MMX:
284        compute_mmx_offset(op);
285        goto load_vector;
286
287    case X86_OP_SSE:
288        compute_xmm_offset(op);
289    load_vector:
290        if (op->has_ea) {
291            bool aligned = sse_needs_alignment(s, decode, op->ot);
292            gen_load_sse(s, v, op->ot, op->offset, aligned);
293        }
294        break;
295
296    default:
297        g_assert_not_reached();
298    }
299}
300
301static TCGv_ptr op_ptr(X86DecodedInsn *decode, int opn)
302{
303    X86DecodedOp *op = &decode->op[opn];
304
305    assert(op->unit == X86_OP_MMX || op->unit == X86_OP_SSE);
306    if (op->v_ptr) {
307        return op->v_ptr;
308    }
309    op->v_ptr = tcg_temp_new_ptr();
310
311    /* The temporary points to the MMXReg or ZMMReg.  */
312    tcg_gen_addi_ptr(op->v_ptr, tcg_env, vector_reg_offset(op));
313    return op->v_ptr;
314}
315
316#define OP_PTR0 op_ptr(decode, 0)
317#define OP_PTR1 op_ptr(decode, 1)
318#define OP_PTR2 op_ptr(decode, 2)
319
320static void gen_writeback(DisasContext *s, X86DecodedInsn *decode, int opn, TCGv v)
321{
322    X86DecodedOp *op = &decode->op[opn];
323    switch (op->unit) {
324    case X86_OP_SKIP:
325        break;
326    case X86_OP_SEG:
327        /* Note that gen_movl_seg takes care of interrupt shadow and TF.  */
328        gen_movl_seg(s, op->n, s->T0);
329        break;
330    case X86_OP_INT:
331        if (op->has_ea) {
332            gen_op_st_v(s, op->ot, v, s->A0);
333        } else {
334            gen_op_mov_reg_v(s, op->ot, op->n, v);
335        }
336        break;
337    case X86_OP_MMX:
338        break;
339    case X86_OP_SSE:
340        if (!op->has_ea && (s->prefix & PREFIX_VEX) && op->ot <= MO_128) {
341            tcg_gen_gvec_dup_imm(MO_64,
342                                 offsetof(CPUX86State, xmm_regs[op->n].ZMM_X(1)),
343                                 16, 16, 0);
344        }
345        break;
346    case X86_OP_CR:
347    case X86_OP_DR:
348    default:
349        g_assert_not_reached();
350    }
351    op->unit = X86_OP_SKIP;
352}
353
354static inline int vector_len(DisasContext *s, X86DecodedInsn *decode)
355{
356    if (decode->e.special == X86_SPECIAL_MMX &&
357        !(s->prefix & (PREFIX_DATA | PREFIX_REPZ | PREFIX_REPNZ))) {
358        return 8;
359    }
360    return s->vex_l ? 32 : 16;
361}
362
363static void prepare_update1_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
364{
365    decode->cc_dst = s->T0;
366    decode->cc_op = op;
367}
368
369static void prepare_update2_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op)
370{
371    decode->cc_src = s->T1;
372    decode->cc_dst = s->T0;
373    decode->cc_op = op;
374}
375
376static void prepare_update_cc_incdec(X86DecodedInsn *decode, DisasContext *s, CCOp op)
377{
378    gen_compute_eflags_c(s, s->T1);
379    prepare_update2_cc(decode, s, op);
380}
381
382static void prepare_update3_cc(X86DecodedInsn *decode, DisasContext *s, CCOp op, TCGv reg)
383{
384    decode->cc_src2 = reg;
385    decode->cc_src = s->T1;
386    decode->cc_dst = s->T0;
387    decode->cc_op = op;
388}
389
390static void gen_store_sse(DisasContext *s, X86DecodedInsn *decode, int src_ofs)
391{
392    MemOp ot = decode->op[0].ot;
393    int vec_len = vector_len(s, decode);
394    bool aligned = sse_needs_alignment(s, decode, ot);
395
396    if (!decode->op[0].has_ea) {
397        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, vec_len, vec_len);
398        return;
399    }
400
401    switch (ot) {
402    case MO_64:
403        gen_stq_env_A0(s, src_ofs);
404        break;
405    case MO_128:
406        gen_sto_env_A0(s, src_ofs, aligned);
407        break;
408    case MO_256:
409        gen_sty_env_A0(s, src_ofs, aligned);
410        break;
411    default:
412        g_assert_not_reached();
413    }
414}
415
416static void gen_helper_pavgusb(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b)
417{
418    gen_helper_pavgb_mmx(env, reg_a, reg_a, reg_b);
419}
420
421#define FN_3DNOW_MOVE ((SSEFunc_0_epp) (uintptr_t) 1)
422static const SSEFunc_0_epp fns_3dnow[] = {
423    [0x0c] = gen_helper_pi2fw,
424    [0x0d] = gen_helper_pi2fd,
425    [0x1c] = gen_helper_pf2iw,
426    [0x1d] = gen_helper_pf2id,
427    [0x8a] = gen_helper_pfnacc,
428    [0x8e] = gen_helper_pfpnacc,
429    [0x90] = gen_helper_pfcmpge,
430    [0x94] = gen_helper_pfmin,
431    [0x96] = gen_helper_pfrcp,
432    [0x97] = gen_helper_pfrsqrt,
433    [0x9a] = gen_helper_pfsub,
434    [0x9e] = gen_helper_pfadd,
435    [0xa0] = gen_helper_pfcmpgt,
436    [0xa4] = gen_helper_pfmax,
437    [0xa6] = FN_3DNOW_MOVE, /* PFRCPIT1; no need to actually increase precision */
438    [0xa7] = FN_3DNOW_MOVE, /* PFRSQIT1 */
439    [0xb6] = FN_3DNOW_MOVE, /* PFRCPIT2 */
440    [0xaa] = gen_helper_pfsubr,
441    [0xae] = gen_helper_pfacc,
442    [0xb0] = gen_helper_pfcmpeq,
443    [0xb4] = gen_helper_pfmul,
444    [0xb7] = gen_helper_pmulhrw_mmx,
445    [0xbb] = gen_helper_pswapd,
446    [0xbf] = gen_helper_pavgusb,
447};
448
449static void gen_3dnow(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
450{
451    uint8_t b = decode->immediate;
452    SSEFunc_0_epp fn = b < ARRAY_SIZE(fns_3dnow) ? fns_3dnow[b] : NULL;
453
454    if (!fn) {
455        gen_illegal_opcode(s);
456        return;
457    }
458    if (s->flags & HF_TS_MASK) {
459        gen_NM_exception(s);
460        return;
461    }
462    if (s->flags & HF_EM_MASK) {
463        gen_illegal_opcode(s);
464        return;
465    }
466
467    gen_helper_enter_mmx(tcg_env);
468    if (fn == FN_3DNOW_MOVE) {
469       tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset);
470       tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset);
471    } else {
472       fn(tcg_env, OP_PTR0, OP_PTR1);
473    }
474}
475
476/*
477 * 00 = v*ps Vps, Hps, Wpd
478 * 66 = v*pd Vpd, Hpd, Wps
479 * f3 = v*ss Vss, Hss, Wps
480 * f2 = v*sd Vsd, Hsd, Wps
481 */
482static inline void gen_unary_fp_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
483                              SSEFunc_0_epp pd_xmm, SSEFunc_0_epp ps_xmm,
484                              SSEFunc_0_epp pd_ymm, SSEFunc_0_epp ps_ymm,
485                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
486{
487    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
488        SSEFunc_0_eppp fn = s->prefix & PREFIX_REPZ ? ss : sd;
489        if (!fn) {
490            gen_illegal_opcode(s);
491            return;
492        }
493        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
494    } else {
495        SSEFunc_0_epp ps, pd, fn;
496        ps = s->vex_l ? ps_ymm : ps_xmm;
497        pd = s->vex_l ? pd_ymm : pd_xmm;
498        fn = s->prefix & PREFIX_DATA ? pd : ps;
499        if (!fn) {
500            gen_illegal_opcode(s);
501            return;
502        }
503        fn(tcg_env, OP_PTR0, OP_PTR2);
504    }
505}
506#define UNARY_FP_SSE(uname, lname)                                                 \
507static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
508{                                                                                  \
509    gen_unary_fp_sse(s, env, decode,                                               \
510                     gen_helper_##lname##pd_xmm,                                   \
511                     gen_helper_##lname##ps_xmm,                                   \
512                     gen_helper_##lname##pd_ymm,                                   \
513                     gen_helper_##lname##ps_ymm,                                   \
514                     gen_helper_##lname##sd,                                       \
515                     gen_helper_##lname##ss);                                      \
516}
517UNARY_FP_SSE(VSQRT, sqrt)
518
519/*
520 * 00 = v*ps Vps, Hps, Wpd
521 * 66 = v*pd Vpd, Hpd, Wps
522 * f3 = v*ss Vss, Hss, Wps
523 * f2 = v*sd Vsd, Hsd, Wps
524 */
525static inline void gen_fp_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
526                              SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
527                              SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm,
528                              SSEFunc_0_eppp sd, SSEFunc_0_eppp ss)
529{
530    SSEFunc_0_eppp ps, pd, fn;
531    if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) != 0) {
532        fn = s->prefix & PREFIX_REPZ ? ss : sd;
533    } else {
534        ps = s->vex_l ? ps_ymm : ps_xmm;
535        pd = s->vex_l ? pd_ymm : pd_xmm;
536        fn = s->prefix & PREFIX_DATA ? pd : ps;
537    }
538    if (fn) {
539        fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
540    } else {
541        gen_illegal_opcode(s);
542    }
543}
544
545#define FP_SSE(uname, lname)                                                       \
546static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
547{                                                                                  \
548    gen_fp_sse(s, env, decode,                                                     \
549               gen_helper_##lname##pd_xmm,                                         \
550               gen_helper_##lname##ps_xmm,                                         \
551               gen_helper_##lname##pd_ymm,                                         \
552               gen_helper_##lname##ps_ymm,                                         \
553               gen_helper_##lname##sd,                                             \
554               gen_helper_##lname##ss);                                            \
555}
556FP_SSE(VADD, add)
557FP_SSE(VMUL, mul)
558FP_SSE(VSUB, sub)
559FP_SSE(VMIN, min)
560FP_SSE(VDIV, div)
561FP_SSE(VMAX, max)
562
563#define FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, even, odd)                         \
564static void gen_##uname##Px(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
565{                                                                                  \
566    SSEFunc_0_eppppii xmm = s->vex_w ? gen_helper_fma4pd_xmm : gen_helper_fma4ps_xmm; \
567    SSEFunc_0_eppppii ymm = s->vex_w ? gen_helper_fma4pd_ymm : gen_helper_fma4ps_ymm; \
568    SSEFunc_0_eppppii fn = s->vex_l ? ymm : xmm;                                   \
569                                                                                   \
570    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
571       tcg_constant_i32(even),                                                     \
572       tcg_constant_i32((even) ^ (odd)));                                          \
573}
574
575#define FMA_SSE(uname, ptr0, ptr1, ptr2, flags)                                    \
576FMA_SSE_PACKED(uname, ptr0, ptr1, ptr2, flags, flags)                              \
577static void gen_##uname##Sx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
578{                                                                                  \
579    SSEFunc_0_eppppi fn = s->vex_w ? gen_helper_fma4sd : gen_helper_fma4ss;        \
580                                                                                   \
581    fn(tcg_env, OP_PTR0, ptr0, ptr1, ptr2,                                         \
582       tcg_constant_i32(flags));                                                   \
583}                                                                                  \
584
585FMA_SSE(VFMADD231,  OP_PTR1, OP_PTR2, OP_PTR0, 0)
586FMA_SSE(VFMADD213,  OP_PTR1, OP_PTR0, OP_PTR2, 0)
587FMA_SSE(VFMADD132,  OP_PTR0, OP_PTR2, OP_PTR1, 0)
588
589FMA_SSE(VFNMADD231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_product)
590FMA_SSE(VFNMADD213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_product)
591FMA_SSE(VFNMADD132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_product)
592
593FMA_SSE(VFMSUB231,  OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c)
594FMA_SSE(VFMSUB213,  OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c)
595FMA_SSE(VFMSUB132,  OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c)
596
597FMA_SSE(VFNMSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c|float_muladd_negate_product)
598FMA_SSE(VFNMSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c|float_muladd_negate_product)
599FMA_SSE(VFNMSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c|float_muladd_negate_product)
600
601FMA_SSE_PACKED(VFMADDSUB231, OP_PTR1, OP_PTR2, OP_PTR0, float_muladd_negate_c, 0)
602FMA_SSE_PACKED(VFMADDSUB213, OP_PTR1, OP_PTR0, OP_PTR2, float_muladd_negate_c, 0)
603FMA_SSE_PACKED(VFMADDSUB132, OP_PTR0, OP_PTR2, OP_PTR1, float_muladd_negate_c, 0)
604
605FMA_SSE_PACKED(VFMSUBADD231, OP_PTR1, OP_PTR2, OP_PTR0, 0, float_muladd_negate_c)
606FMA_SSE_PACKED(VFMSUBADD213, OP_PTR1, OP_PTR0, OP_PTR2, 0, float_muladd_negate_c)
607FMA_SSE_PACKED(VFMSUBADD132, OP_PTR0, OP_PTR2, OP_PTR1, 0, float_muladd_negate_c)
608
609#define FP_UNPACK_SSE(uname, lname)                                                \
610static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
611{                                                                                  \
612    /* PS maps to the DQ integer instruction, PD maps to QDQ.  */                  \
613    gen_fp_sse(s, env, decode,                                                     \
614               gen_helper_##lname##qdq_xmm,                                        \
615               gen_helper_##lname##dq_xmm,                                         \
616               gen_helper_##lname##qdq_ymm,                                        \
617               gen_helper_##lname##dq_ymm,                                         \
618               NULL, NULL);                                                        \
619}
620FP_UNPACK_SSE(VUNPCKLPx, punpckl)
621FP_UNPACK_SSE(VUNPCKHPx, punpckh)
622
623/*
624 * 00 = v*ps Vps, Wpd
625 * f3 = v*ss Vss, Wps
626 */
627static inline void gen_unary_fp32_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
628                                      SSEFunc_0_epp ps_xmm,
629                                      SSEFunc_0_epp ps_ymm,
630                                      SSEFunc_0_eppp ss)
631{
632    if ((s->prefix & (PREFIX_DATA | PREFIX_REPNZ)) != 0) {
633        goto illegal_op;
634    } else if (s->prefix & PREFIX_REPZ) {
635        if (!ss) {
636            goto illegal_op;
637        }
638        ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
639    } else {
640        SSEFunc_0_epp fn = s->vex_l ? ps_ymm : ps_xmm;
641        if (!fn) {
642            goto illegal_op;
643        }
644        fn(tcg_env, OP_PTR0, OP_PTR2);
645    }
646    return;
647
648illegal_op:
649    gen_illegal_opcode(s);
650}
651#define UNARY_FP32_SSE(uname, lname)                                               \
652static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
653{                                                                                  \
654    gen_unary_fp32_sse(s, env, decode,                                             \
655                       gen_helper_##lname##ps_xmm,                                 \
656                       gen_helper_##lname##ps_ymm,                                 \
657                       gen_helper_##lname##ss);                                    \
658}
659UNARY_FP32_SSE(VRSQRT, rsqrt)
660UNARY_FP32_SSE(VRCP, rcp)
661
662/*
663 * 66 = v*pd Vpd, Hpd, Wpd
664 * f2 = v*ps Vps, Hps, Wps
665 */
666static inline void gen_horizontal_fp_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
667                                         SSEFunc_0_eppp pd_xmm, SSEFunc_0_eppp ps_xmm,
668                                         SSEFunc_0_eppp pd_ymm, SSEFunc_0_eppp ps_ymm)
669{
670    SSEFunc_0_eppp ps, pd, fn;
671    ps = s->vex_l ? ps_ymm : ps_xmm;
672    pd = s->vex_l ? pd_ymm : pd_xmm;
673    fn = s->prefix & PREFIX_DATA ? pd : ps;
674    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
675}
676#define HORIZONTAL_FP_SSE(uname, lname)                                            \
677static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
678{                                                                                  \
679    gen_horizontal_fp_sse(s, env, decode,                                          \
680                          gen_helper_##lname##pd_xmm, gen_helper_##lname##ps_xmm,  \
681                          gen_helper_##lname##pd_ymm, gen_helper_##lname##ps_ymm); \
682}
683HORIZONTAL_FP_SSE(VHADD, hadd)
684HORIZONTAL_FP_SSE(VHSUB, hsub)
685HORIZONTAL_FP_SSE(VADDSUB, addsub)
686
687static inline void gen_ternary_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
688                                   int op3, SSEFunc_0_epppp xmm, SSEFunc_0_epppp ymm)
689{
690    SSEFunc_0_epppp fn = s->vex_l ? ymm : xmm;
691    TCGv_ptr ptr3 = tcg_temp_new_ptr();
692
693    /* The format of the fourth input is Lx */
694    tcg_gen_addi_ptr(ptr3, tcg_env, ZMM_OFFSET(op3));
695    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, ptr3);
696}
697#define TERNARY_SSE(uname, uvname, lname)                                          \
698static void gen_##uvname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
699{                                                                                  \
700    gen_ternary_sse(s, env, decode, (uint8_t)decode->immediate >> 4,               \
701                    gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);           \
702}                                                                                  \
703static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
704{                                                                                  \
705    gen_ternary_sse(s, env, decode, 0,                                             \
706                  gen_helper_##lname##_xmm, gen_helper_##lname##_ymm);             \
707}
708TERNARY_SSE(BLENDVPS, VBLENDVPS, blendvps)
709TERNARY_SSE(BLENDVPD, VBLENDVPD, blendvpd)
710TERNARY_SSE(PBLENDVB, VPBLENDVB, pblendvb)
711
712static inline void gen_binary_imm_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
713                                      SSEFunc_0_epppi xmm, SSEFunc_0_epppi ymm)
714{
715    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
716    if (!s->vex_l) {
717        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
718    } else {
719        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
720    }
721}
722
723#define BINARY_IMM_SSE(uname, lname)                                               \
724static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
725{                                                                                  \
726    gen_binary_imm_sse(s, env, decode,                                             \
727                       gen_helper_##lname##_xmm,                                   \
728                       gen_helper_##lname##_ymm);                                  \
729}
730
731BINARY_IMM_SSE(VBLENDPD,   blendpd)
732BINARY_IMM_SSE(VBLENDPS,   blendps)
733BINARY_IMM_SSE(VPBLENDW,   pblendw)
734BINARY_IMM_SSE(VDDPS,      dpps)
735#define gen_helper_dppd_ymm NULL
736BINARY_IMM_SSE(VDDPD,      dppd)
737BINARY_IMM_SSE(VMPSADBW,   mpsadbw)
738BINARY_IMM_SSE(PCLMULQDQ,  pclmulqdq)
739
740
741#define UNARY_INT_GVEC(uname, func, ...)                                           \
742static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
743{                                                                                  \
744    int vec_len = vector_len(s, decode);                                          \
745                                                                                   \
746    func(__VA_ARGS__, decode->op[0].offset,                                        \
747         decode->op[2].offset, vec_len, vec_len);                                  \
748}
749UNARY_INT_GVEC(PABSB,          tcg_gen_gvec_abs, MO_8)
750UNARY_INT_GVEC(PABSW,          tcg_gen_gvec_abs, MO_16)
751UNARY_INT_GVEC(PABSD,          tcg_gen_gvec_abs, MO_32)
752UNARY_INT_GVEC(VBROADCASTx128, tcg_gen_gvec_dup_mem, MO_128)
753UNARY_INT_GVEC(VPBROADCASTB,   tcg_gen_gvec_dup_mem, MO_8)
754UNARY_INT_GVEC(VPBROADCASTW,   tcg_gen_gvec_dup_mem, MO_16)
755UNARY_INT_GVEC(VPBROADCASTD,   tcg_gen_gvec_dup_mem, MO_32)
756UNARY_INT_GVEC(VPBROADCASTQ,   tcg_gen_gvec_dup_mem, MO_64)
757
758
759#define BINARY_INT_GVEC(uname, func, ...)                                          \
760static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
761{                                                                                  \
762    int vec_len = vector_len(s, decode);                                          \
763                                                                                   \
764    func(__VA_ARGS__,                                                              \
765         decode->op[0].offset, decode->op[1].offset,                               \
766         decode->op[2].offset, vec_len, vec_len);                                  \
767}
768
769BINARY_INT_GVEC(PADDB,   tcg_gen_gvec_add, MO_8)
770BINARY_INT_GVEC(PADDW,   tcg_gen_gvec_add, MO_16)
771BINARY_INT_GVEC(PADDD,   tcg_gen_gvec_add, MO_32)
772BINARY_INT_GVEC(PADDQ,   tcg_gen_gvec_add, MO_64)
773BINARY_INT_GVEC(PADDSB,  tcg_gen_gvec_ssadd, MO_8)
774BINARY_INT_GVEC(PADDSW,  tcg_gen_gvec_ssadd, MO_16)
775BINARY_INT_GVEC(PADDUSB, tcg_gen_gvec_usadd, MO_8)
776BINARY_INT_GVEC(PADDUSW, tcg_gen_gvec_usadd, MO_16)
777BINARY_INT_GVEC(PAND,    tcg_gen_gvec_and, MO_64)
778BINARY_INT_GVEC(PCMPEQB, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_8)
779BINARY_INT_GVEC(PCMPEQD, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_32)
780BINARY_INT_GVEC(PCMPEQW, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_16)
781BINARY_INT_GVEC(PCMPEQQ, tcg_gen_gvec_cmp, TCG_COND_EQ, MO_64)
782BINARY_INT_GVEC(PCMPGTB, tcg_gen_gvec_cmp, TCG_COND_GT, MO_8)
783BINARY_INT_GVEC(PCMPGTW, tcg_gen_gvec_cmp, TCG_COND_GT, MO_16)
784BINARY_INT_GVEC(PCMPGTD, tcg_gen_gvec_cmp, TCG_COND_GT, MO_32)
785BINARY_INT_GVEC(PCMPGTQ, tcg_gen_gvec_cmp, TCG_COND_GT, MO_64)
786BINARY_INT_GVEC(PMAXSB,  tcg_gen_gvec_smax, MO_8)
787BINARY_INT_GVEC(PMAXSW,  tcg_gen_gvec_smax, MO_16)
788BINARY_INT_GVEC(PMAXSD,  tcg_gen_gvec_smax, MO_32)
789BINARY_INT_GVEC(PMAXUB,  tcg_gen_gvec_umax, MO_8)
790BINARY_INT_GVEC(PMAXUW,  tcg_gen_gvec_umax, MO_16)
791BINARY_INT_GVEC(PMAXUD,  tcg_gen_gvec_umax, MO_32)
792BINARY_INT_GVEC(PMINSB,  tcg_gen_gvec_smin, MO_8)
793BINARY_INT_GVEC(PMINSW,  tcg_gen_gvec_smin, MO_16)
794BINARY_INT_GVEC(PMINSD,  tcg_gen_gvec_smin, MO_32)
795BINARY_INT_GVEC(PMINUB,  tcg_gen_gvec_umin, MO_8)
796BINARY_INT_GVEC(PMINUW,  tcg_gen_gvec_umin, MO_16)
797BINARY_INT_GVEC(PMINUD,  tcg_gen_gvec_umin, MO_32)
798BINARY_INT_GVEC(PMULLW,  tcg_gen_gvec_mul, MO_16)
799BINARY_INT_GVEC(PMULLD,  tcg_gen_gvec_mul, MO_32)
800BINARY_INT_GVEC(POR,     tcg_gen_gvec_or, MO_64)
801BINARY_INT_GVEC(PSUBB,   tcg_gen_gvec_sub, MO_8)
802BINARY_INT_GVEC(PSUBW,   tcg_gen_gvec_sub, MO_16)
803BINARY_INT_GVEC(PSUBD,   tcg_gen_gvec_sub, MO_32)
804BINARY_INT_GVEC(PSUBQ,   tcg_gen_gvec_sub, MO_64)
805BINARY_INT_GVEC(PSUBSB,  tcg_gen_gvec_sssub, MO_8)
806BINARY_INT_GVEC(PSUBSW,  tcg_gen_gvec_sssub, MO_16)
807BINARY_INT_GVEC(PSUBUSB, tcg_gen_gvec_ussub, MO_8)
808BINARY_INT_GVEC(PSUBUSW, tcg_gen_gvec_ussub, MO_16)
809BINARY_INT_GVEC(PXOR,    tcg_gen_gvec_xor, MO_64)
810
811
812/*
813 * 00 = p*  Pq, Qq (if mmx not NULL; no VEX)
814 * 66 = vp* Vx, Hx, Wx
815 *
816 * These are really the same encoding, because 1) V is the same as P when VEX.V
817 * is not present 2) P and Q are the same as H and W apart from MM/XMM
818 */
819static inline void gen_binary_int_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
820                                      SSEFunc_0_eppp mmx, SSEFunc_0_eppp xmm, SSEFunc_0_eppp ymm)
821{
822    assert(!!mmx == !!(decode->e.special == X86_SPECIAL_MMX));
823
824    if (mmx && (s->prefix & PREFIX_VEX) && !(s->prefix & PREFIX_DATA)) {
825        /* VEX encoding is not applicable to MMX instructions.  */
826        gen_illegal_opcode(s);
827        return;
828    }
829    if (!(s->prefix & PREFIX_DATA)) {
830        mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
831    } else if (!s->vex_l) {
832        xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
833    } else {
834        ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
835    }
836}
837
838
839#define BINARY_INT_MMX(uname, lname)                                               \
840static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
841{                                                                                  \
842    gen_binary_int_sse(s, env, decode,                                             \
843                          gen_helper_##lname##_mmx,                                \
844                          gen_helper_##lname##_xmm,                                \
845                          gen_helper_##lname##_ymm);                               \
846}
847BINARY_INT_MMX(PUNPCKLBW,  punpcklbw)
848BINARY_INT_MMX(PUNPCKLWD,  punpcklwd)
849BINARY_INT_MMX(PUNPCKLDQ,  punpckldq)
850BINARY_INT_MMX(PACKSSWB,   packsswb)
851BINARY_INT_MMX(PACKUSWB,   packuswb)
852BINARY_INT_MMX(PUNPCKHBW,  punpckhbw)
853BINARY_INT_MMX(PUNPCKHWD,  punpckhwd)
854BINARY_INT_MMX(PUNPCKHDQ,  punpckhdq)
855BINARY_INT_MMX(PACKSSDW,   packssdw)
856
857BINARY_INT_MMX(PAVGB,   pavgb)
858BINARY_INT_MMX(PAVGW,   pavgw)
859BINARY_INT_MMX(PMADDWD, pmaddwd)
860BINARY_INT_MMX(PMULHUW, pmulhuw)
861BINARY_INT_MMX(PMULHW,  pmulhw)
862BINARY_INT_MMX(PMULUDQ, pmuludq)
863BINARY_INT_MMX(PSADBW,  psadbw)
864
865BINARY_INT_MMX(PSLLW_r, psllw)
866BINARY_INT_MMX(PSLLD_r, pslld)
867BINARY_INT_MMX(PSLLQ_r, psllq)
868BINARY_INT_MMX(PSRLW_r, psrlw)
869BINARY_INT_MMX(PSRLD_r, psrld)
870BINARY_INT_MMX(PSRLQ_r, psrlq)
871BINARY_INT_MMX(PSRAW_r, psraw)
872BINARY_INT_MMX(PSRAD_r, psrad)
873
874BINARY_INT_MMX(PHADDW,    phaddw)
875BINARY_INT_MMX(PHADDSW,   phaddsw)
876BINARY_INT_MMX(PHADDD,    phaddd)
877BINARY_INT_MMX(PHSUBW,    phsubw)
878BINARY_INT_MMX(PHSUBSW,   phsubsw)
879BINARY_INT_MMX(PHSUBD,    phsubd)
880BINARY_INT_MMX(PMADDUBSW, pmaddubsw)
881BINARY_INT_MMX(PSHUFB,    pshufb)
882BINARY_INT_MMX(PSIGNB,    psignb)
883BINARY_INT_MMX(PSIGNW,    psignw)
884BINARY_INT_MMX(PSIGND,    psignd)
885BINARY_INT_MMX(PMULHRSW,  pmulhrsw)
886
887/* Instructions with no MMX equivalent.  */
888#define BINARY_INT_SSE(uname, lname)                                               \
889static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
890{                                                                                  \
891    gen_binary_int_sse(s, env, decode,                                             \
892                          NULL,                                                    \
893                          gen_helper_##lname##_xmm,                                \
894                          gen_helper_##lname##_ymm);                               \
895}
896
897/* Instructions with no MMX equivalent.  */
898BINARY_INT_SSE(PUNPCKLQDQ, punpcklqdq)
899BINARY_INT_SSE(PUNPCKHQDQ, punpckhqdq)
900BINARY_INT_SSE(VPACKUSDW,  packusdw)
901BINARY_INT_SSE(VPERMILPS,  vpermilps)
902BINARY_INT_SSE(VPERMILPD,  vpermilpd)
903BINARY_INT_SSE(VMASKMOVPS, vpmaskmovd)
904BINARY_INT_SSE(VMASKMOVPD, vpmaskmovq)
905
906BINARY_INT_SSE(PMULDQ,    pmuldq)
907
908BINARY_INT_SSE(VAESDEC, aesdec)
909BINARY_INT_SSE(VAESDECLAST, aesdeclast)
910BINARY_INT_SSE(VAESENC, aesenc)
911BINARY_INT_SSE(VAESENCLAST, aesenclast)
912
913#define UNARY_CMP_SSE(uname, lname)                                                \
914static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
915{                                                                                  \
916    if (!s->vex_l) {                                                               \
917        gen_helper_##lname##_xmm(tcg_env, OP_PTR1, OP_PTR2);                       \
918    } else {                                                                       \
919        gen_helper_##lname##_ymm(tcg_env, OP_PTR1, OP_PTR2);                       \
920    }                                                                              \
921    set_cc_op(s, CC_OP_EFLAGS);                                                    \
922}
923UNARY_CMP_SSE(VPTEST,     ptest)
924UNARY_CMP_SSE(VTESTPS,    vtestps)
925UNARY_CMP_SSE(VTESTPD,    vtestpd)
926
927static inline void gen_unary_int_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
928                                     SSEFunc_0_epp xmm, SSEFunc_0_epp ymm)
929{
930    if (!s->vex_l) {
931        xmm(tcg_env, OP_PTR0, OP_PTR2);
932    } else {
933        ymm(tcg_env, OP_PTR0, OP_PTR2);
934    }
935}
936
937#define UNARY_INT_SSE(uname, lname)                                                \
938static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
939{                                                                                  \
940    gen_unary_int_sse(s, env, decode,                                              \
941                      gen_helper_##lname##_xmm,                                    \
942                      gen_helper_##lname##_ymm);                                   \
943}
944
945UNARY_INT_SSE(VPMOVSXBW,    pmovsxbw)
946UNARY_INT_SSE(VPMOVSXBD,    pmovsxbd)
947UNARY_INT_SSE(VPMOVSXBQ,    pmovsxbq)
948UNARY_INT_SSE(VPMOVSXWD,    pmovsxwd)
949UNARY_INT_SSE(VPMOVSXWQ,    pmovsxwq)
950UNARY_INT_SSE(VPMOVSXDQ,    pmovsxdq)
951
952UNARY_INT_SSE(VPMOVZXBW,    pmovzxbw)
953UNARY_INT_SSE(VPMOVZXBD,    pmovzxbd)
954UNARY_INT_SSE(VPMOVZXBQ,    pmovzxbq)
955UNARY_INT_SSE(VPMOVZXWD,    pmovzxwd)
956UNARY_INT_SSE(VPMOVZXWQ,    pmovzxwq)
957UNARY_INT_SSE(VPMOVZXDQ,    pmovzxdq)
958
959UNARY_INT_SSE(VMOVSLDUP,    pmovsldup)
960UNARY_INT_SSE(VMOVSHDUP,    pmovshdup)
961UNARY_INT_SSE(VMOVDDUP,     pmovdldup)
962
963UNARY_INT_SSE(VCVTDQ2PD, cvtdq2pd)
964UNARY_INT_SSE(VCVTPD2DQ, cvtpd2dq)
965UNARY_INT_SSE(VCVTTPD2DQ, cvttpd2dq)
966UNARY_INT_SSE(VCVTDQ2PS, cvtdq2ps)
967UNARY_INT_SSE(VCVTPS2DQ, cvtps2dq)
968UNARY_INT_SSE(VCVTTPS2DQ, cvttps2dq)
969UNARY_INT_SSE(VCVTPH2PS, cvtph2ps)
970
971
972static inline void gen_unary_imm_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
973                                     SSEFunc_0_ppi xmm, SSEFunc_0_ppi ymm)
974{
975    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
976    if (!s->vex_l) {
977        xmm(OP_PTR0, OP_PTR1, imm);
978    } else {
979        ymm(OP_PTR0, OP_PTR1, imm);
980    }
981}
982
983#define UNARY_IMM_SSE(uname, lname)                                                \
984static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
985{                                                                                  \
986    gen_unary_imm_sse(s, env, decode,                                              \
987                      gen_helper_##lname##_xmm,                                    \
988                      gen_helper_##lname##_ymm);                                   \
989}
990
991UNARY_IMM_SSE(PSHUFD,     pshufd)
992UNARY_IMM_SSE(PSHUFHW,    pshufhw)
993UNARY_IMM_SSE(PSHUFLW,    pshuflw)
994#define gen_helper_vpermq_xmm NULL
995UNARY_IMM_SSE(VPERMQ,      vpermq)
996UNARY_IMM_SSE(VPERMILPS_i, vpermilps_imm)
997UNARY_IMM_SSE(VPERMILPD_i, vpermilpd_imm)
998
999static inline void gen_unary_imm_fp_sse(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
1000                                        SSEFunc_0_eppi xmm, SSEFunc_0_eppi ymm)
1001{
1002    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
1003    if (!s->vex_l) {
1004        xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
1005    } else {
1006        ymm(tcg_env, OP_PTR0, OP_PTR1, imm);
1007    }
1008}
1009
1010#define UNARY_IMM_FP_SSE(uname, lname)                                             \
1011static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
1012{                                                                                  \
1013    gen_unary_imm_fp_sse(s, env, decode,                                           \
1014                      gen_helper_##lname##_xmm,                                    \
1015                      gen_helper_##lname##_ymm);                                   \
1016}
1017
1018UNARY_IMM_FP_SSE(VROUNDPS,    roundps)
1019UNARY_IMM_FP_SSE(VROUNDPD,    roundpd)
1020
1021static inline void gen_vexw_avx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
1022                                SSEFunc_0_eppp d_xmm, SSEFunc_0_eppp q_xmm,
1023                                SSEFunc_0_eppp d_ymm, SSEFunc_0_eppp q_ymm)
1024{
1025    SSEFunc_0_eppp d = s->vex_l ? d_ymm : d_xmm;
1026    SSEFunc_0_eppp q = s->vex_l ? q_ymm : q_xmm;
1027    SSEFunc_0_eppp fn = s->vex_w ? q : d;
1028    fn(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
1029}
1030
1031/* VEX.W affects whether to operate on 32- or 64-bit elements.  */
1032#define VEXW_AVX(uname, lname)                                                     \
1033static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
1034{                                                                                  \
1035    gen_vexw_avx(s, env, decode,                                                   \
1036                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1037                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1038}
1039VEXW_AVX(VPSLLV,    vpsllv)
1040VEXW_AVX(VPSRLV,    vpsrlv)
1041VEXW_AVX(VPSRAV,    vpsrav)
1042VEXW_AVX(VPMASKMOV, vpmaskmov)
1043
1044/* Same as above, but with extra arguments to the helper.  */
1045static inline void gen_vsib_avx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
1046                                SSEFunc_0_epppti d_xmm, SSEFunc_0_epppti q_xmm,
1047                                SSEFunc_0_epppti d_ymm, SSEFunc_0_epppti q_ymm)
1048{
1049    SSEFunc_0_epppti d = s->vex_l ? d_ymm : d_xmm;
1050    SSEFunc_0_epppti q = s->vex_l ? q_ymm : q_xmm;
1051    SSEFunc_0_epppti fn = s->vex_w ? q : d;
1052    TCGv_i32 scale = tcg_constant_i32(decode->mem.scale);
1053    TCGv_ptr index = tcg_temp_new_ptr();
1054
1055    /* Pass third input as (index, base, scale) */
1056    tcg_gen_addi_ptr(index, tcg_env, ZMM_OFFSET(decode->mem.index));
1057    fn(tcg_env, OP_PTR0, OP_PTR1, index, s->A0, scale);
1058
1059    /*
1060     * There are two output operands, so zero OP1's high 128 bits
1061     * in the VEX.128 case.
1062     */
1063    if (!s->vex_l) {
1064        int ymmh_ofs = vector_elem_offset(&decode->op[1], MO_128, 1);
1065        tcg_gen_gvec_dup_imm(MO_64, ymmh_ofs, 16, 16, 0);
1066    }
1067}
1068#define VSIB_AVX(uname, lname)                                                     \
1069static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode) \
1070{                                                                                  \
1071    gen_vsib_avx(s, env, decode,                                                   \
1072                 gen_helper_##lname##d_xmm, gen_helper_##lname##q_xmm,             \
1073                 gen_helper_##lname##d_ymm, gen_helper_##lname##q_ymm);            \
1074}
1075VSIB_AVX(VPGATHERD, vpgatherd)
1076VSIB_AVX(VPGATHERQ, vpgatherq)
1077
1078static void gen_AAA(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1079{
1080    gen_update_cc_op(s);
1081    gen_helper_aaa(tcg_env);
1082    set_cc_op(s, CC_OP_EFLAGS);
1083}
1084
1085static void gen_AAD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1086{
1087    gen_helper_aad(tcg_env, tcg_constant_i32(decode->immediate));
1088    set_cc_op(s, CC_OP_LOGICB);
1089}
1090
1091static void gen_AAM(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1092{
1093    if (decode->immediate == 0) {
1094        gen_exception(s, EXCP00_DIVZ);
1095    } else {
1096        gen_helper_aam(tcg_env, tcg_constant_i32(decode->immediate));
1097        set_cc_op(s, CC_OP_LOGICB);
1098    }
1099}
1100
1101static void gen_AAS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1102{
1103    gen_update_cc_op(s);
1104    gen_helper_aas(tcg_env);
1105    set_cc_op(s, CC_OP_EFLAGS);
1106}
1107
1108static void gen_ADC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1109{
1110    MemOp ot = decode->op[1].ot;
1111    TCGv c_in = tcg_temp_new();
1112
1113    gen_compute_eflags_c(s, c_in);
1114    if (s->prefix & PREFIX_LOCK) {
1115        tcg_gen_add_tl(s->T0, c_in, s->T1);
1116        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
1117                                    s->mem_index, ot | MO_LE);
1118    } else {
1119        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1120        tcg_gen_add_tl(s->T0, s->T0, c_in);
1121    }
1122    prepare_update3_cc(decode, s, CC_OP_ADCB + ot, c_in);
1123}
1124
1125/* ADCX/ADOX do not have memory operands and can use set_cc_op.  */
1126static void gen_ADCOX(DisasContext *s, CPUX86State *env, MemOp ot, int cc_op)
1127{
1128    int opposite_cc_op;
1129    TCGv carry_in = NULL;
1130    TCGv carry_out = (cc_op == CC_OP_ADCX ? cpu_cc_dst : cpu_cc_src2);
1131    TCGv zero;
1132
1133    if (cc_op == s->cc_op || s->cc_op == CC_OP_ADCOX) {
1134        /* Re-use the carry-out from a previous round.  */
1135        carry_in = carry_out;
1136    } else {
1137        /* We don't have a carry-in, get it out of EFLAGS.  */
1138        if (s->cc_op != CC_OP_ADCX && s->cc_op != CC_OP_ADOX) {
1139            gen_compute_eflags(s);
1140        }
1141        carry_in = s->tmp0;
1142        tcg_gen_extract_tl(carry_in, cpu_cc_src,
1143            ctz32(cc_op == CC_OP_ADCX ? CC_C : CC_O), 1);
1144    }
1145
1146    switch (ot) {
1147#ifdef TARGET_X86_64
1148    case MO_32:
1149        /* If TL is 64-bit just do everything in 64-bit arithmetic.  */
1150        tcg_gen_ext32u_tl(s->T0, s->T0);
1151        tcg_gen_ext32u_tl(s->T1, s->T1);
1152        tcg_gen_add_i64(s->T0, s->T0, s->T1);
1153        tcg_gen_add_i64(s->T0, s->T0, carry_in);
1154        tcg_gen_shri_i64(carry_out, s->T0, 32);
1155        break;
1156#endif
1157    default:
1158        zero = tcg_constant_tl(0);
1159        tcg_gen_add2_tl(s->T0, carry_out, s->T0, zero, carry_in, zero);
1160        tcg_gen_add2_tl(s->T0, carry_out, s->T0, carry_out, s->T1, zero);
1161        break;
1162    }
1163
1164    opposite_cc_op = cc_op == CC_OP_ADCX ? CC_OP_ADOX : CC_OP_ADCX;
1165    if (s->cc_op == CC_OP_ADCOX || s->cc_op == opposite_cc_op) {
1166        /* Merge with the carry-out from the opposite instruction.  */
1167        set_cc_op(s, CC_OP_ADCOX);
1168    } else {
1169        set_cc_op(s, cc_op);
1170    }
1171}
1172
1173static void gen_ADCX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1174{
1175    gen_ADCOX(s, env, decode->op[0].ot, CC_OP_ADCX);
1176}
1177
1178static void gen_ADD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1179{
1180    MemOp ot = decode->op[1].ot;
1181
1182    if (s->prefix & PREFIX_LOCK) {
1183        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1184                                    s->mem_index, ot | MO_LE);
1185    } else {
1186        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1187    }
1188    prepare_update2_cc(decode, s, CC_OP_ADDB + ot);
1189}
1190
1191static void gen_ADOX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1192{
1193    gen_ADCOX(s, env, decode->op[0].ot, CC_OP_ADOX);
1194}
1195
1196static void gen_AND(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1197{
1198    MemOp ot = decode->op[1].ot;
1199
1200    if (s->prefix & PREFIX_LOCK) {
1201        tcg_gen_atomic_and_fetch_tl(s->T0, s->A0, s->T1,
1202                                    s->mem_index, ot | MO_LE);
1203    } else {
1204        tcg_gen_and_tl(s->T0, s->T0, s->T1);
1205    }
1206    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1207}
1208
1209static void gen_ANDN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1210{
1211    MemOp ot = decode->op[0].ot;
1212
1213    tcg_gen_andc_tl(s->T0, s->T1, s->T0);
1214    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1215}
1216
1217static void gen_ARPL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1218{
1219    TCGv zf = tcg_temp_new();
1220    TCGv flags = tcg_temp_new();
1221
1222    gen_mov_eflags(s, flags);
1223
1224    /* Compute adjusted DST in T1, merging in SRC[RPL].  */
1225    tcg_gen_deposit_tl(s->T1, s->T0, s->T1, 0, 2);
1226
1227    /* Z flag set if DST[RPL] < SRC[RPL] */
1228    tcg_gen_setcond_tl(TCG_COND_LTU, zf, s->T0, s->T1);
1229    tcg_gen_deposit_tl(flags, flags, zf, ctz32(CC_Z), 1);
1230
1231    /* Place maximum RPL in DST */
1232    tcg_gen_umax_tl(s->T0, s->T0, s->T1);
1233
1234    decode->cc_src = flags;
1235    decode->cc_op = CC_OP_EFLAGS;
1236}
1237
1238static void gen_BEXTR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1239{
1240    MemOp ot = decode->op[0].ot;
1241    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1242    TCGv zero = tcg_constant_tl(0);
1243    TCGv mone = tcg_constant_tl(-1);
1244
1245    /*
1246     * Extract START, and shift the operand.
1247     * Shifts larger than operand size get zeros.
1248     */
1249    tcg_gen_ext8u_tl(s->A0, s->T1);
1250    tcg_gen_shr_tl(s->T0, s->T0, s->A0);
1251
1252    tcg_gen_movcond_tl(TCG_COND_LEU, s->T0, s->A0, bound, s->T0, zero);
1253
1254    /*
1255     * Extract the LEN into an inverse mask.  Lengths larger than
1256     * operand size get all zeros, length 0 gets all ones.
1257     */
1258    tcg_gen_extract_tl(s->A0, s->T1, 8, 8);
1259    tcg_gen_shl_tl(s->T1, mone, s->A0);
1260    tcg_gen_movcond_tl(TCG_COND_LEU, s->T1, s->A0, bound, s->T1, zero);
1261    tcg_gen_andc_tl(s->T0, s->T0, s->T1);
1262
1263    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
1264}
1265
1266/* BLSI do not have memory operands and can use set_cc_op.  */
1267static void gen_BLSI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1268{
1269    MemOp ot = decode->op[0].ot;
1270
1271    tcg_gen_mov_tl(cpu_cc_src, s->T0);
1272    tcg_gen_neg_tl(s->T1, s->T0);
1273    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1274    tcg_gen_mov_tl(cpu_cc_dst, s->T0);
1275    set_cc_op(s, CC_OP_BMILGB + ot);
1276}
1277
1278/* BLSMSK do not have memory operands and can use set_cc_op.  */
1279static void gen_BLSMSK(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1280{
1281    MemOp ot = decode->op[0].ot;
1282
1283    tcg_gen_mov_tl(cpu_cc_src, s->T0);
1284    tcg_gen_subi_tl(s->T1, s->T0, 1);
1285    tcg_gen_xor_tl(s->T0, s->T0, s->T1);
1286    tcg_gen_mov_tl(cpu_cc_dst, s->T0);
1287    set_cc_op(s, CC_OP_BMILGB + ot);
1288}
1289
1290/* BLSR do not have memory operands and can use set_cc_op.  */
1291static void gen_BLSR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1292{
1293    MemOp ot = decode->op[0].ot;
1294
1295    tcg_gen_mov_tl(cpu_cc_src, s->T0);
1296    tcg_gen_subi_tl(s->T1, s->T0, 1);
1297    tcg_gen_and_tl(s->T0, s->T0, s->T1);
1298    tcg_gen_mov_tl(cpu_cc_dst, s->T0);
1299    set_cc_op(s, CC_OP_BMILGB + ot);
1300}
1301
1302static void gen_BOUND(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1303{
1304    TCGv_i32 op = tcg_temp_new_i32();
1305    tcg_gen_trunc_tl_i32(op, s->T0);
1306    if (decode->op[1].ot == MO_16) {
1307        gen_helper_boundw(tcg_env, s->A0, op);
1308    } else {
1309        gen_helper_boundl(tcg_env, s->A0, op);
1310    }
1311}
1312
1313static void gen_BSWAP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1314{
1315#ifdef TARGET_X86_64
1316    if (s->dflag == MO_64) {
1317        tcg_gen_bswap64_i64(s->T0, s->T0);
1318        return;
1319    }
1320#endif
1321    tcg_gen_bswap32_tl(s->T0, s->T0, TCG_BSWAP_OZ);
1322}
1323
1324static void gen_BZHI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1325{
1326    MemOp ot = decode->op[0].ot;
1327    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
1328    TCGv zero = tcg_constant_tl(0);
1329    TCGv mone = tcg_constant_tl(-1);
1330
1331    tcg_gen_ext8u_tl(s->T1, s->T1);
1332
1333    tcg_gen_shl_tl(s->A0, mone, s->T1);
1334    tcg_gen_movcond_tl(TCG_COND_LEU, s->A0, s->T1, bound, s->A0, zero);
1335    tcg_gen_andc_tl(s->T0, s->T0, s->A0);
1336    /*
1337     * Note that since we're using BMILG (in order to get O
1338     * cleared) we need to store the inverse into C.
1339     */
1340    tcg_gen_setcond_tl(TCG_COND_LEU, s->T1, s->T1, bound);
1341    prepare_update2_cc(decode, s, CC_OP_BMILGB + ot);
1342}
1343
1344static void gen_CALL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1345{
1346    gen_push_v(s, eip_next_tl(s));
1347    gen_JMP(s, env, decode);
1348}
1349
1350static void gen_CALL_m(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1351{
1352    gen_push_v(s, eip_next_tl(s));
1353    gen_JMP_m(s, env, decode);
1354}
1355
1356static void gen_CALLF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1357{
1358    gen_far_call(s);
1359}
1360
1361static void gen_CALLF_m(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1362{
1363    MemOp ot = decode->op[2].ot;
1364
1365    gen_op_ld_v(s, ot, s->T0, s->A0);
1366    gen_add_A0_im(s, 1 << ot);
1367    gen_op_ld_v(s, MO_16, s->T1, s->A0);
1368    gen_far_call(s);
1369}
1370
1371static void gen_CBW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1372{
1373    MemOp src_ot = decode->op[0].ot - 1;
1374
1375    tcg_gen_ext_tl(s->T0, s->T0, src_ot | MO_SIGN);
1376}
1377
1378static void gen_CLC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1379{
1380    gen_compute_eflags(s);
1381    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, ~CC_C);
1382}
1383
1384static void gen_CLD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1385{
1386    tcg_gen_st_i32(tcg_constant_i32(1), tcg_env, offsetof(CPUX86State, df));
1387}
1388
1389static void gen_CLI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1390{
1391    gen_reset_eflags(s, IF_MASK);
1392}
1393
1394static void gen_CMC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1395{
1396    gen_compute_eflags(s);
1397    tcg_gen_xori_tl(cpu_cc_src, cpu_cc_src, CC_C);
1398}
1399
1400static void gen_CMOVcc(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1401{
1402    gen_cmovcc1(s, decode->b & 0xf, s->T0, s->T1);
1403}
1404
1405static void gen_CMPccXADD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1406{
1407    TCGLabel *label_top = gen_new_label();
1408    TCGLabel *label_bottom = gen_new_label();
1409    TCGv oldv = tcg_temp_new();
1410    TCGv newv = tcg_temp_new();
1411    TCGv cmpv = tcg_temp_new();
1412    TCGCond cond;
1413
1414    TCGv cmp_lhs, cmp_rhs;
1415    MemOp ot, ot_full;
1416
1417    int jcc_op = (decode->b >> 1) & 7;
1418    static const TCGCond cond_table[8] = {
1419        [JCC_O] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1420        [JCC_B] = TCG_COND_LTU,
1421        [JCC_Z] = TCG_COND_EQ,
1422        [JCC_BE] = TCG_COND_LEU,
1423        [JCC_S] = TCG_COND_LT,  /* test sign bit by comparing against 0 */
1424        [JCC_P] = TCG_COND_TSTEQ,  /* even parity - tests low bit of popcount */
1425        [JCC_L] = TCG_COND_LT,
1426        [JCC_LE] = TCG_COND_LE,
1427    };
1428
1429    cond = cond_table[jcc_op];
1430    if (decode->b & 1) {
1431        cond = tcg_invert_cond(cond);
1432    }
1433
1434    ot = decode->op[0].ot;
1435    ot_full = ot | MO_LE;
1436    if (jcc_op >= JCC_S) {
1437        /*
1438         * Sign-extend values before subtracting for S, P (zero/sign extension
1439         * does not matter there) L, LE and their inverses.
1440         */
1441        ot_full |= MO_SIGN;
1442    }
1443
1444    /*
1445     * cmpv will be moved to cc_src *after* cpu_regs[] is written back, so use
1446     * tcg_gen_ext_tl instead of gen_ext_tl.
1447     */
1448    tcg_gen_ext_tl(cmpv, cpu_regs[decode->op[1].n], ot_full);
1449
1450    /*
1451     * Cmpxchg loop starts here.
1452     * - s->T1: addition operand (from decoder)
1453     * - s->A0: dest address (from decoder)
1454     * - s->cc_srcT: memory operand (lhs for comparison)
1455     * - cmpv: rhs for comparison
1456     */
1457    gen_set_label(label_top);
1458    gen_op_ld_v(s, ot_full, s->cc_srcT, s->A0);
1459    tcg_gen_sub_tl(s->T0, s->cc_srcT, cmpv);
1460
1461    /* Compute the comparison result by hand, to avoid clobbering cc_*.  */
1462    switch (jcc_op) {
1463    case JCC_O:
1464        /* (src1 ^ src2) & (src1 ^ dst). newv is only used here for a moment */
1465        tcg_gen_xor_tl(newv, s->cc_srcT, s->T0);
1466        tcg_gen_xor_tl(s->tmp0, s->cc_srcT, cmpv);
1467        tcg_gen_and_tl(s->tmp0, s->tmp0, newv);
1468        tcg_gen_sextract_tl(s->tmp0, s->tmp0, 0, 8 << ot);
1469        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(0);
1470        break;
1471
1472    case JCC_P:
1473        tcg_gen_ext8u_tl(s->tmp0, s->T0);
1474        tcg_gen_ctpop_tl(s->tmp0, s->tmp0);
1475        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(1);
1476        break;
1477
1478    case JCC_S:
1479        tcg_gen_sextract_tl(s->tmp0, s->T0, 0, 8 << ot);
1480        cmp_lhs = s->tmp0, cmp_rhs = tcg_constant_tl(0);
1481        break;
1482
1483    default:
1484        cmp_lhs = s->cc_srcT, cmp_rhs = cmpv;
1485        break;
1486    }
1487
1488    /* Compute new value: if condition does not hold, just store back s->cc_srcT */
1489    tcg_gen_add_tl(newv, s->cc_srcT, s->T1);
1490    tcg_gen_movcond_tl(cond, newv, cmp_lhs, cmp_rhs, newv, s->cc_srcT);
1491    tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, s->cc_srcT, newv, s->mem_index, ot_full);
1492
1493    /* Exit unconditionally if cmpxchg succeeded.  */
1494    tcg_gen_brcond_tl(TCG_COND_EQ, oldv, s->cc_srcT, label_bottom);
1495
1496    /* Try again if there was actually a store to make.  */
1497    tcg_gen_brcond_tl(cond, cmp_lhs, cmp_rhs, label_top);
1498    gen_set_label(label_bottom);
1499
1500    /* Store old value to registers only after a successful store.  */
1501    gen_writeback(s, decode, 1, s->cc_srcT);
1502
1503    decode->cc_dst = s->T0;
1504    decode->cc_src = cmpv;
1505    decode->cc_op = CC_OP_SUBB + ot;
1506}
1507
1508static void gen_CMPS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1509{
1510    MemOp ot = decode->op[2].ot;
1511    if (s->prefix & PREFIX_REPNZ) {
1512        gen_repz_cmps(s, ot, 1);
1513    } else if (s->prefix & PREFIX_REPZ) {
1514        gen_repz_cmps(s, ot, 0);
1515    } else {
1516        gen_cmps(s, ot);
1517    }
1518}
1519
1520static void gen_CRC32(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1521{
1522    MemOp ot = decode->op[2].ot;
1523
1524    tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
1525    gen_helper_crc32(s->T0, s->tmp2_i32, s->T1, tcg_constant_i32(8 << ot));
1526}
1527
1528static void gen_CVTPI2Px(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1529{
1530    gen_helper_enter_mmx(tcg_env);
1531    if (s->prefix & PREFIX_DATA) {
1532        gen_helper_cvtpi2pd(tcg_env, OP_PTR0, OP_PTR2);
1533    } else {
1534        gen_helper_cvtpi2ps(tcg_env, OP_PTR0, OP_PTR2);
1535    }
1536}
1537
1538static void gen_CVTPx2PI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1539{
1540    gen_helper_enter_mmx(tcg_env);
1541    if (s->prefix & PREFIX_DATA) {
1542        gen_helper_cvtpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1543    } else {
1544        gen_helper_cvtps2pi(tcg_env, OP_PTR0, OP_PTR2);
1545    }
1546}
1547
1548static void gen_CVTTPx2PI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1549{
1550    gen_helper_enter_mmx(tcg_env);
1551    if (s->prefix & PREFIX_DATA) {
1552        gen_helper_cvttpd2pi(tcg_env, OP_PTR0, OP_PTR2);
1553    } else {
1554        gen_helper_cvttps2pi(tcg_env, OP_PTR0, OP_PTR2);
1555    }
1556}
1557
1558static void gen_CWD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1559{
1560    int shift = 8 << decode->op[0].ot;
1561
1562    tcg_gen_sextract_tl(s->T0, s->T0, shift - 1, 1);
1563}
1564
1565static void gen_DAA(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1566{
1567    gen_update_cc_op(s);
1568    gen_helper_daa(tcg_env);
1569    set_cc_op(s, CC_OP_EFLAGS);
1570}
1571
1572static void gen_DAS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1573{
1574    gen_update_cc_op(s);
1575    gen_helper_das(tcg_env);
1576    set_cc_op(s, CC_OP_EFLAGS);
1577}
1578
1579static void gen_DEC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1580{
1581    MemOp ot = decode->op[1].ot;
1582
1583    tcg_gen_movi_tl(s->T1, -1);
1584    if (s->prefix & PREFIX_LOCK) {
1585        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1586                                    s->mem_index, ot | MO_LE);
1587    } else {
1588        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1589    }
1590    prepare_update_cc_incdec(decode, s, CC_OP_DECB + ot);
1591}
1592
1593static void gen_DIV(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1594{
1595    MemOp ot = decode->op[2].ot;
1596
1597    switch(ot) {
1598    case MO_8:
1599        gen_helper_divb_AL(tcg_env, s->T1);
1600        break;
1601    case MO_16:
1602        gen_helper_divw_AX(tcg_env, s->T1);
1603        break;
1604    default:
1605    case MO_32:
1606        gen_helper_divl_EAX(tcg_env, s->T1);
1607        break;
1608#ifdef TARGET_X86_64
1609    case MO_64:
1610        gen_helper_divq_EAX(tcg_env, s->T1);
1611        break;
1612#endif
1613    }
1614}
1615
1616static void gen_EMMS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1617{
1618    gen_helper_emms(tcg_env);
1619}
1620
1621static void gen_ENTER(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1622{
1623   gen_enter(s, decode->op[1].imm, decode->op[2].imm);
1624}
1625
1626static void gen_EXTRQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1627{
1628    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
1629    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
1630
1631    gen_helper_extrq_i(tcg_env, OP_PTR0, index, length);
1632}
1633
1634static void gen_EXTRQ_r(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1635{
1636    gen_helper_extrq_r(tcg_env, OP_PTR0, OP_PTR2);
1637}
1638
1639static void gen_HLT(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1640{
1641#ifdef CONFIG_SYSTEM_ONLY
1642    gen_update_cc_op(s);
1643    gen_update_eip_cur(s);
1644    gen_helper_hlt(tcg_env, cur_insn_len_i32(s));
1645    s->base.is_jmp = DISAS_NORETURN;
1646#endif
1647}
1648
1649static void gen_IDIV(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1650{
1651    MemOp ot = decode->op[2].ot;
1652
1653    switch(ot) {
1654    case MO_8:
1655        gen_helper_idivb_AL(tcg_env, s->T1);
1656        break;
1657    case MO_16:
1658        gen_helper_idivw_AX(tcg_env, s->T1);
1659        break;
1660    default:
1661    case MO_32:
1662        gen_helper_idivl_EAX(tcg_env, s->T1);
1663        break;
1664#ifdef TARGET_X86_64
1665    case MO_64:
1666        gen_helper_idivq_EAX(tcg_env, s->T1);
1667        break;
1668#endif
1669    }
1670}
1671
1672static void gen_IMUL3(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1673{
1674    MemOp ot = decode->op[0].ot;
1675    TCGv cc_src_rhs;
1676
1677    switch (ot) {
1678    case MO_16:
1679        /* s->T0 already sign-extended */
1680        tcg_gen_ext16s_tl(s->T1, s->T1);
1681        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1682        /* Compare the full result to the extension of the truncated result.  */
1683        tcg_gen_ext16s_tl(s->T1, s->T0);
1684        cc_src_rhs = s->T0;
1685        break;
1686
1687    case MO_32:
1688#ifdef TARGET_X86_64
1689        if (TCG_TARGET_REG_BITS == 64) {
1690            /*
1691             * This produces fewer TCG ops, and better code if flags are needed,
1692             * but it requires a 64-bit multiply even if they are not.  Use it
1693             * only if the target has 64-bits registers.
1694             *
1695             * s->T0 is already sign-extended.
1696             */
1697            tcg_gen_ext32s_tl(s->T1, s->T1);
1698            tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1699            /* Compare the full result to the extension of the truncated result.  */
1700            tcg_gen_ext32s_tl(s->T1, s->T0);
1701            cc_src_rhs = s->T0;
1702        } else {
1703            /* Variant that only needs a 32-bit widening multiply.  */
1704            TCGv_i32 hi = tcg_temp_new_i32();
1705            TCGv_i32 lo = tcg_temp_new_i32();
1706            tcg_gen_trunc_tl_i32(lo, s->T0);
1707            tcg_gen_trunc_tl_i32(hi, s->T1);
1708            tcg_gen_muls2_i32(lo, hi, lo, hi);
1709            tcg_gen_extu_i32_tl(s->T0, lo);
1710
1711            cc_src_rhs = tcg_temp_new();
1712            tcg_gen_extu_i32_tl(cc_src_rhs, hi);
1713            /* Compare the high part to the sign bit of the truncated result */
1714            tcg_gen_sari_i32(lo, lo, 31);
1715            tcg_gen_extu_i32_tl(s->T1, lo);
1716        }
1717        break;
1718
1719    case MO_64:
1720#endif
1721        cc_src_rhs = tcg_temp_new();
1722        tcg_gen_muls2_tl(s->T0, cc_src_rhs, s->T0, s->T1);
1723        /* Compare the high part to the sign bit of the truncated result */
1724        tcg_gen_sari_tl(s->T1, s->T0, TARGET_LONG_BITS - 1);
1725        break;
1726
1727    default:
1728        g_assert_not_reached();
1729    }
1730
1731    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
1732    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
1733}
1734
1735static void gen_IMUL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1736{
1737    MemOp ot = decode->op[1].ot;
1738    TCGv cc_src_rhs;
1739
1740    switch (ot) {
1741    case MO_8:
1742        /* s->T0 already sign-extended */
1743        tcg_gen_ext8s_tl(s->T1, s->T1);
1744        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1745        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
1746        /* Compare the full result to the extension of the truncated result.  */
1747        tcg_gen_ext8s_tl(s->T1, s->T0);
1748        cc_src_rhs = s->T0;
1749        break;
1750
1751    case MO_16:
1752        /* s->T0 already sign-extended */
1753        tcg_gen_ext16s_tl(s->T1, s->T1);
1754        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1755        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
1756        tcg_gen_shri_tl(s->T1, s->T0, 16);
1757        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
1758        /* Compare the full result to the extension of the truncated result.  */
1759        tcg_gen_ext16s_tl(s->T1, s->T0);
1760        cc_src_rhs = s->T0;
1761        break;
1762
1763    case MO_32:
1764#ifdef TARGET_X86_64
1765        /* s->T0 already sign-extended */
1766        tcg_gen_ext32s_tl(s->T1, s->T1);
1767        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
1768        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
1769        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
1770        /* Compare the full result to the extension of the truncated result.  */
1771        tcg_gen_ext32s_tl(s->T1, s->T0);
1772        cc_src_rhs = s->T0;
1773        break;
1774
1775    case MO_64:
1776#endif
1777        tcg_gen_muls2_tl(s->T0, cpu_regs[R_EDX], s->T0, s->T1);
1778        tcg_gen_mov_tl(cpu_regs[R_EAX], s->T0);
1779
1780        /* Compare the high part to the sign bit of the truncated result */
1781        tcg_gen_negsetcondi_tl(TCG_COND_LT, s->T1, s->T0, 0);
1782        cc_src_rhs = cpu_regs[R_EDX];
1783        break;
1784
1785    default:
1786        g_assert_not_reached();
1787    }
1788
1789    tcg_gen_sub_tl(s->T1, s->T1, cc_src_rhs);
1790    prepare_update2_cc(decode, s, CC_OP_MULB + ot);
1791}
1792
1793static void gen_IN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1794{
1795    MemOp ot = decode->op[0].ot;
1796    TCGv_i32 port = tcg_temp_new_i32();
1797
1798    tcg_gen_trunc_tl_i32(port, s->T1);
1799    tcg_gen_ext16u_i32(port, port);
1800    if (!gen_check_io(s, ot, port, SVM_IOIO_TYPE_MASK)) {
1801        return;
1802    }
1803    translator_io_start(&s->base);
1804    gen_helper_in_func(ot, s->T0, port);
1805    gen_writeback(s, decode, 0, s->T0);
1806    gen_bpt_io(s, port, ot);
1807}
1808
1809static void gen_INC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1810{
1811    MemOp ot = decode->op[1].ot;
1812
1813    tcg_gen_movi_tl(s->T1, 1);
1814    if (s->prefix & PREFIX_LOCK) {
1815        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T1,
1816                                    s->mem_index, ot | MO_LE);
1817    } else {
1818        tcg_gen_add_tl(s->T0, s->T0, s->T1);
1819    }
1820    prepare_update_cc_incdec(decode, s, CC_OP_INCB + ot);
1821}
1822
1823static void gen_INS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1824{
1825    MemOp ot = decode->op[1].ot;
1826    TCGv_i32 port = tcg_temp_new_i32();
1827
1828    tcg_gen_trunc_tl_i32(port, s->T1);
1829    tcg_gen_ext16u_i32(port, port);
1830    if (!gen_check_io(s, ot, port,
1831                      SVM_IOIO_TYPE_MASK | SVM_IOIO_STR_MASK)) {
1832        return;
1833    }
1834
1835    translator_io_start(&s->base);
1836    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
1837        gen_repz_ins(s, ot);
1838    } else {
1839        gen_ins(s, ot);
1840    }
1841}
1842
1843static void gen_INSERTQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1844{
1845    TCGv_i32 length = tcg_constant_i32(decode->immediate & 63);
1846    TCGv_i32 index = tcg_constant_i32((decode->immediate >> 8) & 63);
1847
1848    gen_helper_insertq_i(tcg_env, OP_PTR0, OP_PTR1, index, length);
1849}
1850
1851static void gen_INSERTQ_r(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1852{
1853    gen_helper_insertq_r(tcg_env, OP_PTR0, OP_PTR2);
1854}
1855
1856static void gen_INT(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1857{
1858    gen_interrupt(s, decode->immediate);
1859}
1860
1861static void gen_INT1(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1862{
1863    gen_exception(s, EXCP01_DB);
1864}
1865
1866static void gen_INT3(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1867{
1868    gen_interrupt(s, EXCP03_INT3);
1869}
1870
1871static void gen_INTO(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1872{
1873    gen_update_cc_op(s);
1874    gen_update_eip_cur(s);
1875    gen_helper_into(tcg_env, cur_insn_len_i32(s));
1876}
1877
1878static void gen_IRET(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1879{
1880    if (!PE(s) || VM86(s)) {
1881        gen_helper_iret_real(tcg_env, tcg_constant_i32(s->dflag - 1));
1882    } else {
1883        gen_helper_iret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
1884                                  eip_next_i32(s));
1885    }
1886    set_cc_op(s, CC_OP_EFLAGS);
1887    s->base.is_jmp = DISAS_EOB_ONLY;
1888}
1889
1890static void gen_Jcc(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1891{
1892    gen_bnd_jmp(s);
1893    gen_jcc(s, decode->b & 0xf, decode->immediate);
1894}
1895
1896static void gen_JCXZ(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1897{
1898    TCGLabel *taken = gen_new_label();
1899
1900    gen_update_cc_op(s);
1901    gen_op_jz_ecx(s, taken);
1902    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
1903}
1904
1905static void gen_JMP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1906{
1907    gen_update_cc_op(s);
1908    gen_jmp_rel(s, s->dflag, decode->immediate, 0);
1909}
1910
1911static void gen_JMP_m(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1912{
1913    gen_op_jmp_v(s, s->T0);
1914    gen_bnd_jmp(s);
1915    s->base.is_jmp = DISAS_JUMP;
1916}
1917
1918static void gen_JMPF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1919{
1920    gen_far_jmp(s);
1921}
1922
1923static void gen_JMPF_m(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1924{
1925    MemOp ot = decode->op[2].ot;
1926
1927    gen_op_ld_v(s, ot, s->T0, s->A0);
1928    gen_add_A0_im(s, 1 << ot);
1929    gen_op_ld_v(s, MO_16, s->T1, s->A0);
1930    gen_far_jmp(s);
1931}
1932
1933static void gen_LAHF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1934{
1935    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
1936        return gen_illegal_opcode(s);
1937    }
1938    gen_compute_eflags(s);
1939    /* Note: gen_compute_eflags() only gives the condition codes */
1940    tcg_gen_ori_tl(s->T0, cpu_cc_src, 0x02);
1941    tcg_gen_deposit_tl(cpu_regs[R_EAX], cpu_regs[R_EAX], s->T0, 8, 8);
1942}
1943
1944static void gen_LDMXCSR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1945{
1946    tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T1);
1947    gen_helper_ldmxcsr(tcg_env, s->tmp2_i32);
1948}
1949
1950static void gen_lxx_seg(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode, int seg)
1951{
1952    MemOp ot = decode->op[0].ot;
1953
1954    /* Offset already in s->T0.  */
1955    gen_add_A0_im(s, 1 << ot);
1956    gen_op_ld_v(s, MO_16, s->T1, s->A0);
1957
1958    /* load the segment here to handle exceptions properly */
1959    gen_movl_seg(s, seg, s->T1);
1960}
1961
1962static void gen_LDS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1963{
1964    gen_lxx_seg(s, env, decode, R_DS);
1965}
1966
1967static void gen_LEA(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1968{
1969    tcg_gen_mov_tl(s->T0, s->A0);
1970}
1971
1972static void gen_LEAVE(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1973{
1974    gen_leave(s);
1975}
1976
1977static void gen_LES(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1978{
1979    gen_lxx_seg(s, env, decode, R_ES);
1980}
1981
1982static void gen_LFS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1983{
1984    gen_lxx_seg(s, env, decode, R_FS);
1985}
1986
1987static void gen_LGS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1988{
1989    gen_lxx_seg(s, env, decode, R_GS);
1990}
1991
1992static void gen_LODS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
1993{
1994    MemOp ot = decode->op[2].ot;
1995    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
1996        gen_repz_lods(s, ot);
1997    } else {
1998        gen_lods(s, ot);
1999    }
2000}
2001
2002static void gen_LOOP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2003{
2004    TCGLabel *taken = gen_new_label();
2005
2006    gen_update_cc_op(s);
2007    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2008    gen_op_jnz_ecx(s, taken);
2009    gen_conditional_jump_labels(s, decode->immediate, NULL, taken);
2010}
2011
2012static void gen_LOOPE(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2013{
2014    TCGLabel *taken = gen_new_label();
2015    TCGLabel *not_taken = gen_new_label();
2016
2017    gen_update_cc_op(s);
2018    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2019    gen_op_jz_ecx(s, not_taken);
2020    gen_jcc1(s, (JCC_Z << 1), taken); /* jz taken */
2021    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2022}
2023
2024static void gen_LOOPNE(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2025{
2026    TCGLabel *taken = gen_new_label();
2027    TCGLabel *not_taken = gen_new_label();
2028
2029    gen_update_cc_op(s);
2030    gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
2031    gen_op_jz_ecx(s, not_taken);
2032    gen_jcc1(s, (JCC_Z << 1) | 1, taken); /* jnz taken */
2033    gen_conditional_jump_labels(s, decode->immediate, not_taken, taken);
2034}
2035
2036static void gen_LSS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2037{
2038    gen_lxx_seg(s, env, decode, R_SS);
2039}
2040
2041static void gen_MOV(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2042{
2043    /* nothing to do! */
2044}
2045#define gen_NOP gen_MOV
2046
2047static void gen_MASKMOV(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2048{
2049    gen_lea_v_seg(s, s->aflag, cpu_regs[R_EDI], R_DS, s->override);
2050
2051    if (s->prefix & PREFIX_DATA) {
2052        gen_helper_maskmov_xmm(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2053    } else {
2054        gen_helper_maskmov_mmx(tcg_env, OP_PTR1, OP_PTR2, s->A0);
2055    }
2056}
2057
2058static void gen_MOVBE(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2059{
2060    MemOp ot = decode->op[0].ot;
2061
2062    /* M operand type does not load/store */
2063    if (decode->e.op0 == X86_TYPE_M) {
2064        tcg_gen_qemu_st_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2065    } else {
2066        tcg_gen_qemu_ld_tl(s->T0, s->A0, s->mem_index, ot | MO_BE);
2067    }
2068}
2069
2070static void gen_MOVD_from(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2071{
2072    MemOp ot = decode->op[2].ot;
2073
2074    switch (ot) {
2075    case MO_32:
2076#ifdef TARGET_X86_64
2077        tcg_gen_ld32u_tl(s->T0, tcg_env, decode->op[2].offset);
2078        break;
2079    case MO_64:
2080#endif
2081        tcg_gen_ld_tl(s->T0, tcg_env, decode->op[2].offset);
2082        break;
2083    default:
2084        abort();
2085    }
2086}
2087
2088static void gen_MOVD_to(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2089{
2090    MemOp ot = decode->op[2].ot;
2091    int vec_len = vector_len(s, decode);
2092    int lo_ofs = vector_elem_offset(&decode->op[0], ot, 0);
2093
2094    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2095
2096    switch (ot) {
2097    case MO_32:
2098#ifdef TARGET_X86_64
2099        tcg_gen_st32_tl(s->T1, tcg_env, lo_ofs);
2100        break;
2101    case MO_64:
2102#endif
2103        tcg_gen_st_tl(s->T1, tcg_env, lo_ofs);
2104        break;
2105    default:
2106        g_assert_not_reached();
2107    }
2108}
2109
2110static void gen_MOVDQ(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2111{
2112    gen_store_sse(s, decode, decode->op[2].offset);
2113}
2114
2115static void gen_MOVMSK(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2116{
2117    typeof(gen_helper_movmskps_ymm) *ps, *pd, *fn;
2118    ps = s->vex_l ? gen_helper_movmskps_ymm : gen_helper_movmskps_xmm;
2119    pd = s->vex_l ? gen_helper_movmskpd_ymm : gen_helper_movmskpd_xmm;
2120    fn = s->prefix & PREFIX_DATA ? pd : ps;
2121    fn(s->tmp2_i32, tcg_env, OP_PTR2);
2122    tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
2123}
2124
2125static void gen_MOVQ(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2126{
2127    int vec_len = vector_len(s, decode);
2128    int lo_ofs = vector_elem_offset(&decode->op[0], MO_64, 0);
2129
2130    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
2131    if (decode->op[0].has_ea) {
2132        tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
2133    } else {
2134        /*
2135         * tcg_gen_gvec_dup_i64(MO_64, op0.offset, 8, vec_len, s->tmp1_64) would
2136         * seem to work, but it does not on big-endian platforms; the cleared parts
2137         * are always at higher addresses, but cross-endian emulation inverts the
2138         * byte order so that the cleared parts need to be at *lower* addresses.
2139         * Because oprsz is 8, we see this here even for SSE; but more in general,
2140         * it disqualifies using oprsz < maxsz to emulate VEX128.
2141         */
2142        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2143        tcg_gen_st_i64(s->tmp1_i64, tcg_env, lo_ofs);
2144    }
2145}
2146
2147static void gen_MOVq_dq(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2148{
2149    gen_helper_enter_mmx(tcg_env);
2150    /* Otherwise the same as any other movq.  */
2151    return gen_MOVQ(s, env, decode);
2152}
2153
2154static void gen_MOVS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2155{
2156    MemOp ot = decode->op[2].ot;
2157    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2158        gen_repz_movs(s, ot);
2159    } else {
2160        gen_movs(s, ot);
2161    }
2162}
2163
2164static void gen_MUL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2165{
2166    MemOp ot = decode->op[1].ot;
2167
2168    switch (ot) {
2169    case MO_8:
2170        /* s->T0 already zero-extended */
2171        tcg_gen_ext8u_tl(s->T1, s->T1);
2172        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2173        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2174        tcg_gen_andi_tl(s->T1, s->T0, 0xff00);
2175        decode->cc_dst = s->T0;
2176        decode->cc_src = s->T1;
2177        break;
2178
2179    case MO_16:
2180        /* s->T0 already zero-extended */
2181        tcg_gen_ext16u_tl(s->T1, s->T1);
2182        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2183        gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
2184        tcg_gen_shri_tl(s->T1, s->T0, 16);
2185        gen_op_mov_reg_v(s, MO_16, R_EDX, s->T1);
2186        decode->cc_dst = s->T0;
2187        decode->cc_src = s->T1;
2188        break;
2189
2190    case MO_32:
2191#ifdef TARGET_X86_64
2192        /* s->T0 already zero-extended */
2193        tcg_gen_ext32u_tl(s->T1, s->T1);
2194        tcg_gen_mul_tl(s->T0, s->T0, s->T1);
2195        tcg_gen_ext32u_tl(cpu_regs[R_EAX], s->T0);
2196        tcg_gen_shri_tl(cpu_regs[R_EDX], s->T0, 32);
2197        decode->cc_dst = cpu_regs[R_EAX];
2198        decode->cc_src = cpu_regs[R_EDX];
2199        break;
2200
2201    case MO_64:
2202#endif
2203        tcg_gen_mulu2_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], s->T0, s->T1);
2204        decode->cc_dst = cpu_regs[R_EAX];
2205        decode->cc_src = cpu_regs[R_EDX];
2206        break;
2207
2208    default:
2209        g_assert_not_reached();
2210    }
2211
2212    decode->cc_op = CC_OP_MULB + ot;
2213}
2214
2215static void gen_MULX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2216{
2217    MemOp ot = decode->op[0].ot;
2218
2219    /* low part of result in VEX.vvvv, high in MODRM */
2220    switch (ot) {
2221    case MO_32:
2222#ifdef TARGET_X86_64
2223        tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
2224        tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
2225        tcg_gen_mulu2_i32(s->tmp2_i32, s->tmp3_i32,
2226                          s->tmp2_i32, s->tmp3_i32);
2227        tcg_gen_extu_i32_tl(cpu_regs[s->vex_v], s->tmp2_i32);
2228        tcg_gen_extu_i32_tl(s->T0, s->tmp3_i32);
2229        break;
2230
2231    case MO_64:
2232#endif
2233        tcg_gen_mulu2_tl(cpu_regs[s->vex_v], s->T0, s->T0, s->T1);
2234        break;
2235
2236    default:
2237        g_assert_not_reached();
2238    }
2239}
2240
2241static void gen_NEG(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2242{
2243    MemOp ot = decode->op[0].ot;
2244    TCGv oldv = tcg_temp_new();
2245
2246    if (s->prefix & PREFIX_LOCK) {
2247        TCGv newv = tcg_temp_new();
2248        TCGv cmpv = tcg_temp_new();
2249        TCGLabel *label1 = gen_new_label();
2250
2251        gen_set_label(label1);
2252        gen_op_ld_v(s, ot, oldv, s->A0);
2253        tcg_gen_neg_tl(newv, oldv);
2254        tcg_gen_atomic_cmpxchg_tl(cmpv, s->A0, oldv, newv,
2255                                  s->mem_index, ot | MO_LE);
2256        tcg_gen_brcond_tl(TCG_COND_NE, oldv, cmpv, label1);
2257    } else {
2258        tcg_gen_mov_tl(oldv, s->T0);
2259    }
2260    tcg_gen_neg_tl(s->T0, oldv);
2261
2262    decode->cc_dst = s->T0;
2263    decode->cc_src = oldv;
2264    tcg_gen_movi_tl(s->cc_srcT, 0);
2265    decode->cc_op = CC_OP_SUBB + ot;
2266}
2267
2268static void gen_NOT(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2269{
2270    MemOp ot = decode->op[0].ot;
2271
2272    if (s->prefix & PREFIX_LOCK) {
2273        tcg_gen_movi_tl(s->T0, ~0);
2274        tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T0,
2275                                    s->mem_index, ot | MO_LE);
2276    } else {
2277        tcg_gen_not_tl(s->T0, s->T0);
2278    }
2279}
2280
2281static void gen_OR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2282{
2283    MemOp ot = decode->op[1].ot;
2284
2285    if (s->prefix & PREFIX_LOCK) {
2286        tcg_gen_atomic_or_fetch_tl(s->T0, s->A0, s->T1,
2287                                   s->mem_index, ot | MO_LE);
2288    } else {
2289        tcg_gen_or_tl(s->T0, s->T0, s->T1);
2290    }
2291    prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
2292}
2293
2294static void gen_OUT(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2295{
2296    MemOp ot = decode->op[1].ot;
2297    TCGv_i32 port = tcg_temp_new_i32();
2298    TCGv_i32 value = tcg_temp_new_i32();
2299
2300    tcg_gen_trunc_tl_i32(port, s->T1);
2301    tcg_gen_ext16u_i32(port, port);
2302    if (!gen_check_io(s, ot, port, 0)) {
2303        return;
2304    }
2305    tcg_gen_trunc_tl_i32(value, s->T0);
2306    translator_io_start(&s->base);
2307    gen_helper_out_func(ot, port, value);
2308    gen_bpt_io(s, port, ot);
2309}
2310
2311static void gen_OUTS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2312{
2313    MemOp ot = decode->op[1].ot;
2314    TCGv_i32 port = tcg_temp_new_i32();
2315
2316    tcg_gen_trunc_tl_i32(port, s->T1);
2317    tcg_gen_ext16u_i32(port, port);
2318    if (!gen_check_io(s, ot, port, SVM_IOIO_STR_MASK)) {
2319        return;
2320    }
2321
2322    translator_io_start(&s->base);
2323    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
2324        gen_repz_outs(s, ot);
2325    } else {
2326        gen_outs(s, ot);
2327    }
2328}
2329
2330static void gen_PALIGNR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2331{
2332    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2333    if (!(s->prefix & PREFIX_DATA)) {
2334        gen_helper_palignr_mmx(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2335    } else if (!s->vex_l) {
2336        gen_helper_palignr_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2337    } else {
2338        gen_helper_palignr_ymm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
2339    }
2340}
2341
2342static void gen_PANDN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2343{
2344    int vec_len = vector_len(s, decode);
2345
2346    /* Careful, operand order is reversed!  */
2347    tcg_gen_gvec_andc(MO_64,
2348                      decode->op[0].offset, decode->op[2].offset,
2349                      decode->op[1].offset, vec_len, vec_len);
2350}
2351
2352static void gen_PCMPESTRI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2353{
2354    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2355    gen_helper_pcmpestri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2356    set_cc_op(s, CC_OP_EFLAGS);
2357}
2358
2359static void gen_PCMPESTRM(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2360{
2361    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2362    gen_helper_pcmpestrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2363    set_cc_op(s, CC_OP_EFLAGS);
2364    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2365        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2366                             16, 16, 0);
2367    }
2368}
2369
2370static void gen_PCMPISTRI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2371{
2372    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2373    gen_helper_pcmpistri_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2374    set_cc_op(s, CC_OP_EFLAGS);
2375}
2376
2377static void gen_PCMPISTRM(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2378{
2379    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2380    gen_helper_pcmpistrm_xmm(tcg_env, OP_PTR1, OP_PTR2, imm);
2381    set_cc_op(s, CC_OP_EFLAGS);
2382    if ((s->prefix & PREFIX_VEX) && !s->vex_l) {
2383        tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_regs[0].ZMM_X(1)),
2384                             16, 16, 0);
2385    }
2386}
2387
2388static void gen_PDEP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2389{
2390    gen_helper_pdep(s->T0, s->T0, s->T1);
2391}
2392
2393static void gen_PEXT(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2394{
2395    gen_helper_pext(s->T0, s->T0, s->T1);
2396}
2397
2398static inline void gen_pextr(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode, MemOp ot)
2399{
2400    int vec_len = vector_len(s, decode);
2401    int mask = (vec_len >> ot) - 1;
2402    int val = decode->immediate & mask;
2403
2404    switch (ot) {
2405    case MO_8:
2406        tcg_gen_ld8u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2407        break;
2408    case MO_16:
2409        tcg_gen_ld16u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2410        break;
2411    case MO_32:
2412#ifdef TARGET_X86_64
2413        tcg_gen_ld32u_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2414        break;
2415    case MO_64:
2416#endif
2417        tcg_gen_ld_tl(s->T0, tcg_env, vector_elem_offset(&decode->op[1], ot, val));
2418        break;
2419    default:
2420        abort();
2421    }
2422}
2423
2424static void gen_PEXTRB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2425{
2426    gen_pextr(s, env, decode, MO_8);
2427}
2428
2429static void gen_PEXTRW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2430{
2431    gen_pextr(s, env, decode, MO_16);
2432}
2433
2434static void gen_PEXTR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2435{
2436    MemOp ot = decode->op[0].ot;
2437    gen_pextr(s, env, decode, ot);
2438}
2439
2440static inline void gen_pinsr(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode, MemOp ot)
2441{
2442    int vec_len = vector_len(s, decode);
2443    int mask = (vec_len >> ot) - 1;
2444    int val = decode->immediate & mask;
2445
2446    if (decode->op[1].offset != decode->op[0].offset) {
2447        assert(vec_len == 16);
2448        gen_store_sse(s, decode, decode->op[1].offset);
2449    }
2450
2451    switch (ot) {
2452    case MO_8:
2453        tcg_gen_st8_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2454        break;
2455    case MO_16:
2456        tcg_gen_st16_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2457        break;
2458    case MO_32:
2459#ifdef TARGET_X86_64
2460        tcg_gen_st32_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2461        break;
2462    case MO_64:
2463#endif
2464        tcg_gen_st_tl(s->T1, tcg_env, vector_elem_offset(&decode->op[0], ot, val));
2465        break;
2466    default:
2467        abort();
2468    }
2469}
2470
2471static void gen_PINSRB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2472{
2473    gen_pinsr(s, env, decode, MO_8);
2474}
2475
2476static void gen_PINSRW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2477{
2478    gen_pinsr(s, env, decode, MO_16);
2479}
2480
2481static void gen_PINSR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2482{
2483    gen_pinsr(s, env, decode, decode->op[2].ot);
2484}
2485
2486static void gen_pmovmskb_i64(TCGv_i64 d, TCGv_i64 s)
2487{
2488    TCGv_i64 t = tcg_temp_new_i64();
2489
2490    tcg_gen_andi_i64(d, s, 0x8080808080808080ull);
2491
2492    /*
2493     * After each shift+or pair:
2494     * 0:  a.......b.......c.......d.......e.......f.......g.......h.......
2495     * 7:  ab......bc......cd......de......ef......fg......gh......h.......
2496     * 14: abcd....bcde....cdef....defg....efgh....fgh.....gh......h.......
2497     * 28: abcdefghbcdefgh.cdefgh..defgh...efgh....fgh.....gh......h.......
2498     * The result is left in the high bits of the word.
2499     */
2500    tcg_gen_shli_i64(t, d, 7);
2501    tcg_gen_or_i64(d, d, t);
2502    tcg_gen_shli_i64(t, d, 14);
2503    tcg_gen_or_i64(d, d, t);
2504    tcg_gen_shli_i64(t, d, 28);
2505    tcg_gen_or_i64(d, d, t);
2506}
2507
2508static void gen_pmovmskb_vec(unsigned vece, TCGv_vec d, TCGv_vec s)
2509{
2510    TCGv_vec t = tcg_temp_new_vec_matching(d);
2511    TCGv_vec m = tcg_constant_vec_matching(d, MO_8, 0x80);
2512
2513    /* See above */
2514    tcg_gen_and_vec(vece, d, s, m);
2515    tcg_gen_shli_vec(vece, t, d, 7);
2516    tcg_gen_or_vec(vece, d, d, t);
2517    tcg_gen_shli_vec(vece, t, d, 14);
2518    tcg_gen_or_vec(vece, d, d, t);
2519    tcg_gen_shli_vec(vece, t, d, 28);
2520    tcg_gen_or_vec(vece, d, d, t);
2521}
2522
2523static void gen_PMOVMSKB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2524{
2525    static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
2526    static const GVecGen2 g = {
2527        .fni8 = gen_pmovmskb_i64,
2528        .fniv = gen_pmovmskb_vec,
2529        .opt_opc = vecop_list,
2530        .vece = MO_64,
2531        .prefer_i64 = TCG_TARGET_REG_BITS == 64
2532    };
2533    MemOp ot = decode->op[2].ot;
2534    int vec_len = vector_len(s, decode);
2535    TCGv t = tcg_temp_new();
2536
2537    tcg_gen_gvec_2(offsetof(CPUX86State, xmm_t0) + xmm_offset(ot), decode->op[2].offset,
2538                   vec_len, vec_len, &g);
2539    tcg_gen_ld8u_tl(s->T0, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
2540    while (vec_len > 8) {
2541        vec_len -= 8;
2542        if (TCG_TARGET_HAS_extract2_tl) {
2543            /*
2544             * Load the next byte of the result into the high byte of T.
2545             * TCG does a similar expansion of deposit to shl+extract2; by
2546             * loading the whole word, the shift left is avoided.
2547             */
2548#ifdef TARGET_X86_64
2549            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_Q((vec_len - 1) / 8)));
2550#else
2551            tcg_gen_ld_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L((vec_len - 1) / 4)));
2552#endif
2553
2554            tcg_gen_extract2_tl(s->T0, t, s->T0, TARGET_LONG_BITS - 8);
2555        } else {
2556            /*
2557             * The _previous_ value is deposited into bits 8 and higher of t.  Because
2558             * those bits are known to be zero after ld8u, this becomes a shift+or
2559             * if deposit is not available.
2560             */
2561            tcg_gen_ld8u_tl(t, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_B(vec_len - 1)));
2562            tcg_gen_deposit_tl(s->T0, t, s->T0, 8, TARGET_LONG_BITS - 8);
2563        }
2564    }
2565}
2566
2567static void gen_POP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2568{
2569    MemOp ot = gen_pop_T0(s);
2570    if (decode->op[0].has_ea) {
2571        /* NOTE: order is important for MMU exceptions */
2572        gen_op_st_v(s, ot, s->T0, s->A0);
2573        decode->op[0].unit = X86_OP_SKIP;
2574    }
2575    /* NOTE: writing back registers after update is important for pop %sp */
2576    gen_pop_update(s, ot);
2577}
2578
2579static void gen_POPA(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2580{
2581    gen_popa(s);
2582}
2583
2584static void gen_POPF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2585{
2586    MemOp ot;
2587    int mask = TF_MASK | AC_MASK | ID_MASK | NT_MASK;
2588
2589    if (CPL(s) == 0) {
2590        mask |= IF_MASK | IOPL_MASK;
2591    } else if (CPL(s) <= IOPL(s)) {
2592        mask |= IF_MASK;
2593    }
2594    if (s->dflag == MO_16) {
2595        mask &= 0xffff;
2596    }
2597
2598    ot = gen_pop_T0(s);
2599    gen_helper_write_eflags(tcg_env, s->T0, tcg_constant_i32(mask));
2600    gen_pop_update(s, ot);
2601    set_cc_op(s, CC_OP_EFLAGS);
2602    /* abort translation because TF/AC flag may change */
2603    s->base.is_jmp = DISAS_EOB_NEXT;
2604}
2605
2606static void gen_PSHUFW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2607{
2608    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
2609    gen_helper_pshufw_mmx(OP_PTR0, OP_PTR1, imm);
2610}
2611
2612static void gen_PSRLW_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2613{
2614    int vec_len = vector_len(s, decode);
2615
2616    if (decode->immediate >= 16) {
2617        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2618    } else {
2619        tcg_gen_gvec_shri(MO_16,
2620                          decode->op[0].offset, decode->op[1].offset,
2621                          decode->immediate, vec_len, vec_len);
2622    }
2623}
2624
2625static void gen_PSLLW_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2626{
2627    int vec_len = vector_len(s, decode);
2628
2629    if (decode->immediate >= 16) {
2630        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2631    } else {
2632        tcg_gen_gvec_shli(MO_16,
2633                          decode->op[0].offset, decode->op[1].offset,
2634                          decode->immediate, vec_len, vec_len);
2635    }
2636}
2637
2638static void gen_PSRAW_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2639{
2640    int vec_len = vector_len(s, decode);
2641
2642    if (decode->immediate >= 16) {
2643        decode->immediate = 15;
2644    }
2645    tcg_gen_gvec_sari(MO_16,
2646                      decode->op[0].offset, decode->op[1].offset,
2647                      decode->immediate, vec_len, vec_len);
2648}
2649
2650static void gen_PSRLD_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2651{
2652    int vec_len = vector_len(s, decode);
2653
2654    if (decode->immediate >= 32) {
2655        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2656    } else {
2657        tcg_gen_gvec_shri(MO_32,
2658                          decode->op[0].offset, decode->op[1].offset,
2659                          decode->immediate, vec_len, vec_len);
2660    }
2661}
2662
2663static void gen_PSLLD_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2664{
2665    int vec_len = vector_len(s, decode);
2666
2667    if (decode->immediate >= 32) {
2668        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2669    } else {
2670        tcg_gen_gvec_shli(MO_32,
2671                          decode->op[0].offset, decode->op[1].offset,
2672                          decode->immediate, vec_len, vec_len);
2673    }
2674}
2675
2676static void gen_PSRAD_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2677{
2678    int vec_len = vector_len(s, decode);
2679
2680    if (decode->immediate >= 32) {
2681        decode->immediate = 31;
2682    }
2683    tcg_gen_gvec_sari(MO_32,
2684                      decode->op[0].offset, decode->op[1].offset,
2685                      decode->immediate, vec_len, vec_len);
2686}
2687
2688static void gen_PSRLQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2689{
2690    int vec_len = vector_len(s, decode);
2691
2692    if (decode->immediate >= 64) {
2693        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2694    } else {
2695        tcg_gen_gvec_shri(MO_64,
2696                          decode->op[0].offset, decode->op[1].offset,
2697                          decode->immediate, vec_len, vec_len);
2698    }
2699}
2700
2701static void gen_PSLLQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2702{
2703    int vec_len = vector_len(s, decode);
2704
2705    if (decode->immediate >= 64) {
2706        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
2707    } else {
2708        tcg_gen_gvec_shli(MO_64,
2709                          decode->op[0].offset, decode->op[1].offset,
2710                          decode->immediate, vec_len, vec_len);
2711    }
2712}
2713
2714static TCGv_ptr make_imm8u_xmm_vec(uint8_t imm, int vec_len)
2715{
2716    MemOp ot = vec_len == 16 ? MO_128 : MO_256;
2717    TCGv_i32 imm_v = tcg_constant8u_i32(imm);
2718    TCGv_ptr ptr = tcg_temp_new_ptr();
2719
2720    tcg_gen_gvec_dup_imm(MO_64, offsetof(CPUX86State, xmm_t0) + xmm_offset(ot),
2721                         vec_len, vec_len, 0);
2722
2723    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_t0));
2724    tcg_gen_st_i32(imm_v, tcg_env, offsetof(CPUX86State, xmm_t0.ZMM_L(0)));
2725    return ptr;
2726}
2727
2728static void gen_PSRLDQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2729{
2730    int vec_len = vector_len(s, decode);
2731    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
2732
2733    if (s->vex_l) {
2734        gen_helper_psrldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
2735    } else {
2736        gen_helper_psrldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
2737    }
2738}
2739
2740static void gen_PSLLDQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2741{
2742    int vec_len = vector_len(s, decode);
2743    TCGv_ptr imm_vec = make_imm8u_xmm_vec(decode->immediate, vec_len);
2744
2745    if (s->vex_l) {
2746        gen_helper_pslldq_ymm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
2747    } else {
2748        gen_helper_pslldq_xmm(tcg_env, OP_PTR0, OP_PTR1, imm_vec);
2749    }
2750}
2751
2752static void gen_PUSH(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2753{
2754    gen_push_v(s, s->T1);
2755}
2756
2757static void gen_PUSHA(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2758{
2759    gen_pusha(s);
2760}
2761
2762static void gen_PUSHF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2763{
2764    gen_update_cc_op(s);
2765    gen_helper_read_eflags(s->T0, tcg_env);
2766    gen_push_v(s, s->T0);
2767}
2768
2769static MemOp gen_shift_count(DisasContext *s, X86DecodedInsn *decode,
2770                             bool *can_be_zero, TCGv *count)
2771{
2772    MemOp ot = decode->op[0].ot;
2773    int mask = (ot <= MO_32 ? 0x1f : 0x3f);
2774
2775    *can_be_zero = false;
2776    switch (decode->op[2].unit) {
2777    case X86_OP_INT:
2778        *count = tcg_temp_new();
2779        tcg_gen_andi_tl(*count, s->T1, mask);
2780        *can_be_zero = true;
2781        break;
2782
2783    case X86_OP_IMM:
2784        if ((decode->immediate & mask) == 0) {
2785            *count = NULL;
2786            break;
2787        }
2788        *count = tcg_temp_new();
2789        tcg_gen_movi_tl(*count, decode->immediate & mask);
2790        break;
2791
2792    case X86_OP_SKIP:
2793        *count = tcg_temp_new();
2794        tcg_gen_movi_tl(*count, 1);
2795        break;
2796
2797    default:
2798        g_assert_not_reached();
2799    }
2800
2801    return ot;
2802}
2803
2804/*
2805 * Compute existing flags in decode->cc_src, for gen_* functions that wants
2806 * to set the cc_op set to CC_OP_ADCOX.  In particular, this allows rotate
2807 * operations to compute the carry in decode->cc_dst and the overflow in
2808 * decode->cc_src2.
2809 *
2810 * If need_flags is true, decode->cc_dst and decode->cc_src2 are preloaded
2811 * with the value of CF and OF before the instruction, so that it is possible
2812 * to keep the flags unmodified.
2813 *
2814 * Return true if carry could be made available cheaply as a 1-bit value in
2815 * decode->cc_dst (trying a bit harder if want_carry is true).  If false is
2816 * returned, decode->cc_dst is uninitialized and the carry is only available
2817 * as bit 0 of decode->cc_src.
2818 */
2819static bool gen_eflags_adcox(DisasContext *s, X86DecodedInsn *decode, bool want_carry, bool need_flags)
2820{
2821    bool got_cf = false;
2822    bool got_of = false;
2823
2824    decode->cc_dst = tcg_temp_new();
2825    decode->cc_src = tcg_temp_new();
2826    decode->cc_src2 = tcg_temp_new();
2827    decode->cc_op = CC_OP_ADCOX;
2828
2829    /* A lot more cc_ops could be "optimized" to avoid the extracts at
2830     * the end (INC/DEC, BMILG, MUL), but they are all really unlikely
2831     * to be followed by rotations within the same basic block.
2832     */
2833    switch (s->cc_op) {
2834    case CC_OP_ADCOX:
2835        /* No need to compute the full EFLAGS, CF/OF are already isolated.  */
2836        tcg_gen_mov_tl(decode->cc_src, cpu_cc_src);
2837        if (need_flags) {
2838            tcg_gen_mov_tl(decode->cc_src2, cpu_cc_src2);
2839            got_of = true;
2840        }
2841        if (want_carry || need_flags) {
2842            tcg_gen_mov_tl(decode->cc_dst, cpu_cc_dst);
2843            got_cf = true;
2844        }
2845        break;
2846
2847    case CC_OP_LOGICB ... CC_OP_LOGICQ:
2848        /* CF and OF are zero, do it just because it's easy.  */
2849        gen_mov_eflags(s, decode->cc_src);
2850        if (need_flags) {
2851            tcg_gen_movi_tl(decode->cc_src2, 0);
2852            got_of = true;
2853        }
2854        if (want_carry || need_flags) {
2855            tcg_gen_movi_tl(decode->cc_dst, 0);
2856            got_cf = true;
2857        }
2858        break;
2859
2860    case CC_OP_SARB ... CC_OP_SARQ:
2861        /*
2862         * SHR/RCR/SHR/RCR/... is a relatively common occurrence of RCR.
2863         * By computing CF without using eflags, the calls to cc_compute_all
2864         * can be eliminated as dead code (except for the last RCR).
2865         */
2866        if (want_carry || need_flags) {
2867            tcg_gen_andi_tl(decode->cc_dst, cpu_cc_src, 1);
2868            got_cf = true;
2869        }
2870        gen_mov_eflags(s, decode->cc_src);
2871        break;
2872
2873    case CC_OP_SHLB ... CC_OP_SHLQ:
2874        /*
2875         * Likewise for SHL/RCL/SHL/RCL/... but, if CF is not in the sign
2876         * bit, we might as well fish CF out of EFLAGS and save a shift.
2877         */
2878        if (want_carry && (!need_flags || s->cc_op == CC_OP_SHLB + MO_TL)) {
2879            tcg_gen_shri_tl(decode->cc_dst, cpu_cc_src, (8 << (s->cc_op - CC_OP_SHLB)) - 1);
2880            got_cf = true;
2881        }
2882        gen_mov_eflags(s, decode->cc_src);
2883        break;
2884
2885    default:
2886        gen_mov_eflags(s, decode->cc_src);
2887        break;
2888    }
2889
2890    if (need_flags) {
2891        /* If the flags could be left unmodified, always load them.  */
2892        if (!got_of) {
2893            tcg_gen_extract_tl(decode->cc_src2, decode->cc_src, ctz32(CC_O), 1);
2894            got_of = true;
2895        }
2896        if (!got_cf) {
2897            tcg_gen_extract_tl(decode->cc_dst, decode->cc_src, ctz32(CC_C), 1);
2898            got_cf = true;
2899        }
2900    }
2901    return got_cf;
2902}
2903
2904static void gen_rot_overflow(X86DecodedInsn *decode, TCGv result, TCGv old, TCGv count)
2905{
2906    MemOp ot = decode->op[0].ot;
2907    TCGv temp = count ? tcg_temp_new() : decode->cc_src2;
2908
2909    tcg_gen_xor_tl(temp, old, result);
2910    tcg_gen_extract_tl(temp, temp, (8 << ot) - 1, 1);
2911    if (count) {
2912        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src2, count, tcg_constant_tl(0),
2913                           decode->cc_src2, temp);
2914    }
2915}
2916
2917/*
2918 * RCx operations are invariant modulo 8*operand_size+1.  For 8 and 16-bit operands,
2919 * this is less than 0x1f (the mask applied by gen_shift_count) so reduce further.
2920 */
2921static void gen_rotc_mod(MemOp ot, TCGv count)
2922{
2923    TCGv temp;
2924
2925    switch (ot) {
2926    case MO_8:
2927        temp = tcg_temp_new();
2928        tcg_gen_subi_tl(temp, count, 18);
2929        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
2930        tcg_gen_subi_tl(temp, count, 9);
2931        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
2932        break;
2933
2934    case MO_16:
2935        temp = tcg_temp_new();
2936        tcg_gen_subi_tl(temp, count, 17);
2937        tcg_gen_movcond_tl(TCG_COND_GE, count, temp, tcg_constant_tl(0), temp, count);
2938        break;
2939
2940    default:
2941        break;
2942    }
2943}
2944
2945/*
2946 * The idea here is that the bit to the right of the new bit 0 is the
2947 * new carry, and the bit to the right of the old bit 0 is the old carry.
2948 * Just like a regular rotation, the result of the rotation is composed
2949 * from a right shifted part and a left shifted part of s->T0.  The new carry
2950 * is extracted from the right-shifted portion, and the old carry is
2951 * inserted at the end of the left-shifted portion.
2952 *
2953 * Because of the separate shifts involving the carry, gen_RCL and gen_RCR
2954 * mostly operate on count-1.  This also comes in handy when computing
2955 * length - count, because (length-1) - (count-1) can be computed with
2956 * a XOR, and that is commutative unlike subtraction.
2957 */
2958static void gen_RCL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
2959{
2960    bool have_1bit_cin, can_be_zero;
2961    TCGv count;
2962    TCGLabel *zero_label = NULL;
2963    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count);
2964    TCGv low, high, low_count;
2965
2966    if (!count) {
2967        return;
2968    }
2969
2970    low = tcg_temp_new();
2971    high = tcg_temp_new();
2972    low_count = tcg_temp_new();
2973
2974    gen_rotc_mod(ot, count);
2975    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
2976    if (can_be_zero) {
2977        zero_label = gen_new_label();
2978        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
2979    }
2980
2981    /* Compute high part, including incoming carry.  */
2982    if (!have_1bit_cin || TCG_TARGET_deposit_tl_valid(1, TARGET_LONG_BITS - 1)) {
2983        /* high = (T0 << 1) | cin */
2984        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
2985        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
2986    } else {
2987        /* Same as above but without deposit; cin in cc_dst.  */
2988        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
2989        tcg_gen_add_tl(high, high, s->T0);
2990    }
2991    tcg_gen_subi_tl(count, count, 1);
2992    tcg_gen_shl_tl(high, high, count);
2993
2994    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
2995    tcg_gen_xori_tl(low_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
2996    tcg_gen_shr_tl(low, s->T0, low_count);
2997    tcg_gen_andi_tl(decode->cc_dst, low, 1);
2998    tcg_gen_shri_tl(low, low, 1);
2999
3000    /* Compute result and outgoing overflow */
3001    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3002    tcg_gen_or_tl(s->T0, low, high);
3003    gen_rot_overflow(decode, s->T0, decode->cc_src2, NULL);
3004
3005    if (zero_label) {
3006        gen_set_label(zero_label);
3007    }
3008}
3009
3010static void gen_RCR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3011{
3012    bool have_1bit_cin, can_be_zero;
3013    TCGv count;
3014    TCGLabel *zero_label = NULL;
3015    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count);
3016    TCGv low, high, high_count;
3017
3018    if (!count) {
3019        return;
3020    }
3021
3022    low = tcg_temp_new();
3023    high = tcg_temp_new();
3024    high_count = tcg_temp_new();
3025
3026    gen_rotc_mod(ot, count);
3027    have_1bit_cin = gen_eflags_adcox(s, decode, true, can_be_zero);
3028    if (can_be_zero) {
3029        zero_label = gen_new_label();
3030        tcg_gen_brcondi_tl(TCG_COND_EQ, count, 0, zero_label);
3031    }
3032
3033    /* Save incoming carry into high, it will be shifted later.  */
3034    if (!have_1bit_cin || TCG_TARGET_deposit_tl_valid(1, TARGET_LONG_BITS - 1)) {
3035        TCGv cin = have_1bit_cin ? decode->cc_dst : decode->cc_src;
3036        tcg_gen_deposit_tl(high, cin, s->T0, 1, TARGET_LONG_BITS - 1);
3037    } else {
3038        /* Same as above but without deposit; cin in cc_dst.  */
3039        tcg_gen_add_tl(high, s->T0, decode->cc_dst);
3040        tcg_gen_add_tl(high, high, s->T0);
3041    }
3042
3043    /* Compute low part and outgoing carry, incoming s->T0 is zero extended */
3044    tcg_gen_subi_tl(count, count, 1);
3045    tcg_gen_shr_tl(low, s->T0, count);
3046    tcg_gen_andi_tl(decode->cc_dst, low, 1);
3047    tcg_gen_shri_tl(low, low, 1);
3048
3049    /* Move high part to the right position */
3050    tcg_gen_xori_tl(high_count, count, (8 << ot) - 1); /* LENGTH - 1 - (count - 1) */
3051    tcg_gen_shl_tl(high, high, high_count);
3052
3053    /* Compute result and outgoing overflow */
3054    tcg_gen_mov_tl(decode->cc_src2, s->T0);
3055    tcg_gen_or_tl(s->T0, low, high);
3056    gen_rot_overflow(decode, s->T0, decode->cc_src2, NULL);
3057
3058    if (zero_label) {
3059        gen_set_label(zero_label);
3060    }
3061}
3062
3063static void gen_RET(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3064{
3065    int16_t adjust = decode->e.op2 == X86_TYPE_I ? decode->immediate : 0;
3066
3067    MemOp ot = gen_pop_T0(s);
3068    gen_stack_update(s, adjust + (1 << ot));
3069    gen_op_jmp_v(s, s->T0);
3070    gen_bnd_jmp(s);
3071    s->base.is_jmp = DISAS_JUMP;
3072}
3073
3074static void gen_RETF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3075{
3076    int16_t adjust = decode->e.op2 == X86_TYPE_I ? decode->immediate : 0;
3077
3078    if (!PE(s) || VM86(s)) {
3079        gen_stack_A0(s);
3080        /* pop offset */
3081        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3082        /* NOTE: keeping EIP updated is not a problem in case of
3083           exception */
3084        gen_op_jmp_v(s, s->T0);
3085        /* pop selector */
3086        gen_add_A0_im(s, 1 << s->dflag);
3087        gen_op_ld_v(s, s->dflag, s->T0, s->A0);
3088        gen_op_movl_seg_real(s, R_CS, s->T0);
3089        /* add stack offset */
3090        gen_stack_update(s, adjust + (2 << s->dflag));
3091    } else {
3092        gen_update_cc_op(s);
3093        gen_update_eip_cur(s);
3094        gen_helper_lret_protected(tcg_env, tcg_constant_i32(s->dflag - 1),
3095                                  tcg_constant_i32(adjust));
3096    }
3097    s->base.is_jmp = DISAS_EOB_ONLY;
3098}
3099
3100/*
3101 * Return non-NULL if a 32-bit rotate works, after possibly replicating the input.
3102 * The input has already been zero-extended upon operand decode.
3103 */
3104static TCGv_i32 gen_rot_replicate(MemOp ot, TCGv in)
3105{
3106    TCGv_i32 temp;
3107    switch (ot) {
3108    case MO_8:
3109        temp = tcg_temp_new_i32();
3110        tcg_gen_trunc_tl_i32(temp, in);
3111        tcg_gen_muli_i32(temp, temp, 0x01010101);
3112        return temp;
3113
3114    case MO_16:
3115        temp = tcg_temp_new_i32();
3116        tcg_gen_trunc_tl_i32(temp, in);
3117        tcg_gen_deposit_i32(temp, temp, temp, 16, 16);
3118        return temp;
3119
3120#ifdef TARGET_X86_64
3121    case MO_32:
3122        temp = tcg_temp_new_i32();
3123        tcg_gen_trunc_tl_i32(temp, in);
3124        return temp;
3125#endif
3126
3127    default:
3128        return NULL;
3129    }
3130}
3131
3132static void gen_rot_carry(X86DecodedInsn *decode, TCGv result, TCGv count, int bit)
3133{
3134    if (count == NULL) {
3135        tcg_gen_extract_tl(decode->cc_dst, result, bit, 1);
3136    } else {
3137        TCGv temp = tcg_temp_new();
3138        tcg_gen_extract_tl(temp, result, bit, 1);
3139        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3140                           decode->cc_dst, temp);
3141    }
3142}
3143
3144static void gen_ROL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3145{
3146    bool can_be_zero;
3147    TCGv count;
3148    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count);
3149    TCGv_i32 temp32, count32;
3150    TCGv old = tcg_temp_new();
3151
3152    if (!count) {
3153        return;
3154    }
3155
3156    gen_eflags_adcox(s, decode, false, can_be_zero);
3157    tcg_gen_mov_tl(old, s->T0);
3158    temp32 = gen_rot_replicate(ot, s->T0);
3159    if (temp32) {
3160        count32 = tcg_temp_new_i32();
3161        tcg_gen_trunc_tl_i32(count32, count);
3162        tcg_gen_rotl_i32(temp32, temp32, count32);
3163        /* Zero extend to facilitate later optimization.  */
3164        tcg_gen_extu_i32_tl(s->T0, temp32);
3165    } else {
3166        tcg_gen_rotl_tl(s->T0, s->T0, count);
3167    }
3168    gen_rot_carry(decode, s->T0, count, 0);
3169    gen_rot_overflow(decode, s->T0, old, count);
3170}
3171
3172static void gen_ROR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3173{
3174    bool can_be_zero;
3175    TCGv count;
3176    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count);
3177    TCGv_i32 temp32, count32;
3178    TCGv old = tcg_temp_new();
3179
3180    if (!count) {
3181        return;
3182    }
3183
3184    gen_eflags_adcox(s, decode, false, can_be_zero);
3185    tcg_gen_mov_tl(old, s->T0);
3186    temp32 = gen_rot_replicate(ot, s->T0);
3187    if (temp32) {
3188        count32 = tcg_temp_new_i32();
3189        tcg_gen_trunc_tl_i32(count32, count);
3190        tcg_gen_rotr_i32(temp32, temp32, count32);
3191        /* Zero extend to facilitate later optimization.  */
3192        tcg_gen_extu_i32_tl(s->T0, temp32);
3193        gen_rot_carry(decode, s->T0, count, 31);
3194    } else {
3195        tcg_gen_rotr_tl(s->T0, s->T0, count);
3196        gen_rot_carry(decode, s->T0, count, TARGET_LONG_BITS - 1);
3197    }
3198    gen_rot_overflow(decode, s->T0, old, count);
3199}
3200
3201static void gen_RORX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3202{
3203    MemOp ot = decode->op[0].ot;
3204    int mask = ot == MO_64 ? 63 : 31;
3205    int b = decode->immediate & mask;
3206
3207    switch (ot) {
3208    case MO_32:
3209#ifdef TARGET_X86_64
3210        tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3211        tcg_gen_rotri_i32(s->tmp2_i32, s->tmp2_i32, b);
3212        tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
3213        break;
3214
3215    case MO_64:
3216#endif
3217        tcg_gen_rotri_tl(s->T0, s->T0, b);
3218        break;
3219
3220    default:
3221        g_assert_not_reached();
3222    }
3223}
3224
3225static void gen_SAHF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3226{
3227    if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM)) {
3228        return gen_illegal_opcode(s);
3229    }
3230    tcg_gen_shri_tl(s->T0, cpu_regs[R_EAX], 8);
3231    gen_compute_eflags(s);
3232    tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, CC_O);
3233    tcg_gen_andi_tl(s->T0, s->T0, CC_S | CC_Z | CC_A | CC_P | CC_C);
3234    tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, s->T0);
3235}
3236
3237static void gen_SALC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3238{
3239    gen_compute_eflags_c(s, s->T0);
3240    tcg_gen_neg_tl(s->T0, s->T0);
3241}
3242
3243static void gen_shift_dynamic_flags(DisasContext *s, X86DecodedInsn *decode, TCGv count, CCOp cc_op)
3244{
3245    TCGv_i32 count32 = tcg_temp_new_i32();
3246    TCGv_i32 old_cc_op;
3247
3248    decode->cc_op = CC_OP_DYNAMIC;
3249    decode->cc_op_dynamic = tcg_temp_new_i32();
3250
3251    assert(decode->cc_dst == s->T0);
3252    if (cc_op_live[s->cc_op] & USES_CC_DST) {
3253        decode->cc_dst = tcg_temp_new();
3254        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_dst, count, tcg_constant_tl(0),
3255                           cpu_cc_dst, s->T0);
3256    }
3257
3258    if (cc_op_live[s->cc_op] & USES_CC_SRC) {
3259        tcg_gen_movcond_tl(TCG_COND_EQ, decode->cc_src, count, tcg_constant_tl(0),
3260                           cpu_cc_src, decode->cc_src);
3261    }
3262
3263    tcg_gen_trunc_tl_i32(count32, count);
3264    if (s->cc_op == CC_OP_DYNAMIC) {
3265        old_cc_op = cpu_cc_op;
3266    } else {
3267        old_cc_op = tcg_constant_i32(s->cc_op);
3268    }
3269    tcg_gen_movcond_i32(TCG_COND_EQ, decode->cc_op_dynamic, count32, tcg_constant_i32(0),
3270                        old_cc_op, tcg_constant_i32(cc_op));
3271}
3272
3273static void gen_SAR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3274{
3275    bool can_be_zero;
3276    TCGv count;
3277    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count);
3278
3279    if (!count) {
3280        return;
3281    }
3282
3283    decode->cc_dst = s->T0;
3284    decode->cc_src = tcg_temp_new();
3285    tcg_gen_subi_tl(decode->cc_src, count, 1);
3286    tcg_gen_sar_tl(decode->cc_src, s->T0, decode->cc_src);
3287    tcg_gen_sar_tl(s->T0, s->T0, count);
3288    if (can_be_zero) {
3289        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3290    } else {
3291        decode->cc_op = CC_OP_SARB + ot;
3292    }
3293}
3294
3295static void gen_SARX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3296{
3297    MemOp ot = decode->op[0].ot;
3298    int mask;
3299
3300    mask = ot == MO_64 ? 63 : 31;
3301    tcg_gen_andi_tl(s->T1, s->T1, mask);
3302    tcg_gen_sar_tl(s->T0, s->T0, s->T1);
3303}
3304
3305static void gen_SBB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3306{
3307    MemOp ot = decode->op[0].ot;
3308    TCGv c_in = tcg_temp_new();
3309
3310    gen_compute_eflags_c(s, c_in);
3311    if (s->prefix & PREFIX_LOCK) {
3312        tcg_gen_add_tl(s->T0, s->T1, c_in);
3313        tcg_gen_neg_tl(s->T0, s->T0);
3314        tcg_gen_atomic_add_fetch_tl(s->T0, s->A0, s->T0,
3315                                    s->mem_index, ot | MO_LE);
3316    } else {
3317        /*
3318         * TODO: SBB reg, reg could use gen_prepare_eflags_c followed by
3319         * negsetcond, and CC_OP_SUBB as the cc_op.
3320         */
3321        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
3322        tcg_gen_sub_tl(s->T0, s->T0, c_in);
3323    }
3324    prepare_update3_cc(decode, s, CC_OP_SBBB + ot, c_in);
3325}
3326
3327static void gen_SCAS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3328{
3329    MemOp ot = decode->op[2].ot;
3330    if (s->prefix & PREFIX_REPNZ) {
3331        gen_repz_scas(s, ot, 1);
3332    } else if (s->prefix & PREFIX_REPZ) {
3333        gen_repz_scas(s, ot, 0);
3334    } else {
3335        gen_scas(s, ot);
3336    }
3337}
3338
3339static void gen_SETcc(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3340{
3341    gen_setcc1(s, decode->b & 0xf, s->T0);
3342}
3343
3344static void gen_SHA1NEXTE(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3345{
3346    gen_helper_sha1nexte(OP_PTR0, OP_PTR1, OP_PTR2);
3347}
3348
3349static void gen_SHA1MSG1(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3350{
3351    gen_helper_sha1msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3352}
3353
3354static void gen_SHA1MSG2(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3355{
3356    gen_helper_sha1msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3357}
3358
3359static void gen_SHA1RNDS4(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3360{
3361    switch(decode->immediate & 3) {
3362    case 0:
3363        gen_helper_sha1rnds4_f0(OP_PTR0, OP_PTR0, OP_PTR1);
3364        break;
3365    case 1:
3366        gen_helper_sha1rnds4_f1(OP_PTR0, OP_PTR0, OP_PTR1);
3367        break;
3368    case 2:
3369        gen_helper_sha1rnds4_f2(OP_PTR0, OP_PTR0, OP_PTR1);
3370        break;
3371    case 3:
3372        gen_helper_sha1rnds4_f3(OP_PTR0, OP_PTR0, OP_PTR1);
3373        break;
3374    }
3375}
3376
3377static void gen_SHA256MSG1(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3378{
3379    gen_helper_sha256msg1(OP_PTR0, OP_PTR1, OP_PTR2);
3380}
3381
3382static void gen_SHA256MSG2(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3383{
3384    gen_helper_sha256msg2(OP_PTR0, OP_PTR1, OP_PTR2);
3385}
3386
3387static void gen_SHA256RNDS2(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3388{
3389    TCGv_i32 wk0 = tcg_temp_new_i32();
3390    TCGv_i32 wk1 = tcg_temp_new_i32();
3391
3392    tcg_gen_ld_i32(wk0, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(0)));
3393    tcg_gen_ld_i32(wk1, tcg_env, ZMM_OFFSET(0) + offsetof(ZMMReg, ZMM_L(1)));
3394
3395    gen_helper_sha256rnds2(OP_PTR0, OP_PTR1, OP_PTR2, wk0, wk1);
3396}
3397
3398static void gen_SHL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3399{
3400    bool can_be_zero;
3401    TCGv count;
3402    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count);
3403
3404    if (!count) {
3405        return;
3406    }
3407
3408    decode->cc_dst = s->T0;
3409    decode->cc_src = tcg_temp_new();
3410    tcg_gen_subi_tl(decode->cc_src, count, 1);
3411    tcg_gen_shl_tl(decode->cc_src, s->T0, decode->cc_src);
3412    tcg_gen_shl_tl(s->T0, s->T0, count);
3413    if (can_be_zero) {
3414        gen_shift_dynamic_flags(s, decode, count, CC_OP_SHLB + ot);
3415    } else {
3416        decode->cc_op = CC_OP_SHLB + ot;
3417    }
3418}
3419
3420static void gen_SHLX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3421{
3422    MemOp ot = decode->op[0].ot;
3423    int mask;
3424
3425    mask = ot == MO_64 ? 63 : 31;
3426    tcg_gen_andi_tl(s->T1, s->T1, mask);
3427    tcg_gen_shl_tl(s->T0, s->T0, s->T1);
3428}
3429
3430static void gen_SHR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3431{
3432    bool can_be_zero;
3433    TCGv count;
3434    MemOp ot = gen_shift_count(s, decode, &can_be_zero, &count);
3435
3436    if (!count) {
3437        return;
3438    }
3439
3440    decode->cc_dst = s->T0;
3441    decode->cc_src = tcg_temp_new();
3442    tcg_gen_subi_tl(decode->cc_src, count, 1);
3443    tcg_gen_shr_tl(decode->cc_src, s->T0, decode->cc_src);
3444    tcg_gen_shr_tl(s->T0, s->T0, count);
3445    if (can_be_zero) {
3446        gen_shift_dynamic_flags(s, decode, count, CC_OP_SARB + ot);
3447    } else {
3448        decode->cc_op = CC_OP_SARB + ot;
3449    }
3450}
3451
3452static void gen_SHRX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3453{
3454    MemOp ot = decode->op[0].ot;
3455    int mask;
3456
3457    mask = ot == MO_64 ? 63 : 31;
3458    tcg_gen_andi_tl(s->T1, s->T1, mask);
3459    tcg_gen_shr_tl(s->T0, s->T0, s->T1);
3460}
3461
3462static void gen_STC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3463{
3464    gen_compute_eflags(s);
3465    tcg_gen_ori_tl(cpu_cc_src, cpu_cc_src, CC_C);
3466}
3467
3468static void gen_STD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3469{
3470    tcg_gen_st_i32(tcg_constant_i32(-1), tcg_env, offsetof(CPUX86State, df));
3471}
3472
3473static void gen_STI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3474{
3475    gen_set_eflags(s, IF_MASK);
3476    /* interruptions are enabled only the first insn after sti */
3477    gen_update_eip_next(s);
3478    gen_eob_inhibit_irq(s);
3479}
3480
3481static void gen_VAESKEYGEN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3482{
3483    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
3484    assert(!s->vex_l);
3485    gen_helper_aeskeygenassist_xmm(tcg_env, OP_PTR0, OP_PTR1, imm);
3486}
3487
3488static void gen_STMXCSR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3489{
3490    gen_helper_update_mxcsr(tcg_env);
3491    tcg_gen_ld32u_tl(s->T0, tcg_env, offsetof(CPUX86State, mxcsr));
3492}
3493
3494static void gen_STOS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3495{
3496    MemOp ot = decode->op[1].ot;
3497    if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
3498        gen_repz_stos(s, ot);
3499    } else {
3500        gen_stos(s, ot);
3501    }
3502}
3503
3504static void gen_SUB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3505{
3506    MemOp ot = decode->op[1].ot;
3507
3508    if (s->prefix & PREFIX_LOCK) {
3509        tcg_gen_neg_tl(s->T0, s->T1);
3510        tcg_gen_atomic_fetch_add_tl(s->cc_srcT, s->A0, s->T0,
3511                                    s->mem_index, ot | MO_LE);
3512        tcg_gen_sub_tl(s->T0, s->cc_srcT, s->T1);
3513    } else {
3514        tcg_gen_mov_tl(s->cc_srcT, s->T0);
3515        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
3516    }
3517    prepare_update2_cc(decode, s, CC_OP_SUBB + ot);
3518}
3519
3520static void gen_UD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3521{
3522    gen_illegal_opcode(s);
3523}
3524
3525static void gen_VAESIMC(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3526{
3527    assert(!s->vex_l);
3528    gen_helper_aesimc_xmm(tcg_env, OP_PTR0, OP_PTR2);
3529}
3530
3531/*
3532 * 00 = v*ps Vps, Hps, Wpd
3533 * 66 = v*pd Vpd, Hpd, Wps
3534 * f3 = v*ss Vss, Hss, Wps
3535 * f2 = v*sd Vsd, Hsd, Wps
3536 */
3537#define SSE_CMP(x) { \
3538    gen_helper_ ## x ## ps ## _xmm, gen_helper_ ## x ## pd ## _xmm, \
3539    gen_helper_ ## x ## ss, gen_helper_ ## x ## sd, \
3540    gen_helper_ ## x ## ps ## _ymm, gen_helper_ ## x ## pd ## _ymm}
3541static const SSEFunc_0_eppp gen_helper_cmp_funcs[32][6] = {
3542    SSE_CMP(cmpeq),
3543    SSE_CMP(cmplt),
3544    SSE_CMP(cmple),
3545    SSE_CMP(cmpunord),
3546    SSE_CMP(cmpneq),
3547    SSE_CMP(cmpnlt),
3548    SSE_CMP(cmpnle),
3549    SSE_CMP(cmpord),
3550
3551    SSE_CMP(cmpequ),
3552    SSE_CMP(cmpnge),
3553    SSE_CMP(cmpngt),
3554    SSE_CMP(cmpfalse),
3555    SSE_CMP(cmpnequ),
3556    SSE_CMP(cmpge),
3557    SSE_CMP(cmpgt),
3558    SSE_CMP(cmptrue),
3559
3560    SSE_CMP(cmpeqs),
3561    SSE_CMP(cmpltq),
3562    SSE_CMP(cmpleq),
3563    SSE_CMP(cmpunords),
3564    SSE_CMP(cmpneqq),
3565    SSE_CMP(cmpnltq),
3566    SSE_CMP(cmpnleq),
3567    SSE_CMP(cmpords),
3568
3569    SSE_CMP(cmpequs),
3570    SSE_CMP(cmpngeq),
3571    SSE_CMP(cmpngtq),
3572    SSE_CMP(cmpfalses),
3573    SSE_CMP(cmpnequs),
3574    SSE_CMP(cmpgeq),
3575    SSE_CMP(cmpgtq),
3576    SSE_CMP(cmptrues),
3577};
3578#undef SSE_CMP
3579
3580static void gen_VCMP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3581{
3582    int index = decode->immediate & (s->prefix & PREFIX_VEX ? 31 : 7);
3583    int b =
3584        s->prefix & PREFIX_REPZ  ? 2 /* ss */ :
3585        s->prefix & PREFIX_REPNZ ? 3 /* sd */ :
3586        !!(s->prefix & PREFIX_DATA) /* pd */ + (s->vex_l << 2);
3587
3588    gen_helper_cmp_funcs[index][b](tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
3589}
3590
3591static void gen_VCOMI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3592{
3593    SSEFunc_0_epp fn;
3594    fn = s->prefix & PREFIX_DATA ? gen_helper_comisd : gen_helper_comiss;
3595    fn(tcg_env, OP_PTR1, OP_PTR2);
3596    set_cc_op(s, CC_OP_EFLAGS);
3597}
3598
3599static void gen_VCVTPD2PS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3600{
3601    if (s->vex_l) {
3602        gen_helper_cvtpd2ps_ymm(tcg_env, OP_PTR0, OP_PTR2);
3603    } else {
3604        gen_helper_cvtpd2ps_xmm(tcg_env, OP_PTR0, OP_PTR2);
3605    }
3606}
3607
3608static void gen_VCVTPS2PD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3609{
3610    if (s->vex_l) {
3611        gen_helper_cvtps2pd_ymm(tcg_env, OP_PTR0, OP_PTR2);
3612    } else {
3613        gen_helper_cvtps2pd_xmm(tcg_env, OP_PTR0, OP_PTR2);
3614    }
3615}
3616
3617static void gen_VCVTPS2PH(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3618{
3619    gen_unary_imm_fp_sse(s, env, decode,
3620                      gen_helper_cvtps2ph_xmm,
3621                      gen_helper_cvtps2ph_ymm);
3622    /*
3623     * VCVTPS2PH is the only instruction that performs an operation on a
3624     * register source and then *stores* into memory.
3625     */
3626    if (decode->op[0].has_ea) {
3627        gen_store_sse(s, decode, decode->op[0].offset);
3628    }
3629}
3630
3631static void gen_VCVTSD2SS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3632{
3633    gen_helper_cvtsd2ss(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
3634}
3635
3636static void gen_VCVTSS2SD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3637{
3638    gen_helper_cvtss2sd(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2);
3639}
3640
3641static void gen_VCVTSI2Sx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3642{
3643    int vec_len = vector_len(s, decode);
3644    TCGv_i32 in;
3645
3646    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
3647
3648#ifdef TARGET_X86_64
3649    MemOp ot = decode->op[2].ot;
3650    if (ot == MO_64) {
3651        if (s->prefix & PREFIX_REPNZ) {
3652            gen_helper_cvtsq2sd(tcg_env, OP_PTR0, s->T1);
3653        } else {
3654            gen_helper_cvtsq2ss(tcg_env, OP_PTR0, s->T1);
3655        }
3656        return;
3657    }
3658    in = s->tmp2_i32;
3659    tcg_gen_trunc_tl_i32(in, s->T1);
3660#else
3661    in = s->T1;
3662#endif
3663
3664    if (s->prefix & PREFIX_REPNZ) {
3665        gen_helper_cvtsi2sd(tcg_env, OP_PTR0, in);
3666    } else {
3667        gen_helper_cvtsi2ss(tcg_env, OP_PTR0, in);
3668    }
3669}
3670
3671static inline void gen_VCVTtSx2SI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
3672                                  SSEFunc_i_ep ss2si, SSEFunc_l_ep ss2sq,
3673                                  SSEFunc_i_ep sd2si, SSEFunc_l_ep sd2sq)
3674{
3675    TCGv_i32 out;
3676
3677#ifdef TARGET_X86_64
3678    MemOp ot = decode->op[0].ot;
3679    if (ot == MO_64) {
3680        if (s->prefix & PREFIX_REPNZ) {
3681            sd2sq(s->T0, tcg_env, OP_PTR2);
3682        } else {
3683            ss2sq(s->T0, tcg_env, OP_PTR2);
3684        }
3685        return;
3686    }
3687
3688    out = s->tmp2_i32;
3689#else
3690    out = s->T0;
3691#endif
3692    if (s->prefix & PREFIX_REPNZ) {
3693        sd2si(out, tcg_env, OP_PTR2);
3694    } else {
3695        ss2si(out, tcg_env, OP_PTR2);
3696    }
3697#ifdef TARGET_X86_64
3698    tcg_gen_extu_i32_tl(s->T0, out);
3699#endif
3700}
3701
3702#ifndef TARGET_X86_64
3703#define gen_helper_cvtss2sq NULL
3704#define gen_helper_cvtsd2sq NULL
3705#define gen_helper_cvttss2sq NULL
3706#define gen_helper_cvttsd2sq NULL
3707#endif
3708
3709static void gen_VCVTSx2SI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3710{
3711    gen_VCVTtSx2SI(s, env, decode,
3712                   gen_helper_cvtss2si, gen_helper_cvtss2sq,
3713                   gen_helper_cvtsd2si, gen_helper_cvtsd2sq);
3714}
3715
3716static void gen_VCVTTSx2SI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3717{
3718    gen_VCVTtSx2SI(s, env, decode,
3719                   gen_helper_cvttss2si, gen_helper_cvttss2sq,
3720                   gen_helper_cvttsd2si, gen_helper_cvttsd2sq);
3721}
3722
3723static void gen_VEXTRACTx128(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3724{
3725    int mask = decode->immediate & 1;
3726    int src_ofs = vector_elem_offset(&decode->op[1], MO_128, mask);
3727    if (decode->op[0].has_ea) {
3728        /* VEX-only instruction, no alignment requirements.  */
3729        gen_sto_env_A0(s, src_ofs, false);
3730    } else {
3731        tcg_gen_gvec_mov(MO_64, decode->op[0].offset, src_ofs, 16, 16);
3732    }
3733}
3734
3735static void gen_VEXTRACTPS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3736{
3737    gen_pextr(s, env, decode, MO_32);
3738}
3739
3740static void gen_vinsertps(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3741{
3742    int val = decode->immediate;
3743    int dest_word = (val >> 4) & 3;
3744    int new_mask = (val & 15) | (1 << dest_word);
3745    int vec_len = 16;
3746
3747    assert(!s->vex_l);
3748
3749    if (new_mask == 15) {
3750        /* All zeroes except possibly for the inserted element */
3751        tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3752    } else if (decode->op[1].offset != decode->op[0].offset) {
3753        gen_store_sse(s, decode, decode->op[1].offset);
3754    }
3755
3756    if (new_mask != (val & 15)) {
3757        tcg_gen_st_i32(s->tmp2_i32, tcg_env,
3758                       vector_elem_offset(&decode->op[0], MO_32, dest_word));
3759    }
3760
3761    if (new_mask != 15) {
3762        TCGv_i32 zero = tcg_constant_i32(0); /* float32_zero */
3763        int i;
3764        for (i = 0; i < 4; i++) {
3765            if ((val >> i) & 1) {
3766                tcg_gen_st_i32(zero, tcg_env,
3767                               vector_elem_offset(&decode->op[0], MO_32, i));
3768            }
3769        }
3770    }
3771}
3772
3773static void gen_VINSERTPS_r(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3774{
3775    int val = decode->immediate;
3776    tcg_gen_ld_i32(s->tmp2_i32, tcg_env,
3777                   vector_elem_offset(&decode->op[2], MO_32, (val >> 6) & 3));
3778    gen_vinsertps(s, env, decode);
3779}
3780
3781static void gen_VINSERTPS_m(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3782{
3783    tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
3784    gen_vinsertps(s, env, decode);
3785}
3786
3787static void gen_VINSERTx128(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3788{
3789    int mask = decode->immediate & 1;
3790    tcg_gen_gvec_mov(MO_64,
3791                     decode->op[0].offset + offsetof(YMMReg, YMM_X(mask)),
3792                     decode->op[2].offset + offsetof(YMMReg, YMM_X(0)), 16, 16);
3793    tcg_gen_gvec_mov(MO_64,
3794                     decode->op[0].offset + offsetof(YMMReg, YMM_X(!mask)),
3795                     decode->op[1].offset + offsetof(YMMReg, YMM_X(!mask)), 16, 16);
3796}
3797
3798static inline void gen_maskmov(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode,
3799                               SSEFunc_0_eppt xmm, SSEFunc_0_eppt ymm)
3800{
3801    if (!s->vex_l) {
3802        xmm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
3803    } else {
3804        ymm(tcg_env, OP_PTR2, OP_PTR1, s->A0);
3805    }
3806}
3807
3808static void gen_VMASKMOVPD_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3809{
3810    gen_maskmov(s, env, decode, gen_helper_vpmaskmovq_st_xmm, gen_helper_vpmaskmovq_st_ymm);
3811}
3812
3813static void gen_VMASKMOVPS_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3814{
3815    gen_maskmov(s, env, decode, gen_helper_vpmaskmovd_st_xmm, gen_helper_vpmaskmovd_st_ymm);
3816}
3817
3818static void gen_VMOVHPx_ld(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3819{
3820    gen_ldq_env_A0(s, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
3821    if (decode->op[0].offset != decode->op[1].offset) {
3822        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
3823        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
3824    }
3825}
3826
3827static void gen_VMOVHPx_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3828{
3829    gen_stq_env_A0(s, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
3830}
3831
3832static void gen_VMOVHPx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3833{
3834    if (decode->op[0].offset != decode->op[2].offset) {
3835        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
3836        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
3837    }
3838    if (decode->op[0].offset != decode->op[1].offset) {
3839        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
3840        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
3841    }
3842}
3843
3844static void gen_VMOVHLPS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3845{
3846    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(1)));
3847    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
3848    if (decode->op[0].offset != decode->op[1].offset) {
3849        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(1)));
3850        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
3851    }
3852}
3853
3854static void gen_VMOVLHPS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3855{
3856    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset);
3857    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(1)));
3858    if (decode->op[0].offset != decode->op[1].offset) {
3859        tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[1].offset + offsetof(XMMReg, XMM_Q(0)));
3860        tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
3861    }
3862}
3863
3864/*
3865 * Note that MOVLPx supports 256-bit operation unlike MOVHLPx, MOVLHPx, MOXHPx.
3866 * Use a gvec move to move everything above the bottom 64 bits.
3867 */
3868
3869static void gen_VMOVLPx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3870{
3871    int vec_len = vector_len(s, decode);
3872
3873    tcg_gen_ld_i64(s->tmp1_i64, tcg_env, decode->op[2].offset + offsetof(XMMReg, XMM_Q(0)));
3874    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
3875    tcg_gen_st_i64(s->tmp1_i64, tcg_env, decode->op[0].offset + offsetof(XMMReg, XMM_Q(0)));
3876}
3877
3878static void gen_VMOVLPx_ld(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3879{
3880    int vec_len = vector_len(s, decode);
3881
3882    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
3883    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
3884    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
3885}
3886
3887static void gen_VMOVLPx_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3888{
3889    tcg_gen_ld_i64(s->tmp1_i64, OP_PTR2, offsetof(ZMMReg, ZMM_Q(0)));
3890    tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
3891}
3892
3893static void gen_VMOVSD_ld(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3894{
3895    TCGv_i64 zero = tcg_constant_i64(0);
3896
3897    tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
3898    tcg_gen_st_i64(zero, OP_PTR0, offsetof(ZMMReg, ZMM_Q(1)));
3899    tcg_gen_st_i64(s->tmp1_i64, OP_PTR0, offsetof(ZMMReg, ZMM_Q(0)));
3900}
3901
3902static void gen_VMOVSS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3903{
3904    int vec_len = vector_len(s, decode);
3905
3906    tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
3907    tcg_gen_gvec_mov(MO_64, decode->op[0].offset, decode->op[1].offset, vec_len, vec_len);
3908    tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
3909}
3910
3911static void gen_VMOVSS_ld(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3912{
3913    int vec_len = vector_len(s, decode);
3914
3915    tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
3916    tcg_gen_gvec_dup_imm(MO_64, decode->op[0].offset, vec_len, vec_len, 0);
3917    tcg_gen_st_i32(s->tmp2_i32, OP_PTR0, offsetof(ZMMReg, ZMM_L(0)));
3918}
3919
3920static void gen_VMOVSS_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3921{
3922    tcg_gen_ld_i32(s->tmp2_i32, OP_PTR2, offsetof(ZMMReg, ZMM_L(0)));
3923    tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
3924}
3925
3926static void gen_VPMASKMOV_st(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3927{
3928    if (s->vex_w) {
3929        gen_VMASKMOVPD_st(s, env, decode);
3930    } else {
3931        gen_VMASKMOVPS_st(s, env, decode);
3932    }
3933}
3934
3935static void gen_VPERMD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3936{
3937    assert(s->vex_l);
3938    gen_helper_vpermd_ymm(OP_PTR0, OP_PTR1, OP_PTR2);
3939}
3940
3941static void gen_VPERM2x128(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3942{
3943    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
3944    assert(s->vex_l);
3945    gen_helper_vpermdq_ymm(OP_PTR0, OP_PTR1, OP_PTR2, imm);
3946}
3947
3948static void gen_VPHMINPOSUW(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3949{
3950    assert(!s->vex_l);
3951    gen_helper_phminposuw_xmm(tcg_env, OP_PTR0, OP_PTR2);
3952}
3953
3954static void gen_VROUNDSD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3955{
3956    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
3957    assert(!s->vex_l);
3958    gen_helper_roundsd_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
3959}
3960
3961static void gen_VROUNDSS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3962{
3963    TCGv_i32 imm = tcg_constant8u_i32(decode->immediate);
3964    assert(!s->vex_l);
3965    gen_helper_roundss_xmm(tcg_env, OP_PTR0, OP_PTR1, OP_PTR2, imm);
3966}
3967
3968static void gen_VSHUF(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3969{
3970    TCGv_i32 imm = tcg_constant_i32(decode->immediate);
3971    SSEFunc_0_pppi ps, pd, fn;
3972    ps = s->vex_l ? gen_helper_shufps_ymm : gen_helper_shufps_xmm;
3973    pd = s->vex_l ? gen_helper_shufpd_ymm : gen_helper_shufpd_xmm;
3974    fn = s->prefix & PREFIX_DATA ? pd : ps;
3975    fn(OP_PTR0, OP_PTR1, OP_PTR2, imm);
3976}
3977
3978static void gen_VUCOMI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3979{
3980    SSEFunc_0_epp fn;
3981    fn = s->prefix & PREFIX_DATA ? gen_helper_ucomisd : gen_helper_ucomiss;
3982    fn(tcg_env, OP_PTR1, OP_PTR2);
3983    set_cc_op(s, CC_OP_EFLAGS);
3984}
3985
3986static void gen_VZEROALL(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3987{
3988    TCGv_ptr ptr = tcg_temp_new_ptr();
3989
3990    tcg_gen_addi_ptr(ptr, tcg_env, offsetof(CPUX86State, xmm_regs));
3991    gen_helper_memset(ptr, ptr, tcg_constant_i32(0),
3992                      tcg_constant_ptr(CPU_NB_REGS * sizeof(ZMMReg)));
3993}
3994
3995static void gen_VZEROUPPER(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
3996{
3997    int i;
3998
3999    for (i = 0; i < CPU_NB_REGS; i++) {
4000        int offset = offsetof(CPUX86State, xmm_regs[i].ZMM_X(1));
4001        tcg_gen_gvec_dup_imm(MO_64, offset, 16, 16, 0);
4002    }
4003}
4004
4005static void gen_WAIT(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
4006{
4007    if ((s->flags & (HF_MP_MASK | HF_TS_MASK)) == (HF_MP_MASK | HF_TS_MASK)) {
4008        gen_NM_exception(s);
4009    } else {
4010        /* needs to be treated as I/O because of ferr_irq */
4011        translator_io_start(&s->base);
4012        gen_helper_fwait(tcg_env);
4013    }
4014}
4015
4016static void gen_XCHG(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
4017{
4018    if (decode->b == 0x90 && !REX_B(s)) {
4019        if (s->prefix & PREFIX_REPZ) {
4020            gen_update_cc_op(s);
4021            gen_update_eip_cur(s);
4022            gen_helper_pause(tcg_env, cur_insn_len_i32(s));
4023            s->base.is_jmp = DISAS_NORETURN;
4024        }
4025        /* No writeback.  */
4026        decode->op[0].unit = X86_OP_SKIP;
4027        return;
4028    }
4029
4030    if (s->prefix & PREFIX_LOCK) {
4031        tcg_gen_atomic_xchg_tl(s->T0, s->A0, s->T1,
4032                               s->mem_index, decode->op[0].ot | MO_LE);
4033        /* now store old value into register operand */
4034        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4035    } else {
4036        /* move destination value into source operand, source preserved in T1 */
4037        gen_op_mov_reg_v(s, decode->op[2].ot, decode->op[2].n, s->T0);
4038        tcg_gen_mov_tl(s->T0, s->T1);
4039    }
4040}
4041
4042static void gen_XLAT(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
4043{
4044    /* AL is already zero-extended into s->T0.  */
4045    tcg_gen_add_tl(s->A0, cpu_regs[R_EBX], s->T0);
4046    gen_add_A0_ds_seg(s);
4047    gen_op_ld_v(s, MO_8, s->T0, s->A0);
4048}
4049
4050static void gen_XOR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
4051{
4052    /* special case XOR reg, reg */
4053    if (decode->op[1].unit == X86_OP_INT &&
4054        decode->op[2].unit == X86_OP_INT &&
4055        decode->op[1].n == decode->op[2].n) {
4056        tcg_gen_movi_tl(s->T0, 0);
4057        decode->cc_op = CC_OP_CLR;
4058    } else {
4059        MemOp ot = decode->op[1].ot;
4060
4061        if (s->prefix & PREFIX_LOCK) {
4062            tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T1,
4063                                        s->mem_index, ot | MO_LE);
4064        } else {
4065            tcg_gen_xor_tl(s->T0, s->T0, s->T1);
4066        }
4067        prepare_update1_cc(decode, s, CC_OP_LOGICB + ot);
4068    }
4069}
4070