xref: /openbmc/qemu/tcg/s390x/tcg-target.c.inc (revision a6771f2f)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2009 Ulrich Hecht <uli@suse.de>
5 * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
6 * Copyright (c) 2010 Richard Henderson <rth@twiddle.net>
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
14 *
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 * THE SOFTWARE.
25 */
26
27/* We only support generating code for 64-bit mode.  */
28#if TCG_TARGET_REG_BITS != 64
29#error "unsupported code generation mode"
30#endif
31
32#include "../tcg-ldst.c.inc"
33#include "../tcg-pool.c.inc"
34#include "elf.h"
35
36#define TCG_CT_CONST_S16        (1 << 8)
37#define TCG_CT_CONST_S32        (1 << 9)
38#define TCG_CT_CONST_S33        (1 << 10)
39#define TCG_CT_CONST_ZERO       (1 << 11)
40#define TCG_CT_CONST_P32        (1 << 12)
41#define TCG_CT_CONST_INV        (1 << 13)
42#define TCG_CT_CONST_INVRISBG   (1 << 14)
43
44#define ALL_GENERAL_REGS     MAKE_64BIT_MASK(0, 16)
45#define ALL_VECTOR_REGS      MAKE_64BIT_MASK(32, 32)
46
47/* Several places within the instruction set 0 means "no register"
48   rather than TCG_REG_R0.  */
49#define TCG_REG_NONE    0
50
51/* A scratch register that may be be used throughout the backend.  */
52#define TCG_TMP0        TCG_REG_R1
53
54#ifndef CONFIG_SOFTMMU
55#define TCG_GUEST_BASE_REG TCG_REG_R13
56#endif
57
58/* All of the following instructions are prefixed with their instruction
59   format, and are defined as 8- or 16-bit quantities, even when the two
60   halves of the 16-bit quantity may appear 32 bits apart in the insn.
61   This makes it easy to copy the values from the tables in Appendix B.  */
62typedef enum S390Opcode {
63    RIL_AFI     = 0xc209,
64    RIL_AGFI    = 0xc208,
65    RIL_ALFI    = 0xc20b,
66    RIL_ALGFI   = 0xc20a,
67    RIL_BRASL   = 0xc005,
68    RIL_BRCL    = 0xc004,
69    RIL_CFI     = 0xc20d,
70    RIL_CGFI    = 0xc20c,
71    RIL_CLFI    = 0xc20f,
72    RIL_CLGFI   = 0xc20e,
73    RIL_CLRL    = 0xc60f,
74    RIL_CLGRL   = 0xc60a,
75    RIL_CRL     = 0xc60d,
76    RIL_CGRL    = 0xc608,
77    RIL_IIHF    = 0xc008,
78    RIL_IILF    = 0xc009,
79    RIL_LARL    = 0xc000,
80    RIL_LGFI    = 0xc001,
81    RIL_LGRL    = 0xc408,
82    RIL_LLIHF   = 0xc00e,
83    RIL_LLILF   = 0xc00f,
84    RIL_LRL     = 0xc40d,
85    RIL_MSFI    = 0xc201,
86    RIL_MSGFI   = 0xc200,
87    RIL_NIHF    = 0xc00a,
88    RIL_NILF    = 0xc00b,
89    RIL_OIHF    = 0xc00c,
90    RIL_OILF    = 0xc00d,
91    RIL_SLFI    = 0xc205,
92    RIL_SLGFI   = 0xc204,
93    RIL_XIHF    = 0xc006,
94    RIL_XILF    = 0xc007,
95
96    RI_AGHI     = 0xa70b,
97    RI_AHI      = 0xa70a,
98    RI_BRC      = 0xa704,
99    RI_CHI      = 0xa70e,
100    RI_CGHI     = 0xa70f,
101    RI_IIHH     = 0xa500,
102    RI_IIHL     = 0xa501,
103    RI_IILH     = 0xa502,
104    RI_IILL     = 0xa503,
105    RI_LGHI     = 0xa709,
106    RI_LLIHH    = 0xa50c,
107    RI_LLIHL    = 0xa50d,
108    RI_LLILH    = 0xa50e,
109    RI_LLILL    = 0xa50f,
110    RI_MGHI     = 0xa70d,
111    RI_MHI      = 0xa70c,
112    RI_NIHH     = 0xa504,
113    RI_NIHL     = 0xa505,
114    RI_NILH     = 0xa506,
115    RI_NILL     = 0xa507,
116    RI_OIHH     = 0xa508,
117    RI_OIHL     = 0xa509,
118    RI_OILH     = 0xa50a,
119    RI_OILL     = 0xa50b,
120    RI_TMLL     = 0xa701,
121
122    RIEb_CGRJ    = 0xec64,
123    RIEb_CLGRJ   = 0xec65,
124    RIEb_CLRJ    = 0xec77,
125    RIEb_CRJ     = 0xec76,
126
127    RIEc_CGIJ    = 0xec7c,
128    RIEc_CIJ     = 0xec7e,
129    RIEc_CLGIJ   = 0xec7d,
130    RIEc_CLIJ    = 0xec7f,
131
132    RIEf_RISBG   = 0xec55,
133
134    RIEg_LOCGHI  = 0xec46,
135
136    RRE_AGR     = 0xb908,
137    RRE_ALGR    = 0xb90a,
138    RRE_ALCR    = 0xb998,
139    RRE_ALCGR   = 0xb988,
140    RRE_ALGFR   = 0xb91a,
141    RRE_CGR     = 0xb920,
142    RRE_CLGR    = 0xb921,
143    RRE_DLGR    = 0xb987,
144    RRE_DLR     = 0xb997,
145    RRE_DSGFR   = 0xb91d,
146    RRE_DSGR    = 0xb90d,
147    RRE_FLOGR   = 0xb983,
148    RRE_LGBR    = 0xb906,
149    RRE_LCGR    = 0xb903,
150    RRE_LGFR    = 0xb914,
151    RRE_LGHR    = 0xb907,
152    RRE_LGR     = 0xb904,
153    RRE_LLGCR   = 0xb984,
154    RRE_LLGFR   = 0xb916,
155    RRE_LLGHR   = 0xb985,
156    RRE_LRVR    = 0xb91f,
157    RRE_LRVGR   = 0xb90f,
158    RRE_LTGR    = 0xb902,
159    RRE_MLGR    = 0xb986,
160    RRE_MSGR    = 0xb90c,
161    RRE_MSR     = 0xb252,
162    RRE_NGR     = 0xb980,
163    RRE_OGR     = 0xb981,
164    RRE_SGR     = 0xb909,
165    RRE_SLGR    = 0xb90b,
166    RRE_SLBR    = 0xb999,
167    RRE_SLBGR   = 0xb989,
168    RRE_XGR     = 0xb982,
169
170    RRFa_MGRK   = 0xb9ec,
171    RRFa_MSRKC  = 0xb9fd,
172    RRFa_MSGRKC = 0xb9ed,
173    RRFa_NCRK   = 0xb9f5,
174    RRFa_NCGRK  = 0xb9e5,
175    RRFa_NNRK   = 0xb974,
176    RRFa_NNGRK  = 0xb964,
177    RRFa_NORK   = 0xb976,
178    RRFa_NOGRK  = 0xb966,
179    RRFa_NRK    = 0xb9f4,
180    RRFa_NGRK   = 0xb9e4,
181    RRFa_NXRK   = 0xb977,
182    RRFa_NXGRK  = 0xb967,
183    RRFa_OCRK   = 0xb975,
184    RRFa_OCGRK  = 0xb965,
185    RRFa_ORK    = 0xb9f6,
186    RRFa_OGRK   = 0xb9e6,
187    RRFa_SRK    = 0xb9f9,
188    RRFa_SGRK   = 0xb9e9,
189    RRFa_SLRK   = 0xb9fb,
190    RRFa_SLGRK  = 0xb9eb,
191    RRFa_XRK    = 0xb9f7,
192    RRFa_XGRK   = 0xb9e7,
193
194    RRFam_SELGR = 0xb9e3,
195
196    RRFc_LOCR   = 0xb9f2,
197    RRFc_LOCGR  = 0xb9e2,
198    RRFc_POPCNT = 0xb9e1,
199
200    RR_AR       = 0x1a,
201    RR_ALR      = 0x1e,
202    RR_BASR     = 0x0d,
203    RR_BCR      = 0x07,
204    RR_CLR      = 0x15,
205    RR_CR       = 0x19,
206    RR_DR       = 0x1d,
207    RR_LCR      = 0x13,
208    RR_LR       = 0x18,
209    RR_LTR      = 0x12,
210    RR_NR       = 0x14,
211    RR_OR       = 0x16,
212    RR_SR       = 0x1b,
213    RR_SLR      = 0x1f,
214    RR_XR       = 0x17,
215
216    RSY_RLL     = 0xeb1d,
217    RSY_RLLG    = 0xeb1c,
218    RSY_SLLG    = 0xeb0d,
219    RSY_SLLK    = 0xebdf,
220    RSY_SRAG    = 0xeb0a,
221    RSY_SRAK    = 0xebdc,
222    RSY_SRLG    = 0xeb0c,
223    RSY_SRLK    = 0xebde,
224
225    RS_SLL      = 0x89,
226    RS_SRA      = 0x8a,
227    RS_SRL      = 0x88,
228
229    RXY_AG      = 0xe308,
230    RXY_AY      = 0xe35a,
231    RXY_CG      = 0xe320,
232    RXY_CLG     = 0xe321,
233    RXY_CLY     = 0xe355,
234    RXY_CY      = 0xe359,
235    RXY_LAY     = 0xe371,
236    RXY_LB      = 0xe376,
237    RXY_LG      = 0xe304,
238    RXY_LGB     = 0xe377,
239    RXY_LGF     = 0xe314,
240    RXY_LGH     = 0xe315,
241    RXY_LHY     = 0xe378,
242    RXY_LLGC    = 0xe390,
243    RXY_LLGF    = 0xe316,
244    RXY_LLGH    = 0xe391,
245    RXY_LMG     = 0xeb04,
246    RXY_LRV     = 0xe31e,
247    RXY_LRVG    = 0xe30f,
248    RXY_LRVH    = 0xe31f,
249    RXY_LY      = 0xe358,
250    RXY_NG      = 0xe380,
251    RXY_OG      = 0xe381,
252    RXY_STCY    = 0xe372,
253    RXY_STG     = 0xe324,
254    RXY_STHY    = 0xe370,
255    RXY_STMG    = 0xeb24,
256    RXY_STRV    = 0xe33e,
257    RXY_STRVG   = 0xe32f,
258    RXY_STRVH   = 0xe33f,
259    RXY_STY     = 0xe350,
260    RXY_XG      = 0xe382,
261
262    RX_A        = 0x5a,
263    RX_C        = 0x59,
264    RX_L        = 0x58,
265    RX_LA       = 0x41,
266    RX_LH       = 0x48,
267    RX_ST       = 0x50,
268    RX_STC      = 0x42,
269    RX_STH      = 0x40,
270
271    VRIa_VGBM   = 0xe744,
272    VRIa_VREPI  = 0xe745,
273    VRIb_VGM    = 0xe746,
274    VRIc_VREP   = 0xe74d,
275
276    VRRa_VLC    = 0xe7de,
277    VRRa_VLP    = 0xe7df,
278    VRRa_VLR    = 0xe756,
279    VRRc_VA     = 0xe7f3,
280    VRRc_VCEQ   = 0xe7f8,   /* we leave the m5 cs field 0 */
281    VRRc_VCH    = 0xe7fb,   /* " */
282    VRRc_VCHL   = 0xe7f9,   /* " */
283    VRRc_VERLLV = 0xe773,
284    VRRc_VESLV  = 0xe770,
285    VRRc_VESRAV = 0xe77a,
286    VRRc_VESRLV = 0xe778,
287    VRRc_VML    = 0xe7a2,
288    VRRc_VMN    = 0xe7fe,
289    VRRc_VMNL   = 0xe7fc,
290    VRRc_VMX    = 0xe7ff,
291    VRRc_VMXL   = 0xe7fd,
292    VRRc_VN     = 0xe768,
293    VRRc_VNC    = 0xe769,
294    VRRc_VNN    = 0xe76e,
295    VRRc_VNO    = 0xe76b,
296    VRRc_VNX    = 0xe76c,
297    VRRc_VO     = 0xe76a,
298    VRRc_VOC    = 0xe76f,
299    VRRc_VPKS   = 0xe797,   /* we leave the m5 cs field 0 */
300    VRRc_VS     = 0xe7f7,
301    VRRa_VUPH   = 0xe7d7,
302    VRRa_VUPL   = 0xe7d6,
303    VRRc_VX     = 0xe76d,
304    VRRe_VSEL   = 0xe78d,
305    VRRf_VLVGP  = 0xe762,
306
307    VRSa_VERLL  = 0xe733,
308    VRSa_VESL   = 0xe730,
309    VRSa_VESRA  = 0xe73a,
310    VRSa_VESRL  = 0xe738,
311    VRSb_VLVG   = 0xe722,
312    VRSc_VLGV   = 0xe721,
313
314    VRX_VL      = 0xe706,
315    VRX_VLLEZ   = 0xe704,
316    VRX_VLREP   = 0xe705,
317    VRX_VST     = 0xe70e,
318    VRX_VSTEF   = 0xe70b,
319    VRX_VSTEG   = 0xe70a,
320
321    NOP         = 0x0707,
322} S390Opcode;
323
324#ifdef CONFIG_DEBUG_TCG
325static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
326    "%r0",  "%r1",  "%r2",  "%r3",  "%r4",  "%r5",  "%r6",  "%r7",
327    "%r8",  "%r9",  "%r10", "%r11", "%r12", "%r13", "%r14", "%r15",
328    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
329    "%v0",  "%v1",  "%v2",  "%v3",  "%v4",  "%v5",  "%v6",  "%v7",
330    "%v8",  "%v9",  "%v10", "%v11", "%v12", "%v13", "%v14", "%v15",
331    "%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23",
332    "%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31",
333};
334#endif
335
336/* Since R6 is a potential argument register, choose it last of the
337   call-saved registers.  Likewise prefer the call-clobbered registers
338   in reverse order to maximize the chance of avoiding the arguments.  */
339static const int tcg_target_reg_alloc_order[] = {
340    /* Call saved registers.  */
341    TCG_REG_R13,
342    TCG_REG_R12,
343    TCG_REG_R11,
344    TCG_REG_R10,
345    TCG_REG_R9,
346    TCG_REG_R8,
347    TCG_REG_R7,
348    TCG_REG_R6,
349    /* Call clobbered registers.  */
350    TCG_REG_R14,
351    TCG_REG_R0,
352    TCG_REG_R1,
353    /* Argument registers, in reverse order of allocation.  */
354    TCG_REG_R5,
355    TCG_REG_R4,
356    TCG_REG_R3,
357    TCG_REG_R2,
358
359    /* V8-V15 are call saved, and omitted. */
360    TCG_REG_V0,
361    TCG_REG_V1,
362    TCG_REG_V2,
363    TCG_REG_V3,
364    TCG_REG_V4,
365    TCG_REG_V5,
366    TCG_REG_V6,
367    TCG_REG_V7,
368    TCG_REG_V16,
369    TCG_REG_V17,
370    TCG_REG_V18,
371    TCG_REG_V19,
372    TCG_REG_V20,
373    TCG_REG_V21,
374    TCG_REG_V22,
375    TCG_REG_V23,
376    TCG_REG_V24,
377    TCG_REG_V25,
378    TCG_REG_V26,
379    TCG_REG_V27,
380    TCG_REG_V28,
381    TCG_REG_V29,
382    TCG_REG_V30,
383    TCG_REG_V31,
384};
385
386static const int tcg_target_call_iarg_regs[] = {
387    TCG_REG_R2,
388    TCG_REG_R3,
389    TCG_REG_R4,
390    TCG_REG_R5,
391    TCG_REG_R6,
392};
393
394static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
395{
396    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
397    tcg_debug_assert(slot == 0);
398    return TCG_REG_R2;
399}
400
401#define S390_CC_EQ      8
402#define S390_CC_LT      4
403#define S390_CC_GT      2
404#define S390_CC_OV      1
405#define S390_CC_NE      (S390_CC_LT | S390_CC_GT)
406#define S390_CC_LE      (S390_CC_LT | S390_CC_EQ)
407#define S390_CC_GE      (S390_CC_GT | S390_CC_EQ)
408#define S390_CC_NEVER   0
409#define S390_CC_ALWAYS  15
410
411/* Condition codes that result from a COMPARE and COMPARE LOGICAL.  */
412static const uint8_t tcg_cond_to_s390_cond[] = {
413    [TCG_COND_EQ]  = S390_CC_EQ,
414    [TCG_COND_NE]  = S390_CC_NE,
415    [TCG_COND_LT]  = S390_CC_LT,
416    [TCG_COND_LE]  = S390_CC_LE,
417    [TCG_COND_GT]  = S390_CC_GT,
418    [TCG_COND_GE]  = S390_CC_GE,
419    [TCG_COND_LTU] = S390_CC_LT,
420    [TCG_COND_LEU] = S390_CC_LE,
421    [TCG_COND_GTU] = S390_CC_GT,
422    [TCG_COND_GEU] = S390_CC_GE,
423};
424
425/* Condition codes that result from a LOAD AND TEST.  Here, we have no
426   unsigned instruction variation, however since the test is vs zero we
427   can re-map the outcomes appropriately.  */
428static const uint8_t tcg_cond_to_ltr_cond[] = {
429    [TCG_COND_EQ]  = S390_CC_EQ,
430    [TCG_COND_NE]  = S390_CC_NE,
431    [TCG_COND_LT]  = S390_CC_LT,
432    [TCG_COND_LE]  = S390_CC_LE,
433    [TCG_COND_GT]  = S390_CC_GT,
434    [TCG_COND_GE]  = S390_CC_GE,
435    [TCG_COND_LTU] = S390_CC_NEVER,
436    [TCG_COND_LEU] = S390_CC_EQ,
437    [TCG_COND_GTU] = S390_CC_NE,
438    [TCG_COND_GEU] = S390_CC_ALWAYS,
439};
440
441static const tcg_insn_unit *tb_ret_addr;
442uint64_t s390_facilities[3];
443
444static inline bool is_general_reg(TCGReg r)
445{
446    return r <= TCG_REG_R15;
447}
448
449static inline bool is_vector_reg(TCGReg r)
450{
451    return r >= TCG_REG_V0 && r <= TCG_REG_V31;
452}
453
454static bool patch_reloc(tcg_insn_unit *src_rw, int type,
455                        intptr_t value, intptr_t addend)
456{
457    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
458    intptr_t pcrel2;
459    uint32_t old;
460
461    value += addend;
462    pcrel2 = (tcg_insn_unit *)value - src_rx;
463
464    switch (type) {
465    case R_390_PC16DBL:
466        if (pcrel2 == (int16_t)pcrel2) {
467            tcg_patch16(src_rw, pcrel2);
468            return true;
469        }
470        break;
471    case R_390_PC32DBL:
472        if (pcrel2 == (int32_t)pcrel2) {
473            tcg_patch32(src_rw, pcrel2);
474            return true;
475        }
476        break;
477    case R_390_20:
478        if (value == sextract64(value, 0, 20)) {
479            old = *(uint32_t *)src_rw & 0xf00000ff;
480            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
481            tcg_patch32(src_rw, old);
482            return true;
483        }
484        break;
485    default:
486        g_assert_not_reached();
487    }
488    return false;
489}
490
491static int is_const_p16(uint64_t val)
492{
493    for (int i = 0; i < 4; ++i) {
494        uint64_t mask = 0xffffull << (i * 16);
495        if ((val & ~mask) == 0) {
496            return i;
497        }
498    }
499    return -1;
500}
501
502static int is_const_p32(uint64_t val)
503{
504    if ((val & 0xffffffff00000000ull) == 0) {
505        return 0;
506    }
507    if ((val & 0x00000000ffffffffull) == 0) {
508        return 1;
509    }
510    return -1;
511}
512
513/*
514 * Accept bit patterns like these:
515 *  0....01....1
516 *  1....10....0
517 *  1..10..01..1
518 *  0..01..10..0
519 * Copied from gcc sources.
520 */
521static bool risbg_mask(uint64_t c)
522{
523    uint64_t lsb;
524    /* We don't change the number of transitions by inverting,
525       so make sure we start with the LSB zero.  */
526    if (c & 1) {
527        c = ~c;
528    }
529    /* Reject all zeros or all ones.  */
530    if (c == 0) {
531        return false;
532    }
533    /* Find the first transition.  */
534    lsb = c & -c;
535    /* Invert to look for a second transition.  */
536    c = ~c;
537    /* Erase the first transition.  */
538    c &= -lsb;
539    /* Find the second transition, if any.  */
540    lsb = c & -c;
541    /* Match if all the bits are 1's, or if c is zero.  */
542    return c == -lsb;
543}
544
545/* Test if a constant matches the constraint. */
546static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
547{
548    if (ct & TCG_CT_CONST) {
549        return 1;
550    }
551
552    if (type == TCG_TYPE_I32) {
553        val = (int32_t)val;
554    }
555
556    /* The following are mutually exclusive.  */
557    if (ct & TCG_CT_CONST_S16) {
558        return val == (int16_t)val;
559    } else if (ct & TCG_CT_CONST_S32) {
560        return val == (int32_t)val;
561    } else if (ct & TCG_CT_CONST_S33) {
562        return val >= -0xffffffffll && val <= 0xffffffffll;
563    } else if (ct & TCG_CT_CONST_ZERO) {
564        return val == 0;
565    }
566
567    if (ct & TCG_CT_CONST_INV) {
568        val = ~val;
569    }
570    /*
571     * Note that is_const_p16 is a subset of is_const_p32,
572     * so we don't need both constraints.
573     */
574    if ((ct & TCG_CT_CONST_P32) && is_const_p32(val) >= 0) {
575        return true;
576    }
577    if ((ct & TCG_CT_CONST_INVRISBG) && risbg_mask(~val)) {
578        return true;
579    }
580
581    return 0;
582}
583
584/* Emit instructions according to the given instruction format.  */
585
586static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
587{
588    tcg_out16(s, (op << 8) | (r1 << 4) | r2);
589}
590
591static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
592                             TCGReg r1, TCGReg r2)
593{
594    tcg_out32(s, (op << 16) | (r1 << 4) | r2);
595}
596
597/* RRF-a without the m4 field */
598static void tcg_out_insn_RRFa(TCGContext *s, S390Opcode op,
599                              TCGReg r1, TCGReg r2, TCGReg r3)
600{
601    tcg_out32(s, (op << 16) | (r3 << 12) | (r1 << 4) | r2);
602}
603
604/* RRF-a with the m4 field */
605static void tcg_out_insn_RRFam(TCGContext *s, S390Opcode op,
606                               TCGReg r1, TCGReg r2, TCGReg r3, int m4)
607{
608    tcg_out32(s, (op << 16) | (r3 << 12) | (m4 << 8) | (r1 << 4) | r2);
609}
610
611static void tcg_out_insn_RRFc(TCGContext *s, S390Opcode op,
612                              TCGReg r1, TCGReg r2, int m3)
613{
614    tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
615}
616
617static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
618{
619    tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
620}
621
622static void tcg_out_insn_RIEg(TCGContext *s, S390Opcode op, TCGReg r1,
623                             int i2, int m3)
624{
625    tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3);
626    tcg_out32(s, (i2 << 16) | (op & 0xff));
627}
628
629static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
630{
631    tcg_out16(s, op | (r1 << 4));
632    tcg_out32(s, i2);
633}
634
635static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
636                            TCGReg b2, TCGReg r3, int disp)
637{
638    tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
639              | (disp & 0xfff));
640}
641
642static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
643                             TCGReg b2, TCGReg r3, int disp)
644{
645    tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
646    tcg_out32(s, (op & 0xff) | (b2 << 28)
647              | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
648}
649
650#define tcg_out_insn_RX   tcg_out_insn_RS
651#define tcg_out_insn_RXY  tcg_out_insn_RSY
652
653static int RXB(TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
654{
655    /*
656     * Shift bit 4 of each regno to its corresponding bit of RXB.
657     * RXB itself begins at bit 8 of the instruction so 8 - 4 = 4
658     * is the left-shift of the 4th operand.
659     */
660    return ((v1 & 0x10) << (4 + 3))
661         | ((v2 & 0x10) << (4 + 2))
662         | ((v3 & 0x10) << (4 + 1))
663         | ((v4 & 0x10) << (4 + 0));
664}
665
666static void tcg_out_insn_VRIa(TCGContext *s, S390Opcode op,
667                              TCGReg v1, uint16_t i2, int m3)
668{
669    tcg_debug_assert(is_vector_reg(v1));
670    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
671    tcg_out16(s, i2);
672    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
673}
674
675static void tcg_out_insn_VRIb(TCGContext *s, S390Opcode op,
676                              TCGReg v1, uint8_t i2, uint8_t i3, int m4)
677{
678    tcg_debug_assert(is_vector_reg(v1));
679    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
680    tcg_out16(s, (i2 << 8) | (i3 & 0xff));
681    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
682}
683
684static void tcg_out_insn_VRIc(TCGContext *s, S390Opcode op,
685                              TCGReg v1, uint16_t i2, TCGReg v3, int m4)
686{
687    tcg_debug_assert(is_vector_reg(v1));
688    tcg_debug_assert(is_vector_reg(v3));
689    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
690    tcg_out16(s, i2);
691    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, v3, 0) | (m4 << 12));
692}
693
694static void tcg_out_insn_VRRa(TCGContext *s, S390Opcode op,
695                              TCGReg v1, TCGReg v2, int m3)
696{
697    tcg_debug_assert(is_vector_reg(v1));
698    tcg_debug_assert(is_vector_reg(v2));
699    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
700    tcg_out32(s, (op & 0x00ff) | RXB(v1, v2, 0, 0) | (m3 << 12));
701}
702
703static void tcg_out_insn_VRRc(TCGContext *s, S390Opcode op,
704                              TCGReg v1, TCGReg v2, TCGReg v3, int m4)
705{
706    tcg_debug_assert(is_vector_reg(v1));
707    tcg_debug_assert(is_vector_reg(v2));
708    tcg_debug_assert(is_vector_reg(v3));
709    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
710    tcg_out16(s, v3 << 12);
711    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, 0) | (m4 << 12));
712}
713
714static void tcg_out_insn_VRRe(TCGContext *s, S390Opcode op,
715                              TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
716{
717    tcg_debug_assert(is_vector_reg(v1));
718    tcg_debug_assert(is_vector_reg(v2));
719    tcg_debug_assert(is_vector_reg(v3));
720    tcg_debug_assert(is_vector_reg(v4));
721    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
722    tcg_out16(s, v3 << 12);
723    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, v4) | (v4 << 12));
724}
725
726static void tcg_out_insn_VRRf(TCGContext *s, S390Opcode op,
727                              TCGReg v1, TCGReg r2, TCGReg r3)
728{
729    tcg_debug_assert(is_vector_reg(v1));
730    tcg_debug_assert(is_general_reg(r2));
731    tcg_debug_assert(is_general_reg(r3));
732    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r2);
733    tcg_out16(s, r3 << 12);
734    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0));
735}
736
737static void tcg_out_insn_VRSa(TCGContext *s, S390Opcode op, TCGReg v1,
738                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
739{
740    tcg_debug_assert(is_vector_reg(v1));
741    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
742    tcg_debug_assert(is_general_reg(b2));
743    tcg_debug_assert(is_vector_reg(v3));
744    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
745    tcg_out16(s, b2 << 12 | d2);
746    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, v3, 0) | (m4 << 12));
747}
748
749static void tcg_out_insn_VRSb(TCGContext *s, S390Opcode op, TCGReg v1,
750                              intptr_t d2, TCGReg b2, TCGReg r3, int m4)
751{
752    tcg_debug_assert(is_vector_reg(v1));
753    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
754    tcg_debug_assert(is_general_reg(b2));
755    tcg_debug_assert(is_general_reg(r3));
756    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r3);
757    tcg_out16(s, b2 << 12 | d2);
758    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
759}
760
761static void tcg_out_insn_VRSc(TCGContext *s, S390Opcode op, TCGReg r1,
762                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
763{
764    tcg_debug_assert(is_general_reg(r1));
765    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
766    tcg_debug_assert(is_general_reg(b2));
767    tcg_debug_assert(is_vector_reg(v3));
768    tcg_out16(s, (op & 0xff00) | (r1 << 4) | (v3 & 0xf));
769    tcg_out16(s, b2 << 12 | d2);
770    tcg_out16(s, (op & 0x00ff) | RXB(0, 0, v3, 0) | (m4 << 12));
771}
772
773static void tcg_out_insn_VRX(TCGContext *s, S390Opcode op, TCGReg v1,
774                             TCGReg b2, TCGReg x2, intptr_t d2, int m3)
775{
776    tcg_debug_assert(is_vector_reg(v1));
777    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
778    tcg_debug_assert(is_general_reg(x2));
779    tcg_debug_assert(is_general_reg(b2));
780    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | x2);
781    tcg_out16(s, (b2 << 12) | d2);
782    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
783}
784
785/* Emit an opcode with "type-checking" of the format.  */
786#define tcg_out_insn(S, FMT, OP, ...) \
787    glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
788
789
790/* emit 64-bit shifts */
791static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
792                         TCGReg src, TCGReg sh_reg, int sh_imm)
793{
794    tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
795}
796
797/* emit 32-bit shifts */
798static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
799                         TCGReg sh_reg, int sh_imm)
800{
801    tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
802}
803
804static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
805{
806    if (src == dst) {
807        return true;
808    }
809    switch (type) {
810    case TCG_TYPE_I32:
811        if (likely(is_general_reg(dst) && is_general_reg(src))) {
812            tcg_out_insn(s, RR, LR, dst, src);
813            break;
814        }
815        /* fallthru */
816
817    case TCG_TYPE_I64:
818        if (likely(is_general_reg(dst))) {
819            if (likely(is_general_reg(src))) {
820                tcg_out_insn(s, RRE, LGR, dst, src);
821            } else {
822                tcg_out_insn(s, VRSc, VLGV, dst, 0, 0, src, 3);
823            }
824            break;
825        } else if (is_general_reg(src)) {
826            tcg_out_insn(s, VRSb, VLVG, dst, 0, 0, src, 3);
827            break;
828        }
829        /* fallthru */
830
831    case TCG_TYPE_V64:
832    case TCG_TYPE_V128:
833        tcg_out_insn(s, VRRa, VLR, dst, src, 0);
834        break;
835
836    default:
837        g_assert_not_reached();
838    }
839    return true;
840}
841
842static const S390Opcode li_insns[4] = {
843    RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
844};
845static const S390Opcode oi_insns[4] = {
846    RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
847};
848static const S390Opcode lif_insns[2] = {
849    RIL_LLILF, RIL_LLIHF,
850};
851
852/* load a register with an immediate value */
853static void tcg_out_movi(TCGContext *s, TCGType type,
854                         TCGReg ret, tcg_target_long sval)
855{
856    tcg_target_ulong uval = sval;
857    ptrdiff_t pc_off;
858    int i;
859
860    if (type == TCG_TYPE_I32) {
861        uval = (uint32_t)sval;
862        sval = (int32_t)sval;
863    }
864
865    /* Try all 32-bit insns that can load it in one go.  */
866    if (sval >= -0x8000 && sval < 0x8000) {
867        tcg_out_insn(s, RI, LGHI, ret, sval);
868        return;
869    }
870
871    i = is_const_p16(uval);
872    if (i >= 0) {
873        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
874        return;
875    }
876
877    /* Try all 48-bit insns that can load it in one go.  */
878    if (sval == (int32_t)sval) {
879        tcg_out_insn(s, RIL, LGFI, ret, sval);
880        return;
881    }
882
883    i = is_const_p32(uval);
884    if (i >= 0) {
885        tcg_out_insn_RIL(s, lif_insns[i], ret, uval >> (i * 32));
886        return;
887    }
888
889    /* Try for PC-relative address load.  For odd addresses, add one. */
890    pc_off = tcg_pcrel_diff(s, (void *)sval) >> 1;
891    if (pc_off == (int32_t)pc_off) {
892        tcg_out_insn(s, RIL, LARL, ret, pc_off);
893        if (sval & 1) {
894            tcg_out_insn(s, RI, AGHI, ret, 1);
895        }
896        return;
897    }
898
899    /* Otherwise, load it by parts. */
900    i = is_const_p16((uint32_t)uval);
901    if (i >= 0) {
902        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
903    } else {
904        tcg_out_insn(s, RIL, LLILF, ret, uval);
905    }
906    uval >>= 32;
907    i = is_const_p16(uval);
908    if (i >= 0) {
909        tcg_out_insn_RI(s, oi_insns[i + 2], ret, uval >> (i * 16));
910    } else {
911        tcg_out_insn(s, RIL, OIHF, ret, uval);
912    }
913}
914
915/* Emit a load/store type instruction.  Inputs are:
916   DATA:     The register to be loaded or stored.
917   BASE+OFS: The effective address.
918   OPC_RX:   If the operation has an RX format opcode (e.g. STC), otherwise 0.
919   OPC_RXY:  The RXY format opcode for the operation (e.g. STCY).  */
920
921static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
922                        TCGReg data, TCGReg base, TCGReg index,
923                        tcg_target_long ofs)
924{
925    if (ofs < -0x80000 || ofs >= 0x80000) {
926        /* Combine the low 20 bits of the offset with the actual load insn;
927           the high 44 bits must come from an immediate load.  */
928        tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
929        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
930        ofs = low;
931
932        /* If we were already given an index register, add it in.  */
933        if (index != TCG_REG_NONE) {
934            tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
935        }
936        index = TCG_TMP0;
937    }
938
939    if (opc_rx && ofs >= 0 && ofs < 0x1000) {
940        tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
941    } else {
942        tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
943    }
944}
945
946static void tcg_out_vrx_mem(TCGContext *s, S390Opcode opc_vrx,
947                            TCGReg data, TCGReg base, TCGReg index,
948                            tcg_target_long ofs, int m3)
949{
950    if (ofs < 0 || ofs >= 0x1000) {
951        if (ofs >= -0x80000 && ofs < 0x80000) {
952            tcg_out_insn(s, RXY, LAY, TCG_TMP0, base, index, ofs);
953            base = TCG_TMP0;
954            index = TCG_REG_NONE;
955            ofs = 0;
956        } else {
957            tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs);
958            if (index != TCG_REG_NONE) {
959                tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
960            }
961            index = TCG_TMP0;
962            ofs = 0;
963        }
964    }
965    tcg_out_insn_VRX(s, opc_vrx, data, base, index, ofs, m3);
966}
967
968/* load data without address translation or endianness conversion */
969static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
970                       TCGReg base, intptr_t ofs)
971{
972    switch (type) {
973    case TCG_TYPE_I32:
974        if (likely(is_general_reg(data))) {
975            tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
976            break;
977        }
978        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_32);
979        break;
980
981    case TCG_TYPE_I64:
982        if (likely(is_general_reg(data))) {
983            tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
984            break;
985        }
986        /* fallthru */
987
988    case TCG_TYPE_V64:
989        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_64);
990        break;
991
992    case TCG_TYPE_V128:
993        /* Hint quadword aligned.  */
994        tcg_out_vrx_mem(s, VRX_VL, data, base, TCG_REG_NONE, ofs, 4);
995        break;
996
997    default:
998        g_assert_not_reached();
999    }
1000}
1001
1002static void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
1003                       TCGReg base, intptr_t ofs)
1004{
1005    switch (type) {
1006    case TCG_TYPE_I32:
1007        if (likely(is_general_reg(data))) {
1008            tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
1009        } else {
1010            tcg_out_vrx_mem(s, VRX_VSTEF, data, base, TCG_REG_NONE, ofs, 1);
1011        }
1012        break;
1013
1014    case TCG_TYPE_I64:
1015        if (likely(is_general_reg(data))) {
1016            tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
1017            break;
1018        }
1019        /* fallthru */
1020
1021    case TCG_TYPE_V64:
1022        tcg_out_vrx_mem(s, VRX_VSTEG, data, base, TCG_REG_NONE, ofs, 0);
1023        break;
1024
1025    case TCG_TYPE_V128:
1026        /* Hint quadword aligned.  */
1027        tcg_out_vrx_mem(s, VRX_VST, data, base, TCG_REG_NONE, ofs, 4);
1028        break;
1029
1030    default:
1031        g_assert_not_reached();
1032    }
1033}
1034
1035static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1036                               TCGReg base, intptr_t ofs)
1037{
1038    return false;
1039}
1040
1041static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1042{
1043    return false;
1044}
1045
1046static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1047                             tcg_target_long imm)
1048{
1049    /* This function is only used for passing structs by reference. */
1050    tcg_out_mem(s, RX_LA, RXY_LAY, rd, rs, TCG_REG_NONE, imm);
1051}
1052
1053static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
1054                                 int msb, int lsb, int ofs, int z)
1055{
1056    /* Format RIE-f */
1057    tcg_out16(s, (RIEf_RISBG & 0xff00) | (dest << 4) | src);
1058    tcg_out16(s, (msb << 8) | (z << 7) | lsb);
1059    tcg_out16(s, (ofs << 8) | (RIEf_RISBG & 0xff));
1060}
1061
1062static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1063{
1064    tcg_out_insn(s, RRE, LGBR, dest, src);
1065}
1066
1067static void tcg_out_ext8u(TCGContext *s, TCGReg dest, TCGReg src)
1068{
1069    tcg_out_insn(s, RRE, LLGCR, dest, src);
1070}
1071
1072static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1073{
1074    tcg_out_insn(s, RRE, LGHR, dest, src);
1075}
1076
1077static void tcg_out_ext16u(TCGContext *s, TCGReg dest, TCGReg src)
1078{
1079    tcg_out_insn(s, RRE, LLGHR, dest, src);
1080}
1081
1082static void tcg_out_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
1083{
1084    tcg_out_insn(s, RRE, LGFR, dest, src);
1085}
1086
1087static void tcg_out_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
1088{
1089    tcg_out_insn(s, RRE, LLGFR, dest, src);
1090}
1091
1092static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1093{
1094    tcg_out_ext32s(s, dest, src);
1095}
1096
1097static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1098{
1099    tcg_out_ext32u(s, dest, src);
1100}
1101
1102static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg dest, TCGReg src)
1103{
1104    tcg_out_mov(s, TCG_TYPE_I32, dest, src);
1105}
1106
1107static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
1108{
1109    int msb, lsb;
1110    if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
1111        /* Achieve wraparound by swapping msb and lsb.  */
1112        msb = 64 - ctz64(~val);
1113        lsb = clz64(~val) - 1;
1114    } else {
1115        msb = clz64(val);
1116        lsb = 63 - ctz64(val);
1117    }
1118    tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
1119}
1120
1121static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
1122{
1123    static const S390Opcode ni_insns[4] = {
1124        RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
1125    };
1126    static const S390Opcode nif_insns[2] = {
1127        RIL_NILF, RIL_NIHF
1128    };
1129    uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
1130    int i;
1131
1132    /* Look for the zero-extensions.  */
1133    if ((val & valid) == 0xffffffff) {
1134        tcg_out_ext32u(s, dest, dest);
1135        return;
1136    }
1137    if ((val & valid) == 0xff) {
1138        tcg_out_ext8u(s, dest, dest);
1139        return;
1140    }
1141    if ((val & valid) == 0xffff) {
1142        tcg_out_ext16u(s, dest, dest);
1143        return;
1144    }
1145
1146    i = is_const_p16(~val & valid);
1147    if (i >= 0) {
1148        tcg_out_insn_RI(s, ni_insns[i], dest, val >> (i * 16));
1149        return;
1150    }
1151
1152    i = is_const_p32(~val & valid);
1153    tcg_debug_assert(i == 0 || type != TCG_TYPE_I32);
1154    if (i >= 0) {
1155        tcg_out_insn_RIL(s, nif_insns[i], dest, val >> (i * 32));
1156        return;
1157    }
1158
1159    if (risbg_mask(val)) {
1160        tgen_andi_risbg(s, dest, dest, val);
1161        return;
1162    }
1163
1164    g_assert_not_reached();
1165}
1166
1167static void tgen_ori(TCGContext *s, TCGReg dest, uint64_t val)
1168{
1169    static const S390Opcode oif_insns[2] = {
1170        RIL_OILF, RIL_OIHF
1171    };
1172
1173    int i;
1174
1175    i = is_const_p16(val);
1176    if (i >= 0) {
1177        tcg_out_insn_RI(s, oi_insns[i], dest, val >> (i * 16));
1178        return;
1179    }
1180
1181    i = is_const_p32(val);
1182    if (i >= 0) {
1183        tcg_out_insn_RIL(s, oif_insns[i], dest, val >> (i * 32));
1184        return;
1185    }
1186
1187    g_assert_not_reached();
1188}
1189
1190static void tgen_xori(TCGContext *s, TCGReg dest, uint64_t val)
1191{
1192    switch (is_const_p32(val)) {
1193    case 0:
1194        tcg_out_insn(s, RIL, XILF, dest, val);
1195        break;
1196    case 1:
1197        tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1198        break;
1199    default:
1200        g_assert_not_reached();
1201    }
1202}
1203
1204static int tgen_cmp2(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1205                     TCGArg c2, bool c2const, bool need_carry, int *inv_cc)
1206{
1207    bool is_unsigned = is_unsigned_cond(c);
1208    TCGCond inv_c = tcg_invert_cond(c);
1209    S390Opcode op;
1210
1211    if (c2const) {
1212        if (c2 == 0) {
1213            if (!(is_unsigned && need_carry)) {
1214                if (type == TCG_TYPE_I32) {
1215                    tcg_out_insn(s, RR, LTR, r1, r1);
1216                } else {
1217                    tcg_out_insn(s, RRE, LTGR, r1, r1);
1218                }
1219                *inv_cc = tcg_cond_to_ltr_cond[inv_c];
1220                return tcg_cond_to_ltr_cond[c];
1221            }
1222        }
1223
1224        if (!is_unsigned && c2 == (int16_t)c2) {
1225            op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI);
1226            tcg_out_insn_RI(s, op, r1, c2);
1227            goto exit;
1228        }
1229
1230        if (type == TCG_TYPE_I32) {
1231            op = (is_unsigned ? RIL_CLFI : RIL_CFI);
1232            tcg_out_insn_RIL(s, op, r1, c2);
1233            goto exit;
1234        }
1235
1236        /*
1237         * Constraints are for a signed 33-bit operand, which is a
1238         * convenient superset of this signed/unsigned test.
1239         */
1240        if (c2 == (is_unsigned ? (TCGArg)(uint32_t)c2 : (TCGArg)(int32_t)c2)) {
1241            op = (is_unsigned ? RIL_CLGFI : RIL_CGFI);
1242            tcg_out_insn_RIL(s, op, r1, c2);
1243            goto exit;
1244        }
1245
1246        /* Load everything else into a register. */
1247        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, c2);
1248        c2 = TCG_TMP0;
1249    }
1250
1251    if (type == TCG_TYPE_I32) {
1252        op = (is_unsigned ? RR_CLR : RR_CR);
1253        tcg_out_insn_RR(s, op, r1, c2);
1254    } else {
1255        op = (is_unsigned ? RRE_CLGR : RRE_CGR);
1256        tcg_out_insn_RRE(s, op, r1, c2);
1257    }
1258
1259 exit:
1260    *inv_cc = tcg_cond_to_s390_cond[inv_c];
1261    return tcg_cond_to_s390_cond[c];
1262}
1263
1264static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1265                    TCGArg c2, bool c2const, bool need_carry)
1266{
1267    int inv_cc;
1268    return tgen_cmp2(s, type, c, r1, c2, c2const, need_carry, &inv_cc);
1269}
1270
1271static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1272                         TCGReg dest, TCGReg c1, TCGArg c2, int c2const)
1273{
1274    int cc;
1275
1276    /* With LOC2, we can always emit the minimum 3 insns.  */
1277    if (HAVE_FACILITY(LOAD_ON_COND2)) {
1278        /* Emit: d = 0, d = (cc ? 1 : d).  */
1279        cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1280        tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1281        tcg_out_insn(s, RIEg, LOCGHI, dest, 1, cc);
1282        return;
1283    }
1284
1285 restart:
1286    switch (cond) {
1287    case TCG_COND_NE:
1288        /* X != 0 is X > 0.  */
1289        if (c2const && c2 == 0) {
1290            cond = TCG_COND_GTU;
1291        } else {
1292            break;
1293        }
1294        /* fallthru */
1295
1296    case TCG_COND_GTU:
1297    case TCG_COND_GT:
1298        /* The result of a compare has CC=2 for GT and CC=3 unused.
1299           ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit.  */
1300        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1301        tcg_out_movi(s, type, dest, 0);
1302        tcg_out_insn(s, RRE, ALCGR, dest, dest);
1303        return;
1304
1305    case TCG_COND_EQ:
1306        /* X == 0 is X <= 0.  */
1307        if (c2const && c2 == 0) {
1308            cond = TCG_COND_LEU;
1309        } else {
1310            break;
1311        }
1312        /* fallthru */
1313
1314    case TCG_COND_LEU:
1315    case TCG_COND_LE:
1316        /* As above, but we're looking for borrow, or !carry.
1317           The second insn computes d - d - borrow, or -1 for true
1318           and 0 for false.  So we must mask to 1 bit afterward.  */
1319        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1320        tcg_out_insn(s, RRE, SLBGR, dest, dest);
1321        tgen_andi(s, type, dest, 1);
1322        return;
1323
1324    case TCG_COND_GEU:
1325    case TCG_COND_LTU:
1326    case TCG_COND_LT:
1327    case TCG_COND_GE:
1328        /* Swap operands so that we can use LEU/GTU/GT/LE.  */
1329        if (!c2const) {
1330            TCGReg t = c1;
1331            c1 = c2;
1332            c2 = t;
1333            cond = tcg_swap_cond(cond);
1334            goto restart;
1335        }
1336        break;
1337
1338    default:
1339        g_assert_not_reached();
1340    }
1341
1342    cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1343    /* Emit: d = 0, t = 1, d = (cc ? t : d).  */
1344    tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1345    tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 1);
1346    tcg_out_insn(s, RRFc, LOCGR, dest, TCG_TMP0, cc);
1347}
1348
1349static void tgen_movcond_int(TCGContext *s, TCGType type, TCGReg dest,
1350                             TCGArg v3, int v3const, TCGReg v4,
1351                             int cc, int inv_cc)
1352{
1353    TCGReg src;
1354
1355    if (v3const) {
1356        if (dest == v4) {
1357            if (HAVE_FACILITY(LOAD_ON_COND2)) {
1358                /* Emit: if (cc) dest = v3. */
1359                tcg_out_insn(s, RIEg, LOCGHI, dest, v3, cc);
1360                return;
1361            }
1362            tcg_out_insn(s, RI, LGHI, TCG_TMP0, v3);
1363            src = TCG_TMP0;
1364        } else {
1365            /* LGR+LOCGHI is larger than LGHI+LOCGR. */
1366            tcg_out_insn(s, RI, LGHI, dest, v3);
1367            cc = inv_cc;
1368            src = v4;
1369        }
1370    } else {
1371        if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1372            /* Emit: dest = cc ? v3 : v4. */
1373            tcg_out_insn(s, RRFam, SELGR, dest, v3, v4, cc);
1374            return;
1375        }
1376        if (dest == v4) {
1377            src = v3;
1378        } else {
1379            tcg_out_mov(s, type, dest, v3);
1380            cc = inv_cc;
1381            src = v4;
1382        }
1383    }
1384
1385    /* Emit: if (cc) dest = src. */
1386    tcg_out_insn(s, RRFc, LOCGR, dest, src, cc);
1387}
1388
1389static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest,
1390                         TCGReg c1, TCGArg c2, int c2const,
1391                         TCGArg v3, int v3const, TCGReg v4)
1392{
1393    int cc, inv_cc;
1394
1395    cc = tgen_cmp2(s, type, c, c1, c2, c2const, false, &inv_cc);
1396    tgen_movcond_int(s, type, dest, v3, v3const, v4, cc, inv_cc);
1397}
1398
1399static void tgen_clz(TCGContext *s, TCGReg dest, TCGReg a1,
1400                     TCGArg a2, int a2const)
1401{
1402    /* Since this sets both R and R+1, we have no choice but to store the
1403       result into R0, allowing R1 == TCG_TMP0 to be clobbered as well.  */
1404    QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
1405    tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
1406
1407    if (a2const && a2 == 64) {
1408        tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
1409        return;
1410    }
1411
1412    /*
1413     * Conditions from FLOGR are:
1414     *   2 -> one bit found
1415     *   8 -> no one bit found
1416     */
1417    tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2);
1418}
1419
1420static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1421{
1422    /* With MIE3, and bit 0 of m4 set, we get the complete result. */
1423    if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1424        if (type == TCG_TYPE_I32) {
1425            tcg_out_ext32u(s, dest, src);
1426            src = dest;
1427        }
1428        tcg_out_insn(s, RRFc, POPCNT, dest, src, 8);
1429        return;
1430    }
1431
1432    /* Without MIE3, each byte gets the count of bits for the byte. */
1433    tcg_out_insn(s, RRFc, POPCNT, dest, src, 0);
1434
1435    /* Multiply to sum each byte at the top of the word. */
1436    if (type == TCG_TYPE_I32) {
1437        tcg_out_insn(s, RIL, MSFI, dest, 0x01010101);
1438        tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24);
1439    } else {
1440        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull);
1441        tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0);
1442        tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56);
1443    }
1444}
1445
1446static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1447                         int ofs, int len, int z)
1448{
1449    int lsb = (63 - ofs);
1450    int msb = lsb - (len - 1);
1451    tcg_out_risbg(s, dest, src, msb, lsb, ofs, z);
1452}
1453
1454static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src,
1455                         int ofs, int len)
1456{
1457    tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
1458}
1459
1460static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
1461{
1462    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1463    if (off == (int16_t)off) {
1464        tcg_out_insn(s, RI, BRC, cc, off);
1465    } else if (off == (int32_t)off) {
1466        tcg_out_insn(s, RIL, BRCL, cc, off);
1467    } else {
1468        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1469        tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1470    }
1471}
1472
1473static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1474{
1475    if (l->has_value) {
1476        tgen_gotoi(s, cc, l->u.value_ptr);
1477    } else {
1478        tcg_out16(s, RI_BRC | (cc << 4));
1479        tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1480        s->code_ptr += 1;
1481    }
1482}
1483
1484static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1485                                TCGReg r1, TCGReg r2, TCGLabel *l)
1486{
1487    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1488    /* Format RIE-b */
1489    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1490    tcg_out16(s, 0);
1491    tcg_out16(s, cc << 12 | (opc & 0xff));
1492}
1493
1494static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1495                                    TCGReg r1, int i2, TCGLabel *l)
1496{
1497    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1498    /* Format RIE-c */
1499    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1500    tcg_out16(s, 0);
1501    tcg_out16(s, (i2 << 8) | (opc & 0xff));
1502}
1503
1504static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1505                        TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1506{
1507    int cc;
1508    bool is_unsigned = is_unsigned_cond(c);
1509    bool in_range;
1510    S390Opcode opc;
1511
1512    cc = tcg_cond_to_s390_cond[c];
1513
1514    if (!c2const) {
1515        opc = (type == TCG_TYPE_I32
1516               ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ)
1517               : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ));
1518        tgen_compare_branch(s, opc, cc, r1, c2, l);
1519        return;
1520    }
1521
1522    /*
1523     * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1524     * If the immediate we've been given does not fit that range, we'll
1525     * fall back to separate compare and branch instructions using the
1526     * larger comparison range afforded by COMPARE IMMEDIATE.
1527     */
1528    if (type == TCG_TYPE_I32) {
1529        if (is_unsigned) {
1530            opc = RIEc_CLIJ;
1531            in_range = (uint32_t)c2 == (uint8_t)c2;
1532        } else {
1533            opc = RIEc_CIJ;
1534            in_range = (int32_t)c2 == (int8_t)c2;
1535        }
1536    } else {
1537        if (is_unsigned) {
1538            opc = RIEc_CLGIJ;
1539            in_range = (uint64_t)c2 == (uint8_t)c2;
1540        } else {
1541            opc = RIEc_CGIJ;
1542            in_range = (int64_t)c2 == (int8_t)c2;
1543        }
1544    }
1545    if (in_range) {
1546        tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1547        return;
1548    }
1549
1550    cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1551    tgen_branch(s, cc, l);
1552}
1553
1554static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest)
1555{
1556    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1557    if (off == (int32_t)off) {
1558        tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1559    } else {
1560        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1561        tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1562    }
1563}
1564
1565static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
1566                         const TCGHelperInfo *info)
1567{
1568    tcg_out_call_int(s, dest);
1569}
1570
1571typedef struct {
1572    TCGReg base;
1573    TCGReg index;
1574    int disp;
1575    TCGAtomAlign aa;
1576} HostAddress;
1577
1578bool tcg_target_has_memory_bswap(MemOp memop)
1579{
1580    return true;
1581}
1582
1583static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1584                                   HostAddress h)
1585{
1586    switch (opc & (MO_SSIZE | MO_BSWAP)) {
1587    case MO_UB:
1588        tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp);
1589        break;
1590    case MO_SB:
1591        tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp);
1592        break;
1593
1594    case MO_UW | MO_BSWAP:
1595        /* swapped unsigned halfword load with upper bits zeroed */
1596        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1597        tcg_out_ext16u(s, data, data);
1598        break;
1599    case MO_UW:
1600        tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp);
1601        break;
1602
1603    case MO_SW | MO_BSWAP:
1604        /* swapped sign-extended halfword load */
1605        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1606        tcg_out_ext16s(s, TCG_TYPE_REG, data, data);
1607        break;
1608    case MO_SW:
1609        tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp);
1610        break;
1611
1612    case MO_UL | MO_BSWAP:
1613        /* swapped unsigned int load with upper bits zeroed */
1614        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1615        tcg_out_ext32u(s, data, data);
1616        break;
1617    case MO_UL:
1618        tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp);
1619        break;
1620
1621    case MO_SL | MO_BSWAP:
1622        /* swapped sign-extended int load */
1623        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1624        tcg_out_ext32s(s, data, data);
1625        break;
1626    case MO_SL:
1627        tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp);
1628        break;
1629
1630    case MO_UQ | MO_BSWAP:
1631        tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp);
1632        break;
1633    case MO_UQ:
1634        tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp);
1635        break;
1636
1637    default:
1638        g_assert_not_reached();
1639    }
1640}
1641
1642static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1643                                   HostAddress h)
1644{
1645    switch (opc & (MO_SIZE | MO_BSWAP)) {
1646    case MO_UB:
1647        if (h.disp >= 0 && h.disp < 0x1000) {
1648            tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp);
1649        } else {
1650            tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp);
1651        }
1652        break;
1653
1654    case MO_UW | MO_BSWAP:
1655        tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp);
1656        break;
1657    case MO_UW:
1658        if (h.disp >= 0 && h.disp < 0x1000) {
1659            tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp);
1660        } else {
1661            tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp);
1662        }
1663        break;
1664
1665    case MO_UL | MO_BSWAP:
1666        tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp);
1667        break;
1668    case MO_UL:
1669        if (h.disp >= 0 && h.disp < 0x1000) {
1670            tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp);
1671        } else {
1672            tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp);
1673        }
1674        break;
1675
1676    case MO_UQ | MO_BSWAP:
1677        tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp);
1678        break;
1679    case MO_UQ:
1680        tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp);
1681        break;
1682
1683    default:
1684        g_assert_not_reached();
1685    }
1686}
1687
1688static const TCGLdstHelperParam ldst_helper_param = {
1689    .ntmp = 1, .tmp = { TCG_TMP0 }
1690};
1691
1692static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1693{
1694    MemOp opc = get_memop(lb->oi);
1695
1696    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1697                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1698        return false;
1699    }
1700
1701    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1702    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1703    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1704
1705    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1706    return true;
1707}
1708
1709static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1710{
1711    MemOp opc = get_memop(lb->oi);
1712
1713    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1714                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1715        return false;
1716    }
1717
1718    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1719    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1720
1721    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1722    return true;
1723}
1724
1725/*
1726 * For softmmu, perform the TLB load and compare.
1727 * For useronly, perform any required alignment tests.
1728 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1729 * is required and fill in @h with the host address for the fast path.
1730 */
1731static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1732                                           TCGReg addr_reg, MemOpIdx oi,
1733                                           bool is_ld)
1734{
1735    TCGLabelQemuLdst *ldst = NULL;
1736    MemOp opc = get_memop(oi);
1737    unsigned a_mask;
1738
1739    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1740    a_mask = (1 << h->aa.align) - 1;
1741
1742#ifdef CONFIG_SOFTMMU
1743    unsigned s_bits = opc & MO_SIZE;
1744    unsigned s_mask = (1 << s_bits) - 1;
1745    int mem_index = get_mmuidx(oi);
1746    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
1747    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1748    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1749    int ofs, a_off;
1750    uint64_t tlb_mask;
1751
1752    ldst = new_ldst_label(s);
1753    ldst->is_ld = is_ld;
1754    ldst->oi = oi;
1755    ldst->addrlo_reg = addr_reg;
1756
1757    tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE,
1758                 s->page_bits - CPU_TLB_ENTRY_BITS);
1759
1760    QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1761    QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -(1 << 19));
1762    tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off);
1763    tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off);
1764
1765    /*
1766     * For aligned accesses, we check the first byte and include the alignment
1767     * bits within the address.  For unaligned access, we check that we don't
1768     * cross pages using the address of the last byte of the access.
1769     */
1770    a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask);
1771    tlb_mask = (uint64_t)s->page_mask | a_mask;
1772    if (a_off == 0) {
1773        tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask);
1774    } else {
1775        tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off);
1776        tgen_andi(s, TCG_TYPE_TL, TCG_REG_R0, tlb_mask);
1777    }
1778
1779    if (is_ld) {
1780        ofs = offsetof(CPUTLBEntry, addr_read);
1781    } else {
1782        ofs = offsetof(CPUTLBEntry, addr_write);
1783    }
1784    if (TARGET_LONG_BITS == 32) {
1785        tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1786    } else {
1787        tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1788    }
1789
1790    tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1791    ldst->label_ptr[0] = s->code_ptr++;
1792
1793    h->index = TCG_TMP0;
1794    tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE,
1795                 offsetof(CPUTLBEntry, addend));
1796
1797    if (TARGET_LONG_BITS == 32) {
1798        tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg);
1799        h->base = TCG_REG_NONE;
1800    } else {
1801        h->base = addr_reg;
1802    }
1803    h->disp = 0;
1804#else
1805    if (a_mask) {
1806        ldst = new_ldst_label(s);
1807        ldst->is_ld = is_ld;
1808        ldst->oi = oi;
1809        ldst->addrlo_reg = addr_reg;
1810
1811        /* We are expecting a_bits to max out at 7, much lower than TMLL. */
1812        tcg_debug_assert(a_mask <= 0xffff);
1813        tcg_out_insn(s, RI, TMLL, addr_reg, a_mask);
1814
1815        tcg_out16(s, RI_BRC | (7 << 4)); /* CC in {1,2,3} */
1816        ldst->label_ptr[0] = s->code_ptr++;
1817    }
1818
1819    h->base = addr_reg;
1820    if (TARGET_LONG_BITS == 32) {
1821        tcg_out_ext32u(s, TCG_TMP0, addr_reg);
1822        h->base = TCG_TMP0;
1823    }
1824    if (guest_base < 0x80000) {
1825        h->index = TCG_REG_NONE;
1826        h->disp = guest_base;
1827    } else {
1828        h->index = TCG_GUEST_BASE_REG;
1829        h->disp = 0;
1830    }
1831#endif
1832
1833    return ldst;
1834}
1835
1836static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1837                            MemOpIdx oi, TCGType data_type)
1838{
1839    TCGLabelQemuLdst *ldst;
1840    HostAddress h;
1841
1842    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
1843    tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h);
1844
1845    if (ldst) {
1846        ldst->type = data_type;
1847        ldst->datalo_reg = data_reg;
1848        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1849    }
1850}
1851
1852static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1853                            MemOpIdx oi, TCGType data_type)
1854{
1855    TCGLabelQemuLdst *ldst;
1856    HostAddress h;
1857
1858    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
1859    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
1860
1861    if (ldst) {
1862        ldst->type = data_type;
1863        ldst->datalo_reg = data_reg;
1864        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1865    }
1866}
1867
1868static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1869{
1870    /* Reuse the zeroing that exists for goto_ptr.  */
1871    if (a0 == 0) {
1872        tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue);
1873    } else {
1874        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
1875        tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
1876    }
1877}
1878
1879static void tcg_out_goto_tb(TCGContext *s, int which)
1880{
1881    /*
1882     * Branch displacement must be aligned for atomic patching;
1883     * see if we need to add extra nop before branch
1884     */
1885    if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
1886        tcg_out16(s, NOP);
1887    }
1888    tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
1889    set_jmp_insn_offset(s, which);
1890    s->code_ptr += 2;
1891    set_jmp_reset_offset(s, which);
1892}
1893
1894void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1895                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1896{
1897    if (!HAVE_FACILITY(GEN_INST_EXT)) {
1898        return;
1899    }
1900    /* patch the branch destination */
1901    uintptr_t addr = tb->jmp_target_addr[n];
1902    intptr_t disp = addr - (jmp_rx - 2);
1903    qatomic_set((int32_t *)jmp_rw, disp / 2);
1904    /* no need to flush icache explicitly */
1905}
1906
1907# define OP_32_64(x) \
1908        case glue(glue(INDEX_op_,x),_i32): \
1909        case glue(glue(INDEX_op_,x),_i64)
1910
1911static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1912                              const TCGArg args[TCG_MAX_OP_ARGS],
1913                              const int const_args[TCG_MAX_OP_ARGS])
1914{
1915    S390Opcode op, op2;
1916    TCGArg a0, a1, a2;
1917
1918    switch (opc) {
1919    case INDEX_op_goto_ptr:
1920        a0 = args[0];
1921        tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
1922        break;
1923
1924    OP_32_64(ld8u):
1925        /* ??? LLC (RXY format) is only present with the extended-immediate
1926           facility, whereas LLGC is always present.  */
1927        tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
1928        break;
1929
1930    OP_32_64(ld8s):
1931        /* ??? LB is no smaller than LGB, so no point to using it.  */
1932        tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
1933        break;
1934
1935    OP_32_64(ld16u):
1936        /* ??? LLH (RXY format) is only present with the extended-immediate
1937           facility, whereas LLGH is always present.  */
1938        tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
1939        break;
1940
1941    case INDEX_op_ld16s_i32:
1942        tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
1943        break;
1944
1945    case INDEX_op_ld_i32:
1946        tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1947        break;
1948
1949    OP_32_64(st8):
1950        tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
1951                    TCG_REG_NONE, args[2]);
1952        break;
1953
1954    OP_32_64(st16):
1955        tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
1956                    TCG_REG_NONE, args[2]);
1957        break;
1958
1959    case INDEX_op_st_i32:
1960        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
1961        break;
1962
1963    case INDEX_op_add_i32:
1964        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
1965        if (const_args[2]) {
1966        do_addi_32:
1967            if (a0 == a1) {
1968                if (a2 == (int16_t)a2) {
1969                    tcg_out_insn(s, RI, AHI, a0, a2);
1970                    break;
1971                }
1972                tcg_out_insn(s, RIL, AFI, a0, a2);
1973                break;
1974            }
1975            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
1976        } else if (a0 == a1) {
1977            tcg_out_insn(s, RR, AR, a0, a2);
1978        } else {
1979            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
1980        }
1981        break;
1982    case INDEX_op_sub_i32:
1983        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
1984        if (const_args[2]) {
1985            a2 = -a2;
1986            goto do_addi_32;
1987        } else if (a0 == a1) {
1988            tcg_out_insn(s, RR, SR, a0, a2);
1989        } else {
1990            tcg_out_insn(s, RRFa, SRK, a0, a1, a2);
1991        }
1992        break;
1993
1994    case INDEX_op_and_i32:
1995        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
1996        if (const_args[2]) {
1997            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
1998            tgen_andi(s, TCG_TYPE_I32, a0, a2);
1999        } else if (a0 == a1) {
2000            tcg_out_insn(s, RR, NR, a0, a2);
2001        } else {
2002            tcg_out_insn(s, RRFa, NRK, a0, a1, a2);
2003        }
2004        break;
2005    case INDEX_op_or_i32:
2006        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2007        if (const_args[2]) {
2008            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2009            tgen_ori(s, a0, a2);
2010        } else if (a0 == a1) {
2011            tcg_out_insn(s, RR, OR, a0, a2);
2012        } else {
2013            tcg_out_insn(s, RRFa, ORK, a0, a1, a2);
2014        }
2015        break;
2016    case INDEX_op_xor_i32:
2017        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2018        if (const_args[2]) {
2019            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2020            tcg_out_insn(s, RIL, XILF, a0, a2);
2021        } else if (a0 == a1) {
2022            tcg_out_insn(s, RR, XR, args[0], args[2]);
2023        } else {
2024            tcg_out_insn(s, RRFa, XRK, a0, a1, a2);
2025        }
2026        break;
2027
2028    case INDEX_op_andc_i32:
2029        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2030        if (const_args[2]) {
2031            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2032            tgen_andi(s, TCG_TYPE_I32, a0, (uint32_t)~a2);
2033	} else {
2034            tcg_out_insn(s, RRFa, NCRK, a0, a1, a2);
2035	}
2036        break;
2037    case INDEX_op_orc_i32:
2038        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2039        if (const_args[2]) {
2040            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2041            tgen_ori(s, a0, (uint32_t)~a2);
2042        } else {
2043            tcg_out_insn(s, RRFa, OCRK, a0, a1, a2);
2044        }
2045        break;
2046    case INDEX_op_eqv_i32:
2047        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2048        if (const_args[2]) {
2049            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2050            tcg_out_insn(s, RIL, XILF, a0, ~a2);
2051        } else {
2052            tcg_out_insn(s, RRFa, NXRK, a0, a1, a2);
2053        }
2054        break;
2055    case INDEX_op_nand_i32:
2056        tcg_out_insn(s, RRFa, NNRK, args[0], args[1], args[2]);
2057        break;
2058    case INDEX_op_nor_i32:
2059        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[2]);
2060        break;
2061
2062    case INDEX_op_neg_i32:
2063        tcg_out_insn(s, RR, LCR, args[0], args[1]);
2064        break;
2065    case INDEX_op_not_i32:
2066        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[1]);
2067        break;
2068
2069    case INDEX_op_mul_i32:
2070        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2071        if (const_args[2]) {
2072            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2073            if (a2 == (int16_t)a2) {
2074                tcg_out_insn(s, RI, MHI, a0, a2);
2075            } else {
2076                tcg_out_insn(s, RIL, MSFI, a0, a2);
2077            }
2078        } else if (a0 == a1) {
2079            tcg_out_insn(s, RRE, MSR, a0, a2);
2080        } else {
2081            tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2);
2082        }
2083        break;
2084
2085    case INDEX_op_div2_i32:
2086        tcg_debug_assert(args[0] == args[2]);
2087        tcg_debug_assert(args[1] == args[3]);
2088        tcg_debug_assert((args[1] & 1) == 0);
2089        tcg_debug_assert(args[0] == args[1] + 1);
2090        tcg_out_insn(s, RR, DR, args[1], args[4]);
2091        break;
2092    case INDEX_op_divu2_i32:
2093        tcg_debug_assert(args[0] == args[2]);
2094        tcg_debug_assert(args[1] == args[3]);
2095        tcg_debug_assert((args[1] & 1) == 0);
2096        tcg_debug_assert(args[0] == args[1] + 1);
2097        tcg_out_insn(s, RRE, DLR, args[1], args[4]);
2098        break;
2099
2100    case INDEX_op_shl_i32:
2101        op = RS_SLL;
2102        op2 = RSY_SLLK;
2103    do_shift32:
2104        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2105        if (a0 == a1) {
2106            if (const_args[2]) {
2107                tcg_out_sh32(s, op, a0, TCG_REG_NONE, a2);
2108            } else {
2109                tcg_out_sh32(s, op, a0, a2, 0);
2110            }
2111        } else {
2112            /* Using tcg_out_sh64 here for the format; it is a 32-bit shift.  */
2113            if (const_args[2]) {
2114                tcg_out_sh64(s, op2, a0, a1, TCG_REG_NONE, a2);
2115            } else {
2116                tcg_out_sh64(s, op2, a0, a1, a2, 0);
2117            }
2118        }
2119        break;
2120    case INDEX_op_shr_i32:
2121        op = RS_SRL;
2122        op2 = RSY_SRLK;
2123        goto do_shift32;
2124    case INDEX_op_sar_i32:
2125        op = RS_SRA;
2126        op2 = RSY_SRAK;
2127        goto do_shift32;
2128
2129    case INDEX_op_rotl_i32:
2130        /* ??? Using tcg_out_sh64 here for the format; it is a 32-bit rol.  */
2131        if (const_args[2]) {
2132            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_REG_NONE, args[2]);
2133        } else {
2134            tcg_out_sh64(s, RSY_RLL, args[0], args[1], args[2], 0);
2135        }
2136        break;
2137    case INDEX_op_rotr_i32:
2138        if (const_args[2]) {
2139            tcg_out_sh64(s, RSY_RLL, args[0], args[1],
2140                         TCG_REG_NONE, (32 - args[2]) & 31);
2141        } else {
2142            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2143            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_TMP0, 0);
2144        }
2145        break;
2146
2147    case INDEX_op_bswap16_i32:
2148        a0 = args[0], a1 = args[1], a2 = args[2];
2149        tcg_out_insn(s, RRE, LRVR, a0, a1);
2150        if (a2 & TCG_BSWAP_OS) {
2151            tcg_out_sh32(s, RS_SRA, a0, TCG_REG_NONE, 16);
2152        } else {
2153            tcg_out_sh32(s, RS_SRL, a0, TCG_REG_NONE, 16);
2154        }
2155        break;
2156    case INDEX_op_bswap16_i64:
2157        a0 = args[0], a1 = args[1], a2 = args[2];
2158        tcg_out_insn(s, RRE, LRVGR, a0, a1);
2159        if (a2 & TCG_BSWAP_OS) {
2160            tcg_out_sh64(s, RSY_SRAG, a0, a0, TCG_REG_NONE, 48);
2161        } else {
2162            tcg_out_sh64(s, RSY_SRLG, a0, a0, TCG_REG_NONE, 48);
2163        }
2164        break;
2165
2166    case INDEX_op_bswap32_i32:
2167        tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
2168        break;
2169    case INDEX_op_bswap32_i64:
2170        a0 = args[0], a1 = args[1], a2 = args[2];
2171        tcg_out_insn(s, RRE, LRVR, a0, a1);
2172        if (a2 & TCG_BSWAP_OS) {
2173            tcg_out_ext32s(s, a0, a0);
2174        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2175            tcg_out_ext32u(s, a0, a0);
2176        }
2177        break;
2178
2179    case INDEX_op_add2_i32:
2180        if (const_args[4]) {
2181            tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2182        } else {
2183            tcg_out_insn(s, RR, ALR, args[0], args[4]);
2184        }
2185        tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2186        break;
2187    case INDEX_op_sub2_i32:
2188        if (const_args[4]) {
2189            tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2190        } else {
2191            tcg_out_insn(s, RR, SLR, args[0], args[4]);
2192        }
2193        tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2194        break;
2195
2196    case INDEX_op_br:
2197        tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2198        break;
2199
2200    case INDEX_op_brcond_i32:
2201        tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
2202                    args[1], const_args[1], arg_label(args[3]));
2203        break;
2204    case INDEX_op_setcond_i32:
2205        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2206                     args[2], const_args[2]);
2207        break;
2208    case INDEX_op_movcond_i32:
2209        tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1],
2210                     args[2], const_args[2], args[3], const_args[3], args[4]);
2211        break;
2212
2213    case INDEX_op_qemu_ld_a32_i32:
2214    case INDEX_op_qemu_ld_a64_i32:
2215        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32);
2216        break;
2217    case INDEX_op_qemu_ld_a32_i64:
2218    case INDEX_op_qemu_ld_a64_i64:
2219        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64);
2220        break;
2221    case INDEX_op_qemu_st_a32_i32:
2222    case INDEX_op_qemu_st_a64_i32:
2223        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32);
2224        break;
2225    case INDEX_op_qemu_st_a32_i64:
2226    case INDEX_op_qemu_st_a64_i64:
2227        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64);
2228        break;
2229
2230    case INDEX_op_ld16s_i64:
2231        tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2232        break;
2233    case INDEX_op_ld32u_i64:
2234        tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2235        break;
2236    case INDEX_op_ld32s_i64:
2237        tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2238        break;
2239    case INDEX_op_ld_i64:
2240        tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2241        break;
2242
2243    case INDEX_op_st32_i64:
2244        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2245        break;
2246    case INDEX_op_st_i64:
2247        tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2248        break;
2249
2250    case INDEX_op_add_i64:
2251        a0 = args[0], a1 = args[1], a2 = args[2];
2252        if (const_args[2]) {
2253        do_addi_64:
2254            if (a0 == a1) {
2255                if (a2 == (int16_t)a2) {
2256                    tcg_out_insn(s, RI, AGHI, a0, a2);
2257                    break;
2258                }
2259                if (a2 == (int32_t)a2) {
2260                    tcg_out_insn(s, RIL, AGFI, a0, a2);
2261                    break;
2262                }
2263                if (a2 == (uint32_t)a2) {
2264                    tcg_out_insn(s, RIL, ALGFI, a0, a2);
2265                    break;
2266                }
2267                if (-a2 == (uint32_t)-a2) {
2268                    tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2269                    break;
2270                }
2271            }
2272            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2273        } else if (a0 == a1) {
2274            tcg_out_insn(s, RRE, AGR, a0, a2);
2275        } else {
2276            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2277        }
2278        break;
2279    case INDEX_op_sub_i64:
2280        a0 = args[0], a1 = args[1], a2 = args[2];
2281        if (const_args[2]) {
2282            a2 = -a2;
2283            goto do_addi_64;
2284        } else {
2285            tcg_out_insn(s, RRFa, SGRK, a0, a1, a2);
2286        }
2287        break;
2288
2289    case INDEX_op_and_i64:
2290        a0 = args[0], a1 = args[1], a2 = args[2];
2291        if (const_args[2]) {
2292            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2293            tgen_andi(s, TCG_TYPE_I64, args[0], args[2]);
2294        } else {
2295            tcg_out_insn(s, RRFa, NGRK, a0, a1, a2);
2296        }
2297        break;
2298    case INDEX_op_or_i64:
2299        a0 = args[0], a1 = args[1], a2 = args[2];
2300        if (const_args[2]) {
2301            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2302            tgen_ori(s, a0, a2);
2303        } else {
2304            tcg_out_insn(s, RRFa, OGRK, a0, a1, a2);
2305        }
2306        break;
2307    case INDEX_op_xor_i64:
2308        a0 = args[0], a1 = args[1], a2 = args[2];
2309        if (const_args[2]) {
2310            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2311            tgen_xori(s, a0, a2);
2312        } else {
2313            tcg_out_insn(s, RRFa, XGRK, a0, a1, a2);
2314        }
2315        break;
2316
2317    case INDEX_op_andc_i64:
2318        a0 = args[0], a1 = args[1], a2 = args[2];
2319        if (const_args[2]) {
2320            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2321            tgen_andi(s, TCG_TYPE_I64, a0, ~a2);
2322        } else {
2323            tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2);
2324        }
2325        break;
2326    case INDEX_op_orc_i64:
2327        a0 = args[0], a1 = args[1], a2 = args[2];
2328        if (const_args[2]) {
2329            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2330            tgen_ori(s, a0, ~a2);
2331        } else {
2332            tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2);
2333        }
2334        break;
2335    case INDEX_op_eqv_i64:
2336        a0 = args[0], a1 = args[1], a2 = args[2];
2337        if (const_args[2]) {
2338            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2339            tgen_xori(s, a0, ~a2);
2340        } else {
2341            tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2);
2342        }
2343        break;
2344    case INDEX_op_nand_i64:
2345        tcg_out_insn(s, RRFa, NNGRK, args[0], args[1], args[2]);
2346        break;
2347    case INDEX_op_nor_i64:
2348        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[2]);
2349        break;
2350
2351    case INDEX_op_neg_i64:
2352        tcg_out_insn(s, RRE, LCGR, args[0], args[1]);
2353        break;
2354    case INDEX_op_not_i64:
2355        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[1]);
2356        break;
2357    case INDEX_op_bswap64_i64:
2358        tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
2359        break;
2360
2361    case INDEX_op_mul_i64:
2362        a0 = args[0], a1 = args[1], a2 = args[2];
2363        if (const_args[2]) {
2364            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2365            if (a2 == (int16_t)a2) {
2366                tcg_out_insn(s, RI, MGHI, a0, a2);
2367            } else {
2368                tcg_out_insn(s, RIL, MSGFI, a0, a2);
2369            }
2370        } else if (a0 == a1) {
2371            tcg_out_insn(s, RRE, MSGR, a0, a2);
2372        } else {
2373            tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2);
2374        }
2375        break;
2376
2377    case INDEX_op_div2_i64:
2378        /*
2379         * ??? We get an unnecessary sign-extension of the dividend
2380         * into op0 with this definition, but as we do in fact always
2381         * produce both quotient and remainder using INDEX_op_div_i64
2382         * instead requires jumping through even more hoops.
2383         */
2384        tcg_debug_assert(args[0] == args[2]);
2385        tcg_debug_assert(args[1] == args[3]);
2386        tcg_debug_assert((args[1] & 1) == 0);
2387        tcg_debug_assert(args[0] == args[1] + 1);
2388        tcg_out_insn(s, RRE, DSGR, args[1], args[4]);
2389        break;
2390    case INDEX_op_divu2_i64:
2391        tcg_debug_assert(args[0] == args[2]);
2392        tcg_debug_assert(args[1] == args[3]);
2393        tcg_debug_assert((args[1] & 1) == 0);
2394        tcg_debug_assert(args[0] == args[1] + 1);
2395        tcg_out_insn(s, RRE, DLGR, args[1], args[4]);
2396        break;
2397    case INDEX_op_mulu2_i64:
2398        tcg_debug_assert(args[0] == args[2]);
2399        tcg_debug_assert((args[1] & 1) == 0);
2400        tcg_debug_assert(args[0] == args[1] + 1);
2401        tcg_out_insn(s, RRE, MLGR, args[1], args[3]);
2402        break;
2403    case INDEX_op_muls2_i64:
2404        tcg_debug_assert((args[1] & 1) == 0);
2405        tcg_debug_assert(args[0] == args[1] + 1);
2406        tcg_out_insn(s, RRFa, MGRK, args[1], args[2], args[3]);
2407        break;
2408
2409    case INDEX_op_shl_i64:
2410        op = RSY_SLLG;
2411    do_shift64:
2412        if (const_args[2]) {
2413            tcg_out_sh64(s, op, args[0], args[1], TCG_REG_NONE, args[2]);
2414        } else {
2415            tcg_out_sh64(s, op, args[0], args[1], args[2], 0);
2416        }
2417        break;
2418    case INDEX_op_shr_i64:
2419        op = RSY_SRLG;
2420        goto do_shift64;
2421    case INDEX_op_sar_i64:
2422        op = RSY_SRAG;
2423        goto do_shift64;
2424
2425    case INDEX_op_rotl_i64:
2426        if (const_args[2]) {
2427            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2428                         TCG_REG_NONE, args[2]);
2429        } else {
2430            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], args[2], 0);
2431        }
2432        break;
2433    case INDEX_op_rotr_i64:
2434        if (const_args[2]) {
2435            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2436                         TCG_REG_NONE, (64 - args[2]) & 63);
2437        } else {
2438            /* We can use the smaller 32-bit negate because only the
2439               low 6 bits are examined for the rotate.  */
2440            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2441            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], TCG_TMP0, 0);
2442        }
2443        break;
2444
2445    case INDEX_op_add2_i64:
2446        if (const_args[4]) {
2447            if ((int64_t)args[4] >= 0) {
2448                tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2449            } else {
2450                tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2451            }
2452        } else {
2453            tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2454        }
2455        tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2456        break;
2457    case INDEX_op_sub2_i64:
2458        if (const_args[4]) {
2459            if ((int64_t)args[4] >= 0) {
2460                tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2461            } else {
2462                tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2463            }
2464        } else {
2465            tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2466        }
2467        tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2468        break;
2469
2470    case INDEX_op_brcond_i64:
2471        tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2472                    args[1], const_args[1], arg_label(args[3]));
2473        break;
2474    case INDEX_op_setcond_i64:
2475        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2476                     args[2], const_args[2]);
2477        break;
2478    case INDEX_op_movcond_i64:
2479        tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1],
2480                     args[2], const_args[2], args[3], const_args[3], args[4]);
2481        break;
2482
2483    OP_32_64(deposit):
2484        a0 = args[0], a1 = args[1], a2 = args[2];
2485        if (const_args[1]) {
2486            tgen_deposit(s, a0, a2, args[3], args[4], 1);
2487        } else {
2488            /* Since we can't support "0Z" as a constraint, we allow a1 in
2489               any register.  Fix things up as if a matching constraint.  */
2490            if (a0 != a1) {
2491                TCGType type = (opc == INDEX_op_deposit_i64);
2492                if (a0 == a2) {
2493                    tcg_out_mov(s, type, TCG_TMP0, a2);
2494                    a2 = TCG_TMP0;
2495                }
2496                tcg_out_mov(s, type, a0, a1);
2497            }
2498            tgen_deposit(s, a0, a2, args[3], args[4], 0);
2499        }
2500        break;
2501
2502    OP_32_64(extract):
2503        tgen_extract(s, args[0], args[1], args[2], args[3]);
2504        break;
2505
2506    case INDEX_op_clz_i64:
2507        tgen_clz(s, args[0], args[1], args[2], const_args[2]);
2508        break;
2509
2510    case INDEX_op_ctpop_i32:
2511        tgen_ctpop(s, TCG_TYPE_I32, args[0], args[1]);
2512        break;
2513    case INDEX_op_ctpop_i64:
2514        tgen_ctpop(s, TCG_TYPE_I64, args[0], args[1]);
2515        break;
2516
2517    case INDEX_op_mb:
2518        /* The host memory model is quite strong, we simply need to
2519           serialize the instruction stream.  */
2520        if (args[0] & TCG_MO_ST_LD) {
2521            /* fast-bcr-serialization facility (45) is present */
2522            tcg_out_insn(s, RR, BCR, 14, 0);
2523        }
2524        break;
2525
2526    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
2527    case INDEX_op_mov_i64:
2528    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2529    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2530    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2531    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
2532    case INDEX_op_ext8s_i64:
2533    case INDEX_op_ext8u_i32:
2534    case INDEX_op_ext8u_i64:
2535    case INDEX_op_ext16s_i32:
2536    case INDEX_op_ext16s_i64:
2537    case INDEX_op_ext16u_i32:
2538    case INDEX_op_ext16u_i64:
2539    case INDEX_op_ext32s_i64:
2540    case INDEX_op_ext32u_i64:
2541    case INDEX_op_ext_i32_i64:
2542    case INDEX_op_extu_i32_i64:
2543    case INDEX_op_extrl_i64_i32:
2544    default:
2545        g_assert_not_reached();
2546    }
2547}
2548
2549static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2550                            TCGReg dst, TCGReg src)
2551{
2552    if (is_general_reg(src)) {
2553        /* Replicate general register into two MO_64. */
2554        tcg_out_insn(s, VRRf, VLVGP, dst, src, src);
2555        if (vece == MO_64) {
2556            return true;
2557        }
2558        src = dst;
2559    }
2560
2561    /*
2562     * Recall that the "standard" integer, within a vector, is the
2563     * rightmost element of the leftmost doubleword, a-la VLLEZ.
2564     */
2565    tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece);
2566    return true;
2567}
2568
2569static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2570                             TCGReg dst, TCGReg base, intptr_t offset)
2571{
2572    tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece);
2573    return true;
2574}
2575
2576static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2577                             TCGReg dst, int64_t val)
2578{
2579    int i, mask, msb, lsb;
2580
2581    /* Look for int16_t elements.  */
2582    if (vece <= MO_16 ||
2583        (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) {
2584        tcg_out_insn(s, VRIa, VREPI, dst, val, vece);
2585        return;
2586    }
2587
2588    /* Look for bit masks.  */
2589    if (vece == MO_32) {
2590        if (risbg_mask((int32_t)val)) {
2591            /* Handle wraparound by swapping msb and lsb.  */
2592            if ((val & 0x80000001u) == 0x80000001u) {
2593                msb = 32 - ctz32(~val);
2594                lsb = clz32(~val) - 1;
2595            } else {
2596                msb = clz32(val);
2597                lsb = 31 - ctz32(val);
2598            }
2599            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32);
2600            return;
2601        }
2602    } else {
2603        if (risbg_mask(val)) {
2604            /* Handle wraparound by swapping msb and lsb.  */
2605            if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
2606                /* Handle wraparound by swapping msb and lsb.  */
2607                msb = 64 - ctz64(~val);
2608                lsb = clz64(~val) - 1;
2609            } else {
2610                msb = clz64(val);
2611                lsb = 63 - ctz64(val);
2612            }
2613            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64);
2614            return;
2615        }
2616    }
2617
2618    /* Look for all bytes 0x00 or 0xff.  */
2619    for (i = mask = 0; i < 8; i++) {
2620        uint8_t byte = val >> (i * 8);
2621        if (byte == 0xff) {
2622            mask |= 1 << i;
2623        } else if (byte != 0) {
2624            break;
2625        }
2626    }
2627    if (i == 8) {
2628        tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0);
2629        return;
2630    }
2631
2632    /* Otherwise, stuff it in the constant pool.  */
2633    tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0);
2634    new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2);
2635    tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64);
2636}
2637
2638static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2639                           unsigned vecl, unsigned vece,
2640                           const TCGArg args[TCG_MAX_OP_ARGS],
2641                           const int const_args[TCG_MAX_OP_ARGS])
2642{
2643    TCGType type = vecl + TCG_TYPE_V64;
2644    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
2645
2646    switch (opc) {
2647    case INDEX_op_ld_vec:
2648        tcg_out_ld(s, type, a0, a1, a2);
2649        break;
2650    case INDEX_op_st_vec:
2651        tcg_out_st(s, type, a0, a1, a2);
2652        break;
2653    case INDEX_op_dupm_vec:
2654        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2655        break;
2656
2657    case INDEX_op_abs_vec:
2658        tcg_out_insn(s, VRRa, VLP, a0, a1, vece);
2659        break;
2660    case INDEX_op_neg_vec:
2661        tcg_out_insn(s, VRRa, VLC, a0, a1, vece);
2662        break;
2663    case INDEX_op_not_vec:
2664        tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0);
2665        break;
2666
2667    case INDEX_op_add_vec:
2668        tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece);
2669        break;
2670    case INDEX_op_sub_vec:
2671        tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece);
2672        break;
2673    case INDEX_op_and_vec:
2674        tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0);
2675        break;
2676    case INDEX_op_andc_vec:
2677        tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0);
2678        break;
2679    case INDEX_op_mul_vec:
2680        tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece);
2681        break;
2682    case INDEX_op_or_vec:
2683        tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0);
2684        break;
2685    case INDEX_op_orc_vec:
2686        tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0);
2687        break;
2688    case INDEX_op_xor_vec:
2689        tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0);
2690        break;
2691    case INDEX_op_nand_vec:
2692        tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0);
2693        break;
2694    case INDEX_op_nor_vec:
2695        tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0);
2696        break;
2697    case INDEX_op_eqv_vec:
2698        tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0);
2699        break;
2700
2701    case INDEX_op_shli_vec:
2702        tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece);
2703        break;
2704    case INDEX_op_shri_vec:
2705        tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece);
2706        break;
2707    case INDEX_op_sari_vec:
2708        tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece);
2709        break;
2710    case INDEX_op_rotli_vec:
2711        tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece);
2712        break;
2713    case INDEX_op_shls_vec:
2714        tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece);
2715        break;
2716    case INDEX_op_shrs_vec:
2717        tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece);
2718        break;
2719    case INDEX_op_sars_vec:
2720        tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece);
2721        break;
2722    case INDEX_op_rotls_vec:
2723        tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece);
2724        break;
2725    case INDEX_op_shlv_vec:
2726        tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece);
2727        break;
2728    case INDEX_op_shrv_vec:
2729        tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece);
2730        break;
2731    case INDEX_op_sarv_vec:
2732        tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece);
2733        break;
2734    case INDEX_op_rotlv_vec:
2735        tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece);
2736        break;
2737
2738    case INDEX_op_smin_vec:
2739        tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece);
2740        break;
2741    case INDEX_op_smax_vec:
2742        tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece);
2743        break;
2744    case INDEX_op_umin_vec:
2745        tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece);
2746        break;
2747    case INDEX_op_umax_vec:
2748        tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece);
2749        break;
2750
2751    case INDEX_op_bitsel_vec:
2752        tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1);
2753        break;
2754
2755    case INDEX_op_cmp_vec:
2756        switch ((TCGCond)args[3]) {
2757        case TCG_COND_EQ:
2758            tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece);
2759            break;
2760        case TCG_COND_GT:
2761            tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece);
2762            break;
2763        case TCG_COND_GTU:
2764            tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece);
2765            break;
2766        default:
2767            g_assert_not_reached();
2768        }
2769        break;
2770
2771    case INDEX_op_s390_vuph_vec:
2772        tcg_out_insn(s, VRRa, VUPH, a0, a1, vece);
2773        break;
2774    case INDEX_op_s390_vupl_vec:
2775        tcg_out_insn(s, VRRa, VUPL, a0, a1, vece);
2776        break;
2777    case INDEX_op_s390_vpks_vec:
2778        tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece);
2779        break;
2780
2781    case INDEX_op_mov_vec:   /* Always emitted via tcg_out_mov.  */
2782    case INDEX_op_dup_vec:   /* Always emitted via tcg_out_dup_vec.  */
2783    default:
2784        g_assert_not_reached();
2785    }
2786}
2787
2788int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2789{
2790    switch (opc) {
2791    case INDEX_op_abs_vec:
2792    case INDEX_op_add_vec:
2793    case INDEX_op_and_vec:
2794    case INDEX_op_andc_vec:
2795    case INDEX_op_bitsel_vec:
2796    case INDEX_op_eqv_vec:
2797    case INDEX_op_nand_vec:
2798    case INDEX_op_neg_vec:
2799    case INDEX_op_nor_vec:
2800    case INDEX_op_not_vec:
2801    case INDEX_op_or_vec:
2802    case INDEX_op_orc_vec:
2803    case INDEX_op_rotli_vec:
2804    case INDEX_op_rotls_vec:
2805    case INDEX_op_rotlv_vec:
2806    case INDEX_op_sari_vec:
2807    case INDEX_op_sars_vec:
2808    case INDEX_op_sarv_vec:
2809    case INDEX_op_shli_vec:
2810    case INDEX_op_shls_vec:
2811    case INDEX_op_shlv_vec:
2812    case INDEX_op_shri_vec:
2813    case INDEX_op_shrs_vec:
2814    case INDEX_op_shrv_vec:
2815    case INDEX_op_smax_vec:
2816    case INDEX_op_smin_vec:
2817    case INDEX_op_sub_vec:
2818    case INDEX_op_umax_vec:
2819    case INDEX_op_umin_vec:
2820    case INDEX_op_xor_vec:
2821        return 1;
2822    case INDEX_op_cmp_vec:
2823    case INDEX_op_cmpsel_vec:
2824    case INDEX_op_rotrv_vec:
2825        return -1;
2826    case INDEX_op_mul_vec:
2827        return vece < MO_64;
2828    case INDEX_op_ssadd_vec:
2829    case INDEX_op_sssub_vec:
2830        return vece < MO_64 ? -1 : 0;
2831    default:
2832        return 0;
2833    }
2834}
2835
2836static bool expand_vec_cmp_noinv(TCGType type, unsigned vece, TCGv_vec v0,
2837                                 TCGv_vec v1, TCGv_vec v2, TCGCond cond)
2838{
2839    bool need_swap = false, need_inv = false;
2840
2841    switch (cond) {
2842    case TCG_COND_EQ:
2843    case TCG_COND_GT:
2844    case TCG_COND_GTU:
2845        break;
2846    case TCG_COND_NE:
2847    case TCG_COND_LE:
2848    case TCG_COND_LEU:
2849        need_inv = true;
2850        break;
2851    case TCG_COND_LT:
2852    case TCG_COND_LTU:
2853        need_swap = true;
2854        break;
2855    case TCG_COND_GE:
2856    case TCG_COND_GEU:
2857        need_swap = need_inv = true;
2858        break;
2859    default:
2860        g_assert_not_reached();
2861    }
2862
2863    if (need_inv) {
2864        cond = tcg_invert_cond(cond);
2865    }
2866    if (need_swap) {
2867        TCGv_vec t1;
2868        t1 = v1, v1 = v2, v2 = t1;
2869        cond = tcg_swap_cond(cond);
2870    }
2871
2872    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
2873              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
2874
2875    return need_inv;
2876}
2877
2878static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
2879                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
2880{
2881    if (expand_vec_cmp_noinv(type, vece, v0, v1, v2, cond)) {
2882        tcg_gen_not_vec(vece, v0, v0);
2883    }
2884}
2885
2886static void expand_vec_cmpsel(TCGType type, unsigned vece, TCGv_vec v0,
2887                              TCGv_vec c1, TCGv_vec c2,
2888                              TCGv_vec v3, TCGv_vec v4, TCGCond cond)
2889{
2890    TCGv_vec t = tcg_temp_new_vec(type);
2891
2892    if (expand_vec_cmp_noinv(type, vece, t, c1, c2, cond)) {
2893        /* Invert the sense of the compare by swapping arguments.  */
2894        tcg_gen_bitsel_vec(vece, v0, t, v4, v3);
2895    } else {
2896        tcg_gen_bitsel_vec(vece, v0, t, v3, v4);
2897    }
2898    tcg_temp_free_vec(t);
2899}
2900
2901static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0,
2902                           TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc)
2903{
2904    TCGv_vec h1 = tcg_temp_new_vec(type);
2905    TCGv_vec h2 = tcg_temp_new_vec(type);
2906    TCGv_vec l1 = tcg_temp_new_vec(type);
2907    TCGv_vec l2 = tcg_temp_new_vec(type);
2908
2909    tcg_debug_assert (vece < MO_64);
2910
2911    /* Unpack with sign-extension. */
2912    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
2913              tcgv_vec_arg(h1), tcgv_vec_arg(v1));
2914    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
2915              tcgv_vec_arg(h2), tcgv_vec_arg(v2));
2916
2917    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
2918              tcgv_vec_arg(l1), tcgv_vec_arg(v1));
2919    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
2920              tcgv_vec_arg(l2), tcgv_vec_arg(v2));
2921
2922    /* Arithmetic on a wider element size. */
2923    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1),
2924              tcgv_vec_arg(h1), tcgv_vec_arg(h2));
2925    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1),
2926              tcgv_vec_arg(l1), tcgv_vec_arg(l2));
2927
2928    /* Pack with saturation. */
2929    vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1,
2930              tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1));
2931
2932    tcg_temp_free_vec(h1);
2933    tcg_temp_free_vec(h2);
2934    tcg_temp_free_vec(l1);
2935    tcg_temp_free_vec(l2);
2936}
2937
2938void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
2939                       TCGArg a0, ...)
2940{
2941    va_list va;
2942    TCGv_vec v0, v1, v2, v3, v4, t0;
2943
2944    va_start(va, a0);
2945    v0 = temp_tcgv_vec(arg_temp(a0));
2946    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
2947    v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
2948
2949    switch (opc) {
2950    case INDEX_op_cmp_vec:
2951        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
2952        break;
2953
2954    case INDEX_op_cmpsel_vec:
2955        v3 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
2956        v4 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
2957        expand_vec_cmpsel(type, vece, v0, v1, v2, v3, v4, va_arg(va, TCGArg));
2958        break;
2959
2960    case INDEX_op_rotrv_vec:
2961        t0 = tcg_temp_new_vec(type);
2962        tcg_gen_neg_vec(vece, t0, v2);
2963        tcg_gen_rotlv_vec(vece, v0, v1, t0);
2964        tcg_temp_free_vec(t0);
2965        break;
2966
2967    case INDEX_op_ssadd_vec:
2968        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec);
2969        break;
2970    case INDEX_op_sssub_vec:
2971        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec);
2972        break;
2973
2974    default:
2975        g_assert_not_reached();
2976    }
2977    va_end(va);
2978}
2979
2980static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
2981{
2982    switch (op) {
2983    case INDEX_op_goto_ptr:
2984        return C_O0_I1(r);
2985
2986    case INDEX_op_ld8u_i32:
2987    case INDEX_op_ld8u_i64:
2988    case INDEX_op_ld8s_i32:
2989    case INDEX_op_ld8s_i64:
2990    case INDEX_op_ld16u_i32:
2991    case INDEX_op_ld16u_i64:
2992    case INDEX_op_ld16s_i32:
2993    case INDEX_op_ld16s_i64:
2994    case INDEX_op_ld_i32:
2995    case INDEX_op_ld32u_i64:
2996    case INDEX_op_ld32s_i64:
2997    case INDEX_op_ld_i64:
2998        return C_O1_I1(r, r);
2999
3000    case INDEX_op_st8_i32:
3001    case INDEX_op_st8_i64:
3002    case INDEX_op_st16_i32:
3003    case INDEX_op_st16_i64:
3004    case INDEX_op_st_i32:
3005    case INDEX_op_st32_i64:
3006    case INDEX_op_st_i64:
3007        return C_O0_I2(r, r);
3008
3009    case INDEX_op_add_i32:
3010    case INDEX_op_add_i64:
3011    case INDEX_op_shl_i64:
3012    case INDEX_op_shr_i64:
3013    case INDEX_op_sar_i64:
3014    case INDEX_op_rotl_i32:
3015    case INDEX_op_rotl_i64:
3016    case INDEX_op_rotr_i32:
3017    case INDEX_op_rotr_i64:
3018    case INDEX_op_setcond_i32:
3019        return C_O1_I2(r, r, ri);
3020    case INDEX_op_setcond_i64:
3021        return C_O1_I2(r, r, rA);
3022
3023    case INDEX_op_clz_i64:
3024        return C_O1_I2(r, r, rI);
3025
3026    case INDEX_op_sub_i32:
3027    case INDEX_op_sub_i64:
3028    case INDEX_op_and_i32:
3029    case INDEX_op_or_i32:
3030    case INDEX_op_xor_i32:
3031        return C_O1_I2(r, r, ri);
3032    case INDEX_op_and_i64:
3033        return C_O1_I2(r, r, rNKR);
3034    case INDEX_op_or_i64:
3035    case INDEX_op_xor_i64:
3036        return C_O1_I2(r, r, rK);
3037
3038    case INDEX_op_andc_i32:
3039    case INDEX_op_orc_i32:
3040    case INDEX_op_eqv_i32:
3041        return C_O1_I2(r, r, ri);
3042    case INDEX_op_andc_i64:
3043        return C_O1_I2(r, r, rKR);
3044    case INDEX_op_orc_i64:
3045    case INDEX_op_eqv_i64:
3046        return C_O1_I2(r, r, rNK);
3047
3048    case INDEX_op_nand_i32:
3049    case INDEX_op_nand_i64:
3050    case INDEX_op_nor_i32:
3051    case INDEX_op_nor_i64:
3052        return C_O1_I2(r, r, r);
3053
3054    case INDEX_op_mul_i32:
3055        return (HAVE_FACILITY(MISC_INSN_EXT2)
3056                ? C_O1_I2(r, r, ri)
3057                : C_O1_I2(r, 0, ri));
3058    case INDEX_op_mul_i64:
3059        return (HAVE_FACILITY(MISC_INSN_EXT2)
3060                ? C_O1_I2(r, r, rJ)
3061                : C_O1_I2(r, 0, rJ));
3062
3063    case INDEX_op_shl_i32:
3064    case INDEX_op_shr_i32:
3065    case INDEX_op_sar_i32:
3066        return C_O1_I2(r, r, ri);
3067
3068    case INDEX_op_brcond_i32:
3069        return C_O0_I2(r, ri);
3070    case INDEX_op_brcond_i64:
3071        return C_O0_I2(r, rA);
3072
3073    case INDEX_op_bswap16_i32:
3074    case INDEX_op_bswap16_i64:
3075    case INDEX_op_bswap32_i32:
3076    case INDEX_op_bswap32_i64:
3077    case INDEX_op_bswap64_i64:
3078    case INDEX_op_neg_i32:
3079    case INDEX_op_neg_i64:
3080    case INDEX_op_not_i32:
3081    case INDEX_op_not_i64:
3082    case INDEX_op_ext8s_i32:
3083    case INDEX_op_ext8s_i64:
3084    case INDEX_op_ext8u_i32:
3085    case INDEX_op_ext8u_i64:
3086    case INDEX_op_ext16s_i32:
3087    case INDEX_op_ext16s_i64:
3088    case INDEX_op_ext16u_i32:
3089    case INDEX_op_ext16u_i64:
3090    case INDEX_op_ext32s_i64:
3091    case INDEX_op_ext32u_i64:
3092    case INDEX_op_ext_i32_i64:
3093    case INDEX_op_extu_i32_i64:
3094    case INDEX_op_extract_i32:
3095    case INDEX_op_extract_i64:
3096    case INDEX_op_ctpop_i32:
3097    case INDEX_op_ctpop_i64:
3098        return C_O1_I1(r, r);
3099
3100    case INDEX_op_qemu_ld_a32_i32:
3101    case INDEX_op_qemu_ld_a64_i32:
3102    case INDEX_op_qemu_ld_a32_i64:
3103    case INDEX_op_qemu_ld_a64_i64:
3104        return C_O1_I1(r, r);
3105    case INDEX_op_qemu_st_a32_i64:
3106    case INDEX_op_qemu_st_a64_i64:
3107    case INDEX_op_qemu_st_a32_i32:
3108    case INDEX_op_qemu_st_a64_i32:
3109        return C_O0_I2(r, r);
3110
3111    case INDEX_op_deposit_i32:
3112    case INDEX_op_deposit_i64:
3113        return C_O1_I2(r, rZ, r);
3114
3115    case INDEX_op_movcond_i32:
3116        return C_O1_I4(r, r, ri, rI, r);
3117    case INDEX_op_movcond_i64:
3118        return C_O1_I4(r, r, rA, rI, r);
3119
3120    case INDEX_op_div2_i32:
3121    case INDEX_op_div2_i64:
3122    case INDEX_op_divu2_i32:
3123    case INDEX_op_divu2_i64:
3124        return C_O2_I3(o, m, 0, 1, r);
3125
3126    case INDEX_op_mulu2_i64:
3127        return C_O2_I2(o, m, 0, r);
3128    case INDEX_op_muls2_i64:
3129        return C_O2_I2(o, m, r, r);
3130
3131    case INDEX_op_add2_i32:
3132    case INDEX_op_sub2_i32:
3133        return C_O2_I4(r, r, 0, 1, ri, r);
3134
3135    case INDEX_op_add2_i64:
3136    case INDEX_op_sub2_i64:
3137        return C_O2_I4(r, r, 0, 1, rA, r);
3138
3139    case INDEX_op_st_vec:
3140        return C_O0_I2(v, r);
3141    case INDEX_op_ld_vec:
3142    case INDEX_op_dupm_vec:
3143        return C_O1_I1(v, r);
3144    case INDEX_op_dup_vec:
3145        return C_O1_I1(v, vr);
3146    case INDEX_op_abs_vec:
3147    case INDEX_op_neg_vec:
3148    case INDEX_op_not_vec:
3149    case INDEX_op_rotli_vec:
3150    case INDEX_op_sari_vec:
3151    case INDEX_op_shli_vec:
3152    case INDEX_op_shri_vec:
3153    case INDEX_op_s390_vuph_vec:
3154    case INDEX_op_s390_vupl_vec:
3155        return C_O1_I1(v, v);
3156    case INDEX_op_add_vec:
3157    case INDEX_op_sub_vec:
3158    case INDEX_op_and_vec:
3159    case INDEX_op_andc_vec:
3160    case INDEX_op_or_vec:
3161    case INDEX_op_orc_vec:
3162    case INDEX_op_xor_vec:
3163    case INDEX_op_nand_vec:
3164    case INDEX_op_nor_vec:
3165    case INDEX_op_eqv_vec:
3166    case INDEX_op_cmp_vec:
3167    case INDEX_op_mul_vec:
3168    case INDEX_op_rotlv_vec:
3169    case INDEX_op_rotrv_vec:
3170    case INDEX_op_shlv_vec:
3171    case INDEX_op_shrv_vec:
3172    case INDEX_op_sarv_vec:
3173    case INDEX_op_smax_vec:
3174    case INDEX_op_smin_vec:
3175    case INDEX_op_umax_vec:
3176    case INDEX_op_umin_vec:
3177    case INDEX_op_s390_vpks_vec:
3178        return C_O1_I2(v, v, v);
3179    case INDEX_op_rotls_vec:
3180    case INDEX_op_shls_vec:
3181    case INDEX_op_shrs_vec:
3182    case INDEX_op_sars_vec:
3183        return C_O1_I2(v, v, r);
3184    case INDEX_op_bitsel_vec:
3185        return C_O1_I3(v, v, v, v);
3186
3187    default:
3188        g_assert_not_reached();
3189    }
3190}
3191
3192/*
3193 * Mainline glibc added HWCAP_S390_VX before it was kernel abi.
3194 * Some distros have fixed this up locally, others have not.
3195 */
3196#ifndef HWCAP_S390_VXRS
3197#define HWCAP_S390_VXRS 2048
3198#endif
3199
3200static void query_s390_facilities(void)
3201{
3202    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3203    const char *which;
3204
3205    /* Is STORE FACILITY LIST EXTENDED available?  Honestly, I believe this
3206       is present on all 64-bit systems, but let's check for it anyway.  */
3207    if (hwcap & HWCAP_S390_STFLE) {
3208        register int r0 __asm__("0") = ARRAY_SIZE(s390_facilities) - 1;
3209        register void *r1 __asm__("1") = s390_facilities;
3210
3211        /* stfle 0(%r1) */
3212        asm volatile(".word 0xb2b0,0x1000"
3213                     : "=r"(r0) : "r"(r0), "r"(r1) : "memory", "cc");
3214    }
3215
3216    /*
3217     * Use of vector registers requires os support beyond the facility bit.
3218     * If the kernel does not advertise support, disable the facility bits.
3219     * There is nothing else we currently care about in the 3rd word, so
3220     * disable VECTOR with one store.
3221     */
3222    if (!(hwcap & HWCAP_S390_VXRS)) {
3223        s390_facilities[2] = 0;
3224    }
3225
3226    /*
3227     * Minimum supported cpu revision is z196.
3228     * Check for all required facilities.
3229     * ZARCH_ACTIVE is done via preprocessor check for 64-bit.
3230     */
3231    if (!HAVE_FACILITY(LONG_DISP)) {
3232        which = "long-displacement";
3233        goto fail;
3234    }
3235    if (!HAVE_FACILITY(EXT_IMM)) {
3236        which = "extended-immediate";
3237        goto fail;
3238    }
3239    if (!HAVE_FACILITY(GEN_INST_EXT)) {
3240        which = "general-instructions-extension";
3241        goto fail;
3242    }
3243    /*
3244     * Facility 45 is a big bin that contains: distinct-operands,
3245     * fast-BCR-serialization, high-word, population-count,
3246     * interlocked-access-1, and load/store-on-condition-1
3247     */
3248    if (!HAVE_FACILITY(45)) {
3249        which = "45";
3250        goto fail;
3251    }
3252    return;
3253
3254 fail:
3255    error_report("%s: missing required facility %s", __func__, which);
3256    exit(EXIT_FAILURE);
3257}
3258
3259static void tcg_target_init(TCGContext *s)
3260{
3261    query_s390_facilities();
3262
3263    tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
3264    tcg_target_available_regs[TCG_TYPE_I64] = 0xffff;
3265    if (HAVE_FACILITY(VECTOR)) {
3266        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3267        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3268    }
3269
3270    tcg_target_call_clobber_regs = 0;
3271    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3272    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
3273    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3274    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3275    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3276    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3277    /* The r6 register is technically call-saved, but it's also a parameter
3278       register, so it can get killed by setup for the qemu_st helper.  */
3279    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3280    /* The return register can be considered call-clobbered.  */
3281    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
3282
3283    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3284    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3285    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3286    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3287    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3288    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3289    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3290    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3291    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3292    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3293    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3294    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3295    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V20);
3296    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V21);
3297    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V22);
3298    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V23);
3299    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V24);
3300    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V25);
3301    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V26);
3302    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V27);
3303    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V28);
3304    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V29);
3305    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V30);
3306    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V31);
3307
3308    s->reserved_regs = 0;
3309    tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
3310    /* XXX many insns can't be used with R0, so we better avoid it for now */
3311    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
3312    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
3313}
3314
3315#define FRAME_SIZE  ((int)(TCG_TARGET_CALL_STACK_OFFSET          \
3316                           + TCG_STATIC_CALL_ARGS_SIZE           \
3317                           + CPU_TEMP_BUF_NLONGS * sizeof(long)))
3318
3319static void tcg_target_qemu_prologue(TCGContext *s)
3320{
3321    /* stmg %r6,%r15,48(%r15) (save registers) */
3322    tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
3323
3324    /* aghi %r15,-frame_size */
3325    tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
3326
3327    tcg_set_frame(s, TCG_REG_CALL_STACK,
3328                  TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
3329                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3330
3331#ifndef CONFIG_SOFTMMU
3332    if (guest_base >= 0x80000) {
3333        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
3334        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
3335    }
3336#endif
3337
3338    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3339
3340    /* br %r3 (go to TB) */
3341    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
3342
3343    /*
3344     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3345     * and fall through to the rest of the epilogue.
3346     */
3347    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3348    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0);
3349
3350    /* TB epilogue */
3351    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3352
3353    /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
3354    tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
3355                 FRAME_SIZE + 48);
3356
3357    /* br %r14 (return) */
3358    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
3359}
3360
3361static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3362{
3363    memset(p, 0x07, count * sizeof(tcg_insn_unit));
3364}
3365
3366typedef struct {
3367    DebugFrameHeader h;
3368    uint8_t fde_def_cfa[4];
3369    uint8_t fde_reg_ofs[18];
3370} DebugFrame;
3371
3372/* We're expecting a 2 byte uleb128 encoded value.  */
3373QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3374
3375#define ELF_HOST_MACHINE  EM_S390
3376
3377static const DebugFrame debug_frame = {
3378    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3379    .h.cie.id = -1,
3380    .h.cie.version = 1,
3381    .h.cie.code_align = 1,
3382    .h.cie.data_align = 8,                /* sleb128 8 */
3383    .h.cie.return_column = TCG_REG_R14,
3384
3385    /* Total FDE size does not include the "len" member.  */
3386    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3387
3388    .fde_def_cfa = {
3389        12, TCG_REG_CALL_STACK,         /* DW_CFA_def_cfa %r15, ... */
3390        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3391        (FRAME_SIZE >> 7)
3392    },
3393    .fde_reg_ofs = {
3394        0x86, 6,                        /* DW_CFA_offset, %r6, 48 */
3395        0x87, 7,                        /* DW_CFA_offset, %r7, 56 */
3396        0x88, 8,                        /* DW_CFA_offset, %r8, 64 */
3397        0x89, 9,                        /* DW_CFA_offset, %r92, 72 */
3398        0x8a, 10,                       /* DW_CFA_offset, %r10, 80 */
3399        0x8b, 11,                       /* DW_CFA_offset, %r11, 88 */
3400        0x8c, 12,                       /* DW_CFA_offset, %r12, 96 */
3401        0x8d, 13,                       /* DW_CFA_offset, %r13, 104 */
3402        0x8e, 14,                       /* DW_CFA_offset, %r14, 112 */
3403    }
3404};
3405
3406void tcg_register_jit(const void *buf, size_t buf_size)
3407{
3408    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3409}
3410